Re:Ironwood: The first Google TPU for the age of inference
by muxamilian on Wednesday April 09, 2025 @12:16PM
(#43633782)
In-memory computing (analog or digital). Still doing SIMD matrix multiplication but using more efficient hardware:
https://arxiv.org/html/2401.14428v1
https://www.nature.com/articles/s41565-020-0655-z
Re:Ironwood: The first Google TPU for the age of inference
by gautamcgoel on Wednesday April 09, 2025 @02:39PM
(#43635792)
This is very interesting, but not what the Ironside TPU is doing. The blog post says that the TPU uses conventional HBM RAM.
Re:Ironwood: The first Google TPU for the age of inference
by nsteel on Wednesday April 09, 2025 @03:56PM
(#43637034)
There's been some talk/rumour of next-gen HBMs having some compute capability on the base die. But again, not what they're doing here, this is regular HBM3/HBM3e.
https://semiengineering.com/speeding-down-memory-lane-with-c...
Re:Ironwood: The first Google TPU for the age of inference
by vessenes on Wednesday April 09, 2025 @11:10AM
(#43633034)
One big area the last two years has been algorithmic improvements feeding hardware improvements. Supercomputer folks use f64 for everything, or did. Most training was done at f32 four years ago. As algo teams have shown fp8 can be used for training and inference, hardware has updated to accommodate, yielding big gains.
NB: Hobbyist, take all with a grain of salt
Re:Ironwood: The first Google TPU for the age of inference
by jmalicki on Wednesday April 09, 2025 @01:44PM
(#43634984)
Unlike a lot of supercomputer algorithms, where fp error accumulates as you go, gradient descent based algorithms don't need as much precision since any fp errors will still show up at the next loss function calculation to be corrected, which allows you to make do with much lower precision.
Re:Ironwood: The first Google TPU for the age of inference
by cubefox on Thursday April 10, 2025 @08:43AM
(#43643223)
Much lower indeed. Even Boolean functions (e.g. AND) are differentiable (though not exactly in the Newton/Leibniz sense) which can be used for backpropagation. They allow for an optimizer similar to stochastic gradient descent. There is a paper on it: https://arxiv.org/abs/2405.16339
It seems to me that floating point math (matrix multiplication) will over time mostly disappear from ML chips, as Boolean operations are much faster both in training an inference. But currently they are still optimized for FP rather than Boolean operations.
Re:Ironwood: The first Google TPU for the age of inference
by yeahwhatever10 on Wednesday April 09, 2025 @11:10AM
(#43633032)
Specialization. Ie specialized for inference.
Are there a few big things, many small things...? I'm curious what fruit are left hanging for fast SIMD matrix multiplication.