Heey all,
Lately I've been reading some papers about gpgpu programming and thinking of thinkering with it in the future.
But I do have one question:
I've read that (nVidia)cuda cores are more powerfull then (ATI/AMD)Stream cores. But how much more powerfull?
To be more specific: How many Stream cores equall 1 cuda core?
I'm not expecting a real anwser like '1 cuda == 15 stream) but more like '1 cuda == 10 to 20 stream'
Thanks in advance,
assaintor
🎉 Celebrating 25 Years of GameDev.net! 🎉
Not many can claim 25 years on the Internet! Join us in celebrating this milestone. Learn more about our history, and thank you for being a part of our community!
Stream core vs. Cuda core
Author
Sometimes I forget that there is the internet to find this info.
I've compared some nVidia and AMD cards and found that the range is: 1 CUDA == 2 to 5 stream cores
assainator
I've compared some nVidia and AMD cards and found that the range is: 1 CUDA == 2 to 5 stream cores
assainator
That isn't a strict thing because NV and AMD classify things differently.
While AMD might quote huge numbers of 'cores' the reality is their cause prefer traditional SIMD data as they are grouped into 5 or 4 wide units (4 in the newest card). While they can co-issue to a degree heavy single component usage cuts down the amount which can be done as effective 3 or 4 'cores' are idle.
NV on the other hand use a system where they can deal with scalar input well and effective 'combine' cores to work on SIMD data. This means on heavy independant scalar code they effectively have an advantage.
So, the amount of 'power' very much depends on the work you are throwing at the GPU and how the compiler optimises the code for the GPU.
In a heavy parallel suituation AMD could have the edge, where in a scalar setup NV might.
While AMD might quote huge numbers of 'cores' the reality is their cause prefer traditional SIMD data as they are grouped into 5 or 4 wide units (4 in the newest card). While they can co-issue to a degree heavy single component usage cuts down the amount which can be done as effective 3 or 4 'cores' are idle.
NV on the other hand use a system where they can deal with scalar input well and effective 'combine' cores to work on SIMD data. This means on heavy independant scalar code they effectively have an advantage.
So, the amount of 'power' very much depends on the work you are throwing at the GPU and how the compiler optimises the code for the GPU.
In a heavy parallel suituation AMD could have the edge, where in a scalar setup NV might.
This topic is closed to new replies.
Advertisement
Popular Topics
Advertisement