Amd Vulkan beta driver & Fah

[Bryman]

What makes you think Fah is not using more than 60%? The Fah CPU stress test is known to test a CPU as hard if not harder (and hotter) than OCCT. And the CPU client isn't far behind that. Fah has a special talent for optimizing performance to squeeze every CPU cycle to make the science go faster.

If Fah isn't using full gpu resources (which may be true, but haven't seen proof of that) then some other hardware or software is holding it back. And that will be the next target for development.

GPU... not CPU...

Yes, it's pretty efficient with the cpu.. but with the gpu it has to wait on a bunch of calculations which slows down the video card

Games use a lot more power than folding@home does


Folding@home is probably more like 50% true video card usage, while games are like 60%

I tested 3 different stress test programs... Kombustor took the least power consumption (cause it stresses the ram the most, which slows down the gpu) then furmark then OCCT

OCCT stressed the ram the least though, the ram had the least power consumption... OCCT was about 33 watts, Furmark was 45, and Kombustor was 52

But overall OCCT has the most power consumption, thus the most gpu usage

I don't know of anything that has more power consumption than OCCT, then again.. I don't know a lot of stress test programs either

[Bryman]

It's actually 55 to 110 degrees in 15 seconds

https://www.youtube.com/watch?v=jQ4RN6cpmdI

1.35 volts 1170mhz

[7im]

Yes, GPU, not CPU. I was using the CPU history as an example of how fah will use as much folding resources as possible.

>>OCCT stressed the ram the least though, the ram had the least power consumption... OCCT was about 33 watts, Furmark was 45, and Kombustor was 52

And how many watts does FAH use by comparison (noting that memory speed of a GPU has very little to do with FAH performance)?

[foldy]

@Bryman: Maybe you can stick to your other thread reporting what you are doing with your GPU?
viewtopic.php?f=38&t=28667

[Bryman]

@Bryman: Maybe you can stick to your other thread reporting what you are doing with your GPU?
viewtopic.php?f=38&t=28667

Guess I did say a little more info then I needed to.. basically what I was trying to say is that if you want a program to use the gpu near 100%, then run OCCT, but make sure you turn the voltage up on your gpu first... or your screen will start artifacting the !@#$ out of itself and your drivers will crash/BSODs/computer restarts/lines on screen

All that fun stuff

[bcavnaugh]

What makes you think Fah is not using more than 60%? The Fah CPU stress test is known to test a CPU as hard if not harder (and hotter) than OCCT. And the CPU client isn't far behind that. Fah has a special talent for optimizing performance to squeeze every CPU cycle to make the science go faster.

If Fah isn't using full gpu resources (which may be true, but haven't seen proof of that) then some other hardware or software is holding it back. And that will be the next target for development.

Do you think that this applies to GPU testing as well?

[7im]

What makes you think Fah is not using more than 60%? The Fah CPU stress test is known to test a CPU as hard if not harder (and hotter) than OCCT. And the CPU client isn't far behind that. Fah has a special talent for optimizing performance to squeeze every CPU cycle to make the science go faster.

If Fah isn't using full gpu resources (which may be true, but haven't seen proof of that) then some other hardware or software is holding it back. And that will be the next target for development.

Do you think that this applies to GPU testing as well?

I couldn't say for sure, I only have low and mid-range GPUs to test, and are not fast enough to find other limitations.

As I said...

If Fah isn't using full gpu resources (which may be true, but haven't seen proof of that) then some other hardware or software is holding it back. And that will be the next target for development.

If you or anyone else have some numbers to post showing under utilitzation, please do so. It will help the developers.

[bcavnaugh]

What makes you think Fah is not using more than 60%? The Fah CPU stress test is known to test a CPU as hard if not harder (and hotter) than OCCT. And the CPU client isn't far behind that. Fah has a special talent for optimizing performance to squeeze every CPU cycle to make the science go faster.

If Fah isn't using full gpu resources (which may be true, but haven't seen proof of that) then some other hardware or software is holding it back. And that will be the next target for development.

Do you think that this applies to GPU testing as well?

I couldn't say for sure, I only have low and mid-range GPUs to test, and are not fast enough to find other limitations.

As I said...

If Fah isn't using full gpu resources (which may be true, but haven't seen proof of that) then some other hardware or software is holding it back. And that will be the next target for development.

If you or anyone else have some numbers to post showing under utilitzation, please do so. It will help the developers.

Thank you 7im,
So for now would it be best to use client-type advanced or no client-type at all for my AMD Graphics Cards using this Hotfix Driver 16.4.1 ?

