CoreWeave Data Show Nvidia Vera Rubin Chip Delivers 10 Times More Tokens Per MW Than Blackwell
CoreWeave has released the first publicly measured performance data for Nvidia’s Vera Rubin NVL72 artificial intelligence server rack. Operating on the DeepSeek-R1 model, the initial silicon benchmarks indicate the new platform processes 10 times more tokens per MW of power than the company’s previous Blackwell architecture.
The publication follows Nvidia’s confirmation Tuesday that major clients are already testing the Vera Rubin systems, which remain on schedule for deployment across AI data centers. The company also announced shipments of its newly developed Vera CPU to customers including OpenAI, Anthropic, and SpaceX, asserting the processor outperforms rival chips from AMD and Intel. Additionally, the company detailed its 102.4-terabit-per-second networking switch, noting that co-packaged optics technology is now in volume manufacturing rather than the prototype stage.
From the sources (25 posts)
@tradfi*NVIDIA SAYS MAJOR CUSTOMERS ARE ALREADY TESTING VERA RUBIN GEAR *NVIDIA SAYS NEW VERA PROCESSOR IS FASTER THAN AMD'S TURIN *NVIDIA SAYS CHIPS ARE ON SCHEDULE FOR USE IN AI DATA CENTERS $NVDA
@financialjuiceNvidia: Vera Rubin AI Platform on Schedule, Customer Testing Underway $NVDA $AMD
@wallstengineNvidia says major customers are already testing its next-gen Vera Rubin AI systems ahead of deployment. $NVDA also said the new Vera CPU outperforms AMD’s Turin and that the platform remains on schedule for rollout in AI data centers.
@stockmktnewzNVIDIA JUST CONFIRMED ITS NEXT-GEN CHIPS ARE ON SCHEDULE, AND MADE A DIRECT COMPARISON TO AMD IN THE SAME BREATH Nvidia $NVDA says its chips remain on track for use in AI data centers. The company also says its new Vera processor is faste
@firstsquawkNVIDIA'S VERA RUBIN AI PLATFORM IS ON TRACK AS CUSTOMER TESTING BEGINS. MAJOR CLIENTS ARE NOW TESTING THE VERA RUBIN PLATFORM, WITH CHIPS SET TO BE DEPLOYED IN AI DATA CENTERS AS PLANNED. NVIDIA CLAIMS ITS NEW VERA PROCESSOR IS QUICKER THAN
@cnbcNvidia details its next-generation Vera CPU for AI, setting up challenge to AMD and Intel
@stocksavvyshay$NVDA says its next-generation Vera Rubin platform remains on schedule with major customers already testing the hardware. The company also claims its new Vera CPU is faster than $AMD Turin processor
@amir👀Nvidia says it will be able to produce up to 1,000 Vera Rubin racks ~PER DAY~ If it does, that would generate >$630B in revenue per quarter for Nvidia & manufacturing partners
@polymarketmoneyBREAKING: Nvidia declares Vera Rubin remains on schedule, with major customers already testing it.
@businessNvidia, the semiconductor company at the heart of the artificial intelligence boom, said that its latest chip designs are making their way to customers and will help solidify its leadership in the industry.
@stockmktnewzNvidia $NVDA just posted this "NVIDIA Spectrum-6, a 102.4-terabit-per-second Ethernet switch built for the Vera Rubin platform, is now arriving across the world's gigascale AI factories." Here are the companies Nvidia listed as being some
@stocksavvyshay$NVDA says its 102.4 Tbps Spectrum-6 switch for Vera Rubin is rolling out across gigascale AI factories with $CRWV, $MSFT, $NBIS, $SPCX and $TSLA among the first adopters. Its co-packaged optics architecture pressures traditional pluggable
@gavinsbakerInteresting report from @theinformation that Nvidia will be able to make 1000 Vera Rubin racks per day, which is $630b per quarter. Actually a little hard for me to believe and haven’t checked the math, but wow if true. And Vera CPU racks
@melvininvestsThis is WILD! Nvidia just revealed a manufacturing target so big it implies quarterly revenue bigger than the entire company generates today (Save this). Andrew Bell, Nvidia's senior vice president of hardware engineering, said the roughl
@unusual_whalesNvidia, $NVDA, is officially in full production of its next-generation "Vera Rubin" AI computing systems, per YF
@photoncapNVIDIA launched Vera Rubin NVL72, a rack-scale AI supercomputer combining seven codesigned chips for maximum performance per watt and lowest token cost. CoreWeave benchmarks show 10x higher throughput per megawatt compared to Grace Blackwe
@cnbcNvidia became the most valuable company because of insatiable demand for its graphics processing unit, or GPU, the primary chip used for creating and deploying artificial intelligence. But the chip giant is now shipping its own central pro
@firstsquawkNVIDIA EXPANDS AI HARDWARE STRATEGY BEYOND GPUS BY SHIPPING NEW VERA DATA CENTRE CPU, SETTING UP DIRECT COMPETITION WITH INTEL AND AMD IN AI SERVER MARKET; VERA CPUS DELIVERED IN JUNE TO CUSTOMERS INCLUDING OPENAI, ANTHROPIC AND SPACEX - SO
@photoncapRT @ren_stocks: Hyper bullish CPO $LITE $AAOI $SIVE NVIDIA said its co-packaged optics switch is in volume manufacturing and you are beari…
@photoncapRT @damnang2: The most important takeaway is, NVIDIA described its CPO switch for scale-out networking as being in “volume manufacturing,”…
@teortaxestexThe underrated thing is that they're running R1 at ≈150 tps, about 7 times faster than it was served by DeepSeek proper. And for Rubin, this is a healthy regime, not desperate at all, 60% of peak throughput. Almost 1 t/s/watt. V4 does rough
@teortaxestex1648 TFLOPS/GPU on GB300 sounds really cool, and is. this is with MXFP8. So 33% MFU. DeepSeek themselves, in late 2023, did ≈25% on H800s (they calculate MFU based on BF16 here, but heavily used FP8). A 4.3x gap in per-GPU speed. About 2.5x
@mweinbachRT @CoreWeave: The first-ever measured silicon numbers for @NVIDIA Vera Rubin NVL72 are in 😲 First measured performance shows 10x more tok…
@stocksavvyshay$CRWV reports the first measured $NVDA Vera Rubin NVL72 results with 10x more tokens per second per megawatt than Blackwell on DeepSeek-R1. The results show Rubin’s efficiency improvements are already holding up in a real-world environment
@coreweaveThe first-ever measured silicon numbers for @NVIDIA Vera Rubin NVL72 are in 😲 First measured performance shows 10x more tokens per megawatt than Blackwell. No projections. Real results from live hardware.