bruce wrote:...If you have the hardware that can support Double Precision and OpenCL 2.0 has been installed, please report the results of running FAHBench...
bruce wrote:Foldy has demonstrated that CPUs with OpenCL 1.2 will fail when running FAHBench, which is why FAH does not support older CPUs.
If you have the hardware that can support Double Precision and OpenCL 2.0 has been installed, please report the results of running FAHBench. Otherwise we can close this topic.
rwh202 wrote:It's possible that Folding could be able to run with little or no effort on the majority of recent intel iGPUs.
OpenCL™ 2.0 GPU/CPU driver package for Linux* (64-bit)
The Intel intel-opencl-r4.0 (SRB4) Linux driver package provides access to the GPU and CPU components of these processors:
Intel® 5th, 6th, or 7th generation Intel® Core™ processors
Intel® Celeron® Processor J3000 Series with Intel® HD Graphics 500 (J3455, J3355), Intel® Pentium® Processor J4000 Series with Intel® HD Graphics 505 (J4205), Intel® Celeron® Processor N3000 Series with Intel® HD Graphics 500 (N3350, N3450), Intel® Pentium Processor N4000 Series with Intel® HD Graphics 505 (N4200)
Intel® Xeon® processor v4 or v5 with Intel® Graphics Technology (if enabled by OEM in BIOS and motherboard)
foldy wrote:1) The requirement for double precision comes from OpenMM? Supported since 5th GPU generation, Broadwell.
It sounds like the server assignment code differentiates between projects requiring double precision and assigns to appropriate compatible GPUs.rafwiewiora wrote:Mixed precision is used (GPUs without DP are excluded).
Computers have to round numbers after a certain point (e.g. when tracking the position or velocity of an atom). single precision uses 32-bit floating point numbers (more rounding, less accurate), whereas double uses 64-bit numbers (less rounding, more accurate).
Consumer GPUs are really bad at double precision calculations (so manufacturers can sell more expensive enterprise units to researchers). Luckily, molecular dynamics really only requires single precision to be useful. Folding@Home uses single precision.
foldy wrote:So we need to guess how many donors have a 5th gen intel CPU with iGPU. I guess between 1% to 10% of CPUs.
ComputerGenie wrote:foldy wrote:So we need to guess how many donors have a 5th gen intel CPU with iGPU. I guess between 1% to 10% of CPUs.
Since the 6th gen has been the #1 selling CPU for the last 5 quarters, I'd add ~5,000,000 units to whatever that number is.
Joe_H wrote:ComputerGenie wrote:Since the 6th gen has been the #1 selling CPU for the last 5 quarters, I'd add ~5,000,000 units to whatever that number is.
In my opinion, that is a completely unrealistic approach to estimating the potential iGPU's that would be available to folding. If you added a number more reflective of actual participation rates compared to CPU's sold, it would be likely be less than 0.1% of that. In addition, a large portion of that 5,000,000 units reflect sales of lower end CPU's that have correspondingly lower performance integrated video which will be less likely to meet the computational needs of folding.
foldy wrote:.... A fahcore for the iGPUs will not reach new donors in the first place but increase the performance of the current donors. That is why we calculate how many usable iGPUs the current donors may have.
bruce wrote:Why would anyone be telling them that "nobody cares"?
bruce wrote:The fact is that they care very much about increasing the science that's getting done but their funds are not unlimited. Since they can't do everything, they must establish priorities about the best way to increase the science.
Users browsing this forum: No registered users and 3 guests