# gpu monitoring

Monitoring GPU health, performance, utilization, telemetry, and related operational metrics in computing infrastructure.

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

## Amazon SageMaker AI now detects latency drift, idle GPUs, and missed optimizations on your inference endpoints

DevFeed: [Amazon SageMaker AI now detects latency drift, idle GPUs, and missed optimizations on your inference endpoints](<https://devfeed.tech/articles/amazon-sagemaker-ai-now-detects-latency-drift-idle-gpus-and-missed-optimizations-on-your-inference-endpoints-144471.md>)

Original publisher: [Read original article](<https://aws.amazon.com/blogs/machine-learning/amazon-sagemaker-ai-now-detects-latency-drift-idle-gpus-and-missed-optimizations-on-your-inference-endpoints/>)

Author: Hazim Qudah

Published: 2026-11-19T21:28:18Z

Content type: tutorial

Language: en

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

Topics: [Amazon SageMaker](<https://devfeed.tech/topics/amazon-sagemaker.md>), [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>), [gpu monitoring](<https://devfeed.tech/topics/gpu-monitoring.md>), [Benchmark](<https://devfeed.tech/topics/benchmark.md>), [performance-engineering](<https://devfeed.tech/topics/performance-engineering.md>), [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [GPU](<https://devfeed.tech/topics/gpu.md>), [Amazon Web Services](<https://devfeed.tech/topics/aws.md>), [vllm](<https://devfeed.tech/topics/vllm.md>)

Tags: [advanced-300](<https://devfeed.tech/tags/advanced-300.md>), [amazon-sagemaker-ai](<https://devfeed.tech/tags/amazon-sagemaker-ai.md>), [amazon-sagemaker-studio](<https://devfeed.tech/tags/amazon-sagemaker-studio.md>), [announcements](<https://devfeed.tech/tags/announcements.md>), [aws-user-notifications](<https://devfeed.tech/tags/aws-user-notifications.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [drift](<https://devfeed.tech/tags/drift.md>), [generative-ai](<https://devfeed.tech/tags/generative-ai.md>), [gpu-utilization](<https://devfeed.tech/tags/gpu-utilization.md>), [gpus](<https://devfeed.tech/tags/gpus.md>), [inference-endpoints](<https://devfeed.tech/tags/inference-endpoints.md>), [latency](<https://devfeed.tech/tags/latency.md>), [optimizations](<https://devfeed.tech/tags/optimizations.md>), [performance](<https://devfeed.tech/tags/performance.md>), [serving](<https://devfeed.tech/tags/serving.md>), [studio](<https://devfeed.tech/tags/studio.md>)

### AI overview

Amazon SageMaker AI monitors eligible generative AI inference endpoints by default when they use vLLM serving containers and GPU instances. Using aggregated latency, GPU utilization, and serving configuration signals, it can flag week-over-week latency drift, mostly idle GPUs, or supported optimizations and updates that are missing. Monitoring does not read request or response content and has no GPU cost. Findings arrive through AWS Health and email; users can open them in SageMaker Studio and benchmark alternatives on separate compute, comparing response time, throughput, and cost per million tokens. No endpoint change occurs until the user chooses to apply a result, and keeping the current configuration is also an option. Benchmarks incur standard SageMaker charges or use an ML reservation. Monitoring can be disabled by setting OPT_OUT_TRACKING on a model container; doing so prevents telemetry collection and notifications for that endpoint.

### Source excerpt

Amazon SageMaker AI now monitors your generative AI inference endpoints by default and tells you when a change could improve performance or lower cost. It tracks latency, GPU utilization, and serving configuration, then sends a finding through AWS Health with a one-click path to benchmark alternatives in SageMaker Studio before you apply any change.

## Monitor the efficiency and availability of accelerated computing instances with Amazon EC2 Detailed Monitoring and performance status checks

DevFeed: [Monitor the efficiency and availability of accelerated computing instances with Amazon EC2 Detailed Monitoring and performance status checks](<https://devfeed.tech/articles/monitor-the-efficiency-and-availability-of-accelerated-computing-instances-with-amazon-ec2-detailed-monitoring-and-performance-status-checks-144756.md>)

Original publisher: [Read original article](<https://aws.amazon.com/blogs/compute/monitor-the-efficiency-and-availability-of-accelerated-computing-instances-with-amazon-ec2-detailed-monitoring-and-performance-status-checks/>)

Author: Caroline Frost

Published: 2026-11-19T18:34:57Z

Content type: article

Language: en

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

Topics: [gpu monitoring](<https://devfeed.tech/topics/gpu-monitoring.md>), [infrastructure monitoring](<https://devfeed.tech/topics/infrastructure-monitoring.md>), [Monitoring](<https://devfeed.tech/topics/monitoring.md>), [Amazon CloudWatch](<https://devfeed.tech/topics/amazon-cloudwatch.md>), [Monitoring & Alerting](<https://devfeed.tech/topics/monitoring-alerting.md>), [Availability](<https://devfeed.tech/topics/availability.md>), [Amazon EC2](<https://devfeed.tech/topics/amazon-ec2.md>)

Tags: [accelerated](<https://devfeed.tech/tags/accelerated.md>), [accelerated-computing](<https://devfeed.tech/tags/accelerated-computing.md>), [advanced-300](<https://devfeed.tech/tags/advanced-300.md>), [amazon](<https://devfeed.tech/tags/amazon.md>), [amazon-ec2](<https://devfeed.tech/tags/amazon-ec2.md>), [announcements](<https://devfeed.tech/tags/announcements.md>), [availability](<https://devfeed.tech/tags/availability.md>), [computing](<https://devfeed.tech/tags/computing.md>), [ec2](<https://devfeed.tech/tags/ec2.md>), [efficiency](<https://devfeed.tech/tags/efficiency.md>), [fleet](<https://devfeed.tech/tags/fleet.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [hardware](<https://devfeed.tech/tags/hardware.md>), [infrastructure](<https://devfeed.tech/tags/infrastructure.md>), [metrics](<https://devfeed.tech/tags/metrics.md>), [monitor](<https://devfeed.tech/tags/monitor.md>), [monitoring](<https://devfeed.tech/tags/monitoring.md>), [one](<https://devfeed.tech/tags/one.md>), [performance](<https://devfeed.tech/tags/performance.md>), [reports](<https://devfeed.tech/tags/reports.md>), [status](<https://devfeed.tech/tags/status.md>), [with](<https://devfeed.tech/tags/with.md>)

### AI overview

Amazon EC2 Detailed Monitoring provides one-minute GPU metrics to CloudWatch without an in-guest agent or instance compute, while performance status checks flag recoverable hardware conditions that can impair GPU performance. Metrics include GPU and streaming multiprocessor utilization, memory use, temperature, and interconnect traffic, reported at instance and per-GPU levels; teams can use them to assess capacity and tune workloads. Performance checks appear in the console, `DescribeInstanceStatus`, and CloudWatch, with reboot recommended for a pending GPU memory row remap. Teams can alarm on `StatusCheckFailed_Performance`, then checkpoint work, drain and reboot an affected node, or reset the GPU after stopping its processes. Metrics are limited to supported NVIDIA GPU instance types, carry standard CloudWatch pricing, and complement in-guest tools that provide process, container, pod, and sub-minute detail.

### Source excerpt

Amazon EC2 now reports one-minute per-GPU metrics through Detailed Monitoring and surfaces recoverable hardware conditions with a new performance status check, giving infrastructure teams agentless visibility into accelerated computing efficiency and availability across their fleet.

## Using NVIDIA Run:ai GPU Fractioning on Amazon EKS

DevFeed: [Using NVIDIA Run:ai GPU Fractioning on Amazon EKS](<https://devfeed.tech/articles/using-nvidia-run-ai-gpu-fractioning-on-amazon-eks-143848.md>)

Original publisher: [Read original article](<https://aws.amazon.com/blogs/machine-learning/using-nvidia-runai-gpu-fractioning-on-amazon-eks/>)

Author: Muru Bhaskaran

Published: 2026-11-19T16:13:51Z

Content type: tutorial

Language: en

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

Topics: [Amazon Elastic Kubernetes Service](<https://devfeed.tech/topics/amazon-elastic-kubernetes-service.md>), [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [gpu monitoring](<https://devfeed.tech/topics/gpu-monitoring.md>), [Open Source](<https://devfeed.tech/topics/open-source.md>)

Tags: [advanced-300](<https://devfeed.tech/tags/advanced-300.md>), [ai](<https://devfeed.tech/tags/ai.md>), [amazon](<https://devfeed.tech/tags/amazon.md>), [amazon-ec2](<https://devfeed.tech/tags/amazon-ec2.md>), [amazon-eks](<https://devfeed.tech/tags/amazon-eks.md>), [amazon-elastic-kubernetes-service](<https://devfeed.tech/tags/amazon-elastic-kubernetes-service.md>), [eks](<https://devfeed.tech/tags/eks.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [gpu-memory](<https://devfeed.tech/tags/gpu-memory.md>), [kai-scheduler](<https://devfeed.tech/tags/kai-scheduler.md>), [kubernetes](<https://devfeed.tech/tags/kubernetes.md>), [linux](<https://devfeed.tech/tags/linux.md>), [memory](<https://devfeed.tech/tags/memory.md>), [metrics](<https://devfeed.tech/tags/metrics.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [nvidia-run-ai](<https://devfeed.tech/tags/nvidia-run-ai.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [operator](<https://devfeed.tech/tags/operator.md>), [pod](<https://devfeed.tech/tags/pod.md>), [pods](<https://devfeed.tech/tags/pods.md>), [run-ai](<https://devfeed.tech/tags/run-ai.md>), [scheduler](<https://devfeed.tech/tags/scheduler.md>), [technical-how-to](<https://devfeed.tech/tags/technical-how-to.md>)

### AI overview

NVIDIA Run:ai GPU Fractioning lets two Kubernetes Pods share one NVIDIA A10G on Amazon EKS with enforced per-Pod GPU-memory contracts. In the validated setup, each Pod had an 8 GiB limit and successfully allocated 6 GiB; a 9 GiB allocation was rejected. The approach builds on MPS and adds Kubernetes scheduling, memory enforcement, and per-Pod metrics. The walkthrough uses a custom Amazon Linux 2023 image with an R615 driver, KAI Scheduler, and the GPU Fractioning Operator. It validates memory bounds and placement, not throughput, fault containment, or security isolation. Production use requires compatible hardware and software, capacity and runtime headroom, trusted workloads, and performance and rollback testing. It assumes pre-provisioned GPU capacity; GPU-fraction-aware autoscaling was not configured or validated.

### Source excerpt

Learn how to run two memory-bounded Kubernetes Pods on a single NVIDIA A10G GPU using open-source NVIDIA Run:ai GPU Fractioning on Amazon EKS. This walkthrough builds a custom Amazon Linux 2023 image, installs KAI Scheduler and the GPU Fractioning Operator, and validates per-Pod GPU-memory contracts with enforcement and metrics.

## Automatically recover from GPU failures with Amazon ECS Managed Instances

DevFeed: [Automatically recover from GPU failures with Amazon ECS Managed Instances](<https://devfeed.tech/articles/automatically-recover-from-gpu-failures-with-amazon-ecs-managed-instances-131641.md>)

Original publisher: [Read original article](<https://aws.amazon.com/blogs/containers/automatically-recover-from-gpu-failures-with-amazon-ecs-managed-instances/>)

Author: Piyush Mattoo

Published: 2026-11-12T21:24:16Z

Content type: article

Language: en

Sources: [Containers](<https://devfeed.tech/sources/containers.md>)

Topics: [gpu monitoring](<https://devfeed.tech/topics/gpu-monitoring.md>), [Amazon CloudWatch](<https://devfeed.tech/topics/amazon-cloudwatch.md>), [infrastructure monitoring](<https://devfeed.tech/topics/infrastructure-monitoring.md>), [aws-sns](<https://devfeed.tech/topics/aws-sns.md>)

Tags: [advanced-300](<https://devfeed.tech/tags/advanced-300.md>), [amazon](<https://devfeed.tech/tags/amazon.md>), [amazon-ecs](<https://devfeed.tech/tags/amazon-ecs.md>), [amazon-elastic-container-service](<https://devfeed.tech/tags/amazon-elastic-container-service.md>), [automatic](<https://devfeed.tech/tags/automatic.md>), [automatically](<https://devfeed.tech/tags/automatically.md>), [built-in](<https://devfeed.tech/tags/built-in.md>), [changes](<https://devfeed.tech/tags/changes.md>), [code](<https://devfeed.tech/tags/code.md>), [critical](<https://devfeed.tech/tags/critical.md>), [ecs](<https://devfeed.tech/tags/ecs.md>), [failures](<https://devfeed.tech/tags/failures.md>), [fault-injection](<https://devfeed.tech/tags/fault-injection.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [hardware](<https://devfeed.tech/tags/hardware.md>), [health](<https://devfeed.tech/tags/health.md>), [health-monitoring](<https://devfeed.tech/tags/health-monitoring.md>), [how-to](<https://devfeed.tech/tags/how-to.md>), [inference](<https://devfeed.tech/tags/inference.md>), [managed](<https://devfeed.tech/tags/managed.md>), [recover](<https://devfeed.tech/tags/recover.md>), [technical-how-to](<https://devfeed.tech/tags/technical-how-to.md>), [with](<https://devfeed.tech/tags/with.md>)

### AI overview

Amazon ECS Managed Instances uses NVIDIA DCGM to monitor GPU hardware and automatically replace instances after critical GPU failures. It marks an affected instance IMPAIRED, drains it so no new tasks are scheduled, provisions replacement capacity, then terminates the old instance after its tasks stop. Repairs are limited to at most 20% of a capacity provider's instances at once, or one instance for providers with fewer than nine. Application-level faults such as XID 13 and 31 are surfaced through CloudWatch metrics rather than triggering instance replacement. Health and state changes are available through EventBridge; DCGM metrics require Container Insights with enhanced observability. The article demonstrates end-to-end validation by injecting a synthetic XID 79, which triggers the real repair workflow without damaging GPU hardware. The targeted host is still drained and terminated, so the test is intended for a dedicated cluster or maintenance window. In the sample, detection took about two minutes and replacement about eight; model startup added 7-20 minutes, depending on the tier. These timings are not service guarantees.

### Source excerpt

Amazon ECS Managed Instances provides built-in GPU health monitoring and automatic repair powered by NVIDIA DCGM. This post shows how GPU auto-repair detects critical GPU hardware failures and replaces impaired instances with no code changes, and how to observe and safely validate the workflow using DCGM fault injection on a dual-tier inference pipeline.

## 0% GPU utilization: How SANSAR hides Linux cryptomining

DevFeed: [0% GPU utilization: How SANSAR hides Linux cryptomining](<https://devfeed.tech/articles/0-gpu-utilization-how-sansar-hides-linux-cryptomining-130923.md>)

Original publisher: [Read original article](<https://www.elastic.co/security-labs/threat-command/sansar-linux-cryptomining-malware>)

Author: Salim Bitam

Published: 2026-11-12T10:48:47Z

Content type: news

Language: en

Sources: [Elastic Security Labs](<https://devfeed.tech/sources/elastic-security-labs.md>)

Topics: [Linux](<https://devfeed.tech/topics/linux.md>), [gpu monitoring](<https://devfeed.tech/topics/gpu-monitoring.md>), [Linux Kernel](<https://devfeed.tech/topics/linux-kernel.md>), [tor](<https://devfeed.tech/topics/tor.md>), [Kernel](<https://devfeed.tech/topics/kernel.md>), [systemd](<https://devfeed.tech/topics/systemd.md>)

Tags: [backdoor](<https://devfeed.tech/tags/backdoor.md>), [c2](<https://devfeed.tech/tags/c2.md>), [elastic-security-labs](<https://devfeed.tech/tags/elastic-security-labs.md>), [ethereum](<https://devfeed.tech/tags/ethereum.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [gpu-monitoring](<https://devfeed.tech/tags/gpu-monitoring.md>), [gpu-utilization](<https://devfeed.tech/tags/gpu-utilization.md>), [hooks](<https://devfeed.tech/tags/hooks.md>), [kernel](<https://devfeed.tech/tags/kernel.md>), [linux](<https://devfeed.tech/tags/linux.md>), [malware-analysis](<https://devfeed.tech/tags/malware-analysis.md>), [orchestrator](<https://devfeed.tech/tags/orchestrator.md>), [rootkit](<https://devfeed.tech/tags/rootkit.md>), [threat-intelligence](<https://devfeed.tech/tags/threat-intelligence.md>)

### AI overview

REF3364 combines Linux cryptomining with several methods to conceal activity. After exploiting an internet-reachable Ray job agent that accepted unauthenticated submissions, the actor gained access, established persistence, and deployed the toolkit on three GPU hosts. Its Rust orchestrator SANSAR manages CPU and GPU miners over Tor, gathers host and GPU inventory, and supports miner control and interactive shells. A kernel module hides configured processes, files, and socket entries; an NVML wrapper passes calls to NVIDIA's library but reports 0% GPU utilization and substitutes temperature and power readings. A separate backdoor, netns, obtains an IP address from an Ethereum contract and connects over TCP, then can open a shell connection to a supplied destination; the command loop lacks application authentication and integrity checks. The source recommends checking who can submit jobs and whether GPU monitoring libraries have been modified.

### Source excerpt

Elastic Security Labs analyzes SANSAR, a Linux cryptomining orchestrator that manages miners over Tor, alongside a kernel rootkit, GPU monitoring hooks and a backdoor that retrieves its C2 address from Ethereum.

## Nvidia Tightens AI Chip Tracking in Southeast Asia as China Smuggling Concerns Grow

DevFeed: [Nvidia Tightens AI Chip Tracking in Southeast Asia as China Smuggling Concerns Grow](<https://devfeed.tech/articles/nvidia-tightens-ai-chip-tracking-in-southeast-asia-as-china-smuggling-concerns-grow-122268.md>)

Original publisher: [Read original article](<https://www.techrepublic.com/article/news-nvidia-ai-chip-tracking-apac-singapore-china/>)

Author: Matt Gonzales

Published: 2026-11-06T21:15:43Z

Content type: news

Language: en

Sources: [TechRepublic](<https://devfeed.tech/sources/techrepublic.md>)

Topics: [gpu monitoring](<https://devfeed.tech/topics/gpu-monitoring.md>), [Nvidia](<https://devfeed.tech/topics/nvidia.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-chip](<https://devfeed.tech/tags/ai-chip.md>), [ai-chip-smuggling](<https://devfeed.tech/tags/ai-chip-smuggling.md>), [ai-chips](<https://devfeed.tech/tags/ai-chips.md>), [apac](<https://devfeed.tech/tags/apac.md>), [artificial-intelligence](<https://devfeed.tech/tags/artificial-intelligence.md>), [asia](<https://devfeed.tech/tags/asia.md>), [china](<https://devfeed.tech/tags/china.md>), [cloud](<https://devfeed.tech/tags/cloud.md>), [cloud-computing](<https://devfeed.tech/tags/cloud-computing.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [gpu-tracking](<https://devfeed.tech/tags/gpu-tracking.md>), [hardware](<https://devfeed.tech/tags/hardware.md>), [international](<https://devfeed.tech/tags/international.md>), [monitoring](<https://devfeed.tech/tags/monitoring.md>), [news](<https://devfeed.tech/tags/news.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [semiconductor-export-controls](<https://devfeed.tech/tags/semiconductor-export-controls.md>), [southeast-asia](<https://devfeed.tech/tags/southeast-asia.md>), [tech-industry](<https://devfeed.tech/tags/tech-industry.md>), [tech-regulation](<https://devfeed.tech/tags/tech-regulation.md>), [tracking](<https://devfeed.tech/tags/tracking.md>)

### AI overview

Nvidia reportedly requires Southeast Asian providers applying for Nvidia Cloud Partner status to enroll in Fleet Intelligence, a GPU inventory and monitoring service, amid investigations into alleged diversion of restricted AI processors to China. The service uses an installed agent to send operational telemetry to Nvidia's cloud; it monitors utilization, health, and configuration, and Nvidia describes the agent as read-only. Bloomberg also reports component geolocation collection, but enrollment does not automatically require sharing location data with Nvidia or US regulators, and the article says the requirement does not establish that Nvidia can independently locate every sold GPU. The reported policy does not yet clarify obligations for existing partners or expansion to other regions; monitoring is described as voluntary for other customers. Businesses evaluating providers are advised to ask about hardware provenance, deployment locations, monitoring data, and export compliance. The article notes that reported guidance for future contracts to direct at least 51% of services to non-Chinese customers is separate and has not been enforced as mandatory.

### Source excerpt

Nvidia reportedly requires GPU monitoring for Southeast Asian cloud partners as China AI chip smuggling concerns grow. Here's what businesses should know. The post Nvidia Tightens AI Chip Tracking in Southeast Asia as China Smuggling Concerns Grow appeared first on TechRepublic.

## Google Cloud storage moves to the front of the AI stack as enterprises hit the GPU utilization ceiling

DevFeed: [Google Cloud storage moves to the front of the AI stack as enterprises hit the GPU utilization ceiling](<https://devfeed.tech/articles/google-cloud-storage-moves-to-the-front-of-the-ai-stack-as-enterprises-hit-the-gpu-utilization-ceiling-119606.md>)

Original publisher: [Read original article](<https://siliconangle.com/2026/11/05/enterprise-ai-storage-bottleneck-google-cloud-thecuberesearch-appdevangle/>)

Author: Paul Nashawaty

Published: 2026-11-05T21:40:20Z

Content type: article

Language: en

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

Topics: [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [Google Cloud Platform (GCP)](<https://devfeed.tech/topics/google-cloud.md>), [gpu monitoring](<https://devfeed.tech/topics/gpu-monitoring.md>)

Tags: [accelerators](<https://devfeed.tech/tags/accelerators.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agent-infrastructure](<https://devfeed.tech/tags/agent-infrastructure.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [ai-infrastructure](<https://devfeed.tech/tags/ai-infrastructure.md>), [ai-performance](<https://devfeed.tech/tags/ai-performance.md>), [ai-stack](<https://devfeed.tech/tags/ai-stack.md>), [ai-workloads](<https://devfeed.tech/tags/ai-workloads.md>), [appdevangle](<https://devfeed.tech/tags/appdevangle.md>), [cloud](<https://devfeed.tech/tags/cloud.md>), [cloud-storage](<https://devfeed.tech/tags/cloud-storage.md>), [compute](<https://devfeed.tech/tags/compute.md>), [cube-event-coverage](<https://devfeed.tech/tags/cube-event-coverage.md>), [data-management](<https://devfeed.tech/tags/data-management.md>), [enterprise-ai](<https://devfeed.tech/tags/enterprise-ai.md>), [enterprises](<https://devfeed.tech/tags/enterprises.md>), [google](<https://devfeed.tech/tags/google.md>), [google-cloud](<https://devfeed.tech/tags/google-cloud.md>), [google-cloud-storage](<https://devfeed.tech/tags/google-cloud-storage.md>), [google-kubernetes-engine](<https://devfeed.tech/tags/google-kubernetes-engine.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [gpu-utilization](<https://devfeed.tech/tags/gpu-utilization.md>), [infra](<https://devfeed.tech/tags/infra.md>), [infrastructure](<https://devfeed.tech/tags/infrastructure.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [nasdaq-goog](<https://devfeed.tech/tags/nasdaq-goog.md>), [nasdaq-googl](<https://devfeed.tech/tags/nasdaq-googl.md>), [news](<https://devfeed.tech/tags/news.md>), [platforms](<https://devfeed.tech/tags/platforms.md>), [ronen-schwartz](<https://devfeed.tech/tags/ronen-schwartz.md>), [storage](<https://devfeed.tech/tags/storage.md>), [storage-architecture](<https://devfeed.tech/tags/storage-architecture.md>), [the-latest](<https://devfeed.tech/tags/the-latest.md>), [thecube-research](<https://devfeed.tech/tags/thecube-research.md>)

### AI overview

The excerpt argues that storage can limit GPU utilization after enterprises secure accelerators, because slow data delivery leaves costly GPUs idle. It distinguishes three storage needs: training requires high-throughput sequential reads and fast multi-terabyte checkpoint writes; inference is latency-sensitive; and agent workloads need shared, persistent state and low-latency retrieval. Google Cloud's described approach includes zonal placement near accelerator clusters, Rapid object storage, and Object Enrichment, which adds structured metadata to files and images for agent queries. The article reports Google's figures--including checkpoint writes more than three times faster and restores more than five times faster than standard object storage, plus Rapid Bucket throughput above 15 TB/s--and attributes them to the company; the excerpt does not provide independent validation or a direct comparison with other providers.

### Source excerpt

For the past two years, enterprise AI investment has been dominated by one conversation: compute. Organizations raced to secure GPU allocations, negotiated infrastructure deals and stood up machine learning platforms at speed. What many discovered only after the fact is that the bottleneck they thought they were solving was not the only one in the [...] The post Google Cloud storage moves to the front of the AI stack as enterprises hit the GPU utilization ceiling appeared first on SiliconANGLE.

## Configure, Certify, and Operate Kubernetes GPU Clusters with NVIDIA DSX OS

DevFeed: [Configure, Certify, and Operate Kubernetes GPU Clusters with NVIDIA DSX OS](<https://devfeed.tech/articles/configure-certify-and-operate-kubernetes-gpu-clusters-with-nvidia-dsx-os-119476.md>)

Original publisher: [Read original article](<https://developer.nvidia.com/blog/configure-certify-and-operate-kubernetes-gpu-clusters-with-nvidia-dsx-os/>)

Author: Sowmyan Soman

Published: 2026-11-05T17:00:00Z

Content type: tutorial

Language: en

Sources: [NVIDIA Developer](<https://devfeed.tech/sources/nvidia-developer.md>), [NVIDIA Technical Blog](<https://devfeed.tech/sources/nvidia-technical-blog.md>)

Topics: [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [gpu monitoring](<https://devfeed.tech/topics/gpu-monitoring.md>), [NVIDIA DSX](<https://devfeed.tech/topics/nvidia-dsx.md>), [NCCL](<https://devfeed.tech/topics/nccl.md>), [CI/CD](<https://devfeed.tech/topics/cicd.md>), [Open Source](<https://devfeed.tech/topics/open-source.md>)

Tags: [agentic-ai-generative-ai](<https://devfeed.tech/tags/agentic-ai-generative-ai.md>), [basic](<https://devfeed.tech/tags/basic.md>), [can](<https://devfeed.tech/tags/can.md>), [cloud-services](<https://devfeed.tech/tags/cloud-services.md>), [clusters](<https://devfeed.tech/tags/clusters.md>), [configure](<https://devfeed.tech/tags/configure.md>), [data-center-cloud](<https://devfeed.tech/tags/data-center-cloud.md>), [developer-tools-techniques](<https://devfeed.tech/tags/developer-tools-techniques.md>), [distributed](<https://devfeed.tech/tags/distributed.md>), [distributed-training](<https://devfeed.tech/tags/distributed-training.md>), [dsx](<https://devfeed.tech/tags/dsx.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [gpu-clusters](<https://devfeed.tech/tags/gpu-clusters.md>), [health](<https://devfeed.tech/tags/health.md>), [health-checks](<https://devfeed.tech/tags/health-checks.md>), [job](<https://devfeed.tech/tags/job.md>), [kubernetes](<https://devfeed.tech/tags/kubernetes.md>), [leaving](<https://devfeed.tech/tags/leaving.md>), [node](<https://devfeed.tech/tags/node.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [nvidia-dsx](<https://devfeed.tech/tags/nvidia-dsx.md>), [one](<https://devfeed.tech/tags/one.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [os](<https://devfeed.tech/tags/os.md>), [with](<https://devfeed.tech/tags/with.md>)

### AI overview

NVIDIA DSX OS combines three open source tools to keep Kubernetes GPU clusters configured, workload-certified, and healthy through changes and repairs. AICR builds version-locked configuration recipes, renders them for existing deployers, and can detect drift; it generates artifacts but does not apply them. NVCRE runs GPU, network, storage, and AI workload tests, reports measurements and suspected faulty nodes, and applies performance thresholds only when configured. NVSentinel continuously monitors existing health signals without using GPU time. It starts in reporting-only mode; when configured, it can quarantine or drain unhealthy nodes and coordinate NVCRE revalidation before they return to service. The tools divide responsibility: NVCRE records failures and attribution, while NVSentinel handles quarantine and repair. Lifecycle automation can validate new and remediated nodes, but driver, operator, or fabric changes need a separate pipeline-triggered validation process. The excerpt assumes the tools, cluster dependencies, and NVCRE test provider are installed and configured.

### Source excerpt

One underperforming GPU can slow an entire distributed training job. The node may pass basic health checks and show normal utilization, leaving operators to...

## GHOST Driver

DevFeed: [GHOST Driver](<https://devfeed.tech/articles/ghost-driver-116476.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.03649>)

Author: Azhar Shaikh

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

Content type: article

Language: en

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

Topics: [eBPF](<https://devfeed.tech/topics/ebpf.md>), [gpu monitoring](<https://devfeed.tech/topics/gpu-monitoring.md>), [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [dma](<https://devfeed.tech/topics/dma.md>), [tracing](<https://devfeed.tech/topics/tracing.md>), [OpenTelemetry](<https://devfeed.tech/topics/opentelemetry.md>), [Prometheus](<https://devfeed.tech/topics/prometheus.md>), [Linux](<https://devfeed.tech/topics/linux.md>), [dmaengine](<https://devfeed.tech/topics/dmaengine.md>)

Tags: [ai-inference](<https://devfeed.tech/tags/ai-inference.md>), [attribution](<https://devfeed.tech/tags/attribution.md>), [cs-dc](<https://devfeed.tech/tags/cs-dc.md>), [cs-os](<https://devfeed.tech/tags/cs-os.md>), [dma](<https://devfeed.tech/tags/dma.md>), [driver](<https://devfeed.tech/tags/driver.md>), [drm](<https://devfeed.tech/tags/drm.md>), [ebpf](<https://devfeed.tech/tags/ebpf.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [gpu-clusters](<https://devfeed.tech/tags/gpu-clusters.md>), [gpu-utilization](<https://devfeed.tech/tags/gpu-utilization.md>), [interrupt](<https://devfeed.tech/tags/interrupt.md>), [latency](<https://devfeed.tech/tags/latency.md>), [linux-kernel](<https://devfeed.tech/tags/linux-kernel.md>), [metrics](<https://devfeed.tech/tags/metrics.md>), [mig](<https://devfeed.tech/tags/mig.md>), [opentelemetry](<https://devfeed.tech/tags/opentelemetry.md>), [overhead](<https://devfeed.tech/tags/overhead.md>), [profiling](<https://devfeed.tech/tags/profiling.md>), [prometheus](<https://devfeed.tech/tags/prometheus.md>), [vgpu](<https://devfeed.tech/tags/vgpu.md>)

### AI overview

GHOST-Driver aims to explain why high GPU utilization readings can coexist with lower effective throughput by tracing delays inside the Linux kernel driver, which userspace tools such as CUPTI do not expose. The eBPF framework follows six submission stages, from ioctl entry and GEM buffer validation through fence handling, command emission, DMA scheduling, and interrupt signaling. It correlates traces with tenant vGPU context IDs for attribution in SR-IOV and MIG environments and exports metrics to Prometheus and OpenTelemetry. The abstract reports sub-microsecond resolution and CPU overhead below 2% in production.

### Source excerpt

arXiv:2611.03649v1 Announce Type: cross Abstract: AI inference workloads running on GPU clusters exhibit a well-documented utilization paradox: monitoring tools such as nvidia-smi regularly report GPU utilization figures above 80%, yet effective compute throughput is substantially lower. A significant fraction of reported utilization time is consumed by fence waits, driver submission overhead, DMA scheduling latency, and interrupt coalescing delays, none of which is visible to userspace profiling tools. These hidden latency components are embedded deep inside the Linux kernel GPU driver stack, primarily within the Direct Rendering Manager (DRM) subsystem, GEM buffer object lifecycle, and dma_fence synchronization infrastructure. Existing vendor profiling tools like CUPTI operate at the CUDA userspace boundary and provide no visibility into kernel-side driver paths. We present GHOST-Driver (GPU Host-Side Observability and Tracing for Drivers), a non-intrusive eBPF-based framework that attaches to stable DRM kernel tracepoints and select kprobes to expose the full latency decomposition of the GPU driver submission path with sub-microsecond resolution and CPU overhead below 2% in production. GHOST-Driver tracks six key submission stages (ioctl entry, GEM BO validation, dma_fence dependency resolution, command ring emission, DMA engine scheduling, and interrupt fence signaling) and correlates them with per-tenant vGPU context IDs for multi-tenant attribution in SR-IOV and MIG environments, exporting trace metrics to Prometheus and OpenTelemetry.

## Netflix Treats GPU Reliability as a Distributed Systems Problem

DevFeed: [Netflix Treats GPU Reliability as a Distributed Systems Problem](<https://devfeed.tech/articles/netflix-treats-gpu-reliability-as-a-distributed-systems-problem-107827.md>)

Original publisher: [Read original article](<https://www.infoq.com/news/2026/10/netflix-gpu-distributed-systems/>)

Author: Craig Risi

Published: 2026-10-30T12:00:00Z

Content type: article

Language: en

Sources: [InfoQ](<https://devfeed.tech/sources/infoq.md>)

Topics: [reliability-engineering](<https://devfeed.tech/topics/reliability-engineering.md>), [distributed-training](<https://devfeed.tech/topics/distributed-training.md>), [health check](<https://devfeed.tech/topics/health-check.md>), [gpu monitoring](<https://devfeed.tech/topics/gpu-monitoring.md>), [Production Engineering](<https://devfeed.tech/topics/production-engineering.md>), [AI Architecture](<https://devfeed.tech/topics/ai-architecture.md>), [performance-engineering](<https://devfeed.tech/topics/performance-engineering.md>), [Netflix](<https://devfeed.tech/topics/netflix.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-ml-data-engineering](<https://devfeed.tech/tags/ai-ml-data-engineering.md>), [ai-training](<https://devfeed.tech/tags/ai-training.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [automated-remediation](<https://devfeed.tech/tags/automated-remediation.md>), [availability](<https://devfeed.tech/tags/availability.md>), [aws](<https://devfeed.tech/tags/aws.md>), [capacity](<https://devfeed.tech/tags/capacity.md>), [checkpoint](<https://devfeed.tech/tags/checkpoint.md>), [cloud](<https://devfeed.tech/tags/cloud.md>), [cluster](<https://devfeed.tech/tags/cluster.md>), [devops](<https://devfeed.tech/tags/devops.md>), [distributed](<https://devfeed.tech/tags/distributed.md>), [distributed-systems](<https://devfeed.tech/tags/distributed-systems.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [hardware](<https://devfeed.tech/tags/hardware.md>), [health-checks](<https://devfeed.tech/tags/health-checks.md>), [inference](<https://devfeed.tech/tags/inference.md>), [jobs](<https://devfeed.tech/tags/jobs.md>), [kubernetes](<https://devfeed.tech/tags/kubernetes.md>), [ml-data-engineering](<https://devfeed.tech/tags/ml-data-engineering.md>), [netflix](<https://devfeed.tech/tags/netflix.md>), [netflix-gpu-distributed-systems](<https://devfeed.tech/tags/netflix-gpu-distributed-systems.md>), [news](<https://devfeed.tech/tags/news.md>), [platform](<https://devfeed.tech/tags/platform.md>), [reliability](<https://devfeed.tech/tags/reliability.md>), [remediation](<https://devfeed.tech/tags/remediation.md>), [training](<https://devfeed.tech/tags/training.md>)

### AI overview

Netflix treats GPU reliability as a distributed-systems problem: in lockstep training, one failed or slow worker can stall a job even when its machine appears healthy. Its Kubernetes platform combines NVIDIA DCGM and network-interface checks, fail-closed diagnostics, and node conditions with staged remediation--from blocking new work to draining, resetting, rebooting, or replacing capacity. Checkpointing, elastic workers, retries, and slow-rank detection help jobs recover; application failures are separated to avoid retrying software bugs. Netflix reports GPU-node availability above 99.95% and job success rising from about 98% to above 99.5%. MIG lets smaller inference workloads share GPUs with memory isolation, while Kubernetes DRA work aims to improve topology-aware placement. The comparison says Meta emphasizes hardware fault detection, including silent data corruption, and reports hardware failures caused over 66% of training interruptions in its Llama 3 experience.

### Source excerpt

Netflix has detailed how it is building a reliability platform for large-scale GPU training and inference, using layered health checks, automated remediation, and fault-tolerant training to keep jobs running despite hardware failures. By Craig Risi

## Plenty of Room in the Middle: Revealing Opportunities on a GPU Cluster via Fine-Grained Monitoring

DevFeed: [Plenty of Room in the Middle: Revealing Opportunities on a GPU Cluster via Fine-Grained Monitoring](<https://devfeed.tech/articles/plenty-of-room-in-the-middle-revealing-opportunities-on-a-gpu-cluster-via-fine-grained-monitoring-103117.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.36329>)

Author: Andrew Sheinberg, Samyak Gupta, Kai Li

Published: 2026-10-28T04:00:00Z

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: [gpu monitoring](<https://devfeed.tech/topics/gpu-monitoring.md>), [scheduling](<https://devfeed.tech/topics/scheduling.md>), [infrastructure monitoring](<https://devfeed.tech/topics/infrastructure-monitoring.md>), [Multi-GPU](<https://devfeed.tech/topics/multi-gpu.md>)

Tags: [2](<https://devfeed.tech/tags/2.md>), [3](<https://devfeed.tech/tags/3.md>), [abstract](<https://devfeed.tech/tags/abstract.md>), [academia](<https://devfeed.tech/tags/academia.md>), [ai](<https://devfeed.tech/tags/ai.md>), [application](<https://devfeed.tech/tags/application.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [better](<https://devfeed.tech/tags/better.md>), [capture](<https://devfeed.tech/tags/capture.md>), [cluster](<https://devfeed.tech/tags/cluster.md>), [clusters](<https://devfeed.tech/tags/clusters.md>), [communication](<https://devfeed.tech/tags/communication.md>), [compute](<https://devfeed.tech/tags/compute.md>), [computing](<https://devfeed.tech/tags/computing.md>), [cost](<https://devfeed.tech/tags/cost.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-dc](<https://devfeed.tech/tags/cs-dc.md>), [data](<https://devfeed.tech/tags/data.md>), [fine](<https://devfeed.tech/tags/fine.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [monitoring](<https://devfeed.tech/tags/monitoring.md>), [room](<https://devfeed.tech/tags/room.md>)

### AI overview

A 10-day study of 7,399 jobs on a university GPU cluster found substantial underuse and scheduling mismatch: over 70% of jobs used less than 20% of GPU memory, half averaged below 5% SM utilization, and 80% had a large gap between requested and actual runtime. The researchers collected GPU metrics every 100 ms using a lightweight tool built on Nvidia DCGM, deployed across 89 nodes and 319 GPUs. Examined multi-GPU jobs also showed highly periodic communication. The findings point to potential utilization gains from better scheduling and GPU multitasking, without hardware or application changes; the study covers one cluster and period.

### Source excerpt

arXiv:2610.36329v1 Announce Type: new Abstract: GPU clusters have become essential computing resources for AI and scientific research in academia and industry. However, their high cost and growing demand often result in long queueing delays. While previous monitoring efforts have operated at second- or minute-level frequencies, they primarily informed job-level scheduling for clusters and lacked the detail needed to optimize GPU-level utilization. This paper presents an in-depth study of a university research GPU cluster, analyzing resource usage at 100 ms intervals. We have developed a lightweight monitoring tool that leverages Nvidia's DCGM library to capture fine-grained "compute utilization" metrics within GPUs. Deployed on a cluster with 89 nodes and 319 GPUs, our tool collected data over several months. This paper focuses on a 10-day period with 7,399 jobs submitted by 150 users from 12 departments. Our study reveals three key findings: (1) the GPU cluster exhibits significant underutilization, with over 70% of jobs using less than 20% of GPU memory and 50% averaging below 5% Streaming Multiprocessor (SM) utilization; (2) 80% of jobs have a drastic mismatch between their requested and actual durations, significantly limiting the scheduler's effectiveness; and (3) the multi-GPU jobs we examined show highly periodic communication patterns. These findings suggest there is plenty of room in the middle: substantial utilization gains are available through better job scheduling and GPU multitasking, independent of hardware or application improvements. Our monitoring solution is available as open-source software at https://github.com/als244/gpu_server_monitoring.

## How NVIDIA Minerva Turns Telemetry Chaos Into Peak GPU Yield

DevFeed: [How NVIDIA Minerva Turns Telemetry Chaos Into Peak GPU Yield](<https://devfeed.tech/articles/how-nvidia-minerva-turns-telemetry-chaos-into-peak-gpu-yield-100672.md>)

Original publisher: [Read original article](<https://developer.nvidia.com/blog/how-nvidia-minerva-turns-telemetry-chaos-into-peak-gpu-yield/>)

Author: Elizabeth Goodman

Published: 2026-10-26T16:00:00Z

Content type: article

Language: en

Sources: [NVIDIA Developer](<https://devfeed.tech/sources/nvidia-developer.md>), [NVIDIA Technical Blog](<https://devfeed.tech/sources/nvidia-technical-blog.md>)

Topics: [observability](<https://devfeed.tech/topics/observability.md>), [Monitoring](<https://devfeed.tech/topics/monitoring.md>), [gpu monitoring](<https://devfeed.tech/topics/gpu-monitoring.md>), [NVIDIA Spectrum-X Ethernet](<https://devfeed.tech/topics/nvidia-spectrum-x-ethernet.md>), [NVLink](<https://devfeed.tech/topics/nvlink.md>), [NCCL](<https://devfeed.tech/topics/nccl.md>), [NVIDIA DSX](<https://devfeed.tech/topics/nvidia-dsx.md>), [NVMe](<https://devfeed.tech/topics/nvme.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [ai-data-center](<https://devfeed.tech/tags/ai-data-center.md>), [ai-factories](<https://devfeed.tech/tags/ai-factories.md>), [ai-factory](<https://devfeed.tech/tags/ai-factory.md>), [ai-native](<https://devfeed.tech/tags/ai-native.md>), [ai-networking](<https://devfeed.tech/tags/ai-networking.md>), [ai-workloads](<https://devfeed.tech/tags/ai-workloads.md>), [aiops](<https://devfeed.tech/tags/aiops.md>), [alert](<https://devfeed.tech/tags/alert.md>), [complex](<https://devfeed.tech/tags/complex.md>), [data-center-cloud](<https://devfeed.tech/tags/data-center-cloud.md>), [ethernet](<https://devfeed.tech/tags/ethernet.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [gpus](<https://devfeed.tech/tags/gpus.md>), [infrastructure](<https://devfeed.tech/tags/infrastructure.md>), [mlops](<https://devfeed.tech/tags/mlops.md>), [networking-communications](<https://devfeed.tech/tags/networking-communications.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [nvidia-spectrum-x-ethernet](<https://devfeed.tech/tags/nvidia-spectrum-x-ethernet.md>), [software-defined-data-center](<https://devfeed.tech/tags/software-defined-data-center.md>), [spectrum-ethernet](<https://devfeed.tech/tags/spectrum-ethernet.md>), [spectrum-x](<https://devfeed.tech/tags/spectrum-x.md>), [telemetry](<https://devfeed.tech/tags/telemetry.md>), [yield](<https://devfeed.tech/tags/yield.md>)

### AI overview

NVIDIA Minerva combines telemetry and topology across GPUs, networking, storage, schedulers, and workloads to detect subtle degradation and connect symptoms to affected jobs. Supervised models flag known patterns; unsupervised peer comparisons and multivariate analysis can surface anomalies that static thresholds miss. Its knowledge graph keeps infrastructure and job relationships current, while topology-aware correlation groups dependent alerts into incidents. Sherlock lets operators query live evidence conversationally, and an MCP server exposes that context to external agentic tools; mitigations may require human approval or follow local automation policy. Examples in the source describe identifying a degraded NVLink, rare SuperNIC-path hardware errors, failing NVMe storage, and switch congestion. The post says event reduction applies to supported scenarios and that correlated evidence still needs operator interpretation. It cites internal testing for lower alert noise and repair time, without supplying general performance guarantees.

### Source excerpt

AI Factories are some of the most technologically complex operational environments ever created. Hundreds of thousands of GPUs, NVIDIA Spectrum-X Ethernet...

## Data Center Telemetry in the AI Era

DevFeed: [Data Center Telemetry in the AI Era](<https://devfeed.tech/articles/data-center-telemetry-in-the-ai-era-97279.md>)

Original publisher: [Read original article](<https://www.influxdata.com/blog/data-center-telemetry/>)

Author: Charles Mahler (InfluxData)

Published: 2026-10-23T08:00:00Z

Content type: article

Language: en

Sources: [InfluxData](<https://devfeed.tech/sources/influxdata.md>)

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

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-workloads](<https://devfeed.tech/tags/ai-workloads.md>), [automation](<https://devfeed.tech/tags/automation.md>), [data-center](<https://devfeed.tech/tags/data-center.md>), [developer](<https://devfeed.tech/tags/developer.md>), [power](<https://devfeed.tech/tags/power.md>), [telemetry](<https://devfeed.tech/tags/telemetry.md>), [use-cases](<https://devfeed.tech/tags/use-cases.md>)

### AI overview

AI data center telemetry needs to align facility, hardware, network, and job data because cooling drift, power swings, and slow or failing GPUs can appear in different systems. The excerpt proposes collecting across those layers on a shared clock and in one store, with sampling matched to the signal: around one second for GPU and rack power where needed, and 15-60 seconds for many other signals. Fast protective controls should remain local; telemetry supports actions from seconds to minutes and longer-term planning. Telegraf collects through existing exporters and read-only facility paths, while Telegraf Controller centralizes agent configuration, staged rollouts, health checks, and resolution profiles. InfluxDB 3 is presented for high-cardinality time series and long-term history in object storage, with SQL or InfluxQL queries and optional self-hosting. The article's examples include correlating power swings with training jobs, spotting GPU and cooling issues, and reviewing SLA or billing events. The excerpt also notes tradeoffs: high-resolution collection costs more, facility networks are segmented, and the free Controller tier covers up to 100 agents and 20 configurations.

### Source excerpt

A decade ago, a typical enterprise rack drew five to ten kilowatts and was cooled by air. Today, reference designs for liquid-cooled AI racks run up to 132 kW per rack, and a large training cluster can swing a facility's power draw by megawatts in a few seconds. More than half of operators in Uptime Institute's 2026 survey built or upgraded high-density space in the past year, and 70% plan to in the next year as well. The entire industry is changing as AI-focused workloads become more common. Telemetry is how operators see what all that hardware is doing. Most data centers already collect a great deal of it. The trouble is that this data lives in different systems, at different resolutions, with different clocks, owned by different teams. In a data center running AI workloads, where a cooling problem can surface as a slow training job, that fragmentation makes problems difficult to trace. Handling this new set of problems requires a different approach and different tools. What is data center telemetry? Data center telemetry is the continuous, time-stamped measurement of every layer of a data center, from the utility feed to the training job. It is used for three things: Monitoring: dashboards, threshold alerts, and on-call response Automation: actions taken on the data, like draining an unhealthy node or flagging a rack whose coolant flow is drifting Planning: capacity, power provisioning, forecasting, hardware replacement, and contract evidence, all built on history It helps to think of telemetry in seven layers. The first three belong to the facility (operational technology, or OT). The last four belong to the IT stack. Layer Signals Sources Power Utility feeds, switchgear, UPS state of charge and battery health, and PDU branch-circuit load EPMS, power meters, SNMP, Modbus Cooling Supply and return temperatures, chilled water, CDU flow, pressure, leak detection, coolant conductivity BMS, SCADA Environment Rack inlet temperature, humidity, differential pressure, ai

## Workload-Aware Monitoring for AI-RAN

DevFeed: [Workload-Aware Monitoring for AI-RAN](<https://devfeed.tech/articles/workload-aware-monitoring-for-ai-ran-97343.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.31627>)

Author: Babak Azkaei, Georgios Exarchakos, Kishor Chandra Joshi

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

Content type: article

Language: en

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

Topics: [Monitoring](<https://devfeed.tech/topics/monitoring.md>), [gpu monitoring](<https://devfeed.tech/topics/gpu-monitoring.md>), [Performance Testing](<https://devfeed.tech/topics/performance-testing.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [cpu](<https://devfeed.tech/tags/cpu.md>), [cs-ni](<https://devfeed.tech/tags/cs-ni.md>), [cs-sy](<https://devfeed.tech/tags/cs-sy.md>), [dispatch](<https://devfeed.tech/tags/dispatch.md>), [eess-sy](<https://devfeed.tech/tags/eess-sy.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [monitoring](<https://devfeed.tech/tags/monitoring.md>)

### AI overview

Workload-aware monitoring compares radio timings only after associating workload descriptors with processing paths, timing boundaries, and operating context. Across 16 over-the-air downlink experiments, dispatching transport blocks below 4,096 bytes to the CPU and larger blocks to the GPU produced lower mean host-callback durations than direct GPU dispatch: 29.66% lower without a synthetic GPU co-runner and 22.93% lower with one. Grouping attributed the observed separation to small blocks; large-block callback-duration ranges overlapped. The results suggest complementary CPU/GPU dispatch as a candidate, not a demonstrated general improvement.

### Source excerpt

arXiv:2610.31627v1 Announce Type: new Abstract: AI-RAN platforms are expected to share CPU and GPU resources between radio workloads and AI services. Aggregate telemetry reports operating conditions, but timing changes may reflect different workloads, processors, or measurement boundaries. We propose workload-aware monitoring that associates radio-work descriptors with processing paths, timing boundaries, and operating context before comparing like with like. In 16 over-the-air downlink experiments, hybrid dispatch (CPU for transport blocks (TBs) below 4,096 bytes and GPU otherwise) has 29.66% and 22.93% lower mean host-callback durations than direct GPU dispatch, without and with a synthetic GPU co-runner, respectively. Grouping locates the separation in small TBs, while large-TB callback-duration ranges overlap. These observations identify a candidate for complementary CPU/GPU use.

## CoreWeave Leads the Charge in AI Infrastructure Efficiency, with up to 20% Higher GPU Cluster Performance than Alternative Solutions

DevFeed: [CoreWeave Leads the Charge in AI Infrastructure Efficiency, with up to 20% Higher GPU Cluster Performance than Alternative Solutions](<https://devfeed.tech/articles/coreweave-leads-the-charge-in-ai-infrastructure-efficiency-with-up-to-20-higher-gpu-cluster-performance-than-alternative-solutions-96114.md>)

Original publisher: [Read original article](<https://wf.coreweave.com/blog/coreweave-leads-the-charge-in-ai-infrastructure-efficiency-with-up-to-20-higher-gpu-cluster-performance-than-alternative-solutions>)

Author: CoreWeave

Published: 2026-10-22T16:27:39Z

Content type: article

Language: en

Sources: [Coreweave Blog Posts](<https://devfeed.tech/sources/coreweave-blog-posts.md>)

Topics: [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [performance-engineering](<https://devfeed.tech/topics/performance-engineering.md>), [High-Performance Computing](<https://devfeed.tech/topics/high-performance-computing.md>), [gpu monitoring](<https://devfeed.tech/topics/gpu-monitoring.md>), [infrastructure monitoring](<https://devfeed.tech/topics/infrastructure-monitoring.md>), [Hopper](<https://devfeed.tech/topics/hopper.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-infrastructure](<https://devfeed.tech/tags/ai-infrastructure.md>), [ai-infrastructure-efficiency](<https://devfeed.tech/tags/ai-infrastructure-efficiency.md>), [cluster](<https://devfeed.tech/tags/cluster.md>), [content-type](<https://devfeed.tech/tags/content-type.md>), [coreweave](<https://devfeed.tech/tags/coreweave.md>), [efficiency](<https://devfeed.tech/tags/efficiency.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [infrastructure](<https://devfeed.tech/tags/infrastructure.md>), [performance](<https://devfeed.tech/tags/performance.md>)

### AI overview

CoreWeave reports MFU above 50% on NVIDIA Hopper GPUs and results up to 20% higher than public foundation-model training benchmarks, while preliminary NVIDIA DGX Cloud recipe tests were on par with NVIDIA's reference system. In its 128-H100 comparisons, CoreWeave reports 49.2% MFU versus MosaicML's 41.85% for a 30B model with matched training settings, and 51.9% versus Aleph Alpha's 40.43% for a 30B model; the latter used A100s, so the comparison is qualified by different GPU types. The company attributes performance to an AI-focused stack including InfiniBand networking, bare-metal Kubernetes, fast object storage, cluster health monitoring and topology-aware scheduling. These are company-reported results; the text also describes the DGX testing as preliminary.

### Source excerpt

CoreWeave boosts AI infrastructure efficiency with up to 20% higher GPU cluster performance compared to alternatives, narrowing the AI efficiency gap with optimized, purpose-built solutions.

## Engineering Confidence: 4 Ways to Validate and Strengthen Your AI Infrastructure Resilience

DevFeed: [Engineering Confidence: 4 Ways to Validate and Strengthen Your AI Infrastructure Resilience](<https://devfeed.tech/articles/engineering-confidence-4-ways-to-validate-and-strengthen-your-ai-infrastructure-resilience-96125.md>)

Original publisher: [Read original article](<https://wf.coreweave.com/blog/engineering-confidence-4-ways-to-validate-and-strengthen-your-ai-infrastructure-resilience>)

Author: CoreWeave

Published: 2026-10-22T16:27:39Z

Content type: tutorial

Language: en

Sources: [Coreweave Blog Posts](<https://devfeed.tech/sources/coreweave-blog-posts.md>)

Topics: [Resilience](<https://devfeed.tech/topics/resilience.md>), [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [Production Engineering](<https://devfeed.tech/topics/production-engineering.md>), [observability](<https://devfeed.tech/topics/observability.md>), [gpu monitoring](<https://devfeed.tech/topics/gpu-monitoring.md>), [Chaos Engineering](<https://devfeed.tech/topics/chaos-engineering.md>), [Test automation](<https://devfeed.tech/topics/test-automation.md>), [Learning](<https://devfeed.tech/topics/learning.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-infrastructure](<https://devfeed.tech/tags/ai-infrastructure.md>), [ai-native](<https://devfeed.tech/tags/ai-native.md>), [ai-native-observability](<https://devfeed.tech/tags/ai-native-observability.md>), [ai-observability](<https://devfeed.tech/tags/ai-observability.md>), [alerts](<https://devfeed.tech/tags/alerts.md>), [api](<https://devfeed.tech/tags/api.md>), [application](<https://devfeed.tech/tags/application.md>), [applications](<https://devfeed.tech/tags/applications.md>), [automation](<https://devfeed.tech/tags/automation.md>), [content-type](<https://devfeed.tech/tags/content-type.md>), [engineering](<https://devfeed.tech/tags/engineering.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [infrastructure](<https://devfeed.tech/tags/infrastructure.md>), [observability](<https://devfeed.tech/tags/observability.md>), [practical](<https://devfeed.tech/tags/practical.md>), [resilience](<https://devfeed.tech/tags/resilience.md>), [scale](<https://devfeed.tech/tags/scale.md>), [strategies](<https://devfeed.tech/tags/strategies.md>), [stress](<https://devfeed.tech/tags/stress.md>), [testing](<https://devfeed.tech/tags/testing.md>)

### AI overview

For large GPU training jobs, fast recovery and failure diagnosis help sustain throughput and control costs. The article recommends designing training code to recover from node failures through checkpointing, reproducible containerized environments, job resubmission, and suitable parallelism; instrumenting hardware and application metrics to distinguish infrastructure faults from software errors; automating health checks, hang detection, alerts, and bounded restart policies; and stress-testing with NCCL tests, GPU burn-in, controlled node failures, and system tracing. It reports CoreWeave internal tests found automated recovery approximately three times faster than manual intervention and says ETTR can stay above 96%; these are provider-reported figures. The article advises asking infrastructure providers about their stress testing and suggests running NCCL tests before long jobs.

### Source excerpt

Learn four practical strategies to validate and strengthen AI infrastructure resilience, from observability and automation to stress testing at GPU scale.

## 📉 98% GPU Utilization and a 6x Speedup Left on the Table

DevFeed: [📉 98% GPU Utilization and a 6x Speedup Left on the Table](<https://devfeed.tech/articles/98-gpu-utilization-and-a-6x-speedup-left-on-the-table-91251.md>)

Original publisher: [Read original article](<https://read.bytesizeddesign.com/p/98-percent-gpu-utilization-6x-speedup>)

Author: Byte-Sized Design

Published: 2026-10-20T16:16:53Z

Content type: article

Language: en

Sources: [Byte-Sized Design](<https://devfeed.tech/sources/byte-sized-design.md>)

Topics: [gpu monitoring](<https://devfeed.tech/topics/gpu-monitoring.md>), [qwen](<https://devfeed.tech/topics/qwen.md>), [kernels](<https://devfeed.tech/topics/kernels.md>), [pip](<https://devfeed.tech/topics/pip.md>)

Tags: [activation](<https://devfeed.tech/tags/activation.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [attention](<https://devfeed.tech/tags/attention.md>), [batch](<https://devfeed.tech/tags/batch.md>), [bias](<https://devfeed.tech/tags/bias.md>), [bug](<https://devfeed.tech/tags/bug.md>), [chat](<https://devfeed.tech/tags/chat.md>), [community](<https://devfeed.tech/tags/community.md>), [compute](<https://devfeed.tech/tags/compute.md>), [container](<https://devfeed.tech/tags/container.md>), [cpu](<https://devfeed.tech/tags/cpu.md>), [databricks](<https://devfeed.tech/tags/databricks.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [gpu-utilization](<https://devfeed.tech/tags/gpu-utilization.md>), [kernels](<https://devfeed.tech/tags/kernels.md>), [qwen](<https://devfeed.tech/tags/qwen.md>)

### AI overview

A Qwen 27B fine-tune took 7h44m despite 98% GPU utilization; installing the model's supported `flash-linear-attention` and `causal-conv1d` kernels reduced runtime to 2h05m, then packing examples instead of padding them cut it to 1h17m. A PyTorch trace revealed 269,448 GPU launches per step and CPU dispatch slower than GPU execution; after kernel installation, launches fell 15-fold. Packing uses boundary-reset `position_ids` to preserve attention between examples and reduced batches by about 36% on this dataset, though it makes the input pipeline more complex and mask bugs could silently harm quality. The article recommends tracing a step and tracking throughput alongside utilization, since utilization stayed at 98% throughout. It does not report final tokens per second, GPU count, or CPU/GPU timing after the kernel fix.

### Source excerpt

A Qwen 27B fine-tune showed 98% GPU utilization and ran 6x slower than it should. Two fixes from Databricks: install the fused kernels, pack the batches.

## Orchestrating GPU workloads on Runpod with dstack

DevFeed: [Orchestrating GPU workloads on Runpod with dstack](<https://devfeed.tech/articles/orchestrating-gpu-workloads-on-runpod-with-dstack-90257.md>)

Original publisher: [Read original article](<https://www.runpod.io/blog/orchestrating-gpu-workloads-on-runpod-with-dstack>)

Author: Knarik Avanesyan

Published: 2026-10-19T20:52:38Z

Content type: article

Language: en

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

Topics: [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [gpu monitoring](<https://devfeed.tech/topics/gpu-monitoring.md>), [Open Source](<https://devfeed.tech/topics/open-source.md>)

Tags: [automation](<https://devfeed.tech/tags/automation.md>), [cost-savings](<https://devfeed.tech/tags/cost-savings.md>), [provisioning](<https://devfeed.tech/tags/provisioning.md>)

### AI overview

The article explains how dstack orchestrates GPU workloads on Runpod using declarative configuration and a shared control plane. It describes automatic shutdown, utilization-based termination, spot-instance policies, and multi-cloud routing as ways to reduce wasted compute and manage costs across ML development, training, and inference.

### Source excerpt

dstack is an open-source, GPU-native orchestrator that automates provisioning, scaling, and policies for ML teams, helping cut 3-7x GPU waste while.

## Why GPU Cloud Observability Is Harder Than It Looks

DevFeed: [Why GPU Cloud Observability Is Harder Than It Looks](<https://devfeed.tech/articles/why-gpu-cloud-observability-is-harder-than-it-looks-90154.md>)

Original publisher: [Read original article](<https://hackernoon.com/why-gpu-cloud-observability-is-harder-than-it-looks?source=rss>)

Author: Anil Sharma

Published: 2026-10-19T09:27:04Z

Content type: article

Language: en

Sources: [HackerNoon](<https://devfeed.tech/sources/hackernoon.md>)

Topics: [observability](<https://devfeed.tech/topics/observability.md>), [gpu monitoring](<https://devfeed.tech/topics/gpu-monitoring.md>), [Multi-GPU](<https://devfeed.tech/topics/multi-gpu.md>), [clean energy data centers](<https://devfeed.tech/topics/clean-energy-data-centers.md>)

Tags: [ai-data-centers](<https://devfeed.tech/tags/ai-data-centers.md>), [ai-observability](<https://devfeed.tech/tags/ai-observability.md>), [ai-workloads](<https://devfeed.tech/tags/ai-workloads.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [cloud-operations](<https://devfeed.tech/tags/cloud-operations.md>), [distributed-ai-training](<https://devfeed.tech/tags/distributed-ai-training.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [gpu-cloud](<https://devfeed.tech/tags/gpu-cloud.md>), [gpu-cluster-monitoring](<https://devfeed.tech/tags/gpu-cluster-monitoring.md>), [gpu-infrastructure](<https://devfeed.tech/tags/gpu-infrastructure.md>), [networks](<https://devfeed.tech/tags/networks.md>), [observability](<https://devfeed.tech/tags/observability.md>), [private-cloud-compute-ai](<https://devfeed.tech/tags/private-cloud-compute-ai.md>), [storage](<https://devfeed.tech/tags/storage.md>), [topology-aware-observability](<https://devfeed.tech/tags/topology-aware-observability.md>)

### AI overview

The article explains why GPU cloud operations need observability that connects physical infrastructure, changing job allocations, and telemetry. It outlines a topology-aware approach for investigating slow workloads, capacity and placement problems, and customer impact while preserving uncertainty about causes and using safeguards for remediation.

### Source excerpt

Why is GPU cloud observability so hard? Learn how topology-aware observability connects GPUs, networks, storage, and AI workloads to reveal failures and waste.

## GPU Cloud vs On-Prem Clusters: A Cost Breakdown

DevFeed: [GPU Cloud vs On-Prem Clusters: A Cost Breakdown](<https://devfeed.tech/articles/gpu-cloud-vs-on-prem-clusters-a-cost-breakdown-84641.md>)

Original publisher: [Read original article](<https://www.runpod.io/blog/gpu-cloud-vs-on-prem-cost-savings>)

Author: James Sandy

Published: 2026-10-10T23:20:15Z

Content type: comparison

Language: en

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

Topics: [cloud-infrastructure](<https://devfeed.tech/topics/cloud-infrastructure.md>), [cloud cost management](<https://devfeed.tech/topics/cloud-cost-management.md>), [gpu monitoring](<https://devfeed.tech/topics/gpu-monitoring.md>), [Serverless](<https://devfeed.tech/topics/serverless.md>)

Tags: [cost](<https://devfeed.tech/tags/cost.md>), [data-center](<https://devfeed.tech/tags/data-center.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [gpu-cloud](<https://devfeed.tech/tags/gpu-cloud.md>), [gpu-clusters](<https://devfeed.tech/tags/gpu-clusters.md>), [on-prem](<https://devfeed.tech/tags/on-prem.md>)

### AI overview

The article compares on-prem GPU clusters with Runpod's cloud GPU service for AI and machine learning workloads. It says a three-year comparison for four NVIDIA A100 GPUs found cloud deployment could save over $124,000, citing lower upfront costs, pay-as-you-go usage, and reduced infrastructure maintenance. It also notes that dedicated clusters have fixed monthly costs and that workload needs should guide the choice.

### Source excerpt

We crunched the numbers: deploying 4x A100s on Runpod's GPU cloud can save over $124,000 versus an on-prem cluster across 3 years. Learn why cloud beats.

## Researchers found a flaw in NVIDIA DCGM Exporter that could let attackers crash GPU monitoring

DevFeed: [Researchers found a flaw in NVIDIA DCGM Exporter that could let attackers crash GPU monitoring](<https://devfeed.tech/articles/high-severity-nvidia-vulnerability-lets-unauthenticated-attackers-crash-gpu-monitoring-84056.md>)

Original publisher: [Read original article](<https://www.helpnetsecurity.com/2026/10/09/nvidia-dcgm-exporter-vulnerability-cve-2026-47483/>)

Author: Sinisa Markovic

Published: 2026-10-09T12:26:42Z

Content type: news

Language: en

Sources: [Help Net Security](<https://devfeed.tech/sources/help-net-security.md>)

Topics: [gpu monitoring](<https://devfeed.tech/topics/gpu-monitoring.md>), [GPU](<https://devfeed.tech/topics/gpu.md>), [Monitoring](<https://devfeed.tech/topics/monitoring.md>), [Prometheus](<https://devfeed.tech/topics/prometheus.md>), [AI Networking](<https://devfeed.tech/topics/ai-networking.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-infrastructure](<https://devfeed.tech/tags/ai-infrastructure.md>), [cloud](<https://devfeed.tech/tags/cloud.md>), [cve](<https://devfeed.tech/tags/cve.md>), [don-t-miss](<https://devfeed.tech/tags/don-t-miss.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [gpu-monitoring](<https://devfeed.tech/tags/gpu-monitoring.md>), [hot-stuff](<https://devfeed.tech/tags/hot-stuff.md>), [news](<https://devfeed.tech/tags/news.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [security](<https://devfeed.tech/tags/security.md>), [vulnerability](<https://devfeed.tech/tags/vulnerability.md>)

### AI overview

Researchers found more than 2,000 internet-exposed servers running NVIDIA DCGM Exporter and reported that unauthenticated requests to its profiling endpoint could exhaust memory and crash the exporter. The outage could reduce GPU monitoring visibility and potentially affect workloads on the host. Users should upgrade to version 4.8.2 or later, disable profiling unless needed, and restrict monitoring services to private networks with access controls.

### Source excerpt

Hundreds of internet-exposed graphics processing unit (GPU) servers were open to a high-severity flaw in NVIDIA's DCGM Exporter (CVE-2026-47483) that lets unauthenticated attackers crash the monitoring service and may disrupt AI workloads, according to Lava. Lava reported the flaw to NVIDIA, which rated it 8.2 on the CVSS scale and published a security bulletin on July 28, 2026. GPU servers are computers built around GPUs and are used for AI, machine learning and scientific computing ... More -> The post High-severity NVIDIA vulnerability lets unauthenticated attackers crash GPU monitoring appeared first on Help Net Security.

## Cast.ai report finds average GPU utilization at 5% across 23,000 Kubernetes clusters

DevFeed: [Cast.ai report finds average GPU utilization at 5% across 23,000 Kubernetes clusters](<https://devfeed.tech/articles/your-kubernetes-gpus-are-sitting-98-idle-here-s-the-data-66621.md>)

Original publisher: [Read original article](<https://blog.devops.dev/your-kubernetes-gpus-are-sitting-98-idle-heres-the-data-4a589a809d05?source=rss----33f8b2d9a328---4>)

Author: Vortex 404

Published: 2026-10-07T10:39:16Z

Content type: news

Language: en

Sources: [DevOps.dev](<https://devfeed.tech/sources/devops-dev.md>)

Topics: [gpu monitoring](<https://devfeed.tech/topics/gpu-monitoring.md>), [Amazon Elastic Kubernetes Service](<https://devfeed.tech/topics/amazon-elastic-kubernetes-service.md>), [Azure Kubernetes Service](<https://devfeed.tech/topics/azure-kubernetes-service.md>), [Gke](<https://devfeed.tech/topics/gke.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [aks](<https://devfeed.tech/tags/aks.md>), [artificial-intelligence](<https://devfeed.tech/tags/artificial-intelligence.md>), [aws](<https://devfeed.tech/tags/aws.md>), [azure](<https://devfeed.tech/tags/azure.md>), [cloud](<https://devfeed.tech/tags/cloud.md>), [cloud-computing](<https://devfeed.tech/tags/cloud-computing.md>), [devops](<https://devfeed.tech/tags/devops.md>), [eks](<https://devfeed.tech/tags/eks.md>), [gcp](<https://devfeed.tech/tags/gcp.md>), [gke](<https://devfeed.tech/tags/gke.md>), [google-cloud](<https://devfeed.tech/tags/google-cloud.md>), [gpus](<https://devfeed.tech/tags/gpus.md>), [infrastructure](<https://devfeed.tech/tags/infrastructure.md>), [kubernetes](<https://devfeed.tech/tags/kubernetes.md>)

### AI overview

Cast.ai's 2026 State of Kubernetes Optimization Report analyzes resource use across 23,000 Kubernetes clusters on AWS, Azure, and GCP from 2025 through April 2026. It reports average GPU utilization of 5%, with rates of 2% on AKS, 5% on EKS, and 6% on GKE.

### Source excerpt

23,000 real clusters, analyzed across AWS, Azure, and GCP -- and GPU utilization is getting worse, not better Continue reading on DevOps.dev "

## NVIDIA H100 GPU Benchmark Results: What We Learned From Large-Scale GPU Testing

DevFeed: [NVIDIA H100 GPU Benchmark Results: What We Learned From Large-Scale GPU Testing](<https://devfeed.tech/articles/nvidia-h100-gpu-benchmark-results-what-we-learned-from-large-scale-gpu-testing-78011.md>)

Original publisher: [Read original article](<https://wf.coreweave.com/blog/nvidia-h100-gpu-benchmark-results-what-we-learned-from-large-scale-gpu-testing>)

Author: CoreWeave

Published: 2026-10-02T14:12:41Z

Content type: article

Language: en

Sources: [Coreweave Blog Posts](<https://devfeed.tech/sources/coreweave-blog-posts.md>)

Topics: [gpu monitoring](<https://devfeed.tech/topics/gpu-monitoring.md>), [AI Networking](<https://devfeed.tech/topics/ai-networking.md>), [Multi-GPU](<https://devfeed.tech/topics/multi-gpu.md>), [Benchmark](<https://devfeed.tech/topics/benchmark.md>), [Nvidia](<https://devfeed.tech/topics/nvidia.md>), [NVIDIA Blackwell](<https://devfeed.tech/topics/nvidia-blackwell.md>)

Tags: [architecture](<https://devfeed.tech/tags/architecture.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [benchmarking](<https://devfeed.tech/tags/benchmarking.md>), [bluefield-dpu](<https://devfeed.tech/tags/bluefield-dpu.md>), [content-type](<https://devfeed.tech/tags/content-type.md>), [debugging](<https://devfeed.tech/tags/debugging.md>), [ethernet](<https://devfeed.tech/tags/ethernet.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [gpu-clusters](<https://devfeed.tech/tags/gpu-clusters.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [performance](<https://devfeed.tech/tags/performance.md>), [reliability](<https://devfeed.tech/tags/reliability.md>)

### AI overview

A six-week benchmark of production-scale AI training used up to 1,024 NVIDIA H100 GPUs to test infrastructure performance and reliability. It reports 51-52% model FLOPs utilization, a 3.66-day mean time to failure at 1,024 GPUs, and faster checkpointing, alongside details about its hardware, networking, orchestration, and storage setup.

### Source excerpt

Discover how NVIDIA H100 benchmarks prove GPU clusters can achieve higher reliability, performance, and MFU for large-scale AI training.

## hw-monitor 0.6 Released With Per-Process GPU Usage & Memory Reporting

DevFeed: [hw-monitor 0.6 Released With Per-Process GPU Usage & Memory Reporting](<https://devfeed.tech/articles/hw-monitor-0-6-released-with-per-process-gpu-usage-memory-reporting-61950.md>)

Original publisher: [Read original article](<https://www.phoronix.com/news/hw-monitor-0.6>)

Author: Michael Larabel

Published: 2026-09-29T14:23:58Z

Content type: release

Language: en

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

Topics: [gpu monitoring](<https://devfeed.tech/topics/gpu-monitoring.md>), [Intel Iris Xe](<https://devfeed.tech/topics/intel-iris-xe.md>)

Tags: [bug-fixes](<https://devfeed.tech/tags/bug-fixes.md>), [desktop-linux](<https://devfeed.tech/tags/desktop-linux.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [gpu-memory](<https://devfeed.tech/tags/gpu-memory.md>), [hardware](<https://devfeed.tech/tags/hardware.md>), [linux](<https://devfeed.tech/tags/linux.md>), [linux-benchmarking](<https://devfeed.tech/tags/linux-benchmarking.md>), [linux-desktop](<https://devfeed.tech/tags/linux-desktop.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>), [metrics](<https://devfeed.tech/tags/metrics.md>), [monitoring](<https://devfeed.tech/tags/monitoring.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>), [ubuntu-benchmarks](<https://devfeed.tech/tags/ubuntu-benchmarks.md>), [ubuntu-hardware](<https://devfeed.tech/tags/ubuntu-hardware.md>), [ui](<https://devfeed.tech/tags/ui.md>)

### AI overview

The hw-monitor 0.6 release adds per-process GPU usage and memory reporting for supported drivers, a tab for XDG autostart applications, sensor UI improvements, theme enhancements and bug fixes. Version 0.6.1 followed with additional fixes.

### Source excerpt

For those after in-depth hardware and system monitoring support from the Linux desktop, hw-monitor continues to advance as one of the modern options. Released on Monday was hw-monitor 0.6 with per-process GPU reporting metrics when using capable DRM drivers like the AMDGPU, Intel i915 and Xe, and Nouveau drivers...

[Next page](<https://devfeed.tech/topics/gpu-monitoring.md?cursor=WyIyMDI2LTA5LTI5VDE0OjIzOjU4KzAwOjAwIiwgIjVhOWViMGQ4LTNlYjktNDY3Mi1iMmFmLTFlMjZjNjQ4NDJiOCJd>)