NVIDIA Vera CPU performance shows impressive gains over its x86 rivals

NVIDIA's Vera CPU is built for the agentic AI era, and this week the company has lifted the lid on its internal performance data. And there's a lot to get through with performance data covering memory bandwidth and latency, application-specific performance, IPC performance for Vera's Olympus cores, as well as agentic AI performance. Most comparisons pit Vera against AMD's EPYC Turin 'Zen 5' x86 CPU, with Vera convincingly coming out on top.

8

As for the Arm-based Vera CPU's specs, here's a quick refresher of its impressive architecture. The monolithic die features 88 custom Olympus cores with 176 threads and 164MB of L3 Cache. This is paired with a whopping 1.5TB of SOCAMM LPDDR5X 9600 MT/s memory with 1.2TB/s of memory bandwidth. And when you add in low-latency comms between CPU clusters, shared cache, NVLink-C2C integration for 1.8 TB/s of CPU-to-GPU bandwidth, and high-bandwidth memory, it's an impressive engineering feat.

When it comes to the Vera CPU's performance, NVIDIA notes that the monolithic design and custom architecture are two reasons why it can achieve a level of performance not possible on current chiplet-based x86 chips. And when it comes to one of the most notable CPU benchmarks, IPC (Instructions Per Cycle), Vera offers up to a 1.9X improvement over AMD EPYC Turin when testing various AI workflows.

8

This lead increases to up to 2.3X when looking at Branch Prediction, another staple when it comes to CPU performance. And with fewer wasted cycles, Vera can achieve 3.5X higher Taken-branches per cycle compared to x86.

8

"Workloads such as agent runtimes, interpreters, compilers, graph analytics, and data-processing frameworks often combine large instruction footprints with frequent control-flow changes, which can leave execution resources underutilized," NVIDIA explains. "Olympus addresses these challenges with advanced branch prediction, high-bandwidth instruction fetch, and a 10-wide decode engine that delivers more instructions to the core each cycle. At the center of this approach is the Olympus branch prediction subsystem, which includes a neural branch predictor designed to improve accuracy on difficult, statistically biased branch patterns."

Overall, one of the key reasons Vera outperforms its EPYC Turin 'Zen 5' x86 rival comes down to latency, memory, and core-to-core bandwidth. NVIDIA notes that Vera delivers up to 3X the bandwidth, with 40% lower latency, so it doesn't hit the same bandwidth or memory wall as x86 solutions. So even though EPYC Turin has 128 cores, Vera's 88 cores have significantly more bandwidth to play with, which is extremely important for intensive agentic AI workloads.

Frequently Asked Questions

TweakBot answers common questions about this news using TweakTown's own coverage from this page and related content from our archive. All answers are generated from TweakTown content and not outside sources. Tap a question to reveal the answer, or type your own below related to this content.

Question #1

How does Vera's IPC compare to AMD EPYC Turin across AI workloads and which workloads showed the largest IPC gains?

NVIDIA reports Vera delivers up to a 1.9X IPC improvement versus AMD EPYC Turin across various AI workflows. NVIDIA says the largest gains appear on workloads with large instruction footprints and frequent control-flow changes, specifically agent runtimes, interpreters, compilers, graph analytics, and data-processing frameworks, thanks to Olympus core features like advanced branch prediction and a 10-wide decode engine.

Answered

Question #2

What Olympus core features (like branch prediction and decode width) contribute to Vera's higher instruction throughput?

Click to reveal answer

Question #3

How does Vera's memory bandwidth and latency compare to EPYC Turin and why does that matter for agentic AI workloads?

Vera provides about 1.2 TB/s of memory bandwidth and, according to NVIDIA, delivers up to 3X the bandwidth and about 40% lower memory latency versus AMD EPYC Turin. This higher bandwidth and lower latency prevent the bandwidth and memory wall that chiplet x86 designs hit, giving Vera more effective per-core data throughput. That matters for agentic AI workloads because those workloads combine large instruction footprints, frequent control-flow changes, and heavy data movement, so higher bandwidth and lower latency keep execution resources utilized and improve runtime performance.

Answered

Question #4

How do Vera's core count and memory bandwidth trade off against EPYC Turin's higher core count for AI data-processing tasks?

Vera uses 88 custom Olympus cores versus EPYC Turin's 128 cores, but pairs them with much higher memory bandwidth and lower latency , 1.2 TB/s of LPDDR5X bandwidth and NVIDIA claims up to 3X the bandwidth with 40% lower latency. For AI data-processing tasks this tradeoff favors Vera: its greater per-core bandwidth, lower latency, and higher IPC yield better real-world results (NVIDIA reports 1.5X faster data processing and 1.8X faster Python), so the lower core count is offset by substantially higher memory and execution efficiency.

Answered

Have a question about this content not listed here? Ask below and TweakBot will answer it.

Your questionAsk

Answering your question×

TweakBot is analyzing your question…

Could not get an answer

TweakBot is generating an answer

Read More

Was TweakBot helpful?Thanks for the feedback!

And when it comes to Agentic AI performance, which is what NVIDIA's Vera CPU was purpose-built for, Vera delivers 1.8X faster performance in Python with 1.5X faster Data Processing. NVIDIA says that Vera could become the leading CPU supplied in 2026, which begins to make sense, especially with companies like OpenAI, Anthropic, SpaceX, and Perplexity on board as early adopters.