3 ms·
With 4-bit quantization the requirements are more like 64gb. I expect we'll see more unified memory designs like Apple's 128GB M1 Ultra.
by Legend2440 3y ago
With 4-bit quantization the requirements are more like 64gb.
I expect we'll see more unified memory designs like Apple's 128GB M1 Ultra.
- zacmps 3y agoI doubt it to be honest, desktop GPUs use too much power (and hence produce too much heat) to be integrated in that fashion, and any kind of shared memory will be too high latency.
- nullc 3y agoThere are 'desktop' (well server) cpus with 64GB of HBM memory per socket now. And big LLMs can be run on lower memory bandwidth systems (like zen4 chips with 12x ddr5 per socket) at lower performance, but where 1-2TB of ram is no big deal.