Discover and read the best of Twitter Threads about #GPUs
Most recents (6)
@boblautenbach @Alan_L_Lovejoy @Ryu_Zoku Yes Chris Lattner is very interesting, been following his #LLVM and #SWIFT work.
1/
1/
@boblautenbach @Alan_L_Lovejoy @Ryu_Zoku Actually #ZokuTalk adopts the portion of SWIFT #EBNF for #Unicode as they seems have done good job on that, as well as parts of the EBNF for numerics such as "1_234.00" to expand privative data types entry.
\2
\2
@boblautenbach @Alan_L_Lovejoy @Ryu_Zoku Borrowing C & Assembly data types: 0b0011, 0xFF, keeping 16r0A, plus many others.
Actually the numeric formats are one of more complicated part of the #EBNF of #ZokuTalk and the only non-#MOBS (#Messages #Objects #Block #Syntax) part of the evolved syntax.
\3
Actually the numeric formats are one of more complicated part of the #EBNF of #ZokuTalk and the only non-#MOBS (#Messages #Objects #Block #Syntax) part of the evolved syntax.
\3
A short thread on accelerated computing. How $TSLA's #Dojo, $GOOG's #TPU #Cerebras's #WSE-2, #Graphcore's #IPU, and others compare to $NVDA 's #GPU. This is an extract of work we published a few weeks ago. 👇1/6
It is difficult to compare chips. Designing a chip is all about managing trade-offs between multiple dimensions. A chip has limited potential resources and the architect allocates them over these multiple dimensions. 2/6 
With this framework in mind, one sees alternatives to #GPU have fundamentally different architectures, favoring the flow of data across the chip vs. the flow of data between the memory and the chip. 3/6
The size of #NLP models have increased enormously, growing to millions, or even billions, of parameters, along with a significant increase in the financial cost and carbon emissions.
The cost associated with training large models limits the #AIresearch community's ability to innovate, because a research project often needs a lot of experimentation.
Consider training a top-performing #LanguageModel on the Billion Word benchmark. A single experiment would take 384 GPU days (6 days*64 V100 #GPUs, or as much as $36,000 using AWS on-demand instances)
@GoingParabolic #Verus is a truly free, #decentralized protocol and rent-free #blockchain framework that provides chain #interoperability that can scale to the world. It’s not just a protocol, but an ecosystem of interconnected blockchains.
@GoingParabolic Once on mainnet, the Verus #DeFi system will be a fractal, market-driven, & completely liquid worldwide network of independently valued currencies and protocol-based currency baskets that allows anyone holding Verus to perform cross-chain transactions, launch a fractional reserve
#Huawei is bypassing #US sanctions by leasing #NVIDIA #GPUs and #Dell workstations through third-party companies.
Dear @realDonaldTrump @SecPompeo @FBI @CIA @CommerceGov, please investigate.
I got the above tip from @zhangtianliang. I'll put his original Chinese message below
Dear @realDonaldTrump @SecPompeo @FBI @CIA @CommerceGov, please investigate.
I got the above tip from @zhangtianliang. I'll put his original Chinese message below
Original message from @zhangtianliang:
"Jennifer，有人在我谈华为的节目下方留言说：'美国制裁的中国公司正通过租赁方式获取美国产品和服务，我之前所在公司在美国制裁名单上，但公司通过第三方公司租赁英伟达GPU和戴尔工作站，依然可以使用美国产品，烦请先生可以传递给相关部门进行核实。'”
"Jennifer，有人在我谈华为的节目下方留言说：'美国制裁的中国公司正通过租赁方式获取美国产品和服务，我之前所在公司在美国制裁名单上，但公司通过第三方公司租赁英伟达GPU和戴尔工作站，依然可以使用美国产品，烦请先生可以传递给相关部门进行核实。'”
Translation of @zhangtianliang's message: "Jennifer, someone left a comment under my show about #Huawei:' Chinese companies sanctioned by #US are obtaining US products and services via leasing. My former company is on the U.S. sanctions list, but the company is leasing...
Just had a wild idea, so bear with me:
Imagine an #opensource #cryptocrowdfunding model for software products, the kind you buy from @awscloud @Azure @GCPcloud and also the smaller players like @digitalocean and @Rackspace m, where there are no hardware licenses. 1/
Imagine an #opensource #cryptocrowdfunding model for software products, the kind you buy from @awscloud @Azure @GCPcloud and also the smaller players like @digitalocean and @Rackspace m, where there are no hardware licenses. 1/
You don’t even have to run the software or have access to the hardware running it in this model. Instead of #softwarelicensing, #developers set #crypto #staking thresholds which act as #digitalaccesscontroltokens; #stake x of whichever #Crypto, or y of another, with a discount 2/
when using the project’s #sidechain. So you get staking setup, and essentially get #APIkeys which allow you to join the network in r/w mode. Or you could say to heck with all that, I just want to pay someone. You can pay the network directly to do the work the software does. 3/
