Re:TPUs vs. GPUs and why Google is positioned to win AI race...
by bjourne on Thursday November 27, 2025 @10:43AM
(#46070287)
Except the native width of Tensor Cores are about 8-32 (depending on scalar type), whereas the width of TPUs is up to 256. The difference in scale is massive.
[ Parent
| on HN ]
Re:TPUs vs. GPUs and why Google is positioned to win AI race...
by saagarjha on Thursday November 27, 2025 @08:07PM
(#46074657)
Nobody cares about width; they care about TFLOPs.
Re:TPUs vs. GPUs and why Google is positioned to win AI race...
by fooker on Thursday November 27, 2025 @10:35PM
(#46075278)
If it turns out to be useful, Nvidia can't just tweak a parameter in their verilog and declare victory?
If not, what's fundamentally difficult about doing 32 vs 256 here?
Re:TPUs vs. GPUs and why Google is positioned to win AI race...
by neilmovva on Sunday November 30, 2025 @08:15PM
(#46102336)
I think Hopper's native matmul tile is 64x64, and Blackwell is 128x128.
see this blog for a reference on Blackwell:
https://hazyresearch.stanford.edu/blog/2025-03-15-tk-blackwe...