Folding@Home, Team 239902

[Minotaurtoo]

let's pretend I'm a complete idiot for a moment... I haven't noticed this phenomenon of decreased PPD while CPU folding, of coarse I have pretty much always folding on the CPU and GPU, however, I think I would have noticed others getting significantly more GPU PPD than me... on the 5700 XT side I usually see around 1.2 million and around 400K on the cpu... sometimes I've seen a bit higher like 1.35 million/500K on easy ones when I have the power limits off.. I usually limit my cpu to 100w, no real reason other than keeping it in the efficiency window.... sooo, should I try without the CPU and see?  Or am I strangely lucky and getting what I should expect.

[Bastiaan_NL]

I'm not sure what the reason is, if it has anything to do with the newer units and the Cuda support.

The first time I noticed it was while I was finishing both clients to play a game. With the CPU client finished I saw the GPU client increase in PPD.
It was always the same thing, give your GPU client 1 core and it has enough CPU power to run everything. But for some reason it is not enough anymore.

You can always try it out by pausing or finishing the CPU client and take a closer look at the TPF from the GPU client, at least for 10 frames. 

 

I wrote a post about Process Lasso yesterday explaining that the program works perfectly but there is still a decrease in PPD(a little less than without the program) but I hit refresh by accident.... (insert facepalm)

 

Edited December 22, 2020 by Bastiaan_NL

[Minotaurtoo]

I tried a few times and so far can't tell any significant difference between runs with CPU folding and not... I know the PPD estimates are hot garbage at times, but since starting folding for the competition today I've seen up to 1.8m ppd with high power limits and 1.6m ppd with 65w limit on the 3900x and 150w on the 5700xt... I tend to limit power when running days long computations or like now when it's hot in this room in December lol... supposed to cool down tomorrow so I'll be ramping back up then.

[Bastiaan_NL]

Okay, good to hear. Limiting the power is not a bad thing at all, undervolting and underclocking is something many of the 24/7 folders use to reduce power consumption.

 

I've found a problem on the second system and solved a part of it.
The voltage on the PCIe slots was dropping to 11.4v for the upper slot and 11.5v on the lower slot. Turned out it's a bad idea to use an extension cable

Now it's still dropping to 11.8v but that's .4v less than before with the exact same load on the upper slot, the lower slot is at 12v.
I've got the 5 months old PSU from the main system ready to go in that system to replace the 10 years old PSU and I've installed a new one in the main system.
No time to install the PSU in the second system though, competition time

[axipher]

  On 23/12/2020 at 22:00, Minotaurtoo said:

I tried a few times and so far can't tell any significant difference between runs with CPU folding and not... I know the PPD estimates are hot garbage at times, but since starting folding for the competition today I've seen up to 1.8m ppd with high power limits and 1.6m ppd with 65w limit on the 3900x and 150w on the 5700xt... I tend to limit power when running days long computations or like now when it's hot in this room in December lol... supposed to cool down tomorrow so I'll be ramping back up then.

Expand  

 

As you mentioned, the PPD estimate is bollocks at best, when I'm testing settings for CPU or GPU folding, I relay much more on the TPF after a full 1% of a Work Unit under the ne settings.

[axipher]

  On 23/12/2020 at 23:05, Bastiaan_NL said:

Okay, good to hear. Limiting the power is not a bad thing at all, undervolting and underclocking is something many of the 24/7 folders use to reduce power consumption.

 

I've found a problem on the second system and solved a part of it.
The voltage on the PCIe slots was dropping to 11.4v for the upper slot and 11.5v on the lower slot. Turned out it's a bad idea to use an extension cable

Now it's still dropping to 11.8v but that's .4v less than before with the exact same load on the upper slot, the lower slot is at 12v.
I've got the 5 months old PSU from the main system ready to go in that system to replace the 10 years old PSU and I've installed a new one in the main system.
No time to install the PSU in the second system though, competition time

Expand  

 

Odd to see that level of 12V drop over time, not sure exactly what topology it's using to regulate the voltage, but a drop that far over time is a little worrisome.  If the power supply only put out 11.9 V under load from day one, then I wouldn't worry about it, but if it was 12 V on day one with little to no change under load, then definitely time to retire that as a computer PSU and repurpose it as a benchtop electronics power supply, or for testing fans and water cooling.

[Bastiaan_NL]

I've got to be honest that it's the first time I checked those voltages.

Somebody mentioned something about it under a video on Youtube so I added it to GPU-z (I normally only watch a few things like power consumption and load). So I have no idea what it has been doing from day one, though I never had problems like crashes.
It was still running strong now, only displaying low voltages from the PCIe slot. The extension cable was on the 24 pin, made it a little easier to route it in a previous build and so I also used in this case. The 8 pin cables are all at 12v under load, pulling 450w total on 2 GPU's of which only 90w divided over the two PCIe slots. So imagine maxing out those slots, the voltage would drop even more.

 

The PSU will be fine for the water cooling testing etc. like you said, but after 10 years of service I don't mind putting it aside

[axipher]

Cable extensions shouldn't really make that much of a difference in the grand scheme of things unless they are 18AWG and really low quality connectors.  Also software monitoring is also not perfect, depending on the switching frequency of the power supply's DC/DC conversion, the voltage measurements on the motherboard could get locked in to weird sports on that switching frequency and report some slightly off voltages.  Always best to use a known good voltmeter and test points on the board externally from software for true numbers.

[Bastiaan_NL]

I think the connectors are causing the problems, I don't think 8" long cables will cause that much resistance.
Software monitoring is not perfect, I agree on that. But if I pause one client and it goes from 11.4v back to 12v and goes back down to 11.4v as soon as I continue folding that tells me enough. 

It's a lot better anyways, and as soon as I can shut that system down I'll replace the PSU and see if that fixes the last bit of the voltage drop.
If not I'll put the DMM to good use and try to measure the board under load, though I have no idea if this cheap board has any exposed points for that.

[axipher]

  On 24/12/2020 at 08:03, Bastiaan_NL said:

I think the connectors are causing the problems, I don't think 8" long cables will cause that much resistance.
Software monitoring is not perfect, I agree on that. But if I pause one client and it goes from 11.4v back to 12v and goes back down to 11.4v as soon as I continue folding that tells me enough. 

It's a lot better anyways, and as soon as I can shut that system down I'll replace the PSU and see if that fixes the last bit of the voltage drop.
If not I'll put the DMM to good use and try to measure the board under load, though I have no idea if this cheap board has any exposed points for that.

Expand  

 

If you have a spare PCIe or EPS or Molex connector, you can just use that to test voltage from the PSU.  That is assuming you have a single rail 12V supply, or manages to find the same one the GPU is on.

[Bastiaan_NL]

A little update, both the +12v wires of the 24 pin connector are 12.16v under load and 12.18v with no load.
If I want to check any pins on the back of the GPU I'll have to take the backplate off.

 

So if GPU-Z is reading correct that would mean almost .4v gets lost somewhere in the motherboard between the connector and the PCIe slot, I doubt that.

Once the new PSU is in I'll check again.

[axipher]

  On 24/12/2020 at 21:15, Bastiaan_NL said:

A little update, both the +12v wires of the 24 pin connector are 12.16v under load and 12.18v with no load.
If I want to check any pins on the back of the GPU I'll have to take the backplate off.

 

So if GPU-Z is reading correct that would mean almost .4v gets lost somewhere in the motherboard between the connector and the PCIe slot, I doubt that.

Once the new PSU is in I'll check again.

Expand  

 

Hmm, so software reporting could just be wonky, or the voltage measuring chips and sense resistors are not playing nicely with high current draw parts and their accuracy is off.

[Bastiaan_NL]

As long as I don't measure the voltage on the GPU PCB after taking it apart we'll never know (I won't if it still works)  

Taking the extension out made a big difference in the reading, but maybe the real voltage drop is just a fraction of the displayed voltage drop.

 

 

[axipher]

  On 24/12/2020 at 22:51, Bastiaan_NL said:

As long as I don't measure the voltage on the GPU PCB after taking it apart we'll never know (I won't if it still works)  

Taking the extension out made a big difference in the reading, but maybe the real voltage drop is just a fraction of the displayed voltage drop.

 

 

Expand  

 

Well if the lockdown persists in to the new year for a good while, I was thinking of making some extensions with built-in voltage monitoring to a Arduino.

[Supercrumpet]

Haven't gotten the chance to watch this yet, but it could be interesting:

 

 

[Bastiaan_NL]

I just finished watching and it was interesting indeed, thanks for sharing

 

And a new milestone, a total of 2 billion points

[ENTERPRISE]

  On 07/01/2021 at 20:56, Supercrumpet said:

Haven't gotten the chance to watch this yet, but it could be interesting:

 

 

Expand  

Nice !

  On 08/01/2021 at 18:20, Bastiaan_NL said:

I just finished watching and it was interesting indeed, thanks for sharing

 

And a new milestone, a total of 2 billion points

Expand  

Well played dude !

[Bastiaan_NL]

Sub 300, congrats  
I should have been here.. 

 

Edited January 19, 2021 by Bastiaan_NL

[BWG]

Alright, I'm all switched over. Folding MOAR!

[ENTERPRISE]

  On 09/02/2021 at 00:23, BWG said:

Alright, I'm all switched over. Folding MOAR!

Expand  

Another won over to the light side lol. Nice one bud.

[Avacado]

  On 08/01/2021 at 18:20, Bastiaan_NL said:

I just finished watching and it was interesting indeed, thanks for sharing

 

And a new milestone, a total of 2 billion points

Expand  

Grats my old nemesis. You sure did cover a lot of ground. 

[Bastiaan_NL]

  On 09/02/2021 at 00:23, BWG said:

Alright, I'm all switched over. Folding MOAR!

Expand  

Nice

 

 

  On 10/02/2021 at 12:56, Avacado said:

Grats my old nemesis. You sure did cover a lot of ground. 

Expand  

Thanks! 
It still amazes me what these cards can do. (let's not talk about the heat output though )

[Supercrumpet]

Hit 1 billion points for EHW today! That puts me at just over three billion total:

 

https://stats.foldingathome.org/donor/245283

 

Been quite the journey with this site and with folding overall.

 

EDIT: ayy 500 posts too

Edited February 23, 2021 by Supercrumpet

[Bastiaan_NL]

Nice one!
I totally forgot to keep an eye on your stats, you are closing in on me

[Supercrumpet]

  On 23/02/2021 at 22:09, Bastiaan_NL said:

Nice one!
I totally forgot to keep an eye on your stats, you are closing in on me

Expand  

 

Yep, and then you just had to go and participate in the competition lmao

 

The audacity...