Open jafioti opened 6 months ago
Currently Intel offers an A770 for $300 with 16gb of ram and much better flops than a 4060ti ($500).
From what I hear over at tinygrad, they have much better drivers than AMD.
We should support Intel cards as much as possible. Proper pipeline / tensor parallelism can make these things feasible for larger model inference.
Possible libraries:
Currently Intel offers an A770 for $300 with 16gb of ram and much better flops than a 4060ti ($500).
From what I hear over at tinygrad, they have much better drivers than AMD.
We should support Intel cards as much as possible. Proper pipeline / tensor parallelism can make these things feasible for larger model inference.