
We have developed a new release of our HyperDrive core library including the OpenCL support for GPU computing. It means that some calculation are performed by the graphic card and not by CPU. This approach introduces an incredible reduction of the calculation time as reported in this preliminary test:
- System configuration:
AMD Phenom II X4 955 quad core 3.2 GHz, 4 Gb DDR3 Ram, Sapphire/ATI HD5770 1 Gb DDR5 PCIe card.
- Software for the test:
Windows 7 Professional x64, VEGA ZZ 2.4.0, HyperDrive 2.0 for K8 and more CPUs.
- Type of test:
MEP surface calculation (Type: MEP, Dots, Probe Rad: 0, Density: 10) of the inositol monophosphate dehydrogenase (8598 atoms) which file name is impdh.pdb.bz2 and it's placed in the ...VEGA ZZDemoZZ directory.
The results are summarized in the following table:
Calculation mode | Time (seconds) |
OpenCL GPU | ~1 |
OpenCL CPU (GPU emulation) | 11 |
CPU 4 cores | 28 |
CPU 3 cores | 37 |
CPU 2 cores | 56 |
CPU 1 core | 111 |
A middle-end graphic card as the ATI HD5770 is more than 25 times faster than the quad core CPU and more than 10 time faster than the CPU in GPU emulation.
The new HyperDrive library detects automatically the best device (GPU, accelerator and CPU) to run the calculation, building the OpenCL source code for the specific target or disabling the acceleration when it's unavailable. In this way, the backward compatibility is maintained.
At this time, ATI/AMD and NVIDIA only sell OpenCL-enabled GPUs and to check if your graphic card can run OpenCL application, look the Web site of these manufacturers.
The OpenCL technology will be included in the 2.4.0 release of VEGA ZZ for Windows and VEGA for Linux x86 and x64.