Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Intel Announces Knights Mill: A Xeon Phi for Deep Learning

anandtech.com

Post reply on HN
Print view View original

31–40 of 57 posts

Re: Intel Announces Knights Mill: A Xeon Phi for Deep Learning

#31

Post by marmaduke »

joe_the_user wrote: post #22
duaneb wrote: post #16

Earlier quoted context omitted.

This is also on upstream projects not to lock themselves into CUDA. Yes, it's great, but everyone suffers when there's only one supported API. Even more so when it's closed and locked to a specific vendor, as CUDA is.

I'd love an alternative to CUDA. The problem is that as far as I can see, OpenCL is in no way that. Basically, OpenCL gives me the impression that the oceans of boiler plate required both make development hard and effectively locks you into a specific vendor also since the boiler-plate is going to be setting things up for one's specific vendor.

Nope. I have some substantial simulation code written against OpenCL that runs on Intel OpenCL and NVIDIA without modifications, and rather performant on both of them. The only part of the code specific to vendor is the platform selection, which is one line of code.

OpenCL falls down in terms of standard libraries such as cu{dnn,sparse,blas} but if you're writing everything from scratch it's fine.

Re: Intel Announces Knights Mill: A Xeon Phi for Deep Learning

#32

Post by argonaut »

joe_the_user wrote: post #21
argonaut wrote: post #6

Earlier quoted context omitted.

But being drop-in compatible means supporting CUDA or the CPU interface.

Is there a problem with a system that's CUDA compatible? CUDA seems to me like the only simple SIMD-type computing system that's fairly straightforward to program and understand at this point. Drop-in CUDA compatibility seems like a good thing.

> We badly need an alternative to Nvidia/CUDA for deep learning

The post I am replying to.

Re: Intel Announces Knights Mill: A Xeon Phi for Deep Learning

#33

Post by argonaut »

cs702 wrote: post #7
argonaut wrote: post #6

Earlier quoted context omitted.

But being drop-in compatible means supporting CUDA or the CPU interface.

They could also contribute code to the most popular frameworks, instead of releasing forks like "Intel Caffe."[1] [1] https://github.com/intelcaffe/caffe

Yes, but unless the framework authors maintain the contribution (possibly unlikely depending on the situation) they will have to keep maintaining those contributions.

Re: Intel Announces Knights Mill: A Xeon Phi for Deep Learning

#34

Post by marmaduke »

ThinkBeat wrote: post #9

I wish they had a range that was affordable for a hobbyist. You can buy a cheap Nvidia card to "get your feet wet". I would like to play with these things.

A recent multicore i7 (e.g. 4 core Haswell, 8 way SIMD for single precision = 32 threads) is enough to prototype OpenCL code which you can then run on larger CPUs or GPUs.

Re: Intel Announces Knights Mill: A Xeon Phi for Deep Learning

#35

Post by visarga »

joe_the_user wrote: post #22
duaneb wrote: post #16

Earlier quoted context omitted.

This is also on upstream projects not to lock themselves into CUDA. Yes, it's great, but everyone suffers when there's only one supported API. Even more so when it's closed and locked to a specific vendor, as CUDA is.

I'd love an alternative to CUDA. The problem is that as far as I can see, OpenCL is in no way that. Basically, OpenCL gives me the impression that the oceans of boiler plate required both make development hard and effectively locks you into a specific vendor also since the boiler-plate is going to be setting things up for one's specific vendor.

Can anyone tell me what exactly is missing from OpenCL to be able to run the primitives of deep learning frameworks? Like, does it not have some kind of operation that is essential for matrix manipulation?

Re: Intel Announces Knights Mill: A Xeon Phi for Deep Learning

#36

Post by visarga »

frozenport wrote: post #28

Intel has thousands of employees and a compiler. Get 200 of them in a room and implement CUDA.

They don't even have to implement the latest flavor of "Inception module", they only need to implement matrix vector operations and some math primitives like exponential, log, tangent and such. Why is it so hard to port to Intel? I would have liked to make use of my Macbook's Intel Iris GPU for deep learning, but it's not supported by anything.

Re: Intel Announces Knights Mill: A Xeon Phi for Deep Learning

#37

Post by visarga »

imaginenore wrote: post #12
ThinkBeat wrote: post #9

I wish they had a range that was affordable for a hobbyist. You can buy a cheap Nvidia card to "get your feet wet". I would like to play with these things.

Why not use Google's platform, which runs on their new custom chips (Tensor Processing Unit)?

Because it's not made to accelerate training, just inference. The TPU is an 8-bit fixed point processor less power hungry than GPUs, so it won't help research, only deployment for large projects, running in the cloud.

Re: Intel Announces Knights Mill: A Xeon Phi for Deep Learning

#38

Post by jlebar »

cs702 wrote: post #2

We badly need an alternative to Nvidia/CUDA for deep learning... but realistically, if Intel wants to make headway in the deep learning market, it must offer hardware that can not only compete on performance with Nvidia, but also work out-of-the-box (that is, without requiring lots of one-off tinkering and tweaking) with popular deep/machine learning frameworks like TensorFlow, Caffe, Torch, and Theano. There is a lo…

Tensorflow has a mode that lets it run on CPUs. I'm sure other frameworks are the same. Isn't the whole point of Xeon Phi that it looks basically like an x86 CPU with a ton of cores? If so there is almost nothing to port, just the kernel launching.

Granted, you do have to bother to make a fast x86 / AVX-512 port of your code. But because the shape of GPUs is so different than CPUs -- GPUs have a more complicated memory hierarchy for one thing -- I kind of doubt that "just run your CUDA code in the Phi" is going to work well for nontrivial examples.

(Disclaimer, I work at Google on CUDA support in clang. Which is awesome, you should try it out. :) Google "cuda clang" for instructions.)

Re: Intel Announces Knights Mill: A Xeon Phi for Deep Learning

#39

Post by rbanffy »

cma wrote: post #25
duaneb wrote: post #16

Earlier quoted context omitted.

This is also on upstream projects not to lock themselves into CUDA. Yes, it's great, but everyone suffers when there's only one supported API. Even more so when it's closed and locked to a specific vendor, as CUDA is.

Can companies that don't have agreements/court rulings with Intel implement Xeon Phi ISAs?

I think it's fair to assume any significant HPC solution will have to be recompiled to target its runtime environment. Squeezing out every GFLOPS of a part is normal when you get an expensive specialized computer.

In that context, x86 compatibility is less of an issue than the quality of the compilers. As long as you are easier to program than a GPU, you are good.

Re: Intel Announces Knights Mill: A Xeon Phi for Deep Learning

#40

Post by stonogo »

rbanffy wrote: post #39
cma wrote: post #25

Earlier quoted context omitted.

Can companies that don't have agreements/court rulings with Intel implement Xeon Phi ISAs?

I think it's fair to assume any significant HPC solution will have to be recompiled to target its runtime environment. Squeezing out every GFLOPS of a part is normal when you get an expensive specialized computer. In that context, x86 compatibility is less of an issue than the quality of the compilers. As long as you are easier to program than a GPU, you are good.

I wish this were true, but it's not. Multi-million-dollar contracts have been scuttled because a given customer runs codes that they receive as binaries from a vendor, closed-source, and Company A bid Processor X, but the codes only run on Processor Y...
Post reply on HN