# rocm

ROCm is AMD's open-source GPU computing platform, providing compilers, runtimes, and libraries for AI, high-performance computing, and domain-specific workloads.

This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles.

## AMD ROCm 10.2 Brings hipFile to PyTorch, LVM, and mdraid Volumes, and Adds 18 Ryzen Processors

DevFeed: [AMD ROCm 10.2 Brings hipFile to PyTorch, LVM, and mdraid Volumes, and Adds 18 Ryzen Processors](<https://devfeed.tech/articles/amd-rocm-10-2-brings-hipfile-to-pytorch-lvm-and-mdraid-volumes-and-adds-18-ryzen-processors-146856.md>)

Original publisher: [Read original article](<https://www.storagereview.com/news/amd-rocm-10-2-hipfile-pytorch-lvm-mdraid-aiter-ryzen>)

Author: Harold Fritts

Published: 2026-11-20T16:18:56Z

Content type: release

Language: en

Sources: [StorageReview.com](<https://devfeed.tech/sources/storagereview-com.md>)

Topics: [rocm](<https://devfeed.tech/topics/rocm.md>), [PyTorch](<https://devfeed.tech/topics/pytorch.md>), [vllm](<https://devfeed.tech/topics/vllm.md>), [LVM](<https://devfeed.tech/topics/lvm.md>), [Warp](<https://devfeed.tech/topics/warp.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [amd](<https://devfeed.tech/tags/amd.md>), [enterprise](<https://devfeed.tech/tags/enterprise.md>), [lvm](<https://devfeed.tech/tags/lvm.md>), [pytorch](<https://devfeed.tech/tags/pytorch.md>), [rocm](<https://devfeed.tech/tags/rocm.md>), [ryzen](<https://devfeed.tech/tags/ryzen.md>), [volumes](<https://devfeed.tech/tags/volumes.md>)

### AI overview

ROCm 10.2.0 extends AMD Infinity Storage and hipFile into PyTorch, fio, and ior, with support for LVM volumes and Linux mdraid sets. PyTorch's native `load_storage` and `save_storage` APIs move large tensors between GPU memory and storage; hipFile is integrated in fio 3.43 and ior's main branch for benchmarking that path. The stability-focused release also updates vLLM and SGLang, adds AITER support for Radeon and Ryzen AI, and supports 18 additional Ryzen processors. Other changes include expanded GPU libraries and an experimental unified `rocsys` profiler command. AMD's cited MLPerf and scaling figures are vendor-reported and were not independently audited; `rocsys` is experimental and Radeon AI Pro R9700 Hyperloom support is early access.

### Source excerpt

AMD has released ROCm 10.2.0, a stability-focused update to its open GPU software stack that refreshes the serving frameworks, libraries, and tools developers already run on Instinct, Radeon, and Ryzen hardware. For storage teams, hipFile and AMD Infinity Storage now plug into PyTorch, fio, and ior, and they work on volumes built with LVM and The post AMD ROCm 10.2 Brings hipFile to PyTorch, LVM, and mdraid Volumes, and Adds 18 Ryzen Processors appeared first on StorageReview.com.

## AMD ROCm 10.2 Released With Latest Optimizations, Expanded Ryzen Laptop Support

DevFeed: [AMD ROCm 10.2 Released With Latest Optimizations, Expanded Ryzen Laptop Support](<https://devfeed.tech/articles/amd-rocm-10-2-released-with-latest-optimizations-expanded-ryzen-laptop-support-141973.md>)

Original publisher: [Read original article](<https://www.phoronix.com/news/AMD-ROCm-10.2>)

Author: Michael Larabel

Published: 2026-11-19T02:06:39Z

Content type: release

Language: en

Sources: [Phoronix](<https://devfeed.tech/sources/phoronix.md>)

Topics: [rocm](<https://devfeed.tech/topics/rocm.md>), [Open Source](<https://devfeed.tech/topics/open-source.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [amd](<https://devfeed.tech/tags/amd.md>), [amd-ryzen](<https://devfeed.tech/tags/amd-ryzen.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [cli](<https://devfeed.tech/tags/cli.md>), [compatibility](<https://devfeed.tech/tags/compatibility.md>), [compute](<https://devfeed.tech/tags/compute.md>), [desktop-linux](<https://devfeed.tech/tags/desktop-linux.md>), [developers](<https://devfeed.tech/tags/developers.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [gpu-compute](<https://devfeed.tech/tags/gpu-compute.md>), [hip](<https://devfeed.tech/tags/hip.md>), [improvements](<https://devfeed.tech/tags/improvements.md>), [laptop](<https://devfeed.tech/tags/laptop.md>), [linux-benchmarking](<https://devfeed.tech/tags/linux-benchmarking.md>), [linux-hardware-benchmarks](<https://devfeed.tech/tags/linux-hardware-benchmarks.md>), [linux-hardware-reviews](<https://devfeed.tech/tags/linux-hardware-reviews.md>), [linux-how-to](<https://devfeed.tech/tags/linux-how-to.md>), [linux-performance](<https://devfeed.tech/tags/linux-performance.md>), [linux-server-benchmarks](<https://devfeed.tech/tags/linux-server-benchmarks.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [open-source-graphics](<https://devfeed.tech/tags/open-source-graphics.md>), [optimizations](<https://devfeed.tech/tags/optimizations.md>), [phoronix](<https://devfeed.tech/tags/phoronix.md>), [phoronix-test-suite](<https://devfeed.tech/tags/phoronix-test-suite.md>), [radeon](<https://devfeed.tech/tags/radeon.md>), [release](<https://devfeed.tech/tags/release.md>), [rocm](<https://devfeed.tech/tags/rocm.md>), [ryzen](<https://devfeed.tech/tags/ryzen.md>), [support](<https://devfeed.tech/tags/support.md>), [ubuntu-benchmarks](<https://devfeed.tech/tags/ubuntu-benchmarks.md>), [ubuntu-hardware](<https://devfeed.tech/tags/ubuntu-hardware.md>)

### AI overview

ROCm 10.2 adds official support for 18 more Ryzen processors using gfx1103, including older RDNA3 laptop models, with Windows and Linux compatibility and no custom builds or special configurations required. The excerpt says gfx1103 support was validated across HIP, PyTorch, and vLLM. The release also upgrades SGLang and vLLM support, continues work on Hyperloom, ROCm CLI, and AMD Skills, and includes an RCCL performance improvement for PCIe GPUs. Platform support expands to AMD Infinity Storage (AIS), hipFile, RHEL 10.3 and 9.9, and SLES 16.1; the excerpt also lists the Radeon PRO W7900D.

### Source excerpt

AMD ROCm 10.2 is out today as the newest six-week feature release to this open-source AMD Radeon/Instinct GPU compute stack...

## Keeping Agents Fast: High-Interactivity Decode with FlyDSL MonoKernels in vLLM

DevFeed: [Keeping Agents Fast: High-Interactivity Decode with FlyDSL MonoKernels in vLLM](<https://devfeed.tech/articles/keeping-agents-fast-high-interactivity-decode-with-flydsl-monokernels-in-vllm-140940.md>)

Original publisher: [Read original article](<https://vllm.ai/blog/2026-11-18-flydsl-monokernels>)

Author: Felix Li, Lingpeng Jin, Hattie Wu, Carlus Huang, Fangzhou Ai, Zaidy Ali, Amanzhol Salykov, Hyukjoon Lee, Ziming Huang, Jun Chen, Yifeng Huang, Bin Ding, Xiaobing Zhang, Yi Gan, Vinayak Gokhale, Hongxi

Published: 2026-11-18T00:00:00Z

Content type: article

Language: en

Sources: [vLLM Blog](<https://devfeed.tech/sources/vllm-blog.md>)

Topics: [vllm](<https://devfeed.tech/topics/vllm.md>), [rocm](<https://devfeed.tech/topics/rocm.md>), [mlir](<https://devfeed.tech/topics/mlir.md>), [quantization](<https://devfeed.tech/topics/quantization.md>), [High-Performance Computing](<https://devfeed.tech/topics/high-performance-computing.md>)

Tags: [agents](<https://devfeed.tech/tags/agents.md>), [ai-engineering](<https://devfeed.tech/tags/ai-engineering.md>), [amd](<https://devfeed.tech/tags/amd.md>), [decode](<https://devfeed.tech/tags/decode.md>), [fusion](<https://devfeed.tech/tags/fusion.md>), [handoffs](<https://devfeed.tech/tags/handoffs.md>), [hardware](<https://devfeed.tech/tags/hardware.md>), [instinct](<https://devfeed.tech/tags/instinct.md>), [integration](<https://devfeed.tech/tags/integration.md>), [llm-inference](<https://devfeed.tech/tags/llm-inference.md>), [model](<https://devfeed.tech/tags/model.md>), [model-serving](<https://devfeed.tech/tags/model-serving.md>), [open-source-ai](<https://devfeed.tech/tags/open-source-ai.md>), [performance](<https://devfeed.tech/tags/performance.md>), [vllm](<https://devfeed.tech/tags/vllm.md>), [vllm-blog](<https://devfeed.tech/tags/vllm-blog.md>)

### AI overview

FlyDSL MonoKernels aim to reduce low-concurrency decode latency in vLLM on AMD Instinct GPUs by combining dependent stages into a few persistent kernel launches. Authors choose model-specific boundaries and schedules; FlyDSL compiles them, while vLLM selects the opt-in path for supported steps. Thread blocks can reuse work across stages, prefetch weights, and hand off ready tiles through tagged mailboxes, though dependencies and memory traffic remain. Reported evaluations show gains across several models, with results varying by model, GPU, concurrency, and benchmark; some comparisons use historical baselines. The paths preserve vLLM's scheduler and KV cache, fall back to stock layers for unsupported workloads, and require accuracy checks per configuration. Limits include reduced gains at higher concurrency, maximum step widths, and extra weight copies on MI300X/MI325X.

### Source excerpt

How FlyDSL MonoKernels keep agents responsive with high-interactivity decode in vLLM on AMD Instinct GPUs, using model-specific fusion, data-ready handoffs, and specialized-decode integration.

