7 ms·
The performance of OpenCL has generally been fine for me, particularly on AMD GPUs, but I have to say I think CUDA is a lot simpler to work with. OpenCL is one
by tombert 10y ago
The performance of OpenCL has generally been fine for me, particularly on AMD GPUs, but I have to say I think CUDA is a lot simpler to work with.
OpenCL is one of those things that I never felt fully comfortable working with, but I felt productive in CUDA after a week or two. Granted, I learned them in that order, so it's possible that CUDA got an unfair head-start, but I stand by my initial thesis.
- shawn-butler 10y agoYou might want to take a look at SYCL for higher-level abstractions if that was your preference. There is an open-source implementation to dig into: https://github.com/Xilinx/triSYCL https://github.com/Xilinx/triSYCL
- tombert 10y agoActually, nowadays since I mostly do esoteric functional stuff, I mostly use the wonderful `accelerate` package for Haskell: https://www.stackage.org/package/accelerate https://www.stackage.org/package/accelerate It abstracts away most of the difficult stuff.
- Athas 10y agoOut of curiosity, what are you doing with Accelerate? I've been looking for examples of people using it in anger.
- tombert 10y agoI can't get into the weeds of it because of some NDA chicanery, but there are a couple vector transformations that I have to do on a not-quite-big-data-but-also-too-big-for-regular-CPU set of data. The initial version of the project was doing some basic stuff in Python, but that proved to be slow so I rewrote most of it in Haskell and Accelerate.
- rodburns 10y agoIn addition to triSYCL we also have an implementation of SYCL called ComputeCpp http://computecpp.codeplay.com/ http://computecpp.codeplay.com/
- lettergram 10y agoI tried picking up OpenCL then CUDA as well and had a similar experience. CUDA simply feels less hacked together to me. It's like working with a well thought out and documented code base (CUDA) vs working with some library that has little documentation and contradicting syntax and formats (OpenCL). IMO that's why AMD is losing the "deep learning" battle, it's just not easy to develop using OpenCL. At least, not as easy as it should be.
- happycube 10y agoLosing? When it comes to TensorFlow et al they haven't really shown up for the fight. It's a vicious cycle though, since everything uses CUDA/CuDNN everyone buys nVidia cards, which means that developer interest in getting AMD cards fast isn't there. If AMD GPU cores were in phones etc, they would probably be getting TF support for client work at least. Heck, considering they sold off that dept to Qualcomm, which is getting supported...
- lettergram 10y agoThere's really no reason AMD themselves couldn't hire 5 developers to integrate with TensorFlow. I agree it's a vicious cycle, but I feel if they release support for tensorflow it wouldn't take long to shift people from Nvidia to AMD. AMD has a significantly cheaper offering, and no company wants to be tied by only one supplier. Considering how small of a dent that would make in their bottom line (financially), it's really amazing they haven't yet... The potentially upside is easily in the tens (if not hundreds) of millions of dollars. So definitely, I agree they intentionally have decided to not support deep learning. For the life of me, I don't know why.
- ansible 10y agoThere's really no reason AMD themselves couldn't hire 5 developers to integrate with TensorFlow. The problem is that it isn't just TensorFlow. It is also Caffe, Theano, Torch, etc., etc., etc.. For deep learning, Nvidia GPU acceleration is almost a given.
- tmbsundar 10y agoDoes freebsd support CUDA yet? I tried a year back and it didn't, due to limitations from NVIDIA side. I like freebsd a lot, but for this reason had to switch back to debian/ kubuntu.
- jacquesm 10y agoThe question should be is NVIDIA supporting FreeBSD yet? And the answer is so far not. And likely it won't happen, they are supporting Linux because it is a large enough target for their scientific customers, FreeBSD just does not have the market share there.
- floatboth 10y agoNVIDIA actually releases their proprietary GPU driver for FreeBSD! Has been doing that for a long long time. Also for Solaris.
- jacquesm 10y agoYes, but that's not CUDA, just the basic acceleration. For CUDA supported are Windows, Linux, Mac OSX https://developer.nvidia.com/cuda-downloads https://developer.nvidia.com/cuda-downloads
- astrodust 10y ago