The buried lede here is we could’ve had four major players in the GPU space, if it weren’t for those meddling board members. This fourth one surely would’ve been built with compute in mind from the ground up, which might have finally put out a real competitor to CUDA.
Barriers to entry are a high bar. $Bs in VC capital has flown into that exact space over the last 5-8 years, and there are a bunch of dead startups in the wake. Altman’s resources are no different than any of those others.
The buried part of that is that some Middle East sovereign fund would also control them.
Which if we take the “anti-aging tech, water rights in Arizona adjacent to TSMC, ip3 nuclear tech,petrodollar, and hobby owning of republicans” historical context into account makes this an extremely bad decision for humanity.
In some senses you end up with convergent design, it’s not a GPU, it’s just a control system that commands a bunch of accelerator units with a high-bandwidth memory subsystem. But that could be ARM and an accelerator unit etc. Probably need fast networking.
But it’s overall a crazy proposition to me. Like first off goog and amazon are gonna beat you to market on anything that looks good, and you have no real moat other than “I’m sam altman”, and really there’s no market penetration of the thing (or support in execution let alone actual research) etc. Training is a really hard problem to solve because right now it’s absolutely firmly rooted in the CUDA ecosystem. Supposedly there may be a GPU Ocelot thing once again at some point but like, everyone just works with nvidia because they’re the gpgpu ecosystem that matters.
Like, if you wanted to do this you did like Tesla and have Jim Keller design you a big fancy architecture for training fast at scale (Dojo). I guess they walked away from it or something and just didn’t care anymore? Oops.
But, that’s the problem, it’s expensive to stay at the cutting edge. It’s expensive to get the first chip, and you’ll be going against competitors who have the scale to make their own in-house anyway. it’s a crazy business decision to be throwing yourself on the silicon treadmill against intense competition just to give nvidia the finger. wack, hemad.
The buried lede here is we could’ve had four major players in the GPU space, if it weren’t for those meddling board members. This fourth one surely would’ve been built with compute in mind from the ground up, which might have finally put out a real competitor to CUDA.
Barriers to entry are a high bar. $Bs in VC capital has flown into that exact space over the last 5-8 years, and there are a bunch of dead startups in the wake. Altman’s resources are no different than any of those others.
The buried part of that is that some Middle East sovereign fund would also control them.
Which if we take the “anti-aging tech, water rights in Arizona adjacent to TSMC, ip3 nuclear tech,petrodollar, and hobby owning of republicans” historical context into account makes this an extremely bad decision for humanity.
Unlikely. They’re probably looking for investors all over. The Middle East just happens to be one such destination.
last thing we want is the ME to have tons of AIs and drones
Too late for that…
Only the USA can be trusted to use drones and AI responsibility! /s
There are a large number of startups working to rival Nvidia’s AI prowess. TensTorrent, Cerebras, etc.
https://www.ai-startups.org/top/hardware/
Each of them are five or more years ahead of whatever Altman was raising funds for.
I see zero suggestion that anybody would be entering the GPU space, this would be focused for AI applications only.
In some senses you end up with convergent design, it’s not a GPU, it’s just a control system that commands a bunch of accelerator units with a high-bandwidth memory subsystem. But that could be ARM and an accelerator unit etc. Probably need fast networking.
But it’s overall a crazy proposition to me. Like first off goog and amazon are gonna beat you to market on anything that looks good, and you have no real moat other than “I’m sam altman”, and really there’s no market penetration of the thing (or support in execution let alone actual research) etc. Training is a really hard problem to solve because right now it’s absolutely firmly rooted in the CUDA ecosystem. Supposedly there may be a GPU Ocelot thing once again at some point but like, everyone just works with nvidia because they’re the gpgpu ecosystem that matters.
Like, if you wanted to do this you did like Tesla and have Jim Keller design you a big fancy architecture for training fast at scale (Dojo). I guess they walked away from it or something and just didn’t care anymore? Oops.
But, that’s the problem, it’s expensive to stay at the cutting edge. It’s expensive to get the first chip, and you’ll be going against competitors who have the scale to make their own in-house anyway. it’s a crazy business decision to be throwing yourself on the silicon treadmill against intense competition just to give nvidia the finger. wack, hemad.