## AMD ROCm HRX's Loom 0.2 Expands C++ Support, Improved Targeting

DevFeed: [AMD ROCm HRX's Loom 0.2 Expands C++ Support, Improved Targeting](<https://devfeed.tech/articles/amd-rocm-hrx-s-loom-0-2-expands-c-support-improved-targeting-136380.md>)

Original publisher: [Read original article](<https://www.phoronix.com/news/HRX-Loom-0.2>)

Author: Michael Larabel

Published: 2026-11-17T01:37:15Z

Content type: release

Language: en

Sources: [Phoronix](<https://devfeed.tech/sources/phoronix.md>)

Topics: [rocm](<https://devfeed.tech/topics/rocm.md>), [c/c++](<https://devfeed.tech/topics/c-c-plus-plus.md>)

Tags: [2](<https://devfeed.tech/tags/2.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agent-workflows](<https://devfeed.tech/tags/agent-workflows.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [amd](<https://devfeed.tech/tags/amd.md>), [amd-gpu](<https://devfeed.tech/tags/amd-gpu.md>), [announcement](<https://devfeed.tech/tags/announcement.md>), [authoring](<https://devfeed.tech/tags/authoring.md>), [c-plus-plus](<https://devfeed.tech/tags/c-plus-plus.md>), [capabilities](<https://devfeed.tech/tags/capabilities.md>), [compilation](<https://devfeed.tech/tags/compilation.md>), [compiler](<https://devfeed.tech/tags/compiler.md>), [compute](<https://devfeed.tech/tags/compute.md>), [desktop-linux](<https://devfeed.tech/tags/desktop-linux.md>), [infrastructure](<https://devfeed.tech/tags/infrastructure.md>), [linux-benchmarking](<https://devfeed.tech/tags/linux-benchmarking.md>), [linux-hardware-benchmarks](<https://devfeed.tech/tags/linux-hardware-benchmarks.md>), [linux-hardware-reviews](<https://devfeed.tech/tags/linux-hardware-reviews.md>), [linux-how-to](<https://devfeed.tech/tags/linux-how-to.md>), [linux-performance](<https://devfeed.tech/tags/linux-performance.md>), [linux-server-benchmarks](<https://devfeed.tech/tags/linux-server-benchmarks.md>), [loom](<https://devfeed.tech/tags/loom.md>), [open-source-graphics](<https://devfeed.tech/tags/open-source-graphics.md>), [phoronix](<https://devfeed.tech/tags/phoronix.md>), [phoronix-test-suite](<https://devfeed.tech/tags/phoronix-test-suite.md>), [rocm](<https://devfeed.tech/tags/rocm.md>), [runtime](<https://devfeed.tech/tags/runtime.md>), [serving](<https://devfeed.tech/tags/serving.md>), [ubuntu-benchmarks](<https://devfeed.tech/tags/ubuntu-benchmarks.md>), [ubuntu-hardware](<https://devfeed.tech/tags/ubuntu-hardware.md>), [version](<https://devfeed.tech/tags/version.md>)

### AI overview

Loom 0.2 broadens complete-program C++ support and packaged AMD GPU/NPU and Vulkan execution, while expanding a shared target compiler across x86, AIE2P, AMDGPU, SPIR-V, and WebAssembly. The release describes changes aimed at removing fixed and superlinear costs in JIT compilation, graph construction and updates, GPU launch and replay, pipeline planning, and long-context serving. AMD hardware tooling now uses one libamdf-backed HAL via `--device=amdgpu`; the native path supports RDNA/CDNA, shared CPU/GPU/NPU memory contracts, explicit compute-resource selection, cooperative execution, virtual mappings, and reusable command buffers. The former ROCr-backed driver and its separate runtime and memory ownership paths are removed.

### Source excerpt

AMD ROCm's HRX Loom as the compiler, runtime, and serving infrastructure for supporting AI agent workflows is out today with a new version...

## IBM and AMD Open-Source Pelagos GPU Evaluation Platform for Quantum Error-Correction Decoders

DevFeed: [IBM and AMD Open-Source Pelagos GPU Evaluation Platform for Quantum Error-Correction Decoders](<https://devfeed.tech/articles/ibm-and-amd-open-source-pelagos-gpu-evaluation-platform-for-quantum-error-correction-decoders-136398.md>)

Original publisher: [Read original article](<https://joho-todai.com/en/ibm-amd-pelagos-quantum-error-correction-gpu/>)

Author: Toru Misaki

Published: 2026-11-17T01:21:57Z

Content type: news

Language: en

Sources: [Joho no Todai](<https://devfeed.tech/sources/joho-no-todai.md>)

Topics: [Quantum Computing](<https://devfeed.tech/topics/quantum-computing.md>), [rocm](<https://devfeed.tech/topics/rocm.md>), [ibm](<https://devfeed.tech/topics/ibm.md>), [Open Source](<https://devfeed.tech/topics/open-source.md>)

Tags: [accelerate](<https://devfeed.tech/tags/accelerate.md>), [accelerators](<https://devfeed.tech/tags/accelerators.md>), [amd](<https://devfeed.tech/tags/amd.md>), [amd-instinct-mi355x](<https://devfeed.tech/tags/amd-instinct-mi355x.md>), [computing](<https://devfeed.tech/tags/computing.md>), [cpu](<https://devfeed.tech/tags/cpu.md>), [error-correction](<https://devfeed.tech/tags/error-correction.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [framework](<https://devfeed.tech/tags/framework.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [ibm](<https://devfeed.tech/tags/ibm.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [platform](<https://devfeed.tech/tags/platform.md>), [power](<https://devfeed.tech/tags/power.md>), [quantum](<https://devfeed.tech/tags/quantum.md>), [quantum-computing](<https://devfeed.tech/tags/quantum-computing.md>), [quantum-error-correction](<https://devfeed.tech/tags/quantum-error-correction.md>), [qubits](<https://devfeed.tech/tags/qubits.md>), [research](<https://devfeed.tech/tags/research.md>), [throughput](<https://devfeed.tech/tags/throughput.md>)

### AI overview

Pelagos, an IBM and AMD open-source framework, uses GPUs to speed up validation of quantum error-correction decoders by simulating error patterns. It is implemented in Triton and supports NVIDIA (CUDA), AMD (ROCm/HIP), and Intel (SPIR-V) GPUs. Its Relay-BP-S decoder targets quantum LDPC codes. Reported results vary by workload and metric: eight MI355X GPUs reached 6.23x single-GPU speedup end to end for a one-gross circuit, while the decoding stage alone scaled 7.45x for a two-gross circuit; MI300X results were 5.68x and 7.58x, respectively. The earlier claim of up to 97x compares one GPU with a single CPU core and is not directly comparable. Pelagos requires a GPU; a separate Rust relay-bp implementation is available for CPUs.

### Source excerpt

Practical quantum computing requires not only capable qubits but also classical computing power to detect and correct errors. Pelagos, an open-source framework from IBM and AMD, targets that classical-computing bottleneck and achieved up to 97x the throughput of a single CPU core in earlier research.

## HPE Lux AI Cluster Debuts at No. 14 on the TOP500 With 220.79 Petaflops on AMD Instinct MI355X GPUs

DevFeed: [HPE Lux AI Cluster Debuts at No. 14 on the TOP500 With 220.79 Petaflops on AMD Instinct MI355X GPUs](<https://devfeed.tech/articles/hpe-lux-ai-cluster-debuts-at-no-14-on-the-top500-with-220-79-petaflops-on-amd-instinct-mi355x-gpus-136079.md>)

Original publisher: [Read original article](<https://www.storagereview.com/news/hpe-lux-ai-cluster-top500-no-14-220-79-petaflops-ornl-genesis-mission>)

Author: Harold Fritts

Published: 2026-11-16T19:00:00Z

Content type: news

Language: en

Sources: [StorageReview.com](<https://devfeed.tech/sources/storagereview-com.md>)

Topics: [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [Supercomputing](<https://devfeed.tech/topics/supercomputing.md>), [rocm](<https://devfeed.tech/topics/rocm.md>), [AI Networking](<https://devfeed.tech/topics/ai-networking.md>), [GPU optimization](<https://devfeed.tech/topics/gpu-optimization.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-infrastructure](<https://devfeed.tech/tags/ai-infrastructure.md>), [ai-training-and-inference](<https://devfeed.tech/tags/ai-training-and-inference.md>), [amd](<https://devfeed.tech/tags/amd.md>), [amd-epyc](<https://devfeed.tech/tags/amd-epyc.md>), [amd-instinct-mi355x](<https://devfeed.tech/tags/amd-instinct-mi355x.md>), [cluster](<https://devfeed.tech/tags/cluster.md>), [computing](<https://devfeed.tech/tags/computing.md>), [cpus](<https://devfeed.tech/tags/cpus.md>), [data-center](<https://devfeed.tech/tags/data-center.md>), [deep](<https://devfeed.tech/tags/deep.md>), [deep-learning](<https://devfeed.tech/tags/deep-learning.md>), [department-of-energy](<https://devfeed.tech/tags/department-of-energy.md>), [enterprise](<https://devfeed.tech/tags/enterprise.md>), [general](<https://devfeed.tech/tags/general.md>), [gpus](<https://devfeed.tech/tags/gpus.md>), [hpe](<https://devfeed.tech/tags/hpe.md>), [oak-ridge-national-laboratory](<https://devfeed.tech/tags/oak-ridge-national-laboratory.md>), [performance](<https://devfeed.tech/tags/performance.md>), [supercomputing](<https://devfeed.tech/tags/supercomputing.md>), [top500](<https://devfeed.tech/tags/top500.md>)

### AI overview

HPE says Lux, an AI cluster built with AMD for Oak Ridge National Laboratory, measured 220.79 petaflops on HPL and debuted at No. 14 on the TOP500, ahead of systems designed for traditional supercomputing workloads. It uses direct liquid-cooled HPE ProLiant XD685 servers with AMD Instinct MI355X GPUs, EPYC CPUs and Pensando DPUs, plus 800G Ethernet switches. HPE says network load balancing and congestion management are intended to improve GPU utilization; AMD says the system runs ROCm. Lux provides multi-tenant, cloud-style access for researchers and is the first dedicated AI system online for DOE's Genesis Mission. Half of its 3.5 million annual node-hours are reserved for that mission and half for proprietary use. The excerpt gives no node or GPU count.

### Source excerpt

HPE today announced that Lux, the AI cluster it built with AMD for Oak Ridge National Laboratory (ORNL), has debuted at No. 14 on the TOP500 list with 220.79 petaflops of measured HPL performance. HPE says the system, which was designed for AI training and inference, placed ahead of machines built for traditional supercomputing workloads. The post HPE Lux AI Cluster Debuts at No. 14 on the TOP500 With 220.79 Petaflops on AMD Instinct MI355X GPUs appeared first on StorageReview.com.

## IBM, AMD advance quantum error correction decoding at scale

DevFeed: [IBM, AMD advance quantum error correction decoding at scale](<https://devfeed.tech/articles/ibm-amd-advance-quantum-error-correction-decoding-at-scale-136373.md>)

Original publisher: [Read original article](<https://research.ibm.com/blog/pelagos>)

Author: Blake Johnson; Robert Davis

Published: 2026-11-16T15:59:00Z

Content type: news

Language: en

Sources: [IBM Research](<https://devfeed.tech/sources/ibm-research.md>)

Topics: [Quantum Computing](<https://devfeed.tech/topics/quantum-computing.md>), [Decoding](<https://devfeed.tech/topics/decoding.md>), [rocm](<https://devfeed.tech/topics/rocm.md>), [ibm](<https://devfeed.tech/topics/ibm.md>)

Tags: [acceleration](<https://devfeed.tech/tags/acceleration.md>), [accelerators](<https://devfeed.tech/tags/accelerators.md>), [algorithms](<https://devfeed.tech/tags/algorithms.md>), [amd](<https://devfeed.tech/tags/amd.md>), [decoding](<https://devfeed.tech/tags/decoding.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [ibm](<https://devfeed.tech/tags/ibm.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [quantum](<https://devfeed.tech/tags/quantum.md>), [quantum-error-correction](<https://devfeed.tech/tags/quantum-error-correction.md>), [quantum-research](<https://devfeed.tech/tags/quantum-research.md>), [quantum-software](<https://devfeed.tech/tags/quantum-software.md>), [release](<https://devfeed.tech/tags/release.md>), [scale](<https://devfeed.tech/tags/scale.md>)

### AI overview

IBM and AMD released Pelagos, an open-source framework for evaluating quantum error-correction decoders at scale. Built with Triton, it supports AMD and NVIDIA GPUs and combines GPU acceleration with a reusable software architecture. Multi-GPU work uses process-level orchestration, parallel host-side operations and shared memory to reduce coordination overhead. Tests on AMD Instinct MI300X and MI355X showed up to roughly 7.5x speedup in the GPU decoding stage across eight GPUs versus one; the team also deployed Pelagos on ORNL's Lux HPC system. The excerpt says earlier studies found up to 97x higher throughput than a single CPU core. An accompanying survey reviews decoding methods and platforms including CPUs, GPUs, FPGAs and specialized accelerators, and argues that classical infrastructure and hardware-software co-design are key parts of building fault-tolerant quantum systems.

