TITAN V CEO GPU 32GB RAM
23 Jun 2018 6:12:36 UTC
Topic 215524
(moderation:
TITAN V CEO GPU 32GB RAM
***
Nvidia CEO Jensen Huang has unveiled a new souped-up variant of its $3,000 Titan V GPU, which the company launched last year and billed as the most powerful PC GPU ever.
Huang unveiled the 'Titan V CEO Edition' at the Computer Vision and Pattern Recognition conference in Salt Lake City, Utah, where he gave away 20 of the cards to AI researchers working on robotics and autonomous driving projects. And for now, these are the only people in the world who can get their hands on this limited edition model.
The Titan V is Nvidia's most powerful PC GPU, but while gamers may drool over its power, the $3,000 board is aimed primarily at researchers and scientists. The Titan V is built on Nvidia's Volta technology and features its Tensor cores.
Nvidia hasn't revealed pricing for the limited edition model, but AnandTech has detailed specs that suggest if it is released to retail, it may cost somewhere between the standard Titan V and Nvidia's much more expensive Tesla V100 GPUs.
But for now, the Huang-signed edition is a marketing stunt that lets Nvidia align the brand with AI research and distance it from gaming. According to Nvidia, Huang had them "specially made for AI researchers".
SEE: Sensor'd enterprise: IoT, ML, and big data (ZDNet special report) | Download the report as a PDF (TechRepublic)
The standard Titan V features 12GB memory where as the CEO Edition features 32GB, while its deep-learning performance has increased from 110 teraflops to 125 teraflops.
It's also got a 4,096-bit memory bus width, compared with the standard 3,072-bit, and a 6MB rather than 4.5MB L2 Cache.
As AnandTech notes, the beefed-up Titan V CEO Edition's performance on paper is a lot closer to that of the roughly $10,000 Tesla V100 cards.
The CEO Edition also appears to be using Nvidia's new 8-Hi HBM2 memory stacks found in the 32GB Tesla cards.
The question now is whether and when Nvidia will release the product to retail, given the effort it expended to create this special edition Titan V card.
The giveaway cards went to researchers from the Chinese Academy of Science, DFKI, Peking University, Stanford University, Tsinghua University, University of Toronto, University of Tokyo, and University of Washington.
"There's all kinds of research being done here," Huang said. "As someone who benefits from your work, as a person who is going to enjoy the incredible research you guys do -- solving some of the world's grand challenges -- and to be able to witness artificial intelligence happen in my lifetime, I want to thank all of you guys for that. You guys bring me so much joy."
other tec
)
other tec :
https://www.amd.com/en/where-to-buy/epyc-platforms
https://fuse.wikichip.org/news/1177/amds-zen-cpu-complex-cache-and-smu/
https://fuse.wikichip.org/news/1206/quest-a-tci-based-3d-stacked-sram-neural-processor/
https://fuse.wikichip.org/news/1297/cambricon-reaches-for-the-cloud-with-a-custom-ai-accelerator-talks-7nm-ips/
Of course the Titan V barely
)
Of course the Titan V barely outpaces a Vega 64 Liquid Cooled in FP32 so a bit of a waste for E@H. Milkyway makes good use of the FP64 capability but still can't fully load down the core with only 12GB of VRAM.
The extra choice of a 32GB version is great, but it doesn't erase the original poor decision of not having 16Gb on the base card.