LTT Official Folding Month VI

[BiotechBen]

2 minutes ago, Alex Atkin UK said:

Now you've got me wanting to re-enable iGPU on my desktop to see if not using the dGPU for display is a factor.  I probably wont though, too much hassle.

If it's an Nvidia card, it may not make a difference, FAH was hitting exclusively CUDA on my card.

[dizmo]

Maybe I'll finally get to do this one

[podkall]

7 hours ago, BiotechBen said:

I've got a p600 (effectively a 1050 afaik) and at full 40w it's sitting at 72C on its single slot cooler with 150k PPD (before bonus), and the 1080's pull 1.2MPPD at 185W and 84C (blower cooler)

I had a dual fan cooler on 1050, and my 1080ti already close to 80-90°C on hostpot when average is barely 70°C so, I could probably cook inside the case if I let it full speed

[BiotechBen]

21 minutes ago, podkall said:

I had a dual fan cooler on 1050, and my 1080ti already close to 80-90°C on hostpot when average is barely 70°C so, I could probably cook inside the case if I let it full speed

I used my dual 1080's exhaust to thaw some frozen sushi one time, worked surprisingly well.

Could also try making a greenhouse out of it as well with some cheap ducting.

[Alex Atkin UK]

3 hours ago, BiotechBen said:

I used my dual 1080's exhaust to thaw some frozen sushi one time, worked surprisingly well.

Could also try making a greenhouse out of it as well with some cheap ducting.

Sounds like a recipe for food poisoning to me.

[Alex Atkin UK]

1 minute ago, Schnoz said:

I'm literally using my dual 2080 Tis to heat my entire apartment in place of this place's built in heaters. I need to keep the glass door open all the time (there's a wire mesh on it, fortunately) just to keep my place from overheating lol.

I've done that the last two winters, though unfortunately even with all the GPU power I have it doesn't really cope once it gets close to freezing.

 

I had even considered writing a script to check room temperature and turn F@H on/off accordingly, but it doesn't really work when you want to complete a WU quickly for the points, so I do it manually.

[GOTSpectrum]

On 10/2/2023 at 11:18 PM, GOTSpectrum said:

if anyone is wanting to help with the running of the event drop me a PM asap

Thanks to everyone who reached out, I plan to get back to you in the next day or two. 

 

Next up prizes! who's donating? 

[flibberdipper]

I was looking forward to having my M1 Mac Mini grind away on this but it looks like it gets its ass absolutely blown out, it's barely making 2500PPD with the current WU. 

 

EDIT: Ahhhh technology, just had to complain. Within less than a minute of posting it climbed to 60K and seems to be stabilizing closer to 65.

[podkall]

31 minutes ago, flibberdipper said:

EDIT: Ahhhh technology, just had to complain. Within less than a minute of posting it climbed to 60K and seems to be stabilizing closer to 65.

f@h ETA is very innacurate very early and gets more accurate later

 

for example "your machine has 2.73 days to complete this task"

 

ETA for the WU completion... was 3.0 days

 

now it's roughly 8 hours

[TheLANguy]

6 minutes ago, podkall said:

f@h ETA is very innacurate very early and gets more accurate later

 

for example "your machine has 2.73 days to complete this task"

Folding @ Home is always correct!

[flibberdipper]

1 hour ago, podkall said:

f@h ETA is very innacurate very early and gets more accurate later

 

for example "your machine has 2.73 days to complete this task"

 

ETA for the WU completion... was 3.0 days

 

now it's roughly 8 hours

Yeah I'm not new to it, I just figured it got an absolutely brutal WU to do since it was stuck at such a low number for a good 10 minutes. Wouldn't be the first time I've gotten an insane WU to complete lol

[Mxyzptlk]

23 hours ago, Gorgon said:

I _MIGHT_ have just added a couple more 4070tis to the Fold. Had the boot NVMe drive fail on one of the rigs with 2 x 4070tis and BiFrost is having trouble recognizing more than 2 GPUs but I _THINK_ that's just due to using a Ryzen 3000-series CPU and it not liking the PCIe Gen 3 m.2 to PCIe3 x4 risers. Have to try a 2600x today to confirm that. Otherwise I'll have to run 3 additional Rigs rather than one.

 

@Mxyzptlk are you still running your Mining Frame and is it behaving with 3000 and 4000 NVidia GPUs?

Yes, it is running off and on with a 4070 and a 4070ti w/ no PCI riser issues that I have noticed. But I am running at full PCIe3.0x8 to each of the cards from the X99 / 5960X CPU that has 40 PCIe lanes available. 

 

I have run plenty of 2 GPU (20XX series & 30XX series GPU's) setups on a 3000 series CPU's without an issue. But I was not running any adaptor cables, they were mounted straight to the boards.  

[leadeater]

On 10/3/2023 at 11:18 AM, GOTSpectrum said:

if anyone is wanting to help with the running of the event drop me a PM asap

 

[GOTSpectrum]

6 minutes ago, leadeater said:

 

Your role is none-voluntary 

[LuxorAB]

If I remember correctly there were some causes better suited for low-end PCs. Smaller WUs and better suited for CPU folding. Can someone tell what it was?

 

I think it was Alzheimer's, but can't remember, lol

[TylerD321]

13 minutes ago, Schnoz said:

Anyone got recommendations for used PPD-for-the-buck GPUs that are good options right now? I'm currently looking at the RTX 2080 Ti and RTX 30 series models, but are there any outliers in that pile?

I asked this the other day. I’ll tag you.

[gravitywave]

My T600 was pulling 40W + whatever power the rest of the machine it was in required, and that was for just 300k ppd. So on researched impulse, I grabbed a 4060 at Microcenter for $300. It's been doing ~4mil ppd on most projects, and that's with a 110W TDP. I wasn't set on buying another card, but it was the power efficiency that sold me. Looking at 3070's, 2080Ti's on eBay, they seem to be going for similar money, but pull a lot more power, and generally those will be with unknown history and no warranty. My 3070 does ~5mil ppd, but that's nearly twice the power draw of the 4060. Assuming you care about power efficiency, I don't think you can beat a brand new shiny 4060, unless of course you want to go higher up on the 40series line.

[GOTSpectrum]

So plan for the day: 

 

1) Make the chat thread and invite people to the admin team

 

2) Get the spreadsheet made and shared

 

3) Reach out to @LMGcommunity for shout outs and prizes

[RollinLower]

2nd 4070Ti is crunching!

I'll be waiting on the waterblok for a little while, but that doesn't mean she can't get some PPD in the meantime!

[Gorgon]

2 hours ago, gravitywave said:

My T600 was pulling 40W + whatever power the rest of the machine it was in required, and that was for just 300k ppd. So on researched impulse, I grabbed a 4060 at Microcenter for $300. It's been doing ~4mil ppd on most projects, and that's with a 110W TDP. I wasn't set on buying another card, but it was the power efficiency that sold me. Looking at 3070's, 2080Ti's on eBay, they seem to be going for similar money, but pull a lot more power, and generally those will be with unknown history and no warranty. My 3070 does ~5mil ppd, but that's nearly twice the power draw of the 4060. Assuming you care about power efficiency, I don't think you can beat a brand new shiny 4060, unless of course you want to go higher up on the 40series line.

@Schnoz This is about the best bang for the buck nowadays in terms of PPD/$. If you must go used avoid the 2000-series, they're just not worth it. The 3000-series are about 50% more efficient that the 2000-series but don't hold a candle to the 4000-series even with their extortionate pricing which are 80-100% more efficient than the 3000-series. The 4070ti has much better efficiency (PPD/W) than the 4060(ti)/4070 but at a price premium (better Silicon likely).

 

The 4060 will be even more efficient if you cap it's Graphics Clock to 2205MHz or thereabouts:

nvidia-smi -lgc 0,2205

The Power should drop down to ~70W with only a slight decrease in PPD.

[Heats with Nvidia]

2 hours ago, Gorgon said:

@Schnoz This is about the best bang for the buck nowadays in terms of PPD/$. If you must go used avoid the 2000-series, they're just not worth it. The 3000-series are about 50% more efficient that the 2000-series but don't hold a candle to the 4000-series even with their extortionate pricing which are 80-100% more efficient than the 3000-series. The 4070ti has much better efficiency (PPD/W) than the 4060(ti)/4070 but at a price premium (better Silicon likely).

That means my 2080 is more efficient as a space heater and since thats its current main job, it would be a bad idea to replace it with a 40 series.

[podkall]

I've got a question and good news.

 

First the question, why is my F@H lagging even when not working or when CPU is limited to less than all cores.

 

(2nd smaller question is that when selecting a thread count is it actually selecting the physical core count? because the F@H uses 7,5 theads when I select 4 "threads" in the configure)

 

And good news, seems like the thermal issues on my GPU are partially caused by the fan curve (partially by huge gap between average and hotspot temps),

88% fan speed lets me use 75% power limit, instead of previous 55% and the hotspot temp stays under 80°C.

 

3 hours ago, Gorgon said:

4000-series even with their extortionate pricing which are 80-100% more efficient than the 3000-series.

100% more efficient than 30 series? so 4070 pulls double 3070 at same wattage?

[Imakuni]

18 minutes ago, Schnoz said:

I see! Though, the contract of my apartment makes it so for me, power is flat-fee. With power efficiency aside, what would be the best PPD/$? Maybe something AD102-based?

Edit: Nevermind! Just saw this.

If efficiency is not an issue, it's all a matter of PPD / Cost. If I had to bet, it would be a used 30 series; 40 is blazing fast, but it's got the "it's new!" cost associated.

[justpoet]

On 10/2/2023 at 1:00 PM, Dutch_Master said:

I'm considering partaking, but not owning (due to not needing) a d/recent GPU, never mind more then one, how much CPU-bound tasks will find their way to @home-folders? If CPU power can be utilised, I may throw in some EPYC cores (I have 68 at my disposal, if all boards run)

 

Until last year's event, most of my participation was 10+ year old Xeon's.  CPUs work and are helpful, they just don't turn out current GPU numbers, even if the CPU is current.

 

On 10/4/2023 at 5:42 AM, Alex Atkin UK said:

I've done that the last two winters, though unfortunately even with all the GPU power I have it doesn't really cope once it gets close to freezing.

 

I had even considered writing a script to check room temperature and turn F@H on/off accordingly, but it doesn't really work when you want to complete a WU quickly for the points, so I do it manually.

You could do it like most heating thermostats do, which is to call off the heat (set WUs to Finishing) a degree early, but keep circulating so the heat continues to work out of the boiler/furnace until the correct temp (the Finishing time of working, and the cooler ramping down), or a degree above.  Then when it drops to a degree bellow, start WUs.

[BondiBlue]

So far all is going well regarding getting cards up and running. Right now I'm doing a quick test run with dual Quadro P4000s, a GTX 1080, an RTX 2070 and an RTX 2070 Super at full power, mainly to see what temps are going to be like. I didn't have some of these cards last year, so I'm not super familiar with their thermal performance and thermal output. So far I'm pretty impressed with most of the cards, but the regular RTX 2070 sounds like a jet engine in order to avoid thermal throttling. It's got a very, very cheap blower cooler on it, so I'm not surprised at all. 

 

My next task is going to be getting a few more cards up and running. Right now I'm planning to use one of my old Dell Precision T3500s (Xeon X5675) to run two Quadro M4000s along with two Radeon RX 580s. The 580s will be running externally through PCIe risers and powered from an external PSU. It's a bit of a janky setup, but it does work, just as long as the motherboard doesn't die like last year. 

 

Things are getting exciting again!