### Source excerpt

New open-source tool from IBM and AMD helps researchers evaluate quantum error-correction decoders at scale for future fault-tolerant quantum systems.

## AMD wants gamers to remember Radeon loyalty, days after declaring itself a data center-first company

DevFeed: [AMD wants gamers to remember Radeon loyalty, days after declaring itself a data center-first company](<https://devfeed.tech/articles/amd-wants-gamers-to-remember-radeon-loyalty-days-after-declaring-itself-a-data-center-first-company-132363.md>)

Original publisher: [Read original article](<https://www.techspot.com/news/114547-amd-wants-gamers-remember-radeon-loyalty-days-after.html>)

Author: Rob Thubron

Published: 2026-11-13T13:41:00Z

Content type: news

Language: en

Sources: [TechSpot](<https://devfeed.tech/sources/techspot.md>)

Topics: [radeon-gpu-profiler](<https://devfeed.tech/topics/radeon-gpu-profiler.md>), [rocm](<https://devfeed.tech/topics/rocm.md>), [X (Twitter)](<https://devfeed.tech/topics/twitter.md>), [RTX PRO](<https://devfeed.tech/topics/rtx-pro.md>)

Tags: [accelerators](<https://devfeed.tech/tags/accelerators.md>), [ai](<https://devfeed.tech/tags/ai.md>), [amd](<https://devfeed.tech/tags/amd.md>), [data-center](<https://devfeed.tech/tags/data-center.md>), [epyc](<https://devfeed.tech/tags/epyc.md>), [gaming](<https://devfeed.tech/tags/gaming.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [graphics-card](<https://devfeed.tech/tags/graphics-card.md>), [graphics-cards](<https://devfeed.tech/tags/graphics-cards.md>), [hardware](<https://devfeed.tech/tags/hardware.md>), [jensen-huang](<https://devfeed.tech/tags/jensen-huang.md>), [msrp](<https://devfeed.tech/tags/msrp.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [radeon](<https://devfeed.tech/tags/radeon.md>)

### AI overview

AMD's recent "Radeon loyalty" post contrasts with CEO Lisa Su's description of the company as data center first. In the supplied excerpt, AMD's Q3 revenue reached $13.4 billion, with data center revenue up 97% to $8.5 billion and gaming revenue down 40% to $773 million. The excerpt attributes the gaming decline largely to AI-driven memory price increases. It also reports that Radeon GPUs faced price markups, though the RX 9060 XT 16GB's roughly $200 premium over MSRP was smaller than the RTX 5060 Ti 16GB's roughly $300 premium in a ten-country survey. AMD's listed new desktop releases were limited, including a poorly rated entry-level RX 9050 and wider availability for the previously China-exclusive RX 9070 GRE. These figures and examples frame the social post as a mismatch with AMD's business priorities and recent gaming market activity.

### Source excerpt

AMD's post might be a response to all that's going on at Nvidia right now. With Jensen Huang having just quit as CEO after 33 years at the helm, and questions mounting over what his departure means for the AI industry and the wider economy, perhaps Team Red thought it... Read Entire Article

## Nutanix Powered by AMD: AI Infrastructure with Kubernetes

DevFeed: [Nutanix Powered by AMD: AI Infrastructure with Kubernetes](<https://devfeed.tech/articles/nutanix-powered-by-amd-ai-infrastructure-with-kubernetes-127998.md>)

Original publisher: [Read original article](<https://www.nutanix.com/blog/nutanix-powered-by-amd-simplifying-ai-infrastructure-with-enterprise-kubernetes>)

Author: Fahima Zahir

Published: 2026-11-10T08:00:00Z

Content type: article

Language: en

Sources: [Nutanix](<https://devfeed.tech/sources/nutanix.md>)

Topics: [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [nutanix](<https://devfeed.tech/topics/nutanix.md>), [rocm](<https://devfeed.tech/topics/rocm.md>), [Ubuntu](<https://devfeed.tech/topics/ubuntu.md>)

Tags: [2026](<https://devfeed.tech/tags/2026.md>), [acceleration](<https://devfeed.tech/tags/acceleration.md>), [accelerators](<https://devfeed.tech/tags/accelerators.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-infrastructure](<https://devfeed.tech/tags/ai-infrastructure.md>), [amd](<https://devfeed.tech/tags/amd.md>), [blog-post](<https://devfeed.tech/tags/blog-post.md>), [can](<https://devfeed.tech/tags/can.md>), [cloud-native](<https://devfeed.tech/tags/cloud-native.md>), [cloud-native-aos](<https://devfeed.tech/tags/cloud-native-aos.md>), [epyc](<https://devfeed.tech/tags/epyc.md>), [experience](<https://devfeed.tech/tags/experience.md>), [gpus](<https://devfeed.tech/tags/gpus.md>), [infrastructure](<https://devfeed.tech/tags/infrastructure.md>), [instinct](<https://devfeed.tech/tags/instinct.md>), [kubernetes](<https://devfeed.tech/tags/kubernetes.md>), [metal](<https://devfeed.tech/tags/metal.md>), [multicloud-kubernetes](<https://devfeed.tech/tags/multicloud-kubernetes.md>), [networking](<https://devfeed.tech/tags/networking.md>), [nutanix](<https://devfeed.tech/tags/nutanix.md>), [nutanix-kubernetes-platform-nkp](<https://devfeed.tech/tags/nutanix-kubernetes-platform-nkp.md>), [platform](<https://devfeed.tech/tags/platform.md>), [servers](<https://devfeed.tech/tags/servers.md>), [tof-awareness](<https://devfeed.tech/tags/tof-awareness.md>), [use-cases](<https://devfeed.tech/tags/use-cases.md>), [with](<https://devfeed.tech/tags/with.md>)

### AI overview

Nutanix Kubernetes Platform (NKP) Metal brings AMD Instinct MI350X GPUs, AMD networking and AMD 5th Gen EPYC servers into a shared Kubernetes operating model for deployment, application delivery, lifecycle management and support. The text identifies a validated configuration of eight MI350X GPUs per Cisco UCS C885A M8 server running Ubuntu 24.04 LTS, with the AMD GPU Operator and AMD Network Operator available through the NKP AI Application Catalog, including integration with the AMD Pensando Pollara 400 AI NIC and Multus. Nutanix Foundation, Life Cycle Management and Nutanix Cluster Check provide platform-specific setup, upgrades and health checks. The described goal is to reduce custom integration and maintenance work as teams deploy AI alongside other workloads; support depends on applicable NKP, operator and hardware releases.

### Source excerpt

With Nutanix Kubernetes Platform (NKP) Metal, you can bring AMD Instinct MI350X GPUs, AMD networking, and AMD 5th Gen EPYC servers into a common platform experience.

## Fused-Softmax for efficient quantized attention

DevFeed: [Fused-Softmax for efficient quantized attention](<https://devfeed.tech/articles/fused-softmax-for-efficient-quantized-attention-123376.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.09225>)

Author: Tianxing Wu, Juuso Korhonen, Jesus Carabano Bravo

Published: 2026-11-09T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>), [Tensor Cores](<https://devfeed.tech/topics/tensor-cores.md>), [rocm](<https://devfeed.tech/topics/rocm.md>)

Tags: [amd](<https://devfeed.tech/tags/amd.md>), [amd-instinct-mi355x](<https://devfeed.tech/tags/amd-instinct-mi355x.md>), [attention](<https://devfeed.tech/tags/attention.md>), [bf16](<https://devfeed.tech/tags/bf16.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cs-pf](<https://devfeed.tech/tags/cs-pf.md>), [error](<https://devfeed.tech/tags/error.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [hybrid](<https://devfeed.tech/tags/hybrid.md>), [inference](<https://devfeed.tech/tags/inference.md>), [kernels](<https://devfeed.tech/tags/kernels.md>), [mlperf](<https://devfeed.tech/tags/mlperf.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [performance](<https://devfeed.tech/tags/performance.md>), [quality](<https://devfeed.tech/tags/quality.md>), [ship](<https://devfeed.tech/tags/ship.md>), [throughput](<https://devfeed.tech/tags/throughput.md>), [training](<https://devfeed.tech/tags/training.md>)

### AI overview

Fused-Softmax Attention reduces softmax's bottleneck in quantized attention by folding its dominant operations into attention GEMMs, shifting work from vector units to matrix cores. On AMD CDNA 4, its MXFP8, MXFP6 and MXFP4 kernels reach about 3.0, 4.0 and 4.6 PFLOP/s on an MI355X, with reported speedups of 2.1x, 2.8x and 3.2x over BF16 attention. The algorithm is training-free; its added error is reported below the error floor from quantizing GEMM operands. For Wan2.2-T2V-A14B, a hybrid MXFP6/4 schedule achieves up to 2.05x end-to-end speedup while preserving visual quality. The kernels are available in AMD's open-source AITER library and were used in AMD's MLPerf Inference v6.1 Wan2.2 submission, which was 1.7x faster than its previous-round result on the same hardware.

### Source excerpt

arXiv:2611.09225v1 Announce Type: cross Abstract: Low-precision matrix throughput on modern GPUs is growing far faster than vector-arithmetic throughput. Attention therefore cannot realize the available GEMM performance because softmax has become the pipeline bottleneck. We introduce Fused-Softmax Attention, a training-free algorithm that reformulates and approximates softmax to fold its dominant operations into the attention GEMMs, moving work from vector units to matrix cores. Implemented for the AMD CDNA 4 architecture, our MXFP8, MXFP6 and MXFP4 kernels reach approximately 3.0, 4.0 and 4.6 PFLOP/s on an AMD Instinct MI355X GPU, yielding speedups of $2.1\times$, $2.8\times$ and $3.2\times$ over the BF16 attention. Importantly, the additional error introduced by Fused-Softmax remains below the error floor set by quantizing the GEMM operands. On Wan2.2-T2V-A14B, a hybrid MXFP6/4 schedule delivers up to $2.05\times$ end-to-end speedup while preserving the visual quality of the outputs. The kernels ship in AMD's open-source AITER library, and they powered AMD's MLPerf Inference v6.1 Wan2.2 submission, which ran $1.7\times$ faster than AMD's previous-round result on the same hardware. Our results show that redesigning softmax to utilize the matrix pipeline enables attention to capitalize on continued growth in low-precision matrix throughput.

