Folding Forum

ThWuensche wrote:As for the PPD, I don't care much. What I care about is the volume of calculations required to support scientific results. I just bought another PC and two more Radeon VII to support the scientific results (not the PPD). As I understand the support of project moonshot will need a lot of computational power and it's sad if big part of it stays unused (for the electricity bill it's good, but ...)

What I conclude from the observations from Crawdaddy and BobWilliams supports the second of the reasons mentioned above, latency in interaction with the CPU. For a low end GPU, the GPU is limiting and not the interaction with the CPU, for a high end GPU, the interaction with the CPU becomes limiting. Thus latency would be a big show-stopper for high-end GPUs, but mostly go by unnoticed on smaller GPUs. The latency issue seems also confirmed from Crawdaddy's test with stopped CPU folding. The low memory footprint I observed is good for GPUs with limited memory and memory bandwidth, but leaves the benefit of the 16G fast RAM on the Radeon VII unused. So bigger work packages and less interaction might help to make good use of high-end GPUs (if that is possible from the nature of the WUs). Don't know about behaviour of the bigger NVidia GPUs in this situation, any ideas?

ThWuensche wrote:First possible reason might be that the problem does not scale well to highly parallel architectures, parts of code with parallelization are intermixed with and dependent on parts which can not be parallelized, that way limiting the overall performance by those parts which can not be executed in parallel. This probably would be hard to circumvent unless the algorithm may be changed. Only help in that case would be to run WUs in parallel on a GPU.

_r2w_ben wrote:

ThWuensche wrote:First possible reason might be that the problem does not scale well to highly parallel architectures, parts of code with parallelization are intermixed with and dependent on parts which can not be parallelized, that way limiting the overall performance by those parts which can not be executed in parallel. This probably would be hard to circumvent unless the algorithm may be changed. Only help in that case would be to run WUs in parallel on a GPU.

With 3840 shaders, Radeon VII is a very wide. p13409 only has 4,082 atoms, reducing available parallelization. Theoretically, if the same calculation was done for each atom and each atom received it's own shader, it would take two passes and only 242 of 3840 (5% of the GPU) would be active on the second pass. Combined the two passes would represent 52.5% utilization. As the number of atoms increases, the effect of less than full utilization on the last pass is reduced. For example, you'll probably see above average performance on p14201, which has 453,348 atoms. The same calculation per atom would take 119 passes with 228 of 3840 (6% of the GPU) utilized on the last pass. All passes combined would represent 99.2% utilization.

BobWilliams757's onboard video benefits from the lack of utilization on the higher end GPUs used when p13409 was benchmarked. The low number of shaders results in close to full utilization regardless of the number of atoms.

This is an oversimplification since Molecular Dynamics involves a lot of calculations per atom and between pairs of atoms each timestep but should give an idea of what's happening.

Crawdaddy79 wrote:I take back what I said about points. I got seven of these in the last 12 hours. Not cool.

ajm wrote:EDIT: Strangely, none of my other 3 GPUs working in the same time has received any 13409 WUs.

PantherX wrote:

Crawdaddy79 wrote:I take back what I said about points. I got seven of these in the last 12 hours. Not cool.

My I suggest a suggest perspective that might change your POV... while these 1340X Projects yield a low PPD, they are in fact highly experimental and were specifically created to help the Moonshot project (https://covid.postera.ai/covid). Researchers are working to understand this and make it a more efficient process which would mean that the next set of these Projects could be more efficient. Helping to create a potential drug for COVID-19 would be a historic milestone for humanity given that it would be released to the public for free without any restricts. It might seem frustrating in the short term but IMO, it's totally worth for the long term

Crawdaddy79 wrote:Since 6 AM I've gotten four more of these. The server seems to like my system; I've only had one faulty return and I think it was a legit bad work unit.

Crawdaddy79 wrote:Since 6 AM I've gotten four more of these. The server seems to like my system; I've only had one faulty return and I think it was a legit bad work unit.

PantherX wrote:

Crawdaddy79 wrote:I take back what I said about points. I got seven of these in the last 12 hours. Not cool.

My I suggest a suggest perspective that might change your POV... while these 1340X Projects yield a low PPD, they are in fact highly experimental and were specifically created to help the Moonshot project (https://covid.postera.ai/covid). Researchers are working to understand this and make it a more efficient process which would mean that the next set of these Projects could be more efficient. Helping to create a potential drug for COVID-19 would be a historic milestone for humanity given that it would be released to the public for free without any restricts. It might seem frustrating in the short term but IMO, it's totally worth for the long term

Science...

Just kidding. I did look at the project page last night and I agree it is worthy. Had to turn the temp on my A/C up though because my PC is no longer keeping my basement warm.

Hopefully this will drag some of the teams creeping up on us back some as well.