LTT Official Folding Month VI

[Alex Atkin UK]

9 minutes ago, Gorgon said:

The 4080 should be a better value than the 4070ti but it's price point is (still) too high so the 4070ti tends to be at the price and efficiency "sweet-spots" but the 4070 might be better for dual rigs as this is the first model with 2 slot cards available (excluding that Inno3D 4070ti model that is dear in North America and someone (cough @GOTSpectrum @Alex Atkin UK @RollinLower) seem to have bought all the European stock of.)

Hey, I only bought one - my other 4070 Ti is a commonly available Zotac.   Though I did tell @GOTSpectrum that the card existed, so I guess I'm still to blame.

 

Thinking back, I suspect its a bit of flip/flop between 2x4070 Ti or one 4090 being more efficient for F@H.  As there are definitely some WUs where your PPD would be higher with two 4070 Ti, but it kinda evens itself out over time.

 

But ultimately its easier to fit 1x4090 in a case than 2x4070 Ti, especially as a single card is useful for gaming and other software that doesn't scale across multiple GPUs.  Plus its likely to have a better resale value.

[soccermomisqueen]

@GOTSpectrum I was looking at your blog posts and the spreadsheets, I was wondering if there was a way to add who the highest climber is on the side of the sheet (posted here since I cannot comment on blog posts )?

[Gorgon]

1 hour ago, Schnoz said:

So many people here have bought RTX 40-series GPUs for folding. I wonder how much higher the proportion is compared to, well, non-folders. Cuz outside of compute performance, RTX 40-series GPUs don't really offer too much exciting gaming-related stuff over previous-gen GPUs.

I bough a boatload of 3000-series B-stock EVGA Cards when they were still on Fire Sales at their web site and though I love EVGA Build Quality I quickly regretted this as the 4000-series, if run at the sweet spot (2200-2500MHz) are almost twice as efficient as 3000-series (1440MHz). For gaming they don't make fiscal sense but for folding, if you care about your power bill they are the best.

56 minutes ago, Alex Atkin UK said:

I don't think that is true either, as the INNO3D two slot is only available at scalper prices now in the UK and largely unavailable at all outside Europe.  Plus plenty of cases lack venting in the PSU shroud which prevents fitting one in the second slot.

17 minutes ago, Alex Atkin UK said:

Hey, I only bought one - my other 4070 Ti is a commonly available Zotac.

I have 4 Zotac Trinity OC 4070tis in 2 rigs so 1/2 slot of space between the lower and upper cards. IF you run them clock or power limited they're fine with that little space (48-53C on lower and 66-70C on the upper) but I am considering removing the backplate on the lower cards to get more room to breathe. And, combined, you can feed these two cards from 3 dual 8-ping PCIe PEG Power Cables if your running them with a clock or power-limit (likely a bad idea if you want to run them full-bore).

The easiest way I've found on Linux to automatically apply the limits is via a shell script run at reboot from the root crontab:

cat /root/setpwr.sh

#!/bin/bash
nvidia-smi -pm 1
nvidia-smi -i 0 -pl 175
nvidia-smi -i 0 -lgc 0,2205
nvidia-smi -i 1 -pl 175
nvidia-smi -i 1 -lgc 0,2205

crontab -l

# use /bin/bash to run commands, instead of the default /bin/sh
SHELL=/bin/bash
@reboot /root/setpwr.sh > /dev/null 2>&1

[Gorgon]

Here's the efficiencies I'm seeing: Top 4: 4070ti; Orange: 4070, Yellow: 3080; Purple: 4060ti

It appears the chips used on 4070tis are binned better as my 4070 and 4060ti get much lower efficiency in practice then they should just based on CUDA Cores and FP32 Performance but even those are almost twice as efficient as the 3080.

[SansVarnic]

5 hours ago, GOTSpectrum said:

Reminder that you can get free google cloud credit 
 

 

Wait this doesn't work on F@H though does it?

[cbigfoot]

49 minutes ago, soccermomisqueen said:

@GOTSpectrum I was looking at your blog posts and the spreