Nvidia's Vera CPU scored 1.7× SPECrate integer 2026 vs AMD Epyc 9755. First custom core for agentic AI, H2 2026 release.
Nvidia's Vera CPU scored an estimated 1.7× the SPECrate integer 2026 throughput of AMD's Epyc 9755 in dual-socket configurations, per Nvidia's white paper. Vera is Nvidia's first CPU with a custom core design, the Olympus core, built for agentic AI workloads.
Key facts
- Vera scored 1.7× SPECrate integer 2026 vs AMD Epyc 9755.
- First Nvidia CPU with custom Olympus core design.
- General availability targeted for H2 2026.
- Grace standalone deployments at Meta noted as success.
- Vera targets agentic AI workloads, not legacy cloud.
Nvidia has slowly revealed more details about Vera as it ramps into general availability, which is on track for the back half of this year. Now, we have a full picture of the chip. Nvidia shared its Vera white paper, along with unofficial SPEC CPU 2026 results comparing Vera to AMD's Turin-based Epyc 9755 According to Tom's Hardware.
The SPECrate integer 2026 results
Nvidia tested the SPECrate integer suite, which focuses on system throughput with integer-based workloads. Both chips ran in dual-socket configurations. The run is unofficial because Vera was tested in a reference system before broad availability. Nvidia reported an estimated 1.7× throughput advantage over AMD's Epyc 9755, a Turin-based chip with 128 cores per socket. AMD has not yet reported official SPEC CPU 2026 results for its upcoming Venice chips.
Olympus core architecture
Vera is not a chip built to chip away at the market share of AMD and Intel in the cloud. It's built to grab market share in an expanding market, as hyperscalers look to widen AI infrastructure beyond legacy clouds. As such, it's designed in a much different way than Nvidia's x86 competitors, and it even holds some unique architectural design points compared to the swath of Arm-based designs. The Olympus core is Nvidia's first custom CPU core, optimized for agentic AI workloads that require high memory bandwidth and low latency for inference.
Context and competition
Although Grace has seen some success — most notably with Grace standalone deployments at Meta — Vera is Nvidia's first CPU with a custom core design. The server CPU market is exploding on the back of agentic AI demand, and Vera arrives at an ideal time. Nvidia competes with AMD and Intel in data center CPUs, but Vera targets the AI-specific segment where Nvidia's GPU ecosystem gives it a unique advantage. The Vera Rubin NVL72 rack system, announced earlier, pairs Vera with Nvidia's next-generation GPU architecture.
Key Takeaways
- Nvidia's Vera CPU scored 1.7× SPECrate integer 2026 vs AMD Epyc 9755.
- First custom core for agentic AI, H2 2026 release.
What to watch
Watch for official SPEC CPU 2026 submissions from Nvidia once Vera reaches general availability in H2 2026. AMD's Venice chips may counter with their own benchmarks. Also track Vera Rubin NVL72 rack system deployments at hyperscalers like Meta.
Source: tomshardware.com
[Updated 21 Jul via gn_gpu_cluster]
The Vera Rubin NVL72 system, pairing the Vera CPU with Nvidia's next-gen GPU, achieved a 10× token throughput uplift over the Blackwell GB200 NVL72, delivering 800,000 tokens/s versus 80,000 at the same 150MW power envelope, according to Nvidia's own benchmarks [per Wccftech]. This positions the platform as a leader in inference performance for agentic AI workloads.
[Updated 22 Jul via gn_gpu_cluster]
CoreWeave has published the first measured silicon performance stats for Vera Rubin NVL72, confirming the 10× token throughput uplift over Blackwell GB200 NVL72 at 150MW [per CoreWeave]. OpenAI is already deploying Vera Rubin racks in Q3 2026, and Nvidia's Engineering SuperLab demonstrated the system running OpenAI workloads with 800VDC power delivery [per Tom's Hardware]. The rack-scale supply chain spans 350+ factory sites across 30 countries, making it Nvidia's most mature production ramp to date [per HPCwire].
Originally published on gentic.news
Top comments (0)