3 ms·
Welcome to ASIC land....training going the route of crypto mining
by ai2323 8y ago
Welcome to ASIC land....training going the route of crypto mining
- Maybestring 8y agoIt's just an arithmetic logic unit for tensors. It's not at all like a crypto miner that implements a single algorithm.
- jacksmith21006 8y agoExactly. Drives me crazy when people compare to a single alogrithm chip. It is not.
- deepnotderp 8y agoBut the TPU actually is: it's a systolic array matrix multiply ASIC.
- jacksmith21006 8y agoThe TPUs support all kinds of NN. Not specific to one type.
- deepnotderp 8y agoIndeed, all neural nets that use matrix multiplication. Out just so happens to be all the main types right now.
- jacksmith21006 8y agoGoogle is the leader on AI algorithms. Where GANs and Capsules and so many others came from for example. So things change and Google can implement in silicon far earlier than anyone else as needed. But right now they have what they need. BTW, I would be curious on the TPU 3 design considerations to better support dynamic routing with Capsule networks. I suspect creates different memory access patterns and where the big power savings are at today.
- shaklee3 8y agoYou are constantly comparing this single-purpose chip to a general-purpose GPU. How is that any different?
- ai2323 8y agoYou 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.
- Maybestring 8y agoWell... I'll concede that 2-instructions that aren't reading or writing memory is a pretty narrow set of operations. But it's also much more generally useful than computing one specific hash function.
- ai2323 8y agoCompletely 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.
- wmf 8y agoBitmain already announced TPUs (cue "they'll only sell them after they're finished training Skynet").
- sanxiyn 8y agoFYI, here is Bitmain's AI product homepage: https://www.sophon.ai/ https://www.sophon.ai/ They shipped Deep Learning Accelerating Card SC1 at $589. You can't buy it right now because it is sold out.
- sannee 8y agoFaster matrix multiplication is something which has application in practically every engineering field there is.