Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

Good.

Nvidia stuff 1080ti cost too much for a graduated student.

AMD is working on their own stuff supposely later this quarter we'll get tensorflow and other support on their card.

Google's TPU is not for consumer, only for cloud.

If bitmain can sell ASIC for consumer with reasonable price and support tensorflow, pytorch, etc... then it'll be good for the deep learning community.



I suspect they'd be designing for datacenters rather than consumers. That's essentially what they already do with bitcoin rigs too. It's a rented mining cloud.


> It's a rented mining cloud.

That's not all it is. They do sell actual hardware as well, which you then run yourself.


Sure, that's entirely fair.


>rented mining cloud

Would you like to rephrase this to something more accurate?


What part do you consider inaccurate? You pay for hashes/s in their hardware pool. The phrasing seems pretty sufficient.

e.g. https://pool.viabtc.com/contract/


Hashnest is a similar service run by bitmain.

> Hashnest is a cloud hashing platform. This means you can participate in bitcoin mining without maintaining the hardware yourself.


There are still a lot of software side of things we as a community can do in accelerating compute speed - for example, add OpenCL support for popular AI framework/libraries, which will leverage existing hardware and non-Nivida GPUs' power.

[1]: https://github.com/tensorflow/tensorflow/issues/22


A lot of the frameworks(keras, Theanos) uses Cuda close to the metal, so they would need to write alternative interfaces and the to be supported by these libraries




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: