# NUMA

Non-uniform memory access (NUMA) is a multiprocessor architecture in which memory is divided among nodes and access time depends on the CPU and memory locations; software can optimize scheduling and memory usage for this topology.

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

## Redox OS Gets Golang Working, GIMP Ported & Simplified AI Policy

DevFeed: [Redox OS Gets Golang Working, GIMP Ported & Simplified AI Policy](<https://devfeed.tech/articles/redox-os-gets-golang-working-gimp-ported-simplified-ai-policy-134048.md>)

Original publisher: [Read original article](<https://www.phoronix.com/news/Redox-OS-Golang-GIMP>)

Author: Michael Larabel

Published: 2026-11-14T16:06:43Z

Content type: news

Language: en

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

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

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-policy](<https://devfeed.tech/tags/ai-policy.md>), [desktop-linux](<https://devfeed.tech/tags/desktop-linux.md>), [gimp](<https://devfeed.tech/tags/gimp.md>), [golang](<https://devfeed.tech/tags/golang.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>), [numa](<https://devfeed.tech/tags/numa.md>), [open-source-graphics](<https://devfeed.tech/tags/open-source-graphics.md>), [os](<https://devfeed.tech/tags/os.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

Redox OS's October update reports that Go now works for both cross-compilation and self-hosted compilation, adapted to use Relibc, and that GIMP runs after months of porting work. The update also notes a Servo 0.7 port update, initial NUMA support, a RISC-V dynamic-linking fix, and initial rootless package installation in user-equivalent paths. Redox OS simplified its contribution policy to reject contributions generated by large language models. The excerpt mentions further driver and kernel improvements but does not describe them.

### Source excerpt

The Rust-based Redox OS open-source operating system is now out with its October development summary that highlights recent software ports, progress on NUMA, and other enhancements made to this from-scratch platform...

## Starscream: Reimagining LLM Sharding on AMD Instinct GPUs with GPU Partitioning

DevFeed: [Starscream: Reimagining LLM Sharding on AMD Instinct GPUs with GPU Partitioning](<https://devfeed.tech/articles/starscream-reimagining-llm-sharding-on-amd-instinct-gpus-with-gpu-partitioning-131539.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.15951>)

Author: Sumanth Gudaparthi, Smeet Chheda, Ganesh Dasika

Published: 2026-11-13T05:00:00Z

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: [sharding](<https://devfeed.tech/topics/sharding.md>), [GPU optimization](<https://devfeed.tech/topics/gpu-optimization.md>), [Distributed Systems & Parallel Computing](<https://devfeed.tech/topics/distributed-systems-parallel-computing.md>), [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>), [NUMA](<https://devfeed.tech/topics/numa.md>), [Multi-GPU](<https://devfeed.tech/topics/multi-gpu.md>)

Tags: [amd](<https://devfeed.tech/tags/amd.md>), [attention](<https://devfeed.tech/tags/attention.md>), [chiplet](<https://devfeed.tech/tags/chiplet.md>), [communication](<https://devfeed.tech/tags/communication.md>), [cs-ar](<https://devfeed.tech/tags/cs-ar.md>), [cs-dc](<https://devfeed.tech/tags/cs-dc.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [gpu-partitioning](<https://devfeed.tech/tags/gpu-partitioning.md>), [instinct](<https://devfeed.tech/tags/instinct.md>), [kernel](<https://devfeed.tech/tags/kernel.md>), [llm](<https://devfeed.tech/tags/llm.md>), [locality](<https://devfeed.tech/tags/locality.md>), [memory](<https://devfeed.tech/tags/memory.md>), [numa](<https://devfeed.tech/tags/numa.md>), [parallelism](<https://devfeed.tech/tags/parallelism.md>), [performance](<https://devfeed.tech/tags/performance.md>), [redundancy](<https://devfeed.tech/tags/redundancy.md>), [sharding](<https://devfeed.tech/tags/sharding.md>), [strategies](<https://devfeed.tech/tags/strategies.md>), [tensor-parallelism](<https://devfeed.tech/tags/tensor-parallelism.md>)

### AI overview

Starscream adapts LLM sharding to partitioned, chiplet-based GPUs by adding intra-GPU sharding along an orthogonal dimension instead of extending inter-GPU sharding across the partitions. It targets three problems in conventional approaches: redundant memory use, reduced arithmetic intensity, and communication overhead from flat collectives. Its design principles are to preserve kernel shapes, eliminate memory redundancy, and minimize communication. In a tensor-parallelism case study on AMD MI300X and MI355X, partition-aware designs reportedly achieved up to 4x speedup for attention and 1.43x for fused GEMMs versus chiplet-oblivious baselines. The supplied abstract claims the principles can extend to other sharding strategies and future architectures; it does not provide broader evaluation details.

### Source excerpt

arXiv:2611.15951v1 Announce Type: new Abstract: As GPU accelerators grow larger to meet rising compute demands, they exhibit increasingly pronounced non-uniform memory access (NUMA) characteristics that challenge existing LLM frameworks. Chiplet-based architectures provide a practical path to larger GPUs while mitigating manufacturing yield constraints. Modern GPUs such as AMD Instinct MI300X employs such a multi-die design, and further offer GPU partitioning capabilities, partitioning a single physical GPU into multiple logical devices with NUMA-aware memory domains, promising faster kernel launches, and reduced memory access latencies with improved locality. However, current parallelization strategies, designed for homogeneous GPU topologies, when applied under partitioning, extend its unidimensional sharding causing significant performance degradation. Three bottlenecks emerge: (1) memory redundancy; (2) arithmetic intensity collapse; (3) communication overhead from flat collectives. We present Starscream, a rethinking of LLM sharding for the chiplet era. Rather than stretching inter-GPU sharding thinner, Starscream adds intra-GPU sharding along an orthogonal dimension, grounded in three foundational principles: kernel shape preservation, memory redundancy elimination, and communication overhead minimization. These principles form the basis for extending any sharding strategy into the partitioning regime. We demonstrate their effectiveness using tensor parallelism as a case study. On AMD MI300X and MI355X, exploiting partitioning capabilities yields up to 4x speedup for attention and 1.43x for fused GEMMs compared to chiplet-oblivious baselines, demonstrating the effectiveness of chiplet-aware designs. Starscream's principles generalize across sharding strategies and future architectures.

## AMD EPYC Venice SP7 Review: 256 Cores Per Socket and the Bandwidth to Feed Them

DevFeed: [AMD EPYC Venice SP7 Review: 256 Cores Per Socket and the Bandwidth to Feed Them](<https://devfeed.tech/articles/amd-epyc-venice-sp7-review-256-cores-per-socket-and-the-bandwidth-to-feed-them-127758.md>)

Original publisher: [Read original article](<https://www.storagereview.com/review/amd-epyc-venice-sp7-review>)

Author: StorageReview Enterprise Lab

Published: 2026-11-11T00:57:44Z

Content type: article

Language: en

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

Topics: [NUMA](<https://devfeed.tech/topics/numa.md>), [ddr5](<https://devfeed.tech/topics/ddr5.md>), [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [High-Performance Computing](<https://devfeed.tech/topics/high-performance-computing.md>)

Tags: [amd](<https://devfeed.tech/tags/amd.md>), [amd-epyc](<https://devfeed.tech/tags/amd-epyc.md>), [bandwidth](<https://devfeed.tech/tags/bandwidth.md>), [clock](<https://devfeed.tech/tags/clock.md>), [core](<https://devfeed.tech/tags/core.md>), [counts](<https://devfeed.tech/tags/counts.md>), [cpu](<https://devfeed.tech/tags/cpu.md>), [enterprise](<https://devfeed.tech/tags/enterprise.md>), [epyc](<https://devfeed.tech/tags/epyc.md>), [family](<https://devfeed.tech/tags/family.md>), [memory](<https://devfeed.tech/tags/memory.md>), [review](<https://devfeed.tech/tags/review.md>), [server](<https://devfeed.tech/tags/server.md>), [socket](<https://devfeed.tech/tags/socket.md>), [venice](<https://devfeed.tech/tags/venice.md>)

### AI overview

The tested SP7 EPYC 9996 delivers strong compute throughput, while results show that memory bandwidth, locality and workload scaling shape its advantages. In Dell's two-socket R9825, two 256-core processors beat the tested 384-core EPYC 9965 system by 1.8x in GROMACS benchPEP and 2.6x on a large OpenFOAM mesh; AES-256-GCM throughput reached 6.8 TB/s. Moving the same processors from DDR5-8000 to MRDIMM-12800 raised STREAM Triad bandwidth 58% and cut that mesh solve time 21%. Venice's raw STREAM result trailed NVIDIA Vera's by 42% on DDR5, narrowing to 8% with MRDIMMs, while measured remote-socket bandwidth was only about a quarter of local bandwidth, underscoring the value of NUMA-aware placement. Database results were mixed by engine: the 9996 led Vera on MySQL, tied on VillageSQL, and trailed on MariaDB and PostgreSQL, but had 30%-51% lower PAYMENT p99 latency at matched load. Open-source database engines did not use all 256 cores, limiting peak throughput.

### Source excerpt

AMD's 6th Gen EPYC server CPU family, codenamed Venice, brings Zen 6 and Zen 6c cores to a lineup built to take on every corner of the server market. AMD is pushing core counts and clock speeds higher and giving those cores considerably more bandwidth to work with, through a larger memory subsystem and the The post AMD EPYC Venice SP7 Review: 256 Cores Per Socket and the Bandwidth to Feed Them appeared first on StorageReview.com.

## Dell PowerEdge R9825 Review: 512 Venice Cores in an Air-Cooled 3U Server

DevFeed: [Dell PowerEdge R9825 Review: 512 Venice Cores in an Air-Cooled 3U Server](<https://devfeed.tech/articles/dell-poweredge-r9825-review-512-venice-cores-in-an-air-cooled-3u-server-127585.md>)

Original publisher: [Read original article](<https://www.storagereview.com/review/dell-poweredge-r9825-review>)

Author: Brian Beeler

Published: 2026-11-10T22:13:29Z

Content type: article

Language: en

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

Topics: [systems](<https://devfeed.tech/topics/systems.md>), [dell](<https://devfeed.tech/topics/dell.md>), [ddr5](<https://devfeed.tech/topics/ddr5.md>), [valkey](<https://devfeed.tech/topics/valkey.md>), [NUMA](<https://devfeed.tech/topics/numa.md>), [Parallelism](<https://devfeed.tech/topics/parallelism.md>), [Processor Affinity](<https://devfeed.tech/topics/processor-affinity.md>)

Tags: [7-zip](<https://devfeed.tech/tags/7-zip.md>), [aes](<https://devfeed.tech/tags/aes.md>), [airflow](<https://devfeed.tech/tags/airflow.md>), [amd](<https://devfeed.tech/tags/amd.md>), [amd-epyc](<https://devfeed.tech/tags/amd-epyc.md>), [analytics](<https://devfeed.tech/tags/analytics.md>), [api](<https://devfeed.tech/tags/api.md>), [arm](<https://devfeed.tech/tags/arm.md>), [attestation](<https://devfeed.tech/tags/attestation.md>), [ddr5](<https://devfeed.tech/tags/ddr5.md>), [dell](<https://devfeed.tech/tags/dell.md>), [dell-poweredge](<https://devfeed.tech/tags/dell-poweredge.md>), [enterprise](<https://devfeed.tech/tags/enterprise.md>), [epyc](<https://devfeed.tech/tags/epyc.md>), [memory](<https://devfeed.tech/tags/memory.md>), [processors](<https://devfeed.tech/tags/processors.md>), [review](<https://devfeed.tech/tags/review.md>), [server](<https://devfeed.tech/tags/server.md>), [socket](<https://devfeed.tech/tags/socket.md>), [threads](<https://devfeed.tech/tags/threads.md>), [venice](<https://devfeed.tech/tags/venice.md>)

### AI overview

The Dell PowerEdge R9825 fits two 600W EPYC 9996 processors (512 Zen 6c cores total) into an air-cooled 3U chassis, using stacked processor boards, shaped front intake and five hot-swap fans. It supports up to 8TB of 8,000MT/s RDIMMs or up to 4TB of 12,800MT/s MRDIMMs; testing on an AMD reference platform found about 1.9TB/s STREAM Triad bandwidth with MRDIMMs versus 1.2TB/s on the review unit, making memory choice consequential for bandwidth-bound work. The server offers PCIe Gen6 slots and up to 32 E3.S Gen6 bays, though 32-bay capacity was not available at launch. In measured workloads it substantially outperformed the prior 2U R7725 on GROMACS, OpenFOAM and multithreaded compression, while 3U yields fewer cores per rack unit. The excerpt also reports lower results in thread-count-sensitive JVM and scheduler tests, and weaker remote-socket bandwidth than NVIDIA Vera. Dell supports air cooling up to 35°C inlet across listed processors; higher temperatures restrict memory and GPU configurations, and at 45°C the listed processors require liquid cooling.

### Source excerpt

The Dell PowerEdge R9825 is the first server built on AMD's 6th Gen EPYC to reach our lab, and it is the dual-socket flagship of Dell's 18th generation PowerEdge line. Two EPYC 9996 processors give it 512 Zen 6c cores and 1,024 threads, 32 DDR5 slots take it to 8TB of memory, and the front The post Dell PowerEdge R9825 Review: 512 Venice Cores in an Air-Cooled 3U Server appeared first on StorageReview.com.

## Multi-GPU MoE Inference with Adaptive Expert Offloading

DevFeed: [Multi-GPU MoE Inference with Adaptive Expert Offloading](<https://devfeed.tech/articles/multi-gpu-moe-inference-with-adaptive-expert-offloading-125897.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.10219>)

Author: William X. Zheng, Zhongjie Chen, Shuo Yang, Ziming Mao, Xiaoze Fan, Yi Pan, Kunjun Li, Hanfei Yu, Chenfeng Xu, Yang Zhou, Rishabh Iyer, Ion Stoica

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

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

Topics: [Mixture of Experts (MoE)](<https://devfeed.tech/topics/mixture-of-experts-moe.md>), [Multi-GPU](<https://devfeed.tech/topics/multi-gpu.md>), [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>), [NUMA](<https://devfeed.tech/topics/numa.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [adaptive](<https://devfeed.tech/tags/adaptive.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [batch](<https://devfeed.tech/tags/batch.md>), [capability](<https://devfeed.tech/tags/capability.md>), [co-design](<https://devfeed.tech/tags/co-design.md>), [constraint](<https://devfeed.tech/tags/constraint.md>), [cpu](<https://devfeed.tech/tags/cpu.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [decode](<https://devfeed.tech/tags/decode.md>), [demand](<https://devfeed.tech/tags/demand.md>), [design](<https://devfeed.tech/tags/design.md>), [effectively](<https://devfeed.tech/tags/effectively.md>), [execution](<https://devfeed.tech/tags/execution.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [inference](<https://devfeed.tech/tags/inference.md>), [moe](<https://devfeed.tech/tags/moe.md>), [multi-gpu](<https://devfeed.tech/tags/multi-gpu.md>), [offloading](<https://devfeed.tech/tags/offloading.md>), [with](<https://devfeed.tech/tags/with.md>)

### AI overview

AdaOffload targets MoE inference's GPU-memory limits by offloading some expert weights to host memory and executing them on demand. It coordinates GPU-CPU parallelism and co-execution for both decode and prefill, using online workload observation to adapt to hardware, workload, and batch-size differences. The abstract reports evaluation across five platforms with different GPU, CPU, and NUMA configurations, using three open-weight models ranging from 284B to 753B parameters. Across models, AdaOffload achieved 1.9x-15.5x throughput speedups over state-of-the-art systems.

### Source excerpt

arXiv:2611.10219v1 Announce Type: new Abstract: Modern Mixture-of-Experts (MoE) inference is constrained by GPU memory. This constraint is mitigated by expert offloading, which stores part of the model's weight in host memory and executes on demand. Though MoE architecture has computation sparsity, more GPUs are needed for serving increasingly larger frontier models. However, existing expert offloading systems overlook this need and fail to scale performance with additional GPUs. We present AdaOffload, a general and efficient multi-GPU hybrid expert execution framework that effectively handles heterogeneous hardware and workloads. We co-design GPU-CPU parallelism and GPU-CPU co-execution for decode and prefill with online workload observation that adapts to different hardware, workload, and batch size heterogeneity. We evaluated AdaOffload on five different platforms with varying GPU capability (L4, Pro6000, B200), number of GPUs, CPU capability, and NUMA topology on three different frontier open-weight models (284B to 753B). AdaOffload achieves $1.9\times$ to $15.5\times$ throughput speedups across different models over state-of-the-art systems.

## NUMA-Torch: Compiler-Driven NUMA-Aware LLM Prefill on Multi-Socket CPUs

DevFeed: [NUMA-Torch: Compiler-Driven NUMA-Aware LLM Prefill on Multi-Socket CPUs](<https://devfeed.tech/articles/numa-torch-compiler-driven-numa-aware-llm-prefill-on-multi-socket-cpus-121261.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.07105>)

Author: Thisara Rathnayaka Weerakoon (Department of Computer Science & Engineering, University of Moratuwa, Sri Lanka), Ishan Dewpura (Department of Computer Science & Engineering, University of Moratuwa, Sri

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

Content type: article

Language: en

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

Topics: [NUMA](<https://devfeed.tech/topics/numa.md>), [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [PyTorch](<https://devfeed.tech/topics/pytorch.md>)

Tags: [affinity](<https://devfeed.tech/tags/affinity.md>), [attention](<https://devfeed.tech/tags/attention.md>), [bandwidth](<https://devfeed.tech/tags/bandwidth.md>), [banks](<https://devfeed.tech/tags/banks.md>), [bert](<https://devfeed.tech/tags/bert.md>), [compilation](<https://devfeed.tech/tags/compilation.md>), [compiler](<https://devfeed.tech/tags/compiler.md>), [compute](<https://devfeed.tech/tags/compute.md>), [cpu](<https://devfeed.tech/tags/cpu.md>), [cpu-affinity](<https://devfeed.tech/tags/cpu-affinity.md>), [cpus](<https://devfeed.tech/tags/cpus.md>), [cs-dc](<https://devfeed.tech/tags/cs-dc.md>), [cs-pf](<https://devfeed.tech/tags/cs-pf.md>), [cs-pl](<https://devfeed.tech/tags/cs-pl.md>), [data](<https://devfeed.tech/tags/data.md>), [data-locality](<https://devfeed.tech/tags/data-locality.md>), [dram](<https://devfeed.tech/tags/dram.md>), [floating-point](<https://devfeed.tech/tags/floating-point.md>), [llm](<https://devfeed.tech/tags/llm.md>), [numa](<https://devfeed.tech/tags/numa.md>), [prefill](<https://devfeed.tech/tags/prefill.md>), [socket](<https://devfeed.tech/tags/socket.md>), [torch](<https://devfeed.tech/tags/torch.md>)

### AI overview

NUMA-Torch targets multi-socket CPU LLM prefill bottlenecks caused by weights being accessed across NUMA sockets, which reduces effective DRAM bandwidth. Its NUMA-aware Tensor Parallelism treats each socket as a partition, shards weights across sockets, and uses CPU affinity to keep thread pools and memory allocations local. The abstract reports evaluations on GPT-2, BERT, and Gemma 3 270M, with speedups of up to 2.63x on dual-socket Haswell, 1.87x on dual-socket Cascade Lake, and 2.27x on four-socket Haswell-EX. These are reported maxima; the excerpt does not provide per-model results or broader evaluation details.

### Source excerpt

arXiv:2611.07105v1 Announce Type: new Abstract: For sufficiently large transformer-based Large Language Models (LLMs), whose parameters cannot be retained in on-chip caches, prefill throughput on multi-socket CPUs is often limited by sustainable DRAM bandwidth rather than peak floating-point throughput. In the standard PyTorch compilation (i.e., TorchInductor) path, model weights are allocated without considering the Non-Uniform Memory Access (NUMA) topology: specifically, the affinity between memory banks and compute cores is disregarded. This causes a substantial fraction of weight accesses to traverse the inter-socket interconnect, severely degrading the effective memory bandwidth available to each socket. We address this bottleneck through a compiler-level solution that makes TorchInductor NUMA-aware. Treating each NUMA socket as an independent tensor-parallel partition, we introduce NUMA-aware Tensor Parallelism (NUMA-TP). This approach transparently shards weights across sockets and schedules compute kernels such that thread pools and memory allocations remain strictly local to their respective sockets using CPU affinity. We evaluate the approach on GPT-2 (117M, 345M, 762M) and BERT (Small, Large), which exercise standard multi-head attention and MLP, as well as Gemma 3 270M, which combines Multi-Query Attention with a GeGLU-gated MLP. By strictly enforcing data locality, our approach successfully mitigates cross-socket traffic, achieving prefill speedups of up to 2.63x on a dual-socket Haswell platform, up to 1.87x on a dual-socket Cascade Lake platform, and up to 2.27x on a four-socket Haswell-EX platform.

## Idle Cycles Are Not Wasted Cycles

DevFeed: [Idle Cycles Are Not Wasted Cycles](<https://devfeed.tech/articles/idle-cycles-are-not-wasted-cycles-111763.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2607.16408>)

Author: Matheus Stolet, Simon Peter, Antoine Kaufmann

Published: 2026-11-02T05:00:00Z

Content type: opinion

Language: en

Sources: [arXiv: Networking](<https://devfeed.tech/sources/arxiv-networking.md>)

Topics: [scheduling](<https://devfeed.tech/topics/scheduling.md>), [NUMA](<https://devfeed.tech/topics/numa.md>)

Tags: [amd-epyc](<https://devfeed.tech/tags/amd-epyc.md>), [compute](<https://devfeed.tech/tags/compute.md>), [cs-ar](<https://devfeed.tech/tags/cs-ar.md>), [cs-ni](<https://devfeed.tech/tags/cs-ni.md>), [cs-os](<https://devfeed.tech/tags/cs-os.md>), [power](<https://devfeed.tech/tags/power.md>), [scheduling](<https://devfeed.tech/tags/scheduling.md>), [service](<https://devfeed.tech/tags/service.md>), [systems-design](<https://devfeed.tech/tags/systems-design.md>), [waiting](<https://devfeed.tech/tags/waiting.md>)

### AI overview

The paper argues that idle service cores can return capacity to productive work when active cores are constrained by a shared package power and thermal budget. Measurements on a recent AMD EPYC processor indicate that the value of reclaiming idle cores depends on waiting strategy, processor topology, and idle duration; reclamation may deliver less benefit than commonly assumed while adding scheduling complexity. The authors propose treating power budget, rather than core occupancy, as the central resource and making waiting policy a first-class design choice.

### Source excerpt

arXiv:2607.16408v2 Announce Type: replace-cross Abstract: Idle network service cores are treated as wasted compute. This assumption motivates increasingly sophisticated mechanisms that reclaim idle cores at microsecond timescales. We argue that this view no longer matches modern server hardware. On contemporary multicore processors, active cores compete for a shared package-level power and thermal budget. Once that budget becomes the limiting resource, an idle core that waits efficiently returns compute capacity that hardware can redistribute to productive work. Measurements on a recent AMD EPYC processor show how waiting strategy, processor topology, and idle duration determine this trade-off. Our results suggest that reclaiming idle cores often yields less benefit than commonly assumed while introducing substantial scheduling complexity. We propose a budget-centric view of service-core systems in which power, rather than core occupancy, becomes the fundamental resource and waiting policy becomes a first-class systems design choice.

## NVIDIA Vera Single vs. Dual Socket CPU Performance

DevFeed: [NVIDIA Vera Single vs. Dual Socket CPU Performance](<https://devfeed.tech/articles/nvidia-vera-single-vs-dual-socket-cpu-performance-106354.md>)

Original publisher: [Read original article](<https://www.phoronix.com/review/nvidia-vera-1p-2p>)

Author: Michael Larabel

Published: 2026-10-29T13:36:24Z

Content type: article

Language: en

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

Topics: [NVIDIA Vera](<https://devfeed.tech/topics/nvidia-vera.md>), [NUMA](<https://devfeed.tech/topics/numa.md>)

Tags: [amd-epyc](<https://devfeed.tech/tags/amd-epyc.md>), [benchmarking](<https://devfeed.tech/tags/benchmarking.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [cpu](<https://devfeed.tech/tags/cpu.md>), [cpu-performance](<https://devfeed.tech/tags/cpu-performance.md>), [desktop-linux](<https://devfeed.tech/tags/desktop-linux.md>), [hpc](<https://devfeed.tech/tags/hpc.md>), [intel-xeon](<https://devfeed.tech/tags/intel-xeon.md>), [kernel](<https://devfeed.tech/tags/kernel.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>), [numa](<https://devfeed.tech/tags/numa.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [nvidia-vera](<https://devfeed.tech/tags/nvidia-vera.md>), [nvidia-vera-cpu](<https://devfeed.tech/tags/nvidia-vera-cpu.md>), [open-source-graphics](<https://devfeed.tech/tags/open-source-graphics.md>), [performance](<https://devfeed.tech/tags/performance.md>), [phoronix](<https://devfeed.tech/tags/phoronix.md>), [phoronix-test-suite](<https://devfeed.tech/tags/phoronix-test-suite.md>), [power-and-performance](<https://devfeed.tech/tags/power-and-performance.md>), [scaling](<https://devfeed.tech/tags/scaling.md>), [smt](<https://devfeed.tech/tags/smt.md>), [socket](<https://devfeed.tech/tags/socket.md>), [thread](<https://devfeed.tech/tags/thread.md>), [ubuntu-benchmarks](<https://devfeed.tech/tags/ubuntu-benchmarks.md>), [ubuntu-hardware](<https://devfeed.tech/tags/ubuntu-hardware.md>)

### AI overview

The article reports standalone benchmarks comparing power and performance for one- and two-socket NVIDIA Vera systems across workloads described as scalable to high core and thread counts and NUMA-friendly. A Vera CPU has 88 Olympus cores and SMT for 176 threads; the dual-socket setup is listed as 172 cores and 352 threads. Tests ran remotely on Ubuntu 24.04 LTS with a Linux 6.17 64K kernel. The results are not compared with AMD EPYC or Intel Xeon, since the author lacked matching 88-core products and time was limited.

### Source excerpt

For those curious how NVIDIA Vera is scaling between single and dual socket deployments, here are some benchmarks I carried out a while back exploring the power and performance from one to two Vera CPUs.

## pgCluu v5.0 has been released

DevFeed: [pgCluu v5.0 has been released](<https://devfeed.tech/articles/pgcluu-v5-0-has-been-released-95074.md>)

Original publisher: [Read original article](<https://www.postgresql.org/about/news/pgcluu-v50-has-been-released-3417/>)

Published: 2026-10-22T00:00:00Z

Content type: release

Language: en

Sources: [PostgreSQL news](<https://devfeed.tech/sources/postgresql-news.md>)

Topics: [Tool](<https://devfeed.tech/topics/tool.md>), [IO](<https://devfeed.tech/topics/io.md>), [NUMA](<https://devfeed.tech/topics/numa.md>), [Open Source](<https://devfeed.tech/topics/open-source.md>)

Tags: [2026](<https://devfeed.tech/tags/2026.md>), [audit](<https://devfeed.tech/tags/audit.md>), [collector](<https://devfeed.tech/tags/collector.md>), [hba](<https://devfeed.tech/tags/hba.md>), [i-o](<https://devfeed.tech/tags/i-o.md>), [kernel](<https://devfeed.tech/tags/kernel.md>), [limits](<https://devfeed.tech/tags/limits.md>), [performances](<https://devfeed.tech/tags/performances.md>), [perl](<https://devfeed.tech/tags/perl.md>), [postgresql](<https://devfeed.tech/tags/postgresql.md>), [replication](<https://devfeed.tech/tags/replication.md>), [reports](<https://devfeed.tech/tags/reports.md>)

### AI overview

The excerpt describes pgCluu 6.0, despite the title naming v5.0. pgCluu collects PostgreSQL and system statistics with psql and sar, then builds HTML reports. This release adds reports for replication slots and PostgreSQL I/O, including throughput, latency, WAL, temporary relations, shared-buffer activity, and asynchronous I/O on PostgreSQL 18+. It expands kernel data collection and reports postmaster process limits, cgroup limits, pressure stall information, and kernel memory events. I/O statistics use pg_stat_io on PostgreSQL 16+, while pressure and cgroup details depend on kernel or cgroup versions noted in the excerpt. Wait-event reports now include only events exceeding 0.1% of observed wait time.

### Source excerpt

Antananarivo, Madagascar - October 20, 2026 pgCluu PostgreSQL Cluster Utilization pgCluu is a Perl program used to perform a full audit of a PostgreSQL Cluster performances. It is divided in two parts, a collector used to grab statistics on the PostgreSQL server using psql and sar, a reports builder that will generate all HTML and charts output. Version 6.0 of pgCluu have been released, this is a major release that adds several new features and reports. It also fixes issues reported since the last release. New features: Add report about replication slots retrieved from pg_replication_slots. Add collect of pg_hba_file_rules Limit wait event reporting to those that represent more than 0.1 percent of the total of the wait event observed. Add collect of I/O statistics from pg_stat_io (PG16+) Add reports about: I/O throughput per backend type shared buffers writers and evictions buffer hit ratio and ring buffers I/O latency and size WAL I/O temporary relations I/O asynchronous I/O (PG18+) Add collect more kernel tuning parameters, remove shmmax and shmall vm.dirty_expire_centisecs and vm.dirty_writeback_centisecs, they were not matched by the sysctl filter, and all other vm.dirty_* parameters, vm.min_free_kbytes, vm.watermark_scale_factor and kernel.numa_balancing, net.core.somaxconn, net.ipv4.ip_local_port_range, net.core.rmem_max, net.core.wmem_max, fs.file-max and fs.nr_open, /sys/kernel/mm/transparent_hugepage/shmem_enabled: PostgreSQL shared memory is anonymous shared memory, this is the setting that applies to the shared buffers, not 'enabled', the frequency governor of all CPUs instead of cpu0 only, the cpuidle driver, the intel_pstate status and energy performance preference, and the number of NUMA nodes, read_ahead_kb, nr_requests, max_sectors_kb and wbt_lat_usec of each block device. Collect and report the limits of the postmaster process. pgcluu_collectd now saves in a new POSTMASTER section of sysinfo.txt: the pid of the postmaster, found as the parent of the

## Weight Placement Is Not Enough: Synchronization in Multi-Node CPU Decoding of Ternary LLMs

DevFeed: [Weight Placement Is Not Enough: Synchronization in Multi-Node CPU Decoding of Ternary LLMs](<https://devfeed.tech/articles/weight-placement-is-not-enough-synchronization-in-multi-node-cpu-decoding-of-ternary-llms-93757.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.28620>)

Author: Runguo Li, Dezhen Lu

Published: 2026-10-21T04:00:00Z

Content type: news

Language: en

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

Topics: [synchronization](<https://devfeed.tech/topics/synchronization.md>), [NUMA](<https://devfeed.tech/topics/numa.md>), [CPU Cache](<https://devfeed.tech/topics/cpu-cache.md>)

Tags: [bandwidth](<https://devfeed.tech/tags/bandwidth.md>), [column](<https://devfeed.tech/tags/column.md>), [cpu](<https://devfeed.tech/tags/cpu.md>), [cross](<https://devfeed.tech/tags/cross.md>), [cs-dc](<https://devfeed.tech/tags/cs-dc.md>), [decoding](<https://devfeed.tech/tags/decoding.md>), [dram](<https://devfeed.tech/tags/dram.md>), [energy](<https://devfeed.tech/tags/energy.md>), [kernels](<https://devfeed.tech/tags/kernels.md>), [l3-cache](<https://devfeed.tech/tags/l3-cache.md>), [latency](<https://devfeed.tech/tags/latency.md>), [llms](<https://devfeed.tech/tags/llms.md>), [local](<https://devfeed.tech/tags/local.md>), [megatron](<https://devfeed.tech/tags/megatron.md>), [model](<https://devfeed.tech/tags/model.md>), [numa](<https://devfeed.tech/tags/numa.md>), [split](<https://devfeed.tech/tags/split.md>), [synchronization](<https://devfeed.tech/tags/synchronization.md>), [thread](<https://devfeed.tech/tags/thread.md>), [token](<https://devfeed.tech/tags/token.md>), [weight](<https://devfeed.tech/tags/weight.md>)

### AI overview

On dual-socket, four-NUMA-node Xeon servers, placing ternary LLM weights with the threads that read them does not prevent substantial decode-time synchronization: threads wait at synchronization points for 20-48% of each token. Traces attribute most of that time to slower threads within each node, while barrier latency and serial per-node work contribute less. With identical kernels, a Megatron-style column split outperformed a row-owned split without cross-node reductions in every measured configuration; most of its gain came from reduced waiting, rather than fewer barriers. The resulting engine was 1.5-2.1x faster than ik_llama.cpp using all cores and 1.5-1.9x faster than its fastest thread-count setting for DRAM-bound decoding, with the same accuracy and less energy per token. A small L3-resident model saw larger speedups. Findings are limited to the measured configurations and two Xeon generations.

### Source excerpt

arXiv:2610.28620v1 Announce Type: new Abstract: Ternary LLMs read only a few gigabytes per decoded token, yet on dual-socket servers with four NUMA nodes the fastest ternary runtime we found uses less than half of the memory bandwidth. Placing each weight on the node of the thread that reads it is necessary but not enough: in a decode engine with node-local weights that reuses ik_llama.cpp's kernels, threads on two generations of Xeon servers still spend 20-48% of each token waiting at synchronization points. Per-thread barrier traces show that most of the wait is for the slowest threads of each node; the barriers' own latency (microseconds each, but up to 12% of the token for the smallest model) and serial per-node work are smaller parts. On identical kernels, a Megatron-style column split beats a row-owned split without cross-node reductions in every configuration we measured, and isolating the number of barriers shows that most of its gain comes from its threads waiting less for each other and for their node leaders, and only a minority from having fewer barriers. Where decoding is DRAM-bound, the resulting engine is 1.5-2.1x faster than ik_llama.cpp with all cores and 1.5-1.9x faster than ik_llama.cpp at its fastest thread count (3.2x and 2.8x for a small model that runs from the L3 cache), at the same accuracy and with less energy per token.

## Nvidia Vera's System Architecture: Pushing Bandwidth

DevFeed: [Nvidia Vera's System Architecture: Pushing Bandwidth](<https://devfeed.tech/articles/nvidia-vera-s-system-architecture-pushing-bandwidth-90277.md>)

Original publisher: [Read original article](<https://chipsandcheese.com/p/nvidia-veras-system-architecture>)

Author: Chester Lam

Published: 2026-10-19T18:26:50Z

Content type: article

Language: en

Sources: [Chips and Cheese](<https://devfeed.tech/sources/chips-and-cheese.md>)

Topics: [NVIDIA Vera](<https://devfeed.tech/topics/nvidia-vera.md>), [systems](<https://devfeed.tech/topics/systems.md>), [NUMA](<https://devfeed.tech/topics/numa.md>)

Tags: [architecture](<https://devfeed.tech/tags/architecture.md>), [bandwidth](<https://devfeed.tech/tags/bandwidth.md>), [cache](<https://devfeed.tech/tags/cache.md>), [nvidia-vera](<https://devfeed.tech/tags/nvidia-vera.md>), [performance](<https://devfeed.tech/tags/performance.md>), [server](<https://devfeed.tech/tags/server.md>)

### AI overview

The article examines NVIDIA Vera's server CPU interconnect, memory subsystem, caches, and dual-socket performance. Measurements show strong memory bandwidth and controlled local memory latency under load, but high cross-socket and core-to-core latency. Benchmark comparisons illustrate its tradeoff between strong per-core performance and lower core counts: Vera excels in some memory-intensive workloads, while Graviton 5 leads in more cache-friendly workloads.

### Source excerpt

Nvidia's Vera server chip aims for high per-core performance.

## Microsoft Maia 300 Puts 288GB of HBM4 at 18TB/s on an N2P Chiplet and 96 Accelerators in a 167kW Rack

DevFeed: [Microsoft Maia 300 Puts 288GB of HBM4 at 18TB/s on an N2P Chiplet and 96 Accelerators in a 167kW Rack](<https://devfeed.tech/articles/microsoft-maia-300-puts-288gb-of-hbm4-at-18tb-s-on-an-n2p-chiplet-and-96-accelerators-in-a-167kw-rack-85445.md>)

Original publisher: [Read original article](<https://www.storagereview.com/news/microsoft-maia-300-hbm4-n2p-chiplet-96-accelerator-rack>)

Author: Brian Beeler

Published: 2026-10-12T19:45:29Z

Content type: news

Language: en

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

Topics: [NVIDIA Vera Rubin](<https://devfeed.tech/topics/nvidia-vera-rubin.md>), [Microsoft](<https://devfeed.tech/topics/microsoft.md>), [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [NUMA](<https://devfeed.tech/topics/numa.md>), [Beta channel](<https://devfeed.tech/topics/beta-channel.md>)

Tags: [400g](<https://devfeed.tech/tags/400g.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [ai](<https://devfeed.tech/tags/ai.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [bandwidth](<https://devfeed.tech/tags/bandwidth.md>), [chiplet](<https://devfeed.tech/tags/chiplet.md>), [chiplets](<https://devfeed.tech/tags/chiplets.md>), [cloud](<https://devfeed.tech/tags/cloud.md>), [cluster](<https://devfeed.tech/tags/cluster.md>), [clusters](<https://devfeed.tech/tags/clusters.md>), [compute](<https://devfeed.tech/tags/compute.md>), [configuration](<https://devfeed.tech/tags/configuration.md>), [data](<https://devfeed.tech/tags/data.md>), [deployments](<https://devfeed.tech/tags/deployments.md>), [efficiency](<https://devfeed.tech/tags/efficiency.md>), [enterprise](<https://devfeed.tech/tags/enterprise.md>), [inference](<https://devfeed.tech/tags/inference.md>), [pairs](<https://devfeed.tech/tags/pairs.md>), [tsmc](<https://devfeed.tech/tags/tsmc.md>)

### AI overview

Microsoft announced Maia 300, its third-generation in-house inference accelerator, built from multiple chiplets with eight HBM4 stacks providing 288GB of memory and 18TB/s of bandwidth. A rack holds 96 accelerators and draws 167kW. The article reports Microsoft's performance and efficiency claims while noting that the comparison lacks details about the GPU, model, and serving configuration, and cannot be independently tested outside Microsoft.

### Source excerpt

Microsoft unveiled Azure Maia 300 today at the OCP Global Summit, the third generation of its in-house inference accelerator and the first built as a multi-chiplet part. Each accelerator pairs a TSMC N2P compute die with eight HBM4 stacks for 288GB of memory at 18TB/s inside a 950W TDP, and a Maia 300 rack holds The post Microsoft Maia 300 Puts 288GB of HBM4 at 18TB/s on an N2P Chiplet and 96 Accelerators in a 167kW Rack appeared first on StorageReview.com.

## How Modern Computer Memory Uses Caches, Virtual Memory, and Compressed RAM

DevFeed: [How Modern Computer Memory Uses Caches, Virtual Memory, and Compressed RAM](<https://devfeed.tech/articles/your-computer-memory-is-a-lie-85335.md>)

Original publisher: [Read original article](<https://hackaday.com/2026/10/12/your-computer-memory-is-a-lie/>)

Author: Al Williams

Published: 2026-10-12T14:00:11Z

Content type: article

Language: en

Sources: [Hackaday](<https://devfeed.tech/sources/hackaday.md>)

Topics: [CPU Cache](<https://devfeed.tech/topics/cpu-cache.md>), [NUMA](<https://devfeed.tech/topics/numa.md>), [DRAM price increase](<https://devfeed.tech/topics/dram-price-increase.md>), [upgrade](<https://devfeed.tech/topics/upgrade.md>)

Tags: [architecture](<https://devfeed.tech/tags/architecture.md>), [buffer](<https://devfeed.tech/tags/buffer.md>), [cache](<https://devfeed.tech/tags/cache.md>), [compression](<https://devfeed.tech/tags/compression.md>), [computing](<https://devfeed.tech/tags/computing.md>), [cpu](<https://devfeed.tech/tags/cpu.md>), [cram](<https://devfeed.tech/tags/cram.md>), [cxl](<https://devfeed.tech/tags/cxl.md>), [dram](<https://devfeed.tech/tags/dram.md>), [featured](<https://devfeed.tech/tags/featured.md>), [interest](<https://devfeed.tech/tags/interest.md>), [memory](<https://devfeed.tech/tags/memory.md>), [mmu](<https://devfeed.tech/tags/mmu.md>), [rants](<https://devfeed.tech/tags/rants.md>), [swapfile](<https://devfeed.tech/tags/swapfile.md>), [zram](<https://devfeed.tech/tags/zram.md>), [zswap](<https://devfeed.tech/tags/zswap.md>)

### AI overview

Modern memory systems hide a hierarchy of CPU caches, virtual address translation, NUMA, filesystem caching, and persistent storage behind a simple byte-addressed model. The article also describes Linux zram and zswap, then introduces CRAM, a developmental project that uses hardware-assisted compression exposed as a special NUMA node. Reported benchmarks show promising results for selected workloads, while the project remains unsuitable for everyday use.

### Source excerpt

When you first learn to program, memory seems pretty straightforward. You have a bunch of bytes, each with an address. If you want to read or write something, you specify ...read more

## Deep dive into Azure Maia 300 architecture

DevFeed: [Deep dive into Azure Maia 300 architecture](<https://devfeed.tech/articles/deep-dive-into-azure-maia-300-architecture-92612.md>)

Original publisher: [Read original article](<https://azure.microsoft.com/en-us/blog/deep-dive-into-azure-maia-300-architecture/>)

Author: Rani Borkar and Saurabh Dighe

Published: 2026-10-12T13:00:00Z

Content type: article

Language: en

Sources: [Microsoft Azure](<https://devfeed.tech/sources/microsoft-azure-blog.md>)

Topics: [AI Architecture](<https://devfeed.tech/topics/ai-architecture.md>), [Azure](<https://devfeed.tech/topics/azure.md>), [NUMA](<https://devfeed.tech/topics/numa.md>), [Chip design](<https://devfeed.tech/topics/chip-design.md>), [Cloud Architecture](<https://devfeed.tech/topics/cloud-architecture.md>)

Tags: [accelerators](<https://devfeed.tech/tags/accelerators.md>), [advanced-ai-capabilities](<https://devfeed.tech/tags/advanced-ai-capabilities.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-applications](<https://devfeed.tech/tags/ai-applications.md>), [ai-plus-machine-learning](<https://devfeed.tech/tags/ai-plus-machine-learning.md>), [ai-systems](<https://devfeed.tech/tags/ai-systems.md>), [algorithms](<https://devfeed.tech/tags/algorithms.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [asics](<https://devfeed.tech/tags/asics.md>), [azure](<https://devfeed.tech/tags/azure.md>), [azure-maia-300](<https://devfeed.tech/tags/azure-maia-300.md>), [cloud](<https://devfeed.tech/tags/cloud.md>), [devops](<https://devfeed.tech/tags/devops.md>), [inference](<https://devfeed.tech/tags/inference.md>), [large-language-models-llms](<https://devfeed.tech/tags/large-language-models-llms.md>), [maia](<https://devfeed.tech/tags/maia.md>), [maia-300](<https://devfeed.tech/tags/maia-300.md>), [microsoft](<https://devfeed.tech/tags/microsoft.md>), [microsoft-azure](<https://devfeed.tech/tags/microsoft-azure.md>), [platform](<https://devfeed.tech/tags/platform.md>), [scale](<https://devfeed.tech/tags/scale.md>)

### AI overview

Maia 300 co-designs software, accelerator silicon, networking, rack infrastructure, and datacenter operations for cloud-scale inference. Its software-defined dataflow architecture makes memory placement and data movement explicit; chiplet-local NUMA memory, narrower precision formats, and eight HBM4 stacks (288 GB and 18 TB/s per accelerator) target data movement and bandwidth efficiency. A two-tier Ethernet-based fabric connects 96 accelerators per rack and up to 12,288 per superpod, with electrical in-rack links and optical links across racks. The rack integrates 96 accelerators within 167 kW and uses liquid cooling for over 80% of its thermal load, with facility-water or heat-exchanger deployment options. The article reports, versus Maia 200, up to 7x rack-scale token throughput and more than 2.5x performance per watt; these are vendor-reported figures, and the excerpt provides no independent benchmarks.

### Source excerpt

Maia 300 operates as a single platform to meet the demands of cloud-scale agentic inference. The post Deep dive into Azure Maia 300 architecture appeared first on Microsoft Azure Blog.

## DECing Out the Homelab with an AlphaServer 2100

DevFeed: [DECing Out the Homelab with an AlphaServer 2100](<https://devfeed.tech/articles/decing-out-the-homelab-with-an-alphaserver-2100-85076.md>)

Original publisher: [Read original article](<https://hackaday.com/2026/10/11/decing-out-the-homelab-with-an-alphaserver-2100/>)

Author: Tyler August

Published: 2026-10-12T05:00:36Z

Content type: article

Language: en

Sources: [Hackaday](<https://devfeed.tech/sources/hackaday.md>)

Topics: [NUMA](<https://devfeed.tech/topics/numa.md>), [Ollama](<https://devfeed.tech/topics/ollama.md>)

Tags: [alphaserver](<https://devfeed.tech/tags/alphaserver.md>), [dec-alpha](<https://devfeed.tech/tags/dec-alpha.md>), [hardware](<https://devfeed.tech/tags/hardware.md>), [openvms](<https://devfeed.tech/tags/openvms.md>), [operating-system](<https://devfeed.tech/tags/operating-system.md>), [retrocomputing](<https://devfeed.tech/tags/retrocomputing.md>), [units](<https://devfeed.tech/tags/units.md>)

### AI overview

A tour of a 1994 DEC AlphaServer 2100 highlights its single 64-bit Alpha RISC processor and 256 MB of RAM. The machine powers on, but the tour does not reach an operating system; apparent power problems remain unresolved. The article notes that OpenVMS is promised and that a VAX-descended OS also runs in the cloud and in a virtual machine.

### Source excerpt

Retrocomputer enthusiasts have a bad rap as nerds who just want to play old video games. There's some truth in this, but the reality is that we're nerds who want ...read more

## Meta engineers present Compressed RAM for Linux, with near-native read performance

DevFeed: [Meta engineers present Compressed RAM for Linux, with near-native read performance](<https://devfeed.tech/articles/meta-developing-compressed-ram-cram-for-linux-better-than-zram-and-zswap-84585.md>)

Original publisher: [Read original article](<https://hardware.slashdot.org/story/26/10/10/0631257/meta-developing-compressed-ram-cram-for-linux-better-than-zram-and-zswap>)

Author: EditorDavid

Published: 2026-10-10T12:04:00Z

Content type: news

Language: en

Sources: [Slashdot](<https://devfeed.tech/sources/slashdot.md>)

Topics: [DRAM price increase](<https://devfeed.tech/topics/dram-price-increase.md>), [NUMA](<https://devfeed.tech/topics/numa.md>), [Steam Deck](<https://devfeed.tech/topics/steam-deck.md>)

Tags: [dram](<https://devfeed.tech/tags/dram.md>), [hardware](<https://devfeed.tech/tags/hardware.md>), [linux](<https://devfeed.tech/tags/linux.md>), [numa](<https://devfeed.tech/tags/numa.md>), [phoronix](<https://devfeed.tech/tags/phoronix.md>), [storage](<https://devfeed.tech/tags/storage.md>)

### AI overview

Meta engineers are developing Compressed RAM (CRAM) for Linux using hardware-offloaded compression. The article reports near-native performance for read-only data and faster writes than ZRAM, Zswap, or plain swap, though writes require page faults and folio migration. Most components are already in mainline, with remaining implementation questions before possible kernel adoption.

### Source excerpt

Phoronix reports on a presentation this week by Meta engineer Gregory Price at the Linux Plumbers Conference in Prague: With today's memory shortages, high prices, and ever increasing memory capacity needs, Meta engineers have been working on Compressed RAM ("CRAM") for Linux. This hardware-offloaded compression still allows cacheline/byte access to compressed memory and in turn a better implementation than the likes of ZRAM and Zswap. Very promising is that there is near-native performance to raw DRAM speeds with CRAM... For end-users that may already be relying on the likes of Zswap or ZRAM, CRAM is offering near-native DRAM speeds. Read-only data is shown to be as fast as the raw DRAM performance. And for memory writes, CRAM is much faster than ZRAM or Zswap or plain swap. "Most of the individual pieces needed to support CRAM are already mainline," the article points out. Tom's Hardware writes that "It uses a private NUMA node (essentially, a ghost CPU) instead of pretending to be a block device, and this lets Linux continue using all of its memory semantics, including migration and ballooning, to manage CRAM." Because CRAM is stored in RAM and treated as RAM, with full cacheline/byte access, it can be accessed in a read-only fashion with little delay; just the cost of hardware-offloaded compression... Even when you enable writes, CRAM is still much faster than ZRAM; 5.4x in the worst tested case of 20% writes. That's a huge drop from the 452x read-only case, but keep your context; a 5.4x speedup is still titanic. The massive performance cliff when writes are involved comes down to the need to page fault and migrate folios back to the original NUMA domain, as you can't write directly to compressed data; you'd corrupt everything... While the obvious target of the work here (given its origin at Meta Platforms) is big Linux servers, ZRAM and Zswap are used all over the Linux ecosystem, even on machines as constrained as the Steam Deck. Many distributions enable one

## Upscale AI Partners With Nvidia While Competing With It

DevFeed: [Upscale AI Partners With Nvidia While Competing With It](<https://devfeed.tech/articles/upscale-ai-partners-with-nvidia-while-competing-with-it-84378.md>)

Original publisher: [Read original article](<https://www.nextplatform.com/connect/2026/10/09/upscale-ai-partners-with-nvidia-while-competing-with-it/5302456>)

Author: Timothy Prickett Morgan

Published: 2026-10-09T21:15:34Z

Content type: article

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: [NVLink](<https://devfeed.tech/topics/nvlink.md>), [Supercomputing](<https://devfeed.tech/topics/supercomputing.md>), [NUMA](<https://devfeed.tech/topics/numa.md>), [Spectrum Ethernet](<https://devfeed.tech/topics/spectrum-ethernet.md>), [Nvidia](<https://devfeed.tech/topics/nvidia.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [all](<https://devfeed.tech/tags/all.md>), [choice](<https://devfeed.tech/tags/choice.md>), [cloud-builder](<https://devfeed.tech/tags/cloud-builder.md>), [clusters](<https://devfeed.tech/tags/clusters.md>), [connect](<https://devfeed.tech/tags/connect.md>), [datacenters](<https://devfeed.tech/tags/datacenters.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [nvlink](<https://devfeed.tech/tags/nvlink.md>), [nvswitch](<https://devfeed.tech/tags/nvswitch.md>), [skyfabrix](<https://devfeed.tech/tags/skyfabrix.md>), [skyhammer](<https://devfeed.tech/tags/skyhammer.md>), [upscale-ai](<https://devfeed.tech/tags/upscale-ai.md>)

### AI overview

Upscale AI is developing SkyHammer and SkyFabriX scale-up networking for AI clusters while partnering with Nvidia to build scale-out switches using Spectrum-X ASICs. The article describes bandwidth and accelerator-scale claims, SkyOS support across both network types, and the current limitation that Nvidia GPUs cannot yet use SkyHammer for scale-up.

### Source excerpt

Every CPU maker always had its own and very proprietary way to link multiple CPUs into shared memory ...

## Impactful scheduling for GPU clusters

DevFeed: [Impactful scheduling for GPU clusters](<https://devfeed.tech/articles/impactful-scheduling-for-gpu-clusters-84233.md>)

Original publisher: [Read original article](<https://huggingface.co/blog/allenai/impactful-scheduling>)

Author: Kyle Wiggers

Published: 2026-10-09T15:20:29Z

Content type: article

Language: en

Sources: [Hugging Face](<https://devfeed.tech/sources/hugging-face-blog.md>)

Topics: [scheduling](<https://devfeed.tech/topics/scheduling.md>), [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [NUMA](<https://devfeed.tech/topics/numa.md>)

Tags: [ai-infrastructure](<https://devfeed.tech/tags/ai-infrastructure.md>), [clusters](<https://devfeed.tech/tags/clusters.md>), [distributed](<https://devfeed.tech/tags/distributed.md>), [distributed-training](<https://devfeed.tech/tags/distributed-training.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [gpu-clusters](<https://devfeed.tech/tags/gpu-clusters.md>), [scheduler](<https://devfeed.tech/tags/scheduler.md>), [scheduling](<https://devfeed.tech/tags/scheduling.md>)

### AI overview

Ai2 replaced its priority-based scheduler for research GPU clusters with GPU time budgets, hierarchical fair-share allocation and minimum-runtime contracts. The system maintained 98% occupancy while delivering 98% of allocated GPU hours over a 30-day period; debug-workload p90 wait times fell from two hours to 30 seconds, and repairs requiring human intervention fell by 74%. The article also describes user communication challenges and remaining concerns about capacity fragmentation.

### Source excerpt

*Building a cluster scheduler to prioritize high-impact research while maintaining full occupancy* On the AI Infrastructure team at Ai2, we're responsible for providing the institute's GPU compute capacity, specifically targeting large, distributed training workloads. We think about this task as a pyramid of four metrics that build on each other. The foundation is **availability**: how often the hardware is healthy and ready for work.

## Scaling RL rollouts with Firecracker on Amazon EC2 metal

DevFeed: [Scaling RL rollouts with Firecracker on Amazon EC2 metal](<https://devfeed.tech/articles/scaling-rl-rollouts-with-firecracker-on-amazon-ec2-metal-84346.md>)

Original publisher: [Read original article](<https://aws.amazon.com/blogs/compute/scaling-rl-rollouts-with-firecracker-on-amazon-ec2-metal/>)

Author: Karsten Ploesser

Published: 2026-10-09T14:38:27Z

Content type: tutorial

Language: en

Sources: [AWS Compute Blog](<https://devfeed.tech/sources/aws-compute-blog.md>)

Topics: [Amazon EC2](<https://devfeed.tech/topics/amazon-ec2.md>), [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [virtualization](<https://devfeed.tech/topics/virtualization.md>), [NUMA](<https://devfeed.tech/topics/numa.md>), [grpo](<https://devfeed.tech/topics/grpo.md>)

Tags: [amazon-ec2](<https://devfeed.tech/tags/amazon-ec2.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [expert-400](<https://devfeed.tech/tags/expert-400.md>), [firecracker](<https://devfeed.tech/tags/firecracker.md>), [latency](<https://devfeed.tech/tags/latency.md>), [technical-how-to](<https://devfeed.tech/tags/technical-how-to.md>)

### AI overview

The tutorial explains how to tune Firecracker microVM density on Amazon EC2 metal for reinforcement learning rollouts using the labsweep measurement harness. In the evaluated workloads, tail latency rose sharply above one microVM per vCPU, core pinning improved latency predictability, and SMT benefits depended on workload and density. It provides guidance for reproducing measurements with production workloads; dynamic churn is outside the evaluation's scope.

### Source excerpt

Reinforcement learning at scale runs thousands of isolated rollout sandboxes in parallel, and packing more of them onto each host lowers your cost per training step. This post shows how to tune Firecracker microVM density on Amazon EC2 metal, and introduces labsweep, a harness for mapping your own latency and density curve.

## Avoid Errors by Selecting the Proper Resources for Your Pod

DevFeed: [Avoid Errors by Selecting the Proper Resources for Your Pod](<https://devfeed.tech/articles/avoid-errors-by-selecting-the-proper-resources-for-your-pod-78076.md>)

Original publisher: [Read original article](<https://www.runpod.io/blog/avoid-pod-errors-runpod-resources>)

Author: Brendan McKeag

Published: 2026-10-08T20:47:14Z

Content type: tutorial

Language: en

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

Topics: [Container Image Building](<https://devfeed.tech/topics/container-image-building.md>), [NUMA](<https://devfeed.tech/topics/numa.md>), [Operating system](<https://devfeed.tech/topics/operating-system.md>), [GPU instancing](<https://devfeed.tech/topics/gpu-instancing.md>)

Tags: [all](<https://devfeed.tech/tags/all.md>), [container](<https://devfeed.tech/tags/container.md>), [cuda](<https://devfeed.tech/tags/cuda.md>), [errors](<https://devfeed.tech/tags/errors.md>), [ram](<https://devfeed.tech/tags/ram.md>), [vram](<https://devfeed.tech/tags/vram.md>)

### AI overview

The article explains how to diagnose and address Runpod pod errors caused by insufficient container storage or RAM/VRAM. It covers checking container utilization, increasing volume size, choosing a different GPU configuration, and reducing model memory requirements through quantization.

### Source excerpt

Common errors when spinning up pods often stem from insufficient container space or RAM/VRAM. This post explains how to identify and fix both issues by.

## AMD ROCm 10.1 Targets Storage Bottlenecks with hipFILE Fast-Path, NUMA Memory Allocation, and LLVM 24

DevFeed: [AMD ROCm 10.1 Targets Storage Bottlenecks with hipFILE Fast-Path, NUMA Memory Allocation, and LLVM 24](<https://devfeed.tech/articles/amd-rocm-10-1-targets-storage-bottlenecks-with-hipfile-fast-path-numa-memory-allocation-and-llvm-24-66838.md>)

Original publisher: [Read original article](<https://www.storagereview.com/news/amd-rocm-10-1-hipfile-fast-path-numa-memory-llvm-24>)

Author: Harold Fritts

Published: 2026-10-07T15:46:34Z

Content type: release

Language: en

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

Topics: [rocm](<https://devfeed.tech/topics/rocm.md>), [NUMA](<https://devfeed.tech/topics/numa.md>), [LLVM](<https://devfeed.tech/topics/llvm.md>), [quantization](<https://devfeed.tech/topics/quantization.md>), [restic](<https://devfeed.tech/topics/restic.md>), [Test-driven development](<https://devfeed.tech/topics/tdd.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [amd](<https://devfeed.tech/tags/amd.md>), [bottlenecks](<https://devfeed.tech/tags/bottlenecks.md>), [enterprise](<https://devfeed.tech/tags/enterprise.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [llvm](<https://devfeed.tech/tags/llvm.md>), [numa](<https://devfeed.tech/tags/numa.md>), [release](<https://devfeed.tech/tags/release.md>), [rocm](<https://devfeed.tech/tags/rocm.md>), [storage](<https://devfeed.tech/tags/storage.md>)

### AI overview

AMD ROCm 10.1 expands Infinity Storage with asynchronous hipFILE reads and writes, batched I/O, and NUMA-aware host memory placement to improve storage-to-GPU data movement. The release also updates ROCm CLI and profiling tools, moves the toolchain to LLVM 24, and adds compiler, library, and platform changes.

### Source excerpt

AMD ROCm 10.1, released October 5, puts its weight on the path between storage and the GPU. Checkpoint files, key-value caches, and model weights now routinely outgrow accelerator memory, and when they do, the time spent moving data starts to govern iteration time more than the math does. The release builds out AMD Infinity Storage, The post AMD ROCm 10.1 Targets Storage Bottlenecks with hipFILE Fast-Path, NUMA Memory Allocation, and LLVM 24 appeared first on StorageReview.com.

## AMD Engineer Talks Up New IBS Memory Profiler Feature Found On Zen 6

DevFeed: [AMD Engineer Talks Up New IBS Memory Profiler Feature Found On Zen 6](<https://devfeed.tech/articles/amd-engineer-talks-up-new-ibs-memory-profiler-feature-found-on-zen-6-66847.md>)

Original publisher: [Read original article](<https://www.phoronix.com/news/AMD-Zen-6-IBS-Memory-Profiler>)

Author: Michael Larabel

Published: 2026-10-07T09:47:54Z

Content type: news

Language: en

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

Topics: [NUMA](<https://devfeed.tech/topics/numa.md>), [DRAM price increase](<https://devfeed.tech/topics/dram-price-increase.md>), [QCon San Francisco 2025](<https://devfeed.tech/topics/qcon-san-francisco-2025.md>)

Tags: [amd](<https://devfeed.tech/tags/amd.md>), [amd-epyc](<https://devfeed.tech/tags/amd-epyc.md>), [cxl](<https://devfeed.tech/tags/cxl.md>), [desktop-linux](<https://devfeed.tech/tags/desktop-linux.md>), [epyc](<https://devfeed.tech/tags/epyc.md>), [hardware](<https://devfeed.tech/tags/hardware.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-kernel](<https://devfeed.tech/tags/linux-kernel.md>), [linux-performance](<https://devfeed.tech/tags/linux-performance.md>), [linux-server-benchmarks](<https://devfeed.tech/tags/linux-server-benchmarks.md>), [mainline](<https://devfeed.tech/tags/mainline.md>), [memory](<https://devfeed.tech/tags/memory.md>), [numa](<https://devfeed.tech/tags/numa.md>), [open-source-graphics](<https://devfeed.tech/tags/open-source-graphics.md>), [overhead](<https://devfeed.tech/tags/overhead.md>), [phoronix](<https://devfeed.tech/tags/phoronix.md>), [phoronix-test-suite](<https://devfeed.tech/tags/phoronix-test-suite.md>), [server](<https://devfeed.tech/tags/server.md>), [ubuntu-benchmarks](<https://devfeed.tech/tags/ubuntu-benchmarks.md>), [ubuntu-hardware](<https://devfeed.tech/tags/ubuntu-hardware.md>), [zen-6](<https://devfeed.tech/tags/zen-6.md>)

### AI overview

An AMD engineer presented Zen 6's IBS Memory Profiler, which profiles memory accesses and tracks NUMA nodes. It is intended to support hot-page detection and promotion on Linux systems with tiered memory, including AMD EPYC servers using DRAM and CXL. Work on integrating it with Linux's pghot subsystem is ongoing; reported benchmarks on a 256-core AMD EPYC Venice server indicate low overhead.

### Source excerpt

One of the AMD Zen 6 CPU features that hasn't been talked about too widely beyond appearing in a public AMD tech doc is the IBS Memory Profiler. This week in Prague at LPC 2026 an AMD engineer talked up Zen 6's IBS Memory Profiler and its benefits for hot page detection and promotion on Linux systems...

## WSL 3.0.2 adds automatic virtual NUMA support and experimental sparse VHDs

DevFeed: [WSL 3.0.2 adds automatic virtual NUMA support and experimental sparse VHDs](<https://devfeed.tech/articles/windows-subsystem-for-linux-can-now-run-on-very-large-servers-experimental-sparse-vhd-65668.md>)

Original publisher: [Read original article](<https://www.phoronix.com/news/Microsoft-WSL-3.0.2>)

Author: Michael Larabel

Published: 2026-10-06T10:05:34Z

Content type: news

Language: en

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

Topics: [Windows Subsystem for Linux](<https://devfeed.tech/topics/wsl.md>), [NUMA](<https://devfeed.tech/topics/numa.md>)

Tags: [containers](<https://devfeed.tech/tags/containers.md>), [desktop-linux](<https://devfeed.tech/tags/desktop-linux.md>), [ipv4](<https://devfeed.tech/tags/ipv4.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>), [microsoft](<https://devfeed.tech/tags/microsoft.md>), [numa](<https://devfeed.tech/tags/numa.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>), [storage](<https://devfeed.tech/tags/storage.md>), [ubuntu-benchmarks](<https://devfeed.tech/tags/ubuntu-benchmarks.md>), [ubuntu-hardware](<https://devfeed.tech/tags/ubuntu-hardware.md>), [windows-subsystem-for-linux](<https://devfeed.tech/tags/windows-subsystem-for-linux.md>)

### AI overview

Microsoft's WSL 3.0.2 update adds automatic virtual NUMA support for large WSL virtual machines and experimental sparse VHD creation, which uses disk space as data is written. It also includes IPv4 forwarding and container fixes, adds container health notifications to WSLC events, and includes other fixes.

### Source excerpt

Microsoft yesterday released WSL 3.0.2 as their latest WSL update since last week's WSL3 release with "WSLC" for promoting Linux containers on Windows to general availability. Some notable changes in WSL 3.0.2 include automatic virtual NUMA support and experimental support for sparse VHD storage...

## Redox OS Adds IO\_uring-Like API, NUMA & Gets QEMU Working

DevFeed: [Redox OS Adds IO\_uring-Like API, NUMA & Gets QEMU Working](<https://devfeed.tech/articles/redox-os-adds-io-uring-like-api-numa-gets-qemu-working-60041.md>)

Original publisher: [Read original article](<https://www.phoronix.com/news/Redox-OS-August-2026>)

Author: Michael Larabel

Published: 2026-09-25T15:51:38Z

Content type: news

Language: en

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

Topics: [Operating system](<https://devfeed.tech/topics/operating-system.md>), [Open Source](<https://devfeed.tech/topics/open-source.md>), [io\_uring](<https://devfeed.tech/topics/io-uring.md>), [NUMA](<https://devfeed.tech/topics/numa.md>), [qemu](<https://devfeed.tech/topics/qemu.md>), [Emulator](<https://devfeed.tech/topics/emulator.md>), [Rust](<https://devfeed.tech/topics/rust.md>), [Linux](<https://devfeed.tech/topics/linux.md>)

Tags: [cpu](<https://devfeed.tech/tags/cpu.md>), [desktop-linux](<https://devfeed.tech/tags/desktop-linux.md>), [emulator](<https://devfeed.tech/tags/emulator.md>), [io-uring](<https://devfeed.tech/tags/io-uring.md>), [ipc](<https://devfeed.tech/tags/ipc.md>), [kernel](<https://devfeed.tech/tags/kernel.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>), [numa](<https://devfeed.tech/tags/numa.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [open-source-graphics](<https://devfeed.tech/tags/open-source-graphics.md>), [os](<https://devfeed.tech/tags/os.md>), [phoronix](<https://devfeed.tech/tags/phoronix.md>), [phoronix-test-suite](<https://devfeed.tech/tags/phoronix-test-suite.md>), [qemu](<https://devfeed.tech/tags/qemu.md>), [rust](<https://devfeed.tech/tags/rust.md>), [ubuntu-benchmarks](<https://devfeed.tech/tags/ubuntu-benchmarks.md>), [ubuntu-hardware](<https://devfeed.tech/tags/ubuntu-hardware.md>), [virtual-machine](<https://devfeed.tech/tags/virtual-machine.md>)

### AI overview

Redox OS's August status update reports an IO_uring-like ring-buffer API, initial NUMA support, working QEMU emulation without KVM-like acceleration, Linux dual-boot installation support, and additional kernel, IPC, CPU, and driver improvements.

### Source excerpt

The Redox OS open-source, Rust-based operating system has seen a lot of activity in recent weeks with many notable improvements landing...

[Next page](<https://devfeed.tech/topics/numa.md?cursor=WyIyMDI2LTA5LTI1VDE1OjUxOjM4KzAwOjAwIiwgImQzZjkyN2RhLTc5ZDQtNGE0Zi04ZTA5LTUwM2I3OGZiNDZkNSJd>)