Folding Community Board

[Bitter]

I, the idiot, has figured out my no-post issue with the RX580 and it's in the system, added to fah, but trying to pull work from a 192.0.2.1 address and failing. I ran across some information that this is where the misfit GPU's get assigned a bunk address when they can't fold. Is this an open source default driver issue with OpenCL? I was folding with this exact card under Windows this spring no problems.

 

So the no-post was because my dumbass forgot to force the display to CPU Graphics in BIOS before installing a graphics card with outputs on it. That's a hard face-palm right there.

 

Edit: Oh yea, need the 'Pro' drivers, this is coming back to me now slowly. Currently installing those. Should give me a 1M PPD box and space heater all in one! Need to figure out how to undervolt the RX580 on Ubuntu next.

 

Edit: Not sure the drivers installed, blargh this is annoying. I can't use the P106-100 in Windows and the RX580 doesn't want to work in Linux. LOL!

 

Edit: Got it! Needed to edit a file to say root. Now why doesn't corectrl give me core voltage control? This RX580 can fold at 120W at 1400mhz, not 150W at 1380! All sorted now, forgot about the grub entry that was needed.

[Bitter]

44 minutes ago, yaboistar said:

this is hideously inefficient but i'm doing it to give the lar database some up to date stats for these cards

 

Whats all this then? I've got some odd hardware folding, does this work in Linux easily?

[chaozbandit]

2 hours ago, Bitter said:

Whats all this then? I've got some odd hardware folding, does this work in Linux easily?

 

The display is the v1.1.2 LAR web ui, otherwise its the standard FAH linux installation. I can't say if it works on firefox as I've got it on all my machines via chrome.

[Bitter]

40 minutes ago, chaozbandit said:

 

The display is the v1.1.2 LAR web ui, otherwise its the standard FAH linux installation. I can't say if it works on firefox as I've got it on all my machines via chrome.

If it's a chrome extension it won't work in firefox,but I can install chromium on the folding box and it should work there. Might give it a go later on. I need to move it to it's home in the basement sometime this week anyway and figure out some fans to keep it better fed cool air. It's a  LITTLE cramped in there with the big 580 and P106 both in a very old case, side open but they're still close together and blasting each other with heat all the time. Managed to do a  little about that but I think I'm going to have to do a better job, lower basement air temps might help enough though.

[justpoet]

7 hours ago, Bitter said:

Whats all this then? I've got some odd hardware folding, does this work in Linux easily?

Chrome extension.

https://folding.lar.systems/#get_FITDAT

@LAR_Systems does it also work in the open source chromium on Linux?

[LAR_Systems]

1 hour ago, justpoet said:

Chrome extension.

https://folding.lar.systems/#get_FITDAT

@LAR_Systems does it also work in the open source chromium on Linux?

Not sure about base Chromium, if it can run a Google Web Extension it should work in theory.

My main folding server is also my Linux testing machine that runs Chrome on Linux Mint just fine.

[Shlouski]

On 10/29/2020 at 8:40 PM, shaz2sxy said:

Do you think a 520w psu will be enough for a

 

The wattage alone is meaningless, for starters if you have a bad psu then it likely wont live up to its rating.

 

The 520w watts will be max output, but without knowing the amperage on each rail I could only speculate, it likely would be really on the edge. Calculate how many amps you need on the 12v rail and look at how many your psu has, then add a bit of a buffer.

 

I have a NOX 800w psu and a NOX 600w, which one did I use for my 3080, the one with more amps on the 12v rail, and thats the 600w psu. 

[Shlouski]

On 10/29/2020 at 9:18 PM, marknd59 said:

2070 super power draw is 215w

1070 ti power draw is 180w

 

That puts you at 395w just for your GPU's I'd be looking for a 750w PSU

 

I love that you are trying to help, but this advise in rare circumstances could lead to issues.

 

This PSU below is only 50 watts under what you suggest, and if he were to use this he would be treated to blue smoke in no time. This shows you how meaningless total watts can be, you shouldn't give advice based on it alone, I have seen people run out and buy ones just like this because they were told they needed a certain wattage. The 2070 super needs 18 amps and 1070ti needs 15 amps, then you need to factor in the rest of the system.

 

 

Just for context, this is an average quality 500w psu:

 

 

This psu would stand a better chance of running his system even though its 200 watts less, because it has 36 amps on this 12v rail vs the 700 watt psu that only has 33 amps, the 700w psu has a lot of amps on the 5v and 3.3v rails.

 

Please don't use total watts when recommending psu's.

 

 

[shaz2sxy]

9 hours ago, Shlouski said:

 

I love that you are trying to help, but this advise in rare circumstances could lead to issues.

 

This PSU below is only 50 watts under what you suggest, and if he were to use this he would be treated to blue smoke in no time. This shows you how meaningless total watts can be, you shouldn't give advice based on it alone, I have seen people run out and buy ones just like this because they were told they needed a certain wattage. The 2070 super needs 18 amps and 1070ti needs 15 amps, then you need to factor in the rest of the system.

 

 

Just for context, this is an average quality 500w psu:

 

 

This psu would stand a better chance of running his system even though its 200 watts less, because it has 36 amps on this 12v rail vs the 700 watt psu that only has 33 amps, the 700w psu has a lot of amps on the 5v and 3.3v rails.

 

Please don't use total watts when recommending psu's.

 

 

some good info

 

the 520w PSU has 40A 

 

I did try it and it worked...sort off..

 

had the 2070 super and 1070ti @ 80% but the PSU fan was on MAX and things where getting too hot so didnt keep it like that lol

[cbigfoot]

well about to fire up a GTX 780 TI for sience

[Bitter]

1 hour ago, cbigfoot said:

well about to fire up a GTX 780 TI for sience

For heat!

 

I need a cheap generic cheap inexpensive vertical GPU mount so I can stick the RX580 outside the case. Or I need to quit using the RX580 and get another Nvidia card that's more efficient. Like that GTX 1650 I keep going on about wanting to get for this use. Why aren't 1650's $50 yet?

 

Turned the RX580 (150W) off, wasn't worth the heat/power for the ~500,000 PPD. The Vega56 (175W) does a solid 1M for only about 25W more and the P106-100 does ~700,000 for 25W (125W) less. I'm also letting the V56 finish out it's current WU, warmer week coming up so I'll scale it back for a bit.

 

Vega 56 = 5700 PPW

P106-100 = 5600 PPW

RX 580 = 3300 PPW

Rounding numbers here, varies by work unit obviously, but the gap is too large to be a rounding error. The Vega is a monster of a card and I don't even have it fully tuned in, just a rough underclock/undervolt to keep the power down. I could probably get that closer to 6000 PPW if I sit around and played with it for a while but I just wanted quick and stable. I could probably OC the P106-100 a bit more and get some more effiency out of it but I don't want to fail out WU's from instability. I'm averaging about 1950-1980mhz on it pretty steadily without touching the voltage (unless it's self-bumping that?) at around 125W on average.

 

A GTX 1650 should be around 8000 PPW if it does about 600,000 PPD at 75W, a pair of those would be a solid choice to fold with once they get cheap enough, once either Xe dGPU's start hitting the market at hopefully lower prices or Nvidia/AMD release their new-gen low end stuff. I'm just impatient.

[cbigfoot]

7 hours ago, Bitter said:

For heat!

 

I need a cheap generic cheap inexpensive vertical GPU mount so I can stick the RX580 outside the case. Or I need to quit using the RX580 and get another Nvidia card that's more efficient. Like that GTX 1650 I keep going on about wanting to get for this use. Why aren't 1650's $50 yet?

 

Turned the RX580 (150W) off, wasn't worth the heat/power for the ~500,000 PPD. The Vega56 (175W) does a solid 1M for only about 25W more and the P106-100 does ~700,000 for 25W (125W) less. I'm also letting the V56 finish out it's current WU, warmer week coming up so I'll scale it back for a bit.

 

Vega 56 = 5700 PPW

P106-100 = 5600 PPW

RX 580 = 3300 PPW

Rounding numbers here, varies by work unit obviously, but the gap is too large to be a rounding error. The Vega is a monster of a card and I don't even have it fully tuned in, just a rough underclock/undervolt to keep the power down. I could probably get that closer to 6000 PPW if I sit around and played with it for a while but I just wanted quick and stable. I could probably OC the P106-100 a bit more and get some more effiency out of it but I don't want to fail out WU's from instability. I'm averaging about 1950-1980mhz on it pretty steadily without touching the voltage (unless it's self-bumping that?) at around 125W on average.

 

A GTX 1650 should be around 8000 PPW if it does about 600,000 PPD at 75W, a pair of those would be a solid choice to fold with once they get cheap enough, once either Xe dGPU's start hitting the market at hopefully lower prices or Nvidia/AMD release their new-gen low end stuff. I'm just impatient.

LOL IM folding with what i can get my grubby hands on till the 3070 is orderable. my future should look like Bender 2x 3070 Flexo 2x2060 Calculon 5700xt Lela may not fold after this event too expensive along with fexos rx580 will go for sale at end of event i plan on being 10M+ next event.  with bender being 6M+ should be easy

[TVwazhere]

Hmmm..... Technically The Server is on central time so 6 am, 9a, and 12pm is 7am, 10am and 1pm my time but still.....

[Bitter]

14 minutes ago, TVwazhere said:

Hmmm..... Technically The Server is on central time so 6 am, 9a, and 12pm is 7am, 10am and 1pm my time but still.....

Sadface.png

My p106 misses every other server check-in, my hourly looks like a freakin saw blade! Hopefully the 1650 I have coming takes care of that cheaply.

[Kilrah]

1 hour ago, TVwazhere said:

Hmmm..... Technically The Server is on central time so 6 am, 9a, and 12pm is 7am, 10am and 1pm my time but still.....

EOC might have had some issue, the last update was still "processing" an hour and a half after the update time. 

[SGRacing]

Did some more playing around with OC on the 2080 S and the CPU. I ran 3 WU at 8 threads each for a bit but decided to save 4 of them for background stuff by running 2 at 10 each.

[Bitter]

I could never decide if running 1 WU at all 8 cores or running 2 on 4 and 4 was better, so I quit CPU folding on my 2700X because the power and heat was so much worse than the the GPU for such few points. I know there's a lot of CPU only WU but there's also A LOT more CPU's doing folding out there so I guess I don't feel so bad about it.

GTX 1650 comes in Monday, can't wait to swap the RX580 out and the 1650 in and see how many PPD it does for 75W power use!

[Gorgon]

6 hours ago, SGRacing said:

Did some more playing around with OC on the 2080 S and the CPU. I ran 3 WU at 8 threads each for a bit but decided to save 4 of them for background stuff by running 2 at 10 each.

In theory 1 CPU slot at 20 threads should outperform 2 x 10 threads.

 

You should be able to see this on the fly. Set the soonest to complete slot to finish then once it’s finished note the PPD of the other slot, delete the finished slot then edit the running slot changing it from 10 to 20 and after a few frames note the new PPD.

 

It won’t be exact but it should give you an idea.

[cbigfoot]

4 hours ago, Gorgon said:

In theory 1 CPU slot at 20 threads should outperform 2 x 10 threads.

 

You should be able to see this on the fly. Set the soonest to complete slot to finish then once it’s finished note the PPD of the other slot, delete the finished slot then edit the running slot changing it from 10 to 20 and after a few frames note the new PPD.

 

It won’t be exact but it should give you an idea.

i tried 2x10s and 2x8s and 1x20 and 1x16 on my 3900x and can confirm the more threads per job the better.  reason i tried between 16and 20 was to see what effect if any on GPU...note no discernable difference. but the 2x slots where about 30% less PPD.

[SGRacing]

9 hours ago, cbigfoot said:

i tried 2x10s and 2x8s and 1x20 and 1x16 on my 3900x and can confirm the more threads per job the better.  reason i tried between 16and 20 was to see what effect if any on GPU...note no discernable difference. but the 2x slots where about 30% less PPD.

I opted for this route because I'm not necessarily chasing points, particularly on CPU WUs, and at times I want to have more cores available for what I am working on. At night and while I'm at work both do their thing and when I get home I set the one finishing soonest to finish.

 

I appreciate the input on maximizing PPD on the CPU.

[justpoet]

I've found the opposite effect on my Xeons.  Running half and half in 2 slots (minus a couple threads for normal use), is highest output in general for me.  I also often use just one of the slots, depending on other workloads, so I'm glad it works that way for me.

[shaz2sxy]

I wonder how this will translate to folding? 

 

https://videocardz.com/newz/amd-radeon-rx-6800-allegedly-1-5x-faster-than-geforce-rtx-3090-in-cryptocurrency-mining

 

 

[MrFinn741]

16 hours ago, justpoet said:

I've found the opposite effect on my Xeons.  Running half and half in 2 slots (minus a couple threads for normal use), is highest output in general for me.  I also often use just one of the slots, depending on other workloads, so I'm glad it works that way for me.

Do you have single or multi CPU systems? When testing my dual Xeon server I only got 75-90% utilization when running all threads in one slot, probably because the CPU-CPU communication can't keep up. I did try more slots and fewer cores per slot but same low util, probably have had to force which cores a slot should use. Switched over to Boinc instead where the cores run individually and got 100% util no problem.

[Bitter]

Watching the UPS truck drive all over with my 1650 but it doesn't matter anyway because I won't be home to sign for it.

[Bitter]

Got the 1650! Popped it in, FAH picked it up but wouldn't fold on it, had to set openCL to 0 and Cuda to 1 manually. Took a few minuted but grabbed some work and hammered my CPU because i3, I need to swap in the i5 I guess. Overall PPD took a huge nose dive with both cards running, to about 500K PPD vs the lone P106-100 at 700K PPD. Currently trying to get JUST the 1650 to fold but fahcontrol is being wonky and doesn't seem to like to issue commands down to the fah process properly. It works fine from my laptop remotely, but at the actual PC it's being weird which is new in the past few days. Anyway, finally got the P106 paused and the 1650 folding by itself at 75W is doing a quickly estimated 600-700K PPD and barely making any noise to do so. I really think I'm going to grab a 2nd 1650 of the same cheap OEM style and just run one or two of those for the long term folding box. Now to procure an i5 S lower power CPU, I gave my 4590S away to my dad's PC build.