Would this be possible with OpenCL, too?
KGPU - Augmenting Linux with the CUDA GPU
11–20 of 34 posts
Re: KGPU - Augmenting Linux with the CUDA GPU
#12Where is this useful? The bus speed across to the GPU is so slow, I thought it was only meaningful for near-autonomous operations.
One situation where you might see immediate results is in High volume routing. I think this project [1]. They were using the GPU to saturate multiple 10GbE interfaces with a commodity processor.
Re: KGPU - Augmenting Linux with the CUDA GPU
#13Earlier quoted context omitted.
Right now, the bandwidth to modern GPUs is actually pretty decent (16GB/s bidirectional), but the latency is still horrid. This means that you need rather large operations for offloading to pay off. I think doing raid-5 or full disk encryption with large blocks might just barely be worth it. However, with AMD and Intel integrated GPUs, this is about to change. AMD is doing a lot of work on HSA, which can be summarize…
More and more CPUs have AES instructions and my old lenovo ideapad has a crypto coprocessor. Do you think the GPU offload will be worth it when the sytem has hw accelerated crypto?
The question should be whether dedicated hardware for accelerated crypto will be worth it when GPU offload is suitable for it.
Although in that particular context (security) you might enjoy the isolation of dedicated hardware as opposed to sharing it with others. The GPU solution though of course has the advantage of being able to adapt to new ciphers etc.
Re: KGPU - Augmenting Linux with the CUDA GPU
#14Earlier quoted context omitted.
More and more CPUs have AES instructions and my old lenovo ideapad has a crypto coprocessor. Do you think the GPU offload will be worth it when the sytem has hw accelerated crypto?
The question should be whether dedicated hardware for accelerated crypto will be worth it when GPU offload is suitable for it. Although in that particular context (security) you might enjoy the isolation of dedicated hardware as opposed to sharing it with others. The GPU solution though of course has the advantage of being able to adapt to new ciphers etc.
Adapting/implementing the newest and hottest cipher on the block is not something that the crypto community advocates. Do you really think crypto accelerated hardware is going to fall behind and not support the ciphers that the crypto community (academia/industry) endorses?
Re: KGPU - Augmenting Linux with the CUDA GPU
#15Earlier quoted context omitted.
The question should be whether dedicated hardware for accelerated crypto will be worth it when GPU offload is suitable for it. Although in that particular context (security) you might enjoy the isolation of dedicated hardware as opposed to sharing it with others. The GPU solution though of course has the advantage of being able to adapt to new ciphers etc.
Adapting/implementing the newest and hottest cipher on the block is not something that the crypto community advocates. Do you really think crypto accelerated hardware is going to fall behind and not support the ciphers that the crypto community (academia/industry) endorses?
That depends on which team you're playing for, no?
Re: KGPU - Augmenting Linux with the CUDA GPU
#16Why is this cuda and not OpenCL? There's no reason to legitimize nvidia's proprietary nonsense, it just enables their bad behavior.
Re: KGPU - Augmenting Linux with the CUDA GPU
#17Earlier quoted context omitted.
Adapting/implementing the newest and hottest cipher on the block is not something that the crypto community advocates. Do you really think crypto accelerated hardware is going to fall behind and not support the ciphers that the crypto community (academia/industry) endorses?
That depends on which team you're playing for, no?
I am not trying to be difficult but i have no idea what you are talking about. I looked through your past comments and you seem to be a competent commenter, can you clarify that you meant?
Re: KGPU - Augmenting Linux with the CUDA GPU
#18Earlier quoted context omitted.
More and more CPUs have AES instructions and my old lenovo ideapad has a crypto coprocessor. Do you think the GPU offload will be worth it when the sytem has hw accelerated crypto?
The question should be whether dedicated hardware for accelerated crypto will be worth it when GPU offload is suitable for it. Although in that particular context (security) you might enjoy the isolation of dedicated hardware as opposed to sharing it with others. The GPU solution though of course has the advantage of being able to adapt to new ciphers etc.
Specialized, single-purpose hardware is right now some 10x more energy-efficient for the same task than a GPU, (and some 50x more efficient than a CPU). Given that modern chips are not limited by transistor density but by energy density, we're going to see more special-purpose hardware in our chips, not less.
Re: KGPU - Augmenting Linux with the CUDA GPU
#19Why is this cuda and not OpenCL? There's no reason to legitimize nvidia's proprietary nonsense, it just enables their bad behavior.
CUDA is older and more popular: http://www.google.com/trends/explore#cat=0-5&q=opencl%2C...
Re: KGPU - Augmenting Linux with the CUDA GPU
#20Why is this cuda and not OpenCL? There's no reason to legitimize nvidia's proprietary nonsense, it just enables their bad behavior.
The GPU code seems to be relatively isolated to memory operations in gpuops.cu. Not altogether sure but a quick review suggests such mapping is supported by OpenCL. So one could rewrite that module and the whole thing would work without CUDA. Of course, in terms of compilers it is going to be a while before users can move away from nvcc for nvidia graphics card support.
However, as with the parent I'd really like to see a generic OpenCL vectorization kernel module. Strong suspicion that this work was directly or indirectly underwritten by Nvidia so I guess someone (Intel?) needs to step up and fund similar academic projects.