User Tools
opt:graphics_cards
Differences
This shows you the differences between two versions of the page.
| Both sides previous revision Previous revision Next revision | Previous revision Next revision Both sides next revision | ||
|
opt:graphics_cards [2018/08/31 09:36] pklapetek |
opt:graphics_cards [2018/08/31 12:13] pklapetek |
||
|---|---|---|---|
| Line 12: | Line 12: | ||
| To use GPU for a calculation is not straightforward, unfortunataly. We cannot simply take a conventional PC executable and run it on GPU. Both data processing and memory model is completely different for GPU and for CPU and the part of the code that should be run on GPU (called kernel) must be written to fulfill these conditions. GPU is equipped by several multiprocessors, consisting of a large number of processors. Many hundreds of threads (kernel calls) grouped in thread blocks can be processed simultaneously on GPU, which is the basis of tremendous speedup that we can achieve. Memory available on GPU can be divided into a global memory - accessible by all the multiprocessors, a shared memory - accessible by processors within one multiprocessor, and a local memory - accessible by single processor. All the memories are hardware limited (for each type of GPU differently). We refer to Nvidia CUDA developer zone for further details. | To use GPU for a calculation is not straightforward, unfortunataly. We cannot simply take a conventional PC executable and run it on GPU. Both data processing and memory model is completely different for GPU and for CPU and the part of the code that should be run on GPU (called kernel) must be written to fulfill these conditions. GPU is equipped by several multiprocessors, consisting of a large number of processors. Many hundreds of threads (kernel calls) grouped in thread blocks can be processed simultaneously on GPU, which is the basis of tremendous speedup that we can achieve. Memory available on GPU can be divided into a global memory - accessible by all the multiprocessors, a shared memory - accessible by processors within one multiprocessor, and a local memory - accessible by single processor. All the memories are hardware limited (for each type of GPU differently). We refer to Nvidia CUDA developer zone for further details. | ||
| - | To check if the graphics card installed on your computer is suitable for GSvit calculations and if GSvit was installed with graphics card support at all, you can run the solver with parameter "test 0", e.g. on Linux system: | + | To check if the graphics card installed on your computer is suitable for GSvit calculations and if GSvit was installed with graphics card support at all, you can run the solver with parameter "test 0", e.g. on Linux system (see [[start:tests|all the tests]] for more details): |
| <html> | <html> | ||
| Line 67: | Line 67: | ||
| Even if we try to do our best to have one to one correspondence for everything on CPU and GPU, it can happen that some of the recent algorithms are not yet implemented on GPU. The reference should always be the CPU implementation. We therefore strongly recommend, namely for setting up the new task, to test the task first on CPU, crosscheck it with GPU and then run the various repetitive calculations on GPU to speedup the solution. | Even if we try to do our best to have one to one correspondence for everything on CPU and GPU, it can happen that some of the recent algorithms are not yet implemented on GPU. The reference should always be the CPU implementation. We therefore strongly recommend, namely for setting up the new task, to test the task first on CPU, crosscheck it with GPU and then run the various repetitive calculations on GPU to speedup the solution. | ||
| - | As an example you can test the following parameter files, comparing performance of GSvit on different number of CPU cores and on a single GPU on the above mentioned system. The calculation took on a single core, 1 minute 43 seconds on eight cores, and 54 seconds on the first GPU. | + | As an example you can test the following parameter files, comparing performance of GSvit on different number of CPU cores and on a single GPU on the above mentioned system. The calculation took 8 minutes 37 seconds on a single core, 1 minute 43 seconds on eight cores, and 54 seconds on the first GPU. |
| ---- | ---- | ||
| // | // | ||
opt/graphics_cards.txt ยท Last modified: 2018/09/04 17:24 by pklapetek
Except where otherwise noted, content on this wiki is licensed under the following license: CC Attribution-Noncommercial 4.0 International
