Here is the original funding announcement for NorthPole, which debuted a week ago at Hot Chips 2013. You can get four (4) processors in a 1U.. Price to be determined: https://news.clearancejobs.com/2019/08/16/ibm-research-wins-60m-for-artificial-intelligence-machine-learning-project-with-usaf-dod-daily-contracts/

Here is information about Tesla's Dojo when announced: https://electrek.co/2021/08/20/tesla-dojo-supercomputer-worlds-new-most-powerful-ai-training-machine/

And today (possible paywall): https://www.forbes.com/sites/stevendickens/2023/09/11/teslas-dojo-supercomputer-a-paradigm-shift-in-supercomputing/

Cheers!
Todd B
Fritter Inc

Todd Bezenek wrote:
> Here is the original funding announcement for NorthPole, which debuted a week ago at Hot Chips 2013. You can get four (4) processors in a 1U.. Price to be determined: https://news.clearancejobs.com/2019/08/16/ibm-research-wins-60m-for-artificial-intelligence-machine-learning-project-with-usaf-dod-daily-contracts/
>
> Here is information about Tesla's Dojo when announced: https://electrek.co/2021/08/20/tesla-dojo-supercomputer-worlds-new-most-powerful-ai-training-machine/
>
> And today (possible paywall): https://www.forbes.com/sites/stevendickens/2023/09/11/teslas-dojo-supercomputer-a-paradigm-shift-in-supercomputing/
>
> Cheers!
> Todd B
> Fritter Inc

Forbes is perhaps not the best source for technical info on Dojo D1.

Tesla's D1 chip is RISC-V optimized for Tesla's machine learning.

Inside Tesla’s Innovative And Homegrown “Dojo” AI Supercomputer, 2023-08-23
https://www.nextplatform.com/2022/08/23/inside-teslas-innovative-and-homegrown-dojo-ai-supercomputer/

"And what is most striking are the things that Tesla’s engineers
threw out as they focused on scale."

“We have de-emphasized several mechanisms that you find in typical CPUs,
like coherency, virtual memory, and global lookup directories just because
these mechanisms do not scale very well when we scale up to a very large
system. Instead, we have relied on a very fast and very distributed SRAM
storage throughout the mesh. And this is backed by an order of magnitude
higher speed of interconnect than what you find in a typical distributed
system.”

Basically, D1 looks like a custom, single purpose, embedded architecture,
ASIC chip design, running custom OS and software stacks.
It is never going to do anything other than run Tesla's ML.

That may break Tesla's dependency on Nvidia but I don't see
any technical "Paradigm Shift".

NorthPole: Array of (Vector Matrix Engine+SRAM)
Dojo: Array of Computer

-
s.takano

On Thursday, September 21, 2023 at 4:33:33 PM UTC-6, S.Takano wrote:
> NorthPole: Array of (Vector Matrix Engine+SRAM)
> Dojo: Array of Computer

When I saw this post, I immediately thought of Masashi (Max) Takano,
who is responsible for many of the arrangements of movie themes and
other music for the Yamaha Electone STAGEA organs which led to so
many impressive YouTube videos, like this one:

https://www.youtube.com/watch?v=Yp5HpjhKHKs

John Savard

Last name, 高野(taka-no) is minor :),but I glad such the person is there.

-
s.takano