TPUs vs. GPUs and why Google is positioned to win AI race in the long term

Training is taking an enormous problem and trying to break it into lots of pieces and managing the data dependency between those pieces. It's solving 1 really hard problem. Inference is the opposite, it's lots of small independent problems. All of this "we have X many widgets connected to Y many high bandwidth optical telescopes" is all a training problem that they need to solve. Inference is "I have 20 tokens and I want to throw them at these 5,000,000 matrix multiplies, oh and I don't care about latency".

no https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_proces...

Nvidia doesn't have the software stack to do a TPU.

They could make a systolic array TPU and software, perhaps. But it would mean abandoning 18 years of CUDA.

The top post right now is talking about TPU's colossal advantage in scaling & throughput. Ironwood is massively bigger & faster than what Nvidia is shooting for, already. And that's a huge advantage. But imo that is a replicateable win. Throw gobs more at networking and scaling and nvidia could do similar with their architecture.

The architectural win of what TPU is more interesting. Google sort of has a working super powerful Connection Machine CM-1. The systolic array is a lot of (semi-)independent machines that communicate with nearby chips. There's incredible work going on to figure out how to map problems onto these arrays.

Where-as on a GPU, main memory is used to transfer intermediary results. It doesn't really matter who picks up work, there's lots of worklets with equal access time to that bit of main memory. The actual situation is a little more nuanced (even in consumer gpu's there's really multiple different main memories, which creates some locality), but there's much less need for data locality in the GPU, and much much much much tighter needs, the whole premise of the TPU is to exploit data locality. Because sending data to a neighbor is cheap, sending storing and retrieving data from memory is slower and much more energy intense.

CUDA takes advantage of, relies strongly on the GPU's reliance in main memory being (somewhat) globally accessible. There's plenty of workloads folks do in CUDA that would never work on TPU, on these much more specialized data-passing systolic arrays. That's why TPUs are so amazing, because they are much more constrained devices, that require so much more careful workload planning, to get the work to flow across the 2D array of the chip.

Google's work on projects like XLA and IREE is a wonderful & glorious general pursuit of how to map these big crazy machine learning pipelines down onto specific hardware. Nvidia could make their own or join forces here. And perhaps they will. But the CUDA moat would have to be left behind.

TPUs are also cheaper because GPUs need to be more general purpose whereas TPUs are designed with a focus on LLM workloads meaning there's not wasted silicon. Nothing's there that doesn't need to be there. The potential downside would be if a significantly different architecture arises that would be difficult for TPUs to handle and easier for GPUs (given their more general purpose). But even then Google could probably pivot fairly quickly to a different TPU design.

> Since Google doesn't sell TPUs, they are extremely well-positioned to ensure no one else can profit from any advantages created by TPUs.

First part is true at the moment, not sure the second follows. Microsoft is developing their own “Maia” chips for running AI on Azure with custom hardware, and everyone else is also getting in the game of hardware accelerators. Google is certainly ahead of the curve in making full-stack hardware that’s very very specialized for machine learning. But everyone else is moving in the same direction: lots of action is in buying up other companies that make interconnects and fancy networking equipment, and AMD/NVIDIA continue to hyper specialize their data center chips for neural networks.

Google is in a great position, for sure. But I don’t see how they can stop other players from converging on similar solutions.