Apologies, stand corrected. Conspiracy theories are always more fun:) Guess was just coincidence that after grabbing all this press on GCE v100 finally being available, tpu3 is announced few days later.
Completely agree with you there. Guess the analogy I was making was more about the evolutionary speed we saw there. Google seems to be moving awful fast here.
For all the google hate, nvidia is too blame as well. Their ceo running around saying 'we saving you money every time you spend 10k on a gpu' is getting old. The should have been far more aggressive price wise with this biz considering the resources of their target customers. Almost every 'choose not to use a gpu' use case being touted by msft/goog etc is about cost. Meanwhile nvda busy tweeting stuff like msft seeing ai app for blind powerd by nvda gpu. Soon micron will be tweeting they powering all ai as well, and then arista...all the way down to the power companies. Watching goog today its not looking good for the competition, by the time all these ml/dl accelerators are commoditized google will have one hell of a moat.
You kind of summed it up right there...'it's just an arithmetic logic unit for tensors'...whether its a matrix multiply engine or implmenting sha256 still a custom rapidly iterated chip for narrow very specific use case. Google accomplishment here clearly the software, but doubt they only ones going to crack asic systolic arrays. At some point china inc figures this out in mass...maybe bitmain themselves.
This very typical google. V100 comes out, they don't deploy it in their cloud immediately. Then they launch their tpu cloud. Spend 2 months touting cost/speed in benchmarks. Then a week before io they make v100 available in gce, nearly six months after aws. Then at io they announce tpu3. This is supposed to be nvidia's most captive customer, and now looks like going to be their biggest problem. Would love to see google spend with them and how its changed since they ramped tpu2.
This makes perfect sense. It's all about economics both w respect to intel and nvidia. You can't be paying 8-10k per gpu, and you are not going to pay that when you can make something far cheaper and faster. Google got this done in what two years?? All the debates around the tech comparisons miss the economic picture. Who cares if 4tpu's being compared to one volta. Point is those 4 chips together cheaper than buying one v100. Bitcoin mining no different. Do you care how many chips (196) are in an antminer s9? No! What matters is that it mines 13000x faster than a gpu for roughly double the price. Facebook, Amazon and the likes have every incentive to go down this custom ASIC path. Google is now hiring sales folks for their tpu cloud. There is a reason they are not selling the hardware. It's more valuable to them to get tenants the other hyperscale competition can't match. So think all the hyperscale guys now looking for ways to keep up which obviously is horrible for nvidia. A lot of people have missed how much of nvidia's story been about essentially killing it on one very narrow use case, ml training, in a very short time. This is basically 50%+ of all profit growth they have achieved last two years. Nvidia has got some serious challenges ahead.