4 ms·
Uh no. IIRC were on CUDA 5 or something when imagenet came out and changed the world. They might not have imagined LLMs when they decided to invest in making
by DSingularity 3y ago
Uh no.
IIRC were on CUDA 5 or something when imagenet came out and changed the world.
They might not have imagined LLMs when they decided to invest in making their GPUs programmable but I guarantee you they extrapolated the future compute potential of vector programmable machines and decided it was not a huge risk to enable it as it is simply betting that some important application would be around to tap into it.
- m_mueller 3y agoDidn't even need to extrapolate future potential. The world of supercomputing was massively sunsetting the old vector processors (Cray, Sparc etc.) and switched to commodity x86 hardware back in that timeframe when CUDA came out. Perfect opening for a new vector processor on steroids, for which the chip development was already payed by gamers...
- Gh0stRAT 3y agoIIRC people were already converting their data into texture maps and writing their SIMD instructions as shaders operating on those textures before Nvidia released CUDA.
- mratsim 3y agoAh yeah there was an article on HN about that a couple years ago. GPGPU before Cuda, now onto a treasure hunt ...
- hnlmorg 3y agoI remember some DAWs supporting pluggins to run PCM over GPU back in the early 00s. It was hugely experimental but does demonstrate that the idea of a GPUs being used for compute isn’t a recent one.
- PH95VuimJjqBqy 3y ago^ exactly. What people find contentious is this claim that they specifically did it for ML/AI. That's not why they invested the way they did.