## Zyphra ZYFA: FP8 FlashAttention for AMD MI355X, Up to 1.76x Faster

DevFeed: [Zyphra ZYFA: FP8 FlashAttention for AMD MI355X, Up to 1.76x Faster](<https://devfeed.tech/articles/zyphra-zyfa-fp8-flashattention-for-amd-mi355x-up-to-1-76x-faster-121539.md>)

Original publisher: [Read original article](<https://www.marktechpost.com/2026/11/06/zyphra-zyfa-fp8-flashattention-amd-mi355x/>)

Author: Michal Sutter

Published: 2026-11-06T17:28:04Z

Content type: article

Language: en

Sources: [MarkTechPost](<https://devfeed.tech/sources/marktechpost.md>)

Topics: [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>), [quantization](<https://devfeed.tech/topics/quantization.md>), [rocm](<https://devfeed.tech/topics/rocm.md>), [Open Source](<https://devfeed.tech/topics/open-source.md>)

Tags: [ai-infrastructure](<https://devfeed.tech/tags/ai-infrastructure.md>), [ai-paper-summary](<https://devfeed.tech/tags/ai-paper-summary.md>), [ai-shorts](<https://devfeed.tech/tags/ai-shorts.md>), [amd](<https://devfeed.tech/tags/amd.md>), [amd-instinct-mi355x](<https://devfeed.tech/tags/amd-instinct-mi355x.md>), [applications](<https://devfeed.tech/tags/applications.md>), [artificial-intelligence](<https://devfeed.tech/tags/artificial-intelligence.md>), [attention](<https://devfeed.tech/tags/attention.md>), [editors-pick](<https://devfeed.tech/tags/editors-pick.md>), [error](<https://devfeed.tech/tags/error.md>), [faster](<https://devfeed.tech/tags/faster.md>), [flashattention](<https://devfeed.tech/tags/flashattention.md>), [fp8](<https://devfeed.tech/tags/fp8.md>), [kernels](<https://devfeed.tech/tags/kernels.md>), [language-model](<https://devfeed.tech/tags/language-model.md>), [large-language-model](<https://devfeed.tech/tags/large-language-model.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [new-releases](<https://devfeed.tech/tags/new-releases.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [prefill](<https://devfeed.tech/tags/prefill.md>), [staff](<https://devfeed.tech/tags/staff.md>), [tech-news](<https://devfeed.tech/tags/tech-news.md>), [technology](<https://devfeed.tech/tags/technology.md>), [uncategorized](<https://devfeed.tech/tags/uncategorized.md>)

### AI overview

ZYFA is an open-source Apache 2.0 library of FP8 and BF16 causal prefill attention kernels for AMD Instinct MI355X, with no decode or paged-KV support. On real model activations, its FP8 kernels are 1.16-1.76x faster than AMD's AITER and show 62-78% lower error at head dimension 128; BF16 performance roughly ties. To address MI355X's softmax bottleneck, ZYFA approximates probabilities with one multiply-add and uses a two-wave schedule to overlap softmax and matrix work. Benchmark gains can exceed real-model gains: clock behavior varies with activation values, and ZYFA reports a lower sustained clock than AITER on Mistral-7B despite using fewer cycles. The excerpt specifies MI355X, ROCm 7.2.3, and PyTorch 2.12 support; it reports no NVIDIA measurements.

### Source excerpt

Zyphra released ZYFA, open-source FP8 and BF16 prefill attention kernels for AMD Instinct MI355X. On real model activations, its FP8 kernels run 1.16-1.76x faster than AMD's AITER, with 62-78% lower error at head dimension 128. The post Zyphra ZYFA: FP8 FlashAttention for AMD MI355X, Up to 1.76x Faster appeared first on MarkTechPost.

## Capability Gain from ISA to Shared Execution on AMD MI300A and MI250X

DevFeed: [Capability Gain from ISA to Shared Execution on AMD MI300A and MI250X](<https://devfeed.tech/articles/capability-gain-from-isa-to-shared-execution-on-amd-mi300a-and-mi250x-121265.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.07198>)

Author: Aaron Jarmusch, Sunita Chandrasekaran

Published: 2026-11-06T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Distributed Computing](<https://devfeed.tech/sources/arxiv-distributed-computing.md>)

Topics: [GPU optimization](<https://devfeed.tech/topics/gpu-optimization.md>), [rocm](<https://devfeed.tech/topics/rocm.md>), [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>), [quantization](<https://devfeed.tech/topics/quantization.md>)

Tags: [amd](<https://devfeed.tech/tags/amd.md>), [cs-dc](<https://devfeed.tech/tags/cs-dc.md>), [execution](<https://devfeed.tech/tags/execution.md>), [fp8](<https://devfeed.tech/tags/fp8.md>), [inference](<https://devfeed.tech/tags/inference.md>), [isa](<https://devfeed.tech/tags/isa.md>), [llm](<https://devfeed.tech/tags/llm.md>), [llm-serving](<https://devfeed.tech/tags/llm-serving.md>), [metric](<https://devfeed.tech/tags/metric.md>), [performance](<https://devfeed.tech/tags/performance.md>), [precision](<https://devfeed.tech/tags/precision.md>), [queue](<https://devfeed.tech/tags/queue.md>), [queues](<https://devfeed.tech/tags/queues.md>), [sparsity](<https://devfeed.tech/tags/sparsity.md>), [streams](<https://devfeed.tech/tags/streams.md>), [theoretical](<https://devfeed.tech/tags/theoretical.md>)

### AI overview

The paper defines capability gain as the fraction of an instruction-level speedup retained at higher execution layers, and compares AMD MI300A with MI250X from ISA microbenchmarks through libraries, applications and shared-device workloads. MI300A's FP8 retains most of its speedup in hipBLASLt: an 8192³ GEMM is reported faster than MI250X FP16, though the abstract's speedup value is missing. Unfused quantization makes FP8 slower per layer in LLM inference, and shared LLM serving gains nothing from FP8. 2:4 sparsity retains at most about half its speedup in hipSPARSELt. With queue sharing, FP8 GEMM capability gain exceeds 1; stream fairness depends on hardware queue count rather than precision.

### Source excerpt

arXiv:2611.07198v1 Announce Type: new Abstract: Each new GPU generation is showcased by its peak instruction rates, yet users hardly ever run instructions directly: they reach the hardware's performance through vendor libraries, applications and devices shared with other jobs. How much of a theoretical speedup survives these layers is rarely measured, so a team deciding whether to move to a new generation, or to adopt FP8 or 2:4 sparsity, lacks adequate guidance. AMD's MI300A (CDNA3) adds new FP8 matrix instructions and 2:4 structured-sparse instructions that the predecessor, the MI250X (CDNA2), does not have. We analyze this theoretical speedup gap on AMD's MI300A (CDNA3) against the MI250X (CDNA2) by defining a new metric, capability gain. Capability gain is the fraction of a feature's instruction-level speedup that it keeps at a higher level, such as a library, an application or a shared device. We trace it from ISA microbenchmarks through hipBLASLt and hipSPARSELt to LLM inference and mixed-precision HPL, and onto a device shared by up to eight concurrent streams. FP8 keeps most of its speedup through the library, completing an $8192^3$ GEMM \genCapPkgEndToEndGainEightK{} faster than the MI250X with FP16, but unfused quantization turns it into a per-layer slowdown in an LLM, and 2:4 sparsity keeps at most about half of its speedup in hipSPARSELt. Under queue sharing, FP8's GEMM capability gain rises above 1. How fairly streams share the device depends on the number of hardware queues, not on the numeric precision, yet shared LLM serving gains nothing from FP8.

## ZYFA: Speed and Accuracy of FP8 FlashAttention on AMD CDNA4

DevFeed: [ZYFA: Speed and Accuracy of FP8 FlashAttention on AMD CDNA4](<https://devfeed.tech/articles/zyfa-speed-and-accuracy-of-fp8-flashattention-on-amd-cdna4-120624.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.07496>)

Author: Quentin Anthony, Ganesh Bikshandi, Praneeth Medepalli, Stephen Ebert

Published: 2026-11-06T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Distributed Computing](<https://devfeed.tech/sources/arxiv-distributed-computing.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [quantization](<https://devfeed.tech/topics/quantization.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [rocm](<https://devfeed.tech/topics/rocm.md>), [scheduling](<https://devfeed.tech/topics/scheduling.md>)

Tags: [amd](<https://devfeed.tech/tags/amd.md>), [amd-instinct-mi355x](<https://devfeed.tech/tags/amd-instinct-mi355x.md>), [attention](<https://devfeed.tech/tags/attention.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [blackwell](<https://devfeed.tech/tags/blackwell.md>), [cs-dc](<https://devfeed.tech/tags/cs-dc.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [error](<https://devfeed.tech/tags/error.md>), [flashattention](<https://devfeed.tech/tags/flashattention.md>), [fp8](<https://devfeed.tech/tags/fp8.md>), [gaussian](<https://devfeed.tech/tags/gaussian.md>), [hopper](<https://devfeed.tech/tags/hopper.md>), [kernels](<https://devfeed.tech/tags/kernels.md>), [long-context](<https://devfeed.tech/tags/long-context.md>), [matrix](<https://devfeed.tech/tags/matrix.md>), [ping-pong](<https://devfeed.tech/tags/ping-pong.md>), [power](<https://devfeed.tech/tags/power.md>), [prefill](<https://devfeed.tech/tags/prefill.md>), [quantization](<https://devfeed.tech/tags/quantization.md>), [scaling](<https://devfeed.tech/tags/scaling.md>), [schedule](<https://devfeed.tech/tags/schedule.md>), [speed](<https://devfeed.tech/tags/speed.md>)

### AI overview

ZYFA's MI355X FP8 attention prefill kernels outperform AITER in the reported tests while reducing numerical error at head dimension 128. They replace exponentials in probability conversion with a multiply-add and integer conversion, overlap softmax and matrix work with a ping-pong thread schedule, and scale probability blocks by powers of two before rounding and normalization. On inputs captured from three models, ZYFA reaches 1.16-1.76x AITER throughput; Gaussian kernel benchmarks at 32K tokens show 1.26-1.37x at head dimension 128 and 1.99x for DeepSeek-style multi-head latent attention. At head dimension 128, its numerical error is within a few percent of Composable Kernel and Triton and 62-78% lower than AITER's. Results on random inputs may not predict real-activation throughput because MI355X sustained clock speed varies with operand values.

