Vera Rubin
Vera Rubin is NVIDIA's next-generation AI server and GPU architecture, named after the astronomer Vera Rubin. Compared to the preceding GB200 series, it delivers a substantial increase in token throughput and is designed to reduce the cost of AI inference workloads.
Timeline (last 90 days)
NVIDIA's upcoming Vera Rubin platform relies heavily on FP64 emulation capability based on the Ozaki scheme, recovering double-precision results by splitting operations into lower-precision operations.
CoreWeave has not yet disclosed the price of Rubin during its limited-availability period, and Cognition's rate is also private, so the actual savings remain unknown
Cognition has become the first customer to run its coding agent Devin on NVIDIA Vera Rubin servers
The 4.8x figure refers to the ratio of total tokens processed per GPU per second at the point where single-user response speed is identical, not that answers are 4.8x faster or that 4.8x more users can be served simultaneously
Cognition claims that compared with the previous-generation servers, Rubin delivers up to a 4.8x improvement in tokens processed per second
The Vera Rubin servers on which Cognition runs Devin are rented from cloud provider CoreWeave
According to CoreWeave's comparison chart, Rubin holds approximately a 3x advantage over the GB200 at the slow end, climbing to more than 7x at the fast end
The appendix of this public benchmark only compared the GB200 and GB300, and did not include the upcoming Vera Rubin, which is about to enter mass production and shipment.
NVIDIA's Blackwell and Vera Rubin chips have rack thermal density requirements that exceed the upper limits of traditional air cooling, necessitating liquid cooling or even immersion cooling infrastructure
Vera Rubin is NVIDIA's next-generation computing platform following Blackwell.
5 more timeline events
All Facts (11)
NVIDIA's upcoming Vera Rubin platform relies heavily on FP64 emulation capability based on the Ozaki scheme, recovering double-precision results by splitting operations into lower-precision operations.
50%UnverifiedThe Vera Rubin servers on which Cognition runs Devin are rented from cloud provider CoreWeave
50%UnverifiedCognition claims that compared with the previous-generation servers, Rubin delivers up to a 4.8x improvement in tokens processed per second
50%UnverifiedThe 4.8x figure refers to the ratio of total tokens processed per GPU per second at the point where single-user response speed is identical, not that answers are 4.8x faster or that 4.8x more users can be served simultaneously
50%UnverifiedAccording to CoreWeave's comparison chart, Rubin holds approximately a 3x advantage over the GB200 at the slow end, climbing to more than 7x at the fast end
50%UnverifiedCoreWeave has not yet disclosed the price of Rubin during its limited-availability period, and Cognition's rate is also private, so the actual savings remain unknown
50%UnverifiedNVIDIA's Blackwell and Vera Rubin chips have rack thermal density requirements that exceed the upper limits of traditional air cooling, necessitating liquid cooling or even immersion cooling infrastructure
50%UnverifiedVera Rubin is NVIDIA's next-generation computing platform following Blackwell.
50%UnverifiedThe Vera Rubin platform uses a newly designed GPU and NVIDIA's in-house Arm-architecture CPU to form a superchip solution.
50%UnverifiedVera Rubin is NVIDIA's next-generation computing platform succeeding Blackwell, named after American astronomer Vera Rubin, who discovered key evidence for dark matter.
50%UnverifiedCognition has become the first customer to run its coding agent Devin on NVIDIA Vera Rubin servers
50%