Im curious about it .Is it necessary for deAI using parallel GPU empower?
I have checked ic ai documentation brief.
If onchain GPU already accessible will be great!
I found that using web2 gpt-api with frontend canister is not very difficult to impl.
This got a good example.
Do we have necessary to build fully on chain ai product ?
Since AI is quite trendy these days. guys , do you interesting on this ?
dom talked about AI computation ablity on chain
about 22:00
Explained some canister capability.
But i might missing see the GPUs on chain ?
24:38 mention about wasm vm memory.
wasm32 got 4gb memory.
few month will upgrade to wasm64.
But i got a quesiton :
is this memory on chain got similar performance on a nvidia B200 GPU GDDR6 memory ?
And how ic-replica do parallel computation heavy task like a B200 GPU does ?
You can check out DeAI projects on ICP on the DeAI on ICP Working Group GitHub and Internet Computer Ecosystem Page - AI.
You will see several on-chain LLMs already on ICP. Most notably ICPP-LLM.
Research on and onboarding of GPU-based nodes are ongoing.