Folding Forum

Not sure I'm using the new FAH bench correctly, but I've been playing with over and under clocking the mem / gpu on my 1660 super. I haven't tried pushing this card to the bleeding edge, but instead looked at 3 reviews, found a memory and gpu clock that was achieveable in all of the reviews and went with that for this testing.

WU dhfr
atoms 23558
5 minute run time

Stock Memory (6800) / stock GPU (boost 1905)
Score: 77.936

+600 Memory (7400) / stock (boost 1905)
Score: 79.543

Stock Memory (6800) / +170 GPU (boost 2070)
79.6323

-1000 Memory (5800) / +170 GPU (boost 2070)
score 79.49

+600 Memory / +100 GPU (boost 1995)
Score: 80.73

Juggy wrote:

NoMoreQuarantine wrote:I created a post to help people get HFM.NET setup viewtopic.php?f=16&t=34516

Thank you for this, very handy. I found an article from 2009 with a very old client that I couldn't figure out.

Glad to hear it, let me know if you have any questions it doesn't answer or if you have any suggestions to improve it.

And with the real WU:

WU real
atoms 64614
1 min

Stock Memory (6800) / stock (boost 1905)
score: 51.239
Scaled: 154.621

Stock Memory (6800) / +170 GPU (boost 2070)
Fail

+600 Memory (7400) / stock (boost 1905)
score: 51.976
Scaled: 156.846

-1000 Memory (5800) / +170 GPU (boost 2070)
score: 52.544
Scaled: 158.561

+600 Memory (7400) / +100 GPU (boost 1995)
score: 53.519
Scaled: 161.502


Looks like memory does make a difference.

NoMoreQuarantine wrote:

Juggy wrote:

NoMoreQuarantine wrote:I created a post to help people get HFM.NET setup viewtopic.php?f=16&t=34516

Thank you for this, very handy. I found an article from 2009 with a very old client that I couldn't figure out.

Glad to hear it, let me know if you have any questions it doesn't answer or if you have any suggestions to improve it.

I do have a question, I have done one WU with the CPU so I went to tools, work unit history and did a CSV export. I don't see the hardware type mentioned in the export though. Must I specify that when uploading the CSV?

Endgame124 wrote:Looks like memory does make a difference.

Fascinating. My theory based on these results (and my own which I didn't feel worth posting) is that some portions of the simulation are bandwidth limited while others are not. That would explain why the results increase even when the memory is underclocked while the GPU is overclocked; the parts that are bandwidth limited are made up for by the speed at which the parts that aren't are completed (like a car that is good at straightaways while being bad at cornering, if you'll pardon the analogy). Unfortunately, it is unpredictable at this point to say what percentage of any WU will be bandwidth limited, but I will hazard a guess that it will vary by simulation. Therefore, while it may be true that memory bandwidth is not as important as FP32 TFLOPS, it still has a measurable impact on WU completion time.

Juggy wrote:I do have a question, I have done one WU with the CPU so I went to tools, work unit history and did a CSV export. I don't see the hardware type mentioned in the export though. Must I specify that when uploading the CSV?

Correct, either specify it in the Google Form or modify the csv file and rename the slots. It's unfortunate. foldinghomealone made a feature request on github to allow users to manually name slots though.

Juggy wrote:...I don't see the hardware type mentioned in the export though. Must I specify that when uploading the CSV?

If you have only 1 slot (or 1 GPU and 1 CPU) then you can rename the slot in HFM to like 'RTX 2080Ti - i9900KS' (or whatever HW you're using) and so I can do the rest.

NoMoreQuarantine wrote:

Endgame124 wrote:Looks like memory does make a difference.

Fascinating. My theory based on these results (and my own which I didn't feel worth posting) is that some portions of the simulation are bandwidth limited while others are not. That would explain why the results increase even when the memory is underclocked while the GPU is overclocked; the parts that are bandwidth limited are made up for by the speed at which the parts that aren't are completed (like a car that is good at straightaways while being bad at cornering, if you'll pardon the analogy). Unfortunately, it is unpredictable at this point to say what percentage of any WU will be bandwidth limited, but I will hazard a guess that it will vary by simulation. Therefore, while it may be true that memory bandwidth is not as important as FP32 TFLOPS, it still has a measurable impact on WU completion time.

If we can get someone to run some results with a standard 1660, I think it will help fill in some gaps (though in not sure if system CPU also has an impact). Given how much less bandwidth the standard 1660 has, we can get a better feel for memory impact if they could run from -1000 to +600 (if possible) while keeping the GPU at stock speed. The top end of 1660 memory OC might be close to a 1660 super at -1000.

foldinghomealone wrote:

Juggy wrote:...I don't see the hardware type mentioned in the export though. Must I specify that when uploading the CSV?

If you have only 1 slot (or 1 GPU and 1 CPU) then you can rename the slot in HFM to like 'RTX 2080Ti - i9900KS' (or whatever HW you're using) and so I can do the rest.

Thanks, I will do so.

NoMoreQuarantine wrote:

Juggy wrote:

NoMoreQuarantine wrote:I created a post to help people get HFM.NET setup viewtopic.php?f=16&t=34516

Thank you for this, very handy. I found an article from 2009 with a very old client that I couldn't figure out.

Glad to hear it, let me know if you have any questions it doesn't answer or if you have any suggestions to improve it.

Sorry, another question. How many completed WU's should we do before submitting our sample?

Juggy wrote: Sorry, another question. How many completed WU's should we do before submitting our sample?

In the meantime I have uploaded a single sample for i9-9900KF and RTX 2080 Super so we can get something in the DB.

Juggy wrote:Sorry, another question. How many completed WU's should we do before submitting our sample?

Any samples are better than no samples. Don't be shy submitting them as they are complete. I don't think we're at a volume yet that we'll overwhelm foldinghomealone.

In the beginning upload once per day. But as soon as we have several samples a longer period would be better.
But it's up to you.
As long as you don't send me every WU separately

@NMQ:
Something wrong with your slot names? It says RX 590 but sometimes it's a GPU slot.
But I could figure it out.

If you fold with the same client CPU and GPU then please name it like 'RX 590 - R5 2600' then I will know what to do.

How do you filter previously uploaded results?