News for hackers. Stuff that matters.
Re:Train TensorFlow models faster and at lower cost on Cloud... by puzzle on Wednesday December 12, 2018 @06:02PM (#18668489)
I haven't been at Google for years and never worked on TPUs, but off the top of my head:

They control the system attached to the chips. That includes the kernel and userland (glibc, libstdc++, etc.).

Very limited number of configurations (CPU, RAM, network, motherboard, any RDMA use) to qualify.

Monitoring, profiling, diagnostics, firmware, networking, security and automation follow the internal Google standards.

They can tune cooling to accommodate Google's motherboards and racks, whether it's air (v2) or liquid cooling (v3).

You can bet that they talk directly to the GCS backends through Stubby/gRPC, rather than sending HTTP traffic through the outside network and traversing GFEs.

Then there's all the other "mundane" stuff they don't need to worry about: packaging, manuals, warranties and user-facing RMA, multi-tenancy, etc.

Re:Train TensorFlow models faster and at lower cost on Cloud... by ori_b on Wednesday December 12, 2018 @07:01PM (#18668768)
As far as I'm aware, they're still on OCP, which standardizes the bulk of this. Basically, this comes down to not being willing to be bothered with testing on diverse hardware.

Which, fair enough, is a pain.

Re:Train TensorFlow models faster and at lower cost on Cloud... by kajecounterhack on Wednesday December 12, 2018 @10:38PM (#18669699)
Google isn't on OCP afaik. Their hardware layer is pretty custom. Maybe you were thinking of Facebook?

Google's hardware layer is one thing; as btian mentioned, TPU also is deeply integrated with software infra e.g borg https://ai.google/research/pubs/pub43438

Re:Train TensorFlow models faster and at lower cost on Cloud... by puzzle on Wednesday December 12, 2018 @10:55PM (#18669743)
Years ago, Google announced they'd join OCP. I'd be surprised if even half the fleet made the transition, given how much mileage they like to get out of old hardware.

Besides, for special toys like TPU pods it's not clear that they would value adhering to OCP standards above anything else.

Re:Train TensorFlow models faster and at lower cost on Cloud... by ori_b on Thursday December 13, 2018 @02:40PM (#18675313)
Re:Train TensorFlow models faster and at lower cost on Cloud... by btian on Wednesday December 12, 2018 @07:45PM (#18668976)
Not just hardware. TPU probably was only tested to work on one kernel version, one compiler, one version of GRPC etc.

It probably takes a lot of effort to develop Windows driver, libraries, support for many kernel version etc.

Re:Train TensorFlow models faster and at lower cost on Cloud... by ori_b on Thursday December 13, 2018 @02:42PM (#18675325)
> It probably takes a lot of effort to develop Windows driver, libraries, support for many kernel version etc.

That's easy to solve: "We only support Linux. Testing has only been done on kernel 4.13."