Nvidia just set a 2023 target for 10x higher performance on trillion parameter AI models. Specifically saying 64 module systems that will be able to train trillion parameter models in just 3 dayshttps://t.co/DTzreUJmL4
— David Holz (@DavidSHolz) April 12, 2021
NVIDIAはデータセンター向けCPU製品Graceの構想を発表した。目標として1兆個のパラメーターモデルのトレーニングを1か月から3日に短縮することを目指している。2023年にNVIDIA製品で利用可能になるとのこと。https://t.co/AlI6G7BF7Q
— akira (@AkiraTOSEI) April 12, 2021
GRACE from NVIDIAhttps://t.co/8x8iJs1Cb5
— Rohan Anil ???? (@_arohan_) April 12, 2021
900GB/sec ?
This could unlock new types of distributed training. pic.twitter.com/a5LGbGyOew
CPUs, GPUs & DPUs, oh my!
— Alison B. Lowndes ✿ (@AlisonBLowndes) April 12, 2021
I can say it now that Jensen has, after I read his mind last week & freaked everyone out?
Very proud of the fact we've posthumously named the CPU after Rear Admiral Grace Hopper (1906-1992)
Here's @anandtech's take: https://t.co/zC2Egeokzy #GTC21 pic.twitter.com/Oio0hKtBLU
Wow, turns out I might buy another Nvidia product: https://t.co/Px6upU5BGv
— Ana, Hoverbear ? (@a_hoverbear) April 12, 2021
Nvidia Targets $150 Million Revenue From New Ethereum Mining Processor - https://t.co/AnwA8GMEhr via @_feedspot
— Brrrrr Capital (@InterestRateArb) April 13, 2021
NVIDIA announced a new CPU at GTC today, alongside some new RTX GPUs in the "new" (ish) A-series line: The A5000, A6000, and some others. Learn more here: https://t.co/wMgcrY4Hiu
— GamersNexus (@GamersNexus) April 13, 2021