[Bryman]

*sigh* what I ment is it's running as fast as it can for the calculations it's doing

OCCT can use more of the GPU, is all I meant

Run OCCT and look at the power consumption, compare it with folding@home

I didn't necessarily mean that folding@home wasn't using as much of the GPU as it could



I just tested both folding@home and OCCT with same clock speeds and voltages

folding@home was about 230 watts while OCCT was about 380 watts

960mhz core 1200mhz memory 1.3 volts core 1.5 volts memory both tested at about 80°C

I have a 7970

[bruce]

History lesson:
In the early days of GPU folding, many GPUs had so many threads that it could finish the calculations before the PCIe bus could supply data for the next portion of the calculations. GPUs have gotten more powerful and PCIe buses have gotten faster, but not in lock-step. Also, proteins have gotten (much) bigger.

I don't know if this applies in your situation, but it may be that your GPU is showing the symptoms of data-starvation. Can you set up the same GPU in a machine with a faster PCIe bus and report the results of running another WU from the same project?

[Bryman]

History lesson:
In the early days of GPU folding, many GPUs had so many threads that it could finish the calculations before the PCIe bus could supply data for the next portion of the calculations. GPUs have gotten more powerful and PCIe buses have gotten faster, but not in lock-step. Also, proteins have gotten (much) bigger.

I don't know if this applies in your situation, but it may be that your GPU is showing the symptoms of data-starvation. Can you set up the same GPU in a machine with a faster PCIe bus and report the results of running another WU from the same project?

Are you talking to me?

HWiNFO is reporting the link speed at 8 Gbps

Just tested again, still showing about 230 watts at 80°C... PPD is 160,000 which is one of the highest power consumption WUs

If it was a 130,000 PPD WU then the power usage would only be like 200, instead of 230

OCCT is 380 at the same voltage, clock speed, and temperature

I even paused the CPU WU and it made the wattage go up maybe like 5?

230 for folding@home and 380 for OCCT

And I'm including the memory wattage, which is taking about 30

About 140 with 1.075 volts... man, voltage makes a big difference on power consumption... 20% more volts is like 50% more wattage

[7im]

...snip...

Thank you 7im,
So for now would it be best to use client-type advanced or no client-type at all for my AMD Graphics Cards using this Hotfix Driver 16.4.1 ?

Client type setting is about risk, and not about points. If you are willing to beta test, and closely monitor your systems, and run the latest Core_21 work units that might have coding issues, then beta type is a good choice. If you want less risk, run Advanced. If you want the least risk, run normal fah work units.

Sorry, I don't have any driver specific recommendations at this time, other than run the latest released (non-beta) version of either AMD or NV drivers. Neither one has had significant issues with the fah software lately. Not to where people were recommending very specific versions for very specific types of cards as in the past.

Lately is seems lots of people are willing to fall on the latest beta driver hand grenades and report back. So as of late I have followed what others recommend here in the forum unless I have specifically done testing on my own. I am happy with the 16.3.2 driver on my HD 8570 for the time being.

[7im]

...snip...

I just tested both folding@home and OCCT with same clock speeds and voltages

folding@home was about 230 watts while OCCT was about 380 watts

960mhz core 1200mhz memory 1.3 volts core 1.5 volts memory both tested at about 80°C

I have a 7970

Thanks, that helps us understand better. But when OCCT is using DX as it's testing method on the GPU, it concerns me that doesn't compare well with the compute function that FAH is using. Then a Watts vs Watts comparison is also less meaningful.

I also don't know what those GPU monitoring softwares are measuring for GPU usage. If they measure DX and not OpenCL, then again, the numbers are not meaningful. GPU usage monitors seem like they are designed to monitor gaming performance. Is there even a GPU performance monitor for the compute side? To much unknown to draw good conclusions, IMO.

[bruce]

Are you talking to me?

I'm talking to anyone who is complaining that their GPU isn't showing 100%. It might be drivers or it might be hardware or it might be an OpenCL limitation that needs attention.

[DarkFoss]

I don't know whether you're running a CPU slot alongside you GPU one, but some projects (like p13xxx) don't like when something interfere with the core on the CPU. Try to pause your CPU slot, or free more threads (SMP 6 on a 8 threads CPU for instance).

Sorry for the delay rl called. I'm not running fah on the cpu,right now I'm folding mostly on my second Fury X non crossfired. Using my main Fury for everyday use, gaming vid binging ect nothing running in the background most of the time except for antivirus,HFM ect.