### Source excerpt

arXiv:2611.07496v1 Announce Type: cross Abstract: Attention is a core building block of frontier models and often dominates compute-bound prefill at long context. FP8 matrix instructions offer twice the peak throughput of BF16, but they accelerate only matrix multiplication, and the softmax between the two matrix multiplications of attention runs on vector units that FP8 does not accelerate. On the AMD Instinct MI355X, the vendor's AITER kernels reach 39% of peak FP8 matrix throughput at 32K tokens, compared with 49% in BF16. Attention is also more numerically sensitive to quantization than feed-forward layers, and unlike NVIDIA's Hopper and Blackwell GPUs, the MI355X does not track the completion of matrix instructions in hardware, so overlapping the softmax with the matrix multiplications depends on instruction order. We present ZYFA, prefill kernels for the MI355X built on three techniques: (1) computing each FP8 probability directly from its score with a multiply-add and an integer conversion instead of an exponential; (2) a ping-pong schedule in which one group of threads computes its softmax while another executes its matrix multiplications; and (3) scaling each block of probabilities up by a power of two before rounding, and normalizing by the sum of the rounded probabilities. On attention inputs captured from three models and timed at sustained clock speeds, ZYFA achieves 1.16-1.76x the throughput of AITER. On the Gaussian inputs of kernel benchmarks at 32K tokens, it achieves 1.26-1.37x at head dimension 128 and 1.99x for DeepSeek-style multi-head latent attention. The sustained clock of the MI355X depends on operand values, so throughput on random inputs does not reliably predict throughput on real model activations. At head dimension 128, ZYFA's numerical error is within a few percent of that of Composable Kernel and Triton and 62-78% lower than that of AITER's. The source code is available at https://github.com/Zyphra/zyfa.

## IBM, ORNL, and AMD Co-Locate the 120-Qubit Nighthawk With an MI355X Umbra Cluster for Quantum Error Correction

DevFeed: [IBM, ORNL, and AMD Co-Locate the 120-Qubit Nighthawk With an MI355X Umbra Cluster for Quantum Error Correction](<https://devfeed.tech/articles/ibm-ornl-and-amd-co-locate-the-120-qubit-nighthawk-with-an-mi355x-umbra-cluster-for-quantum-error-correction-119574.md>)

Original publisher: [Read original article](<https://www.storagereview.com/news/ibm-ornl-amd-quantum-error-correction-co-design-platform-nighthawk-umbra-mi355x>)

Author: Harold Fritts

Published: 2026-11-05T16:44:22Z

Content type: news

Language: en

Sources: [StorageReview.com](<https://devfeed.tech/sources/storagereview-com.md>)

Topics: [quantum mechanics](<https://devfeed.tech/topics/quantum-mechanics.md>), [rocm](<https://devfeed.tech/topics/rocm.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [amd](<https://devfeed.tech/tags/amd.md>), [enterprise](<https://devfeed.tech/tags/enterprise.md>), [error](<https://devfeed.tech/tags/error.md>), [error-correction](<https://devfeed.tech/tags/error-correction.md>), [ibm](<https://devfeed.tech/tags/ibm.md>), [nighthawk](<https://devfeed.tech/tags/nighthawk.md>), [ornl](<https://devfeed.tech/tags/ornl.md>), [quantum](<https://devfeed.tech/tags/quantum.md>), [quantum-error-correction](<https://devfeed.tech/tags/quantum-error-correction.md>), [qubit](<https://devfeed.tech/tags/qubit.md>)

### AI overview

IBM, AMD, and ORNL have co-located IBM's 120-qubit Nighthawk processor with ORNL's Umbra system in Poughkeepsie to research quantum error correction. Nighthawk provides square-lattice connectivity, high circuit throughput, and real-time pulse control; Umbra's AMD MI355X GPUs support decoder training and evaluation, while 5th Gen EPYC CPUs handle orchestration. Researchers can test decoding and conditional execution close to the quantum hardware, including mid-circuit measurement and real-time responses, and compare methods by qubit, time, sample, and classical-resource costs. The work spans several code and decoder approaches, with hierarchical error correction among the targets. Connections to Oak Ridge's broader computing infrastructure are planned, but the announcement provides no deployment counts, performance benchmarks, completion date, or external access schedule.

### Source excerpt

IBM, AMD, and the Department of Energy's Quantum Science Center at Oak Ridge National Laboratory unveiled a co-design platform for quantum error correction today, built around quantum and classical hardware sitting in the same facility. At IBM's quantum site in Poughkeepsie, New York, IBM quantum computers including the 120-qubit IBM Quantum Nighthawk processor are co-located The post IBM, ORNL, and AMD Co-Locate the 120-Qubit Nighthawk With an MI355X Umbra Cluster for Quantum Error Correction appeared first on StorageReview.com.

## AMD Cannot Catch Nvidia, But It Sure Can Make Some Money

DevFeed: [AMD Cannot Catch Nvidia, But It Sure Can Make Some Money](<https://devfeed.tech/articles/amd-cannot-catch-nvidia-but-it-sure-can-make-some-money-118817.md>)

Original publisher: [Read original article](<https://www.nextplatform.com/compute/2026/11/04/amd-cannot-catch-nvidia-but-it-sure-can-make-some-money/5307666>)

Author: Timothy Prickett Morgan

Published: 2026-11-04T20:31:15Z

Content type: opinion

Language: en

Sources: [The Next Platform: In-depth coverage of high end computing](<https://devfeed.tech/sources/the-next-platform-in-depth-coverage-of-high-end-computing.md>)

Topics: [Finance](<https://devfeed.tech/topics/finance.md>), [Nvidia](<https://devfeed.tech/topics/nvidia.md>), [rocm](<https://devfeed.tech/topics/rocm.md>)

Tags: [2021](<https://devfeed.tech/tags/2021.md>), [2023](<https://devfeed.tech/tags/2023.md>), [2025](<https://devfeed.tech/tags/2025.md>), [2026](<https://devfeed.tech/tags/2026.md>), [about](<https://devfeed.tech/tags/about.md>), [academic](<https://devfeed.tech/tags/academic.md>), [accelerators](<https://devfeed.tech/tags/accelerators.md>), [acquisition](<https://devfeed.tech/tags/acquisition.md>), [added](<https://devfeed.tech/tags/added.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [amd](<https://devfeed.tech/tags/amd.md>), [but](<https://devfeed.tech/tags/but.md>), [can](<https://devfeed.tech/tags/can.md>), [competition](<https://devfeed.tech/tags/competition.md>), [compute](<https://devfeed.tech/tags/compute.md>), [cpu](<https://devfeed.tech/tags/cpu.md>), [epyc](<https://devfeed.tech/tags/epyc.md>), [function](<https://devfeed.tech/tags/function.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [helios](<https://devfeed.tech/tags/helios.md>), [i](<https://devfeed.tech/tags/i.md>), [instinct](<https://devfeed.tech/tags/instinct.md>), [make](<https://devfeed.tech/tags/make.md>), [money](<https://devfeed.tech/tags/money.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [requires](<https://devfeed.tech/tags/requires.md>)

### AI overview

AMD's datacenter business is growing quickly and delivering strong profits, but limited high bandwidth memory (HBM) allocations constrain Instinct GPU sales, leaving Nvidia's lead wider. The excerpt estimates AMD's Q3 2026 revenue at $13.38 billion, up 44.7%, with Data Center revenue of $8.55 billion, up 96.9%; its CPU sales drove most of that segment's revenue, while modeled Instinct GPU sales were $3.43 billion. The author attributes AMD's gains against Intel in server CPUs to product design and its ability to scale manufacturing and packaging. AMD says it secured more supply and expects server revenue to double year over year in 2027, with datacenter exceeding 80% of total revenue. The author argues AMD can profit from the undersupplied AI market despite Nvidia's much larger scale, and proposes composable, reusable memory architectures as a possible way to ease resource constraints; the piece presents this as a speculative idea, not a planned product.

### Source excerpt

Capitalism requires strong competition to function properly, and because of that, I have always been ...

## "AMD Loves Omarchy" Linux With More Collaboration Coming

DevFeed: ["AMD Loves Omarchy" Linux With More Collaboration Coming](<https://devfeed.tech/articles/amd-loves-omarchy-linux-with-more-collaboration-coming-116095.md>)

Original publisher: [Read original article](<https://www.phoronix.com/news/AMD-Loves-Omarchy>)

Author: Michael Larabel

Published: 2026-11-04T10:50:00Z

Content type: news

Language: en

Sources: [Phoronix](<https://devfeed.tech/sources/phoronix.md>)

Topics: [Linux](<https://devfeed.tech/topics/linux.md>), [rocm](<https://devfeed.tech/topics/rocm.md>)

Tags: [amd](<https://devfeed.tech/tags/amd.md>), [collaboration](<https://devfeed.tech/tags/collaboration.md>), [desktop-linux](<https://devfeed.tech/tags/desktop-linux.md>), [linux](<https://devfeed.tech/tags/linux.md>), [linux-benchmarking](<https://devfeed.tech/tags/linux-benchmarking.md>), [linux-hardware-benchmarks](<https://devfeed.tech/tags/linux-hardware-benchmarks.md>), [linux-hardware-reviews](<https://devfeed.tech/tags/linux-hardware-reviews.md>), [linux-how-to](<https://devfeed.tech/tags/linux-how-to.md>), [linux-performance](<https://devfeed.tech/tags/linux-performance.md>), [linux-server-benchmarks](<https://devfeed.tech/tags/linux-server-benchmarks.md>), [omarchy](<https://devfeed.tech/tags/omarchy.md>), [open-source-graphics](<https://devfeed.tech/tags/open-source-graphics.md>), [phoronix](<https://devfeed.tech/tags/phoronix.md>), [phoronix-test-suite](<https://devfeed.tech/tags/phoronix-test-suite.md>), [ubuntu-benchmarks](<https://devfeed.tech/tags/ubuntu-benchmarks.md>), [ubuntu-hardware](<https://devfeed.tech/tags/ubuntu-hardware.md>)

### AI overview

AMD VP of AI Software Anush Elangovan has praised Omarchy and met its lead, David Heinemeier Hansson, at the Fireworks Forge event, where he supplied him with an AMD-powered HP ZBook. Elangovan said "AMD ❤ Omarchy. Great things are ahead," but the excerpt does not describe specific plans for future collaboration.

### Source excerpt

In addition to organizations like Alibaba, SpaceXSI, DigitalOcean, Meta SuperIntelligence Labs, and others contributing millions of dollars toward the development of the Omarchy Linux distribution, it turns out AMD also has a thing for this Linux distribution built around the Hyprland compositor...

## ROCm InferaSim: Cut LLM Serving Trial Costs and Deploy the Best Configuration

DevFeed: [ROCm InferaSim: Cut LLM Serving Trial Costs and Deploy the Best Configuration](<https://devfeed.tech/articles/rocm-inferasim-cut-llm-serving-trial-costs-and-deploy-the-best-configuration-116480.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.04032>)

Author: Anshu Raina, Yao Fu, Zhang Jiejing, Peyman Razaghi, Shiksha Patel, Ravi Gupta, Deval Shah, Adeem Jassani, Subhajit Dutta Chowdhury, Lorri Rao, Kailash Gogineni, Zhenyu Gu, Onil Gunawardana, Emad Barso

Published: 2026-11-04T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Distributed Computing](<https://devfeed.tech/sources/arxiv-distributed-computing.md>)

Topics: [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>), [rocm](<https://devfeed.tech/topics/rocm.md>), [LLM evaluation / benchmarking](<https://devfeed.tech/topics/llm-evaluation-benchmarking.md>), [vllm](<https://devfeed.tech/topics/vllm.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [amd](<https://devfeed.tech/tags/amd.md>), [amd-instinct-mi355x](<https://devfeed.tech/tags/amd-instinct-mi355x.md>), [anchor](<https://devfeed.tech/tags/anchor.md>), [atom](<https://devfeed.tech/tags/atom.md>), [batch](<https://devfeed.tech/tags/batch.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [best](<https://devfeed.tech/tags/best.md>), [cache](<https://devfeed.tech/tags/cache.md>), [coding](<https://devfeed.tech/tags/coding.md>), [configuration](<https://devfeed.tech/tags/configuration.md>), [costs](<https://devfeed.tech/tags/costs.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-dc](<https://devfeed.tech/tags/cs-dc.md>), [decode](<https://devfeed.tech/tags/decode.md>), [decoding](<https://devfeed.tech/tags/decoding.md>), [deploy](<https://devfeed.tech/tags/deploy.md>), [deployment](<https://devfeed.tech/tags/deployment.md>), [llm](<https://devfeed.tech/tags/llm.md>), [llm-serving](<https://devfeed.tech/tags/llm-serving.md>), [rocm](<https://devfeed.tech/tags/rocm.md>), [serving](<https://devfeed.tech/tags/serving.md>)

### AI overview

ROCm InferaSim aims to reduce LLM serving configuration trials by benchmarking each distinct kernel set once as an "anchor," then analytically projecting performance to deployments that use the same kernels. For multi-node setups, it maps a configuration to a single-node equivalent before scaling the estimate; a simulator replays multi-turn traffic, and a tuning agent searches using the predictions. In tests on AMD Instinct MI355X GPUs with vLLM, SGLang, and ATOM, 11 anchors used 9.1 GPU-hours versus 204 GPU-hours to measure 226 fixed-length configurations. Reported median error for time per output token was 8.1% across 172 configurations; throughput was within 8% on a multi-turn coding workload and within 5.5% for 159 disaggregated deployments that were not anchors. A five-candidate shortlist included a deployment within 12% of the best feasible model and latency-target pairing. These results are limited to the reported setups and workloads.

### Source excerpt

arXiv:2611.04032v1 Announce Type: cross Abstract: Serving an LLM efficiently depends on the engine, how the model is split across GPUs, precision, scheduler settings, speculative decoding, KV-cache management, and prefill-decode disaggregation. The best configuration depends on the model, hardware, workload, and objective. Operators usually find it by trial and error. Exhaustively measuring the candidates in our study would take about 25,000 GPU-hours. ROCm InferaSim predicts serving performance from a few short measurements. Choices such as batch size, prompt length, and GPU count change the amount of work per GPU without changing the kernels. ROCm InferaSim benchmarks each distinct set of kernels once, on at most one node. We call this measurement an anchor. Analytical laws carry that anchor to other deployments that run the same kernels. For multi-node deployments, ROCm InferaSim reduces the configuration to a single-node equivalent, measures it, and scales the result analytically. A simulator replays multi-turn traffic, and a tuning agent searches configurations using the predictions. On AMD Instinct MI355X GPUs with vLLM, SGLang, and ATOM, 11 anchors took 9.1 GPU-hours, against 204 GPU-hours to measure the 226 fixed-length configurations they cover. On 172 configurations, time per output token has 8.1% median error. On a multi-turn coding workload, throughput is within 8%. For 159 disaggregated deployments, none used as an anchor, throughput is within 5.5%. A five-candidate short list contains a deployment within 12% of the best feasible pairing of model and latency target.

## AMD Instinct MI350P First Look: 144GB of HBM3E in a 600W PCIe Card

DevFeed: [AMD Instinct MI350P First Look: 144GB of HBM3E in a 600W PCIe Card](<https://devfeed.tech/articles/amd-instinct-mi350p-first-look-144gb-of-hbm3e-in-a-600w-pcie-card-116363.md>)

Original publisher: [Read original article](<https://www.storagereview.com/review/amd-instinct-mi350p-first-look>)

Author: Brian Beeler

Published: 2026-11-03T17:07:17Z

Content type: article

Language: en

Sources: [StorageReview.com](<https://devfeed.tech/sources/storagereview-com.md>)

Topics: [pcie](<https://devfeed.tech/topics/pcie.md>), [Multi-GPU](<https://devfeed.tech/topics/multi-gpu.md>), [rocm](<https://devfeed.tech/topics/rocm.md>), [vllm](<https://devfeed.tech/topics/vllm.md>), [Hopper](<https://devfeed.tech/topics/hopper.md>), [NVLink](<https://devfeed.tech/topics/nvlink.md>), [ddr5](<https://devfeed.tech/topics/ddr5.md>)

Tags: [2022](<https://devfeed.tech/tags/2022.md>), [ai](<https://devfeed.tech/tags/ai.md>), [all](<https://devfeed.tech/tags/all.md>), [amd](<https://devfeed.tech/tags/amd.md>), [amd-instinct-mi350p](<https://devfeed.tech/tags/amd-instinct-mi350p.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [bandwidth](<https://devfeed.tech/tags/bandwidth.md>), [card](<https://devfeed.tech/tags/card.md>), [company](<https://devfeed.tech/tags/company.md>), [enterprise](<https://devfeed.tech/tags/enterprise.md>), [first](<https://devfeed.tech/tags/first.md>), [first-look](<https://devfeed.tech/tags/first-look.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [hbm3e](<https://devfeed.tech/tags/hbm3e.md>), [instinct](<https://devfeed.tech/tags/instinct.md>), [look](<https://devfeed.tech/tags/look.md>), [memory](<https://devfeed.tech/tags/memory.md>), [memory-bandwidth](<https://devfeed.tech/tags/memory-bandwidth.md>), [mxfp4](<https://devfeed.tech/tags/mxfp4.md>), [pcie](<https://devfeed.tech/tags/pcie.md>), [performance](<https://devfeed.tech/tags/performance.md>), [up](<https://devfeed.tech/tags/up.md>)

### AI overview

The MI350P brings 144GB of HBM3E and 4TB/s bandwidth to a dual-slot PCIe card, and its 450W tests show a strong fit for models that benefit from memory capacity and bandwidth. In matmul tests it outperformed the 600W RTX PRO 6000 Blackwell by 36-38% across tested precisions, while the H200 NVL led it in BF16 and FP8. For chat-sized inference, MI350P beat the RTX PRO 6000 on tested FP8, BF16, and GPT-OSS models; long-prompt results were mixed, and H200 generally led on dense models. Its 144GB capacity lets some models run on one card where the RTX PRO 6000 needs two. Key qualifications: the review unit ran at 450W, four large-model results used ROCm 7.2.3, and the authors say full-power 600W testing remains outstanding.

### Source excerpt

The AMD Instinct MI350P is AMD's first new PCIe Instinct GPU in four years, following the MI210 in 2022. It brings the company's CDNA 4 architecture to a conventional add-in card, with up to 4.6 PFLOPS of MXFP4 performance, 144GB of HBM3E, and 4TB/s of memory bandwidth. All of that fits into a dual-slot, 600W The post AMD Instinct MI350P First Look: 144GB of HBM3E in a 600W PCIe Card appeared first on StorageReview.com.

## Rack-scale AI: How Dell and AMD are building infrastructure

DevFeed: [Rack-scale AI: How Dell and AMD are building infrastructure](<https://devfeed.tech/articles/rack-scale-ai-how-dell-and-amd-are-building-infrastructure-112886.md>)

Original publisher: [Read original article](<https://siliconangle.com/2026/11/02/rack-scale-ai-platforms-amd-dell-dellaidataplatform/>)

Author: Mark Albertson

Published: 2026-11-02T23:33:24Z

Content type: article

Language: en

Sources: [SiliconANGLE](<https://devfeed.tech/sources/siliconangle.md>)

Topics: [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [dell](<https://devfeed.tech/topics/dell.md>), [rocm](<https://devfeed.tech/topics/rocm.md>), [on-prem](<https://devfeed.tech/topics/on-prem.md>), [Advanced AI workflows](<https://devfeed.tech/topics/advanced-ai-workflows.md>)

Tags: [agentic](<https://devfeed.tech/tags/agentic.md>), [agentic-ai](<https://devfeed.tech/tags/agentic-ai.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-infrastructure](<https://devfeed.tech/tags/ai-infrastructure.md>), [ai-platform](<https://devfeed.tech/tags/ai-platform.md>), [amd](<https://devfeed.tech/tags/amd.md>), [amd-epyc](<https://devfeed.tech/tags/amd-epyc.md>), [amd-epyc-processors](<https://devfeed.tech/tags/amd-epyc-processors.md>), [amd-instinct-mi350p](<https://devfeed.tech/tags/amd-instinct-mi350p.md>), [amd-instinct-mi355x](<https://devfeed.tech/tags/amd-instinct-mi355x.md>), [amd-rocm-software](<https://devfeed.tech/tags/amd-rocm-software.md>), [bandwidth](<https://devfeed.tech/tags/bandwidth.md>), [cube-event-coverage](<https://devfeed.tech/tags/cube-event-coverage.md>), [dave-vellante](<https://devfeed.tech/tags/dave-vellante.md>), [dell](<https://devfeed.tech/tags/dell.md>), [dell-ai-data-platform-event-from-ambition-to-ai-at-scale-2026](<https://devfeed.tech/tags/dell-ai-data-platform-event-from-ambition-to-ai-at-scale-2026.md>), [dell-ai-platform-with-amd](<https://devfeed.tech/tags/dell-ai-platform-with-amd.md>), [dell-poweredge-r7725](<https://devfeed.tech/tags/dell-poweredge-r7725.md>), [dell-poweredge-xe7745](<https://devfeed.tech/tags/dell-poweredge-xe7745.md>), [dell-technologies](<https://devfeed.tech/tags/dell-technologies.md>), [dellaidataplatform](<https://devfeed.tech/tags/dellaidataplatform.md>), [graphics-processing-units](<https://devfeed.tech/tags/graphics-processing-units.md>), [hbm3e-memory](<https://devfeed.tech/tags/hbm3e-memory.md>), [infrastructure](<https://devfeed.tech/tags/infrastructure.md>), [memory](<https://devfeed.tech/tags/memory.md>), [mike-darby](<https://devfeed.tech/tags/mike-darby.md>), [network](<https://devfeed.tech/tags/network.md>), [news](<https://devfeed.tech/tags/news.md>), [otel-ai-initiative](<https://devfeed.tech/tags/otel-ai-initiative.md>), [partner](<https://devfeed.tech/tags/partner.md>), [pcie](<https://devfeed.tech/tags/pcie.md>), [poweredge](<https://devfeed.tech/tags/poweredge.md>), [poweredge-xe9785](<https://devfeed.tech/tags/poweredge-xe9785.md>), [powerscale](<https://devfeed.tech/tags/powerscale.md>), [powerswitch](<https://devfeed.tech/tags/powerswitch.md>), [pradyumna-narendra](<https://devfeed.tech/tags/pradyumna-narendra.md>), [rack-scale](<https://devfeed.tech/tags/rack-scale.md>), [robert-hormuth](<https://devfeed.tech/tags/robert-hormuth.md>), [rocm](<https://devfeed.tech/tags/rocm.md>), [server-consolidation](<https://devfeed.tech/tags/server-consolidation.md>), [sovereign-ai](<https://devfeed.tech/tags/sovereign-ai.md>), [telecom](<https://devfeed.tech/tags/telecom.md>), [thecube](<https://devfeed.tech/tags/thecube.md>), [university-of-cambridge](<https://devfeed.tech/tags/university-of-cambridge.md>)

### AI overview

Dell and AMD are assembling enterprise AI infrastructure around server consolidation, modular expansion and a range of GPU deployments. Dell says AMD-powered PowerEdge systems can deliver 70% generational performance gains and consolidate up to 13 prior-generation servers into one. The Dell AI Platform with AMD combines Instinct accelerators, EPYC processors, PowerEdge servers, PowerSwitch networking, PowerScale storage and ROCm; customers can start with smaller configurations and scale on the same validated framework. Dell's support for the PCIe-based MI350P in two PowerEdge models adds 144 GB of HBM3E memory for inference and an air-cooled option for data centers not ready for liquid cooling. The article also cites Cambridge's Zenith AI supercomputer and an AT&T telecom initiative using Dell servers with MI355X GPUs. These are reported capabilities and deployments; the excerpt does not provide independent comparative performance results.

### Source excerpt

Dell Technologies Inc. and its partner Advanced Micro Devices Inc. face similar competitive pressures. Both tech powerhouses operate in a business climate where success is now measured across central processing units, graphics processing units, rack-scale systems, memory capacity, network bandwidth and tokens per dollar. The pressure to compete across those measures underscores what theCUBE Research [...] The post Rack-scale AI: How Dell and AMD are building infrastructure appeared first on SiliconANGLE.

## AMD Helios Racks Are Running at Meta, With a Custom Instinct GPU and a $1 Billion Stake to Follow

DevFeed: [AMD Helios Racks Are Running at Meta, With a Custom Instinct GPU and a $1 Billion Stake to Follow](<https://devfeed.tech/articles/amd-helios-racks-are-running-at-meta-with-a-custom-instinct-gpu-and-a-1-billion-stake-to-follow-106549.md>)

Original publisher: [Read original article](<https://www.storagereview.com/news/amd-helios-racks-running-at-meta-custom-instinct-gpu-6th-gen-epyc-muse>)

Author: Harold Fritts

Published: 2026-10-29T16:41:34Z

Content type: news

Language: en

Sources: [StorageReview.com](<https://devfeed.tech/sources/storagereview-com.md>)

Topics: [rocm](<https://devfeed.tech/topics/rocm.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [amd](<https://devfeed.tech/tags/amd.md>), [cpus](<https://devfeed.tech/tags/cpus.md>), [custom](<https://devfeed.tech/tags/custom.md>), [deployment](<https://devfeed.tech/tags/deployment.md>), [enterprise](<https://devfeed.tech/tags/enterprise.md>), [epyc](<https://devfeed.tech/tags/epyc.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [gpus](<https://devfeed.tech/tags/gpus.md>), [helios](<https://devfeed.tech/tags/helios.md>), [instinct](<https://devfeed.tech/tags/instinct.md>), [meta](<https://devfeed.tech/tags/meta.md>), [muse](<https://devfeed.tech/tags/muse.md>), [networking](<https://devfeed.tech/tags/networking.md>), [pensando](<https://devfeed.tech/tags/pensando.md>), [personal-ai](<https://devfeed.tech/tags/personal-ai.md>), [personal-ai-agent](<https://devfeed.tech/tags/personal-ai-agent.md>), [rackscale](<https://devfeed.tech/tags/rackscale.md>), [roadmaps](<https://devfeed.tech/tags/roadmaps.md>)

### AI overview

AMD says its first Helios rackscale systems are installed and under test at Meta, with a broader deployment planned for the near future; it gives no rack count, site, or deployment date. The partnership spans EPYC CPUs, custom Instinct GPUs, Pensando networking, and ROCm. Meta, which has deployed millions of EPYC processors and is a lead 6th Gen EPYC customer, is also collaborating with AMD on future CPUs including Verano. Meta will deploy a custom Instinct GPU optimized for its workloads; joint software work has optimized recommendation training on MI350X GPUs, with improvements contributed to PyTorch, Triton, and ROCm. AMD also says it provides Day-0 support for Muse Spark Open on Instinct GPUs and Muse Glimmer on Ryzen AI PCs and Radeon GPUs. AMD committed $1 billion to Meta's equity offering.

### Source excerpt

AMD says the first AMD Helios rackscale systems are installed and under test at Meta, with a broader deployment to follow, and that the two companies have aligned product roadmaps around Meta's workloads across EPYC CPUs, Instinct GPUs, Pensando networking, and ROCm. The same announcement puts EPYC behind Meta's Muse personal AI agent, adds Day-0 The post AMD Helios Racks Are Running at Meta, With a Custom Instinct GPU and a $1 Billion Stake to Follow appeared first on StorageReview.com.

## AMD's ROCm HRX Loom v0.1 Released For Targeting AMD CPUs, GPUs & NPUs

DevFeed: [AMD's ROCm HRX Loom v0.1 Released For Targeting AMD CPUs, GPUs & NPUs](<https://devfeed.tech/articles/amd-s-rocm-hrx-loom-v0-1-released-for-targeting-amd-cpus-gpus-npus-106142.md>)

Original publisher: [Read original article](<https://www.phoronix.com/news/AMD-ROCm-HRX-Loom-0.1>)

Author: Michael Larabel

Published: 2026-10-29T10:22:15Z

Content type: release

Language: en

Sources: [Phoronix](<https://devfeed.tech/sources/phoronix.md>)

Topics: [rocm](<https://devfeed.tech/topics/rocm.md>), [GPU optimization](<https://devfeed.tech/topics/gpu-optimization.md>), [SDK](<https://devfeed.tech/topics/sdk.md>)

Tags: [amd](<https://devfeed.tech/tags/amd.md>), [apis](<https://devfeed.tech/tags/apis.md>), [assets](<https://devfeed.tech/tags/assets.md>), [asynchronous](<https://devfeed.tech/tags/asynchronous.md>), [command-line](<https://devfeed.tech/tags/command-line.md>), [compiler](<https://devfeed.tech/tags/compiler.md>), [configuration](<https://devfeed.tech/tags/configuration.md>), [cpus](<https://devfeed.tech/tags/cpus.md>), [desktop-linux](<https://devfeed.tech/tags/desktop-linux.md>), [documentation](<https://devfeed.tech/tags/documentation.md>), [execution](<https://devfeed.tech/tags/execution.md>), [github](<https://devfeed.tech/tags/github.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [gpus](<https://devfeed.tech/tags/gpus.md>), [guide](<https://devfeed.tech/tags/guide.md>), [hip](<https://devfeed.tech/tags/hip.md>), [http](<https://devfeed.tech/tags/http.md>), [linux-benchmarking](<https://devfeed.tech/tags/linux-benchmarking.md>), [linux-hardware-benchmarks](<https://devfeed.tech/tags/linux-hardware-benchmarks.md>), [linux-hardware-reviews](<https://devfeed.tech/tags/linux-hardware-reviews.md>), [linux-how-to](<https://devfeed.tech/tags/linux-how-to.md>), [linux-performance](<https://devfeed.tech/tags/linux-performance.md>), [linux-server-benchmarks](<https://devfeed.tech/tags/linux-server-benchmarks.md>), [loom](<https://devfeed.tech/tags/loom.md>), [open-source-graphics](<https://devfeed.tech/tags/open-source-graphics.md>), [phoronix](<https://devfeed.tech/tags/phoronix.md>), [phoronix-test-suite](<https://devfeed.tech/tags/phoronix-test-suite.md>), [release](<https://devfeed.tech/tags/release.md>), [rocm](<https://devfeed.tech/tags/rocm.md>), [sdk](<https://devfeed.tech/tags/sdk.md>), [ubuntu-benchmarks](<https://devfeed.tech/tags/ubuntu-benchmarks.md>), [ubuntu-hardware](<https://devfeed.tech/tags/ubuntu-hardware.md>)

### AI overview

Loom v0.1 is the first tagged release of an open-source compiler and intermediate-representation stack in AMD's ROCm HRX. It compiles kernels and asynchronous device programs, links model programs, and specializes them for AMD CPUs, GPUs, and NPUs. The SDK includes compiler, runtime, Python APIs, and command-line tools; loom-fetch installs selected model programs and assets, while loom-serve loads them and serves HTTP requests. GPU execution can target AMDGPU HSA or portable SPIR-V/Vulkan; x86 and WebAssembly support is still maturing. The excerpt says the SDK is available through GitHub and as the hrx-loom Python package.

### Source excerpt

At this year's AMD Advancing AI event Hyperloom was announced among the ROCm improvements to expect. Hyperloom aims for agentic workload optimizations and more with quote powerful abilities if everything works out right. As part of the latest round of ROCm advancements, the Loom SDK v0.1 release was released today...

## uGDS: A User-Space Development Kit for High-Performance GPU Direct Storage

DevFeed: [uGDS: A User-Space Development Kit for High-Performance GPU Direct Storage](<https://devfeed.tech/articles/ugds-a-user-space-development-kit-for-high-performance-gpu-direct-storage-97308.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.30868>)

Author: Jing Wei, Guanyi Chen, Qi Chen, Chuyu Zhang, Zheng Zhou, Chutian Wang, Jian Zhang

Published: 2026-10-23T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Distributed Computing](<https://devfeed.tech/sources/arxiv-distributed-computing.md>)

Topics: [GPUDirect](<https://devfeed.tech/topics/gpudirect.md>), [NVMe](<https://devfeed.tech/topics/nvme.md>), [dma](<https://devfeed.tech/topics/dma.md>), [rocm](<https://devfeed.tech/topics/rocm.md>), [gpu-initated](<https://devfeed.tech/topics/gpu-initated.md>)

Tags: [amd](<https://devfeed.tech/tags/amd.md>), [apis](<https://devfeed.tech/tags/apis.md>), [bandwidth](<https://devfeed.tech/tags/bandwidth.md>), [batch](<https://devfeed.tech/tags/batch.md>), [buffer](<https://devfeed.tech/tags/buffer.md>), [command](<https://devfeed.tech/tags/command.md>), [completion](<https://devfeed.tech/tags/completion.md>), [compute](<https://devfeed.tech/tags/compute.md>), [control](<https://devfeed.tech/tags/control.md>), [cpu](<https://devfeed.tech/tags/cpu.md>), [cross](<https://devfeed.tech/tags/cross.md>), [cs-dc](<https://devfeed.tech/tags/cs-dc.md>), [cs-os](<https://devfeed.tech/tags/cs-os.md>), [cuda](<https://devfeed.tech/tags/cuda.md>), [custom](<https://devfeed.tech/tags/custom.md>), [data](<https://devfeed.tech/tags/data.md>), [development-kit](<https://devfeed.tech/tags/development-kit.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [high-performance](<https://devfeed.tech/tags/high-performance.md>), [performance](<https://devfeed.tech/tags/performance.md>), [storage](<https://devfeed.tech/tags/storage.md>), [user](<https://devfeed.tech/tags/user.md>)

### AI overview

uGDS is an open-source GPU Direct Storage library that moves NVMe command management into user space while transferring payloads by PCIe peer-to-peer DMA; a minimal kernel module handles one-time privileged setup. It offers cuFile-compatible synchronous, batched, and stream-ordered APIs for CUDA and HIP, with polling or interrupt completion. Its queue partitioning targets synchronous and batch traffic, and it enforces DMA-safe buffer lifetimes. In the reported A100/Samsung 990 PRO tests, uGDS reached up to 12.12x GDS batch-read and 30.69x batch-write bandwidth at specified small request sizes and batch size 128, and cut 4-KiB synchronous-write latency by 84.1% with two threads. The abstract also reports lower LMCache median time-to-first-token and higher serving goodput, plus up to 1.19x hipFile synchronous-write bandwidth on AMD; these are evaluation-specific results, not general guarantees.

### Source excerpt

arXiv:2610.30868v1 Announce Type: new Abstract: Data-intensive GPU applications increasingly rely on NVMe SSDs to extend GPU memory. NVIDIA GPU Direct Storage (GDS) enables direct SSD-GPU transfers via PCIe peer-to-peer DMA, eliminating host staging. However, GDS retains three key limitations: (1) its control path traverses the kernel for every I/O request, introducing software overhead; (2) its stream-ordered path imposes strict total ordering even when workloads only require batch-level ordering; and (3) its closed-source user-space engine hinders diagnosis and custom scheduling. We present uGDS, https://github.com/ScaleX-IO/uGDS, an open-source user-space GPU Direct Storage library built on four principles: user-space control, unified cross-vendor abstraction, workload-aware scheduling, and end-to-end transparency. The host CPU manages the entire NVMe command lifecycle in user space, while payloads move via PCIe P2P DMA without consuming GPU compute. A minimal kernel module (ugds_drv.ko) is used only for one-time privileged setup. uGDS partitions hardware queues between synchronous and batch traffic, enforces DMA-safe buffer lifetimes, and provides cuFile-compatible synchronous, batched, and stream-ordered APIs with polling or interrupt completion on both NVIDIA CUDA and AMD HIP. Evaluated on an A100 with a Samsung 990 PRO, uGDS achieves up to 12.12x GDS's batch-read bandwidth and 30.69x its batch-write bandwidth (8- and 4-KiB requests, batch size 128), and reduces 4-KiB synchronous write latency by 84.1% with two threads. Its stream-ordered path sustains 5.49 GiB/s on a single stream, where GDS retains only 40.8% of its synchronous bandwidth. In LMCache, uGDS reduces median time-to-first-token (TTFT) by up to 62.7% and raises serving goodput by 2.63x. On AMD, uGDS achieves up to 1.19x hipFile's synchronous-write bandwidth.

## vLLM Delivers Day-0 Support for MiniMax-M3.1-Flash

DevFeed: [vLLM Delivers Day-0 Support for MiniMax-M3.1-Flash](<https://devfeed.tech/articles/vllm-delivers-day-0-support-for-minimax-m3-1-flash-96723.md>)

Original publisher: [Read original article](<https://vllm.ai/blog/2026-10-23-minimax-m31-flash>)

Author: Inferact and the vLLM Team

Published: 2026-10-23T00:00:00Z

Content type: release

Language: en

Sources: [vLLM Blog](<https://devfeed.tech/sources/vllm-blog.md>)

Topics: [vllm](<https://devfeed.tech/topics/vllm.md>), [Model Optimizer](<https://devfeed.tech/topics/model-optimizer.md>), [NVFP4](<https://devfeed.tech/topics/nvfp4.md>), [quantization](<https://devfeed.tech/topics/quantization.md>), [rocm](<https://devfeed.tech/topics/rocm.md>), [NVIDIA Vera Rubin](<https://devfeed.tech/topics/nvidia-vera-rubin.md>)

Tags: [ai-engineering](<https://devfeed.tech/tags/ai-engineering.md>), [amd](<https://devfeed.tech/tags/amd.md>), [blackwell](<https://devfeed.tech/tags/blackwell.md>), [day-0-support](<https://devfeed.tech/tags/day-0-support.md>), [decoding](<https://devfeed.tech/tags/decoding.md>), [dspark](<https://devfeed.tech/tags/dspark.md>), [flash](<https://devfeed.tech/tags/flash.md>), [gpus](<https://devfeed.tech/tags/gpus.md>), [hardware](<https://devfeed.tech/tags/hardware.md>), [kernels](<https://devfeed.tech/tags/kernels.md>), [llm-inference](<https://devfeed.tech/tags/llm-inference.md>), [m3](<https://devfeed.tech/tags/m3.md>), [m3-1-flash](<https://devfeed.tech/tags/m3-1-flash.md>), [minimax](<https://devfeed.tech/tags/minimax.md>), [minimax-m3-1-flash](<https://devfeed.tech/tags/minimax-m3-1-flash.md>), [model-serving](<https://devfeed.tech/tags/model-serving.md>), [nvfp4](<https://devfeed.tech/tags/nvfp4.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [nvidia-blackwell](<https://devfeed.tech/tags/nvidia-blackwell.md>), [nvidia-rubin](<https://devfeed.tech/tags/nvidia-rubin.md>), [open-source-ai](<https://devfeed.tech/tags/open-source-ai.md>), [performance](<https://devfeed.tech/tags/performance.md>), [rl](<https://devfeed.tech/tags/rl.md>), [series](<https://devfeed.tech/tags/series.md>), [support](<https://devfeed.tech/tags/support.md>), [vllm](<https://devfeed.tech/tags/vllm.md>), [vllm-blog](<https://devfeed.tech/tags/vllm-blog.md>)

### AI overview

vLLM provides day-0 MiniMax-M3.1-Flash support on NVIDIA Blackwell and Rubin and AMD MI350-series GPUs, with NVFP4 KV cache and MoE support, DSpark speculative decoding, and RL training from the NVFP4 checkpoint. The model's sparse attention selects at most 16 blocks of 128 tokens per query, limiting attention to 2,048 keys; the article estimates long-context decode HBM traffic at about 12.5% of dense attention. On four B300s, NVFP4 reduces checkpoint size 44.6% and KV-cache bytes per token 43.75%; at 256K context, the stated deployment fits 76 sequences versus 28 for the prior model's FP8 cache. Hardware-specific optimizations include CUDA 13.4 QMUL4 kernels on Blackwell, distinct MSA kernels on Rubin, and table-based NVFP4 conversion, context parallelism, and fused low-latency decoding on AMD. Reported AgentX results vary by platform and metric: Rubin reaches 724.9 tokens/s per user at the latency-sensitive frontier, while AMD MI355X reaches 133.3M total tokens per dollar at the cost-efficient frontier. Day-0 RL uses veRL, FSDP2, rank-32 LoRA, and token-level importance sampling; a 16-GB300, 200-step run raises AIME-2024 accuracy from 36.3% to 67.5%. Prebuilt NVIDIA and AMD vLLM-

### Source excerpt

Day-0 vLLM support for MiniMax-M3.1-Flash on NVIDIA Blackwell, NVIDIA Rubin, and AMD MI350-series GPUs, with NVFP4 MSA kernels, NVFP4 MegaMoE, DSpark speculative decoding, AgentX results, and day-0 RL.

## Accelerating PyTorch on AMD GPUs with FlyDSL

DevFeed: [Accelerating PyTorch on AMD GPUs with FlyDSL](<https://devfeed.tech/articles/accelerating-pytorch-on-amd-gpus-with-flydsl-93059.md>)

Original publisher: [Read original article](<https://pytorch.org/blog/accelerating-pytorch-on-amd-gpus-with-flydsl/>)

Author: AMD: Xiaobing Zhang, Yutao Xu, Rui Li, Heng Yi, Xinya Zhang, Shijie Feng, Felix Li, Jeff Daily, Jiong Gong, Hattie Wu, Peng Sun, Emad Barsoum; META: Jason Ansel and Simon Layton

Published: 2026-10-21T17:45:40Z

Content type: release

Language: en

Sources: [PyTorch](<https://devfeed.tech/sources/blog-pytorch.md>)

Topics: [PyTorch](<https://devfeed.tech/topics/pytorch.md>), [GPU optimization](<https://devfeed.tech/topics/gpu-optimization.md>), [rocm](<https://devfeed.tech/topics/rocm.md>), [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>), [quantization](<https://devfeed.tech/topics/quantization.md>)

Tags: [amd](<https://devfeed.tech/tags/amd.md>), [apis](<https://devfeed.tech/tags/apis.md>), [attention](<https://devfeed.tech/tags/attention.md>), [blog](<https://devfeed.tech/tags/blog.md>), [gpus](<https://devfeed.tech/tags/gpus.md>), [mxfp4](<https://devfeed.tech/tags/mxfp4.md>), [pytorch](<https://devfeed.tech/tags/pytorch.md>)

### AI overview

On MI350-series GPUs (`gfx950`), PyTorch can use FlyDSL for supported GEMMs, RMSNorm, TopK, and attention through existing APIs. `torch.compile` benchmarks eligible GEMM candidates against ATen and Triton and caches a winner; supported eager RMSNorm and TopK calls dispatch automatically when the optional FlyDSL runtime is installed and enabled. Attention runs through AOTriton, which selects among FlyDSL, Triton, and AITER kernels. On MI355X, reported operator-suite geometric-mean gains include 1.10x for dense GEMM, 1.58x-1.68x for MXFP GEMM, 1.20x over Triton for grouped GEMM, and up to 4.82x for small-K TopK over ATen with determinism disabled. End-to-end vLLM results vary by model and concurrency; low-precision tests show larger gains than BF16, whose largest model remains within measured noise. Support has limits: MXFP requires specific quantized operands and layouts, attention does not support FP32, and unsupported workloads retain other PyTorch backends. The excerpt describes current MI350 support; broader hardware and operation coverage are future work.

### Source excerpt

PyTorch users on AMD MI350-series GPUs can now use FlyDSL through existing operator APIs for dense and grouped GEMM, fused multi-head attention (FMHA/Attention), MXFP8/MXFP4 scaled GEMM, RMSNorm, and TopK, without... The post Accelerating PyTorch on AMD GPUs with FlyDSL appeared first on PyTorch.

[Next page](<https://devfeed.tech/topics/rocm.md?cursor=WyIyMDI2LTEwLTIxVDE3OjQ1OjQwKzAwOjAwIiwgIjUzMzg3MzkxLWMyNzAtNDJhYS04MjRhLWI3YTgwY2ZkNjU1MCJd>)