--- title: NVIDIA Vera CPUs double throughput for Modal Sandboxes description: Modal benchmarked NVIDIA's new Arm-based Vera CPU against x86 (AMD Zen 5, Intel GNR) and Arm-based AWS Graviton 4/5 alternatives for agentic Sandbox workloads.... canonical: https://daily.dev/posts/nvidia-vera-cpus-double-throughput-for-modal-sandboxes-uz7r72ho5 twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: NVIDIA Vera CPUs double throughput for Modal Sandboxes | daily.dev og:description: Modal benchmarked NVIDIA's new Arm-based Vera CPU against x86 (AMD Zen 5, Intel GNR) and Arm-based AWS Graviton 4/5 alternatives for agentic Sandbox workloads.... og:url: https://daily.dev/posts/nvidia-vera-cpus-double-throughput-for-modal-sandboxes-uz7r72ho5 og:image: https://api.daily.dev/og/posts/UZ7r72Ho5.png og:image:alt: NVIDIA Vera CPUs double throughput for Modal Sandboxes og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # NVIDIA Vera CPUs double throughput for Modal Sandboxes **[Modal](https://daily.dev/sources/modal_labs)** · 37 min read · 0 upvotes · 0 comments ## Summary Modal benchmarked NVIDIA's new Arm-based Vera CPU against x86 (AMD Zen 5, Intel GNR) and Arm-based AWS Graviton 4/5 alternatives for agentic Sandbox workloads. Using a custom benchmark called Agentless Agent Bench, built from 50 stripped-down Terminal-Bench tasks, Vera showed only modest gains under low load but dramatically outperformed competitors under heavy concurrent load, roughly doubling throughput without the latency collapse typically seen when hosts become contended. This matters for Modal's multi-tenant sandbox scheduling, where workloads are spiky, unpredictable, and prone to noisy-neighbor issues, and where CPU time directly affects costs for background agents and reinforcement learning rollouts waiting on GPU-bound processes. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How does NVIDIA's Vera CPU perform compared to AMD Zen 5 and Intel Granite Rapids under heavy concurrent load? NVIDIA Vera sustains higher throughput than AMD Zen 5 and Intel Granite Rapids (GNR) as concurrency increases, reaching around 4,280 verified task completions per 60 seconds at roughly 99 logical CPUs in a sandbox benchmark, compared to about 1,470-1,570 for AMD Zen 5 and roughly 2,300-2,620 for Intel GNR at similar or higher CPU counts. Unlike the x86 chips, Vera's throughput does not collapse as contention rises. _Teams picking CPUs for containerized agent workloads can follow benchmark comparisons like this on daily.dev._ ### Why do AMD Zen 5 and Intel Granite Rapids throughput numbers drop as more logical CPUs are used in concurrent sandbox benchmarks? Throughput for AMD Zen 5 and Intel Granite Rapids declines past a certain concurrency point because the host becomes fully utilized and task latency suffers from contention, a classic throughput-versus-latency tradeoff in densely packed multi-tenant systems. For example, AMD Zen 5 peaks near 1,560-1,580 completions per 60 seconds around 26-79 logical CPUs, then drops and plateaus around 1,470-1,550 at higher core counts rather than continuing to scale. _Engineers tuning scheduler packing density for bursty workloads can track CPU benchmark findings like this on daily.dev._ ### What benchmark did Modal use to test CPU performance for agentic sandbox workloads? Modal built Agentless Agent Bench, a collection of 50 Terminal-Bench oracle solutions and verifiers stripped of network calls and prewarmed in concurrent Docker containers, isolating CPU-bound tasks like compilation, file grepping, git operations, and device emulation that agents and RL rollouts actually perform, measuring both per-task latency and concurrent throughput under varying container counts. _Developers evaluating compute options for agent workloads can follow benchmarking methodology discussions on daily.dev._ ## Similar posts on daily.dev - [NVIDIA Vera CPU Sets a New Standard for Agentic Workloads in AI Factories](https://daily.dev/posts/nvidia-vera-cpu-sets-a-new-standard-for-agentic-workloads-in-ai-factories-wvjlzbcah) · NVIDIA Developer · 0 upvotes · 0 comments - [AI Innovators Adopt NVIDIA Vera — Why Max Single-Threaded CPU at Scale Matters](https://daily.dev/posts/ai-innovators-adopt-nvidia-vera-why-max-single-threaded-cpu-at-scale-matters-rqsrrsh3y) · NVIDIA · 0 upvotes · 0 comments - [NVIDIA Launches Vera CPU, Purpose-Built for Agentic AI](https://daily.dev/posts/nvidia-launches-vera-cpu-purpose-built-for-agentic-ai-vcgvkme6v) · Hacker News · 1 upvotes · 0 comments --- [View this post on daily.dev](https://daily.dev/posts/nvidia-vera-cpus-double-throughput-for-modal-sandboxes-uz7r72ho5) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"NVIDIA Vera CPUs double throughput for Modal Sandboxes","url":"https://daily.dev/posts/nvidia-vera-cpus-double-throughput-for-modal-sandboxes-uz7r72ho5","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/nvidia-vera-cpus-double-throughput-for-modal-sandboxes-uz7r72ho5"},"datePublished":"2026-10-19T14:05:26.218Z","dateModified":"2026-10-26T17:46:19.455Z","description":"Modal benchmarked NVIDIA's new Arm-based Vera CPU against x86 (AMD Zen 5, Intel GNR) and Arm-based AWS Graviton 4/5 alternatives for agentic Sandbox workloads....","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/478ab40cebc267dfbab996cccfea8204?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/478ab40cebc267dfbab996cccfea8204?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Modal","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Modal","logo":"https://media.daily.dev/image/upload/s--HfKZHzC1--/f_auto/v1750946306/logos/modal_labs","url":"https://daily.dev/sources/modal_labs"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/nvidia-vera-cpus-double-throughput-for-modal-sandboxes-uz7r72ho5","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"","timeRequired":"PT37M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Modal","item":"https://daily.dev/sources/modal_labs"},{"@type":"ListItem","position":3,"name":"NVIDIA Vera CPUs double throughput for Modal Sandboxes"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/nvidia-vera-cpus-double-throughput-for-modal-sandboxes-uz7r72ho5#faq","mainEntity":[{"@type":"Question","name":"How does NVIDIA's Vera CPU perform compared to AMD Zen 5 and Intel Granite Rapids under heavy concurrent load?","acceptedAnswer":{"@type":"Answer","text":"NVIDIA Vera sustains higher throughput than AMD Zen 5 and Intel Granite Rapids (GNR) as concurrency increases, reaching around 4,280 verified task completions per 60 seconds at roughly 99 logical CPUs in a sandbox benchmark, compared to about 1,470-1,570 for AMD Zen 5 and roughly 2,300-2,620 for Intel GNR at similar or higher CPU counts. Unlike the x86 chips, Vera's throughput does not collapse as contention rises. Teams picking CPUs for containerized agent workloads can follow benchmark comparisons like this on daily.dev."}},{"@type":"Question","name":"Why do AMD Zen 5 and Intel Granite Rapids throughput numbers drop as more logical CPUs are used in concurrent sandbox benchmarks?","acceptedAnswer":{"@type":"Answer","text":"Throughput for AMD Zen 5 and Intel Granite Rapids declines past a certain concurrency point because the host becomes fully utilized and task latency suffers from contention, a classic throughput-versus-latency tradeoff in densely packed multi-tenant systems. For example, AMD Zen 5 peaks near 1,560-1,580 completions per 60 seconds around 26-79 logical CPUs, then drops and plateaus around 1,470-1,550 at higher core counts rather than continuing to scale. Engineers tuning scheduler packing density for bursty workloads can track CPU benchmark findings like this on daily.dev."}},{"@type":"Question","name":"What benchmark did Modal use to test CPU performance for agentic sandbox workloads?","acceptedAnswer":{"@type":"Answer","text":"Modal built Agentless Agent Bench, a collection of 50 Terminal-Bench oracle solutions and verifiers stripped of network calls and prewarmed in concurrent Docker containers, isolating CPU-bound tasks like compilation, file grepping, git operations, and device emulation that agents and RL rollouts actually perform, measuring both per-task latency and concurrent throughput under varying container counts. Developers evaluating compute options for agent workloads can follow benchmarking methodology discussions on daily.dev."}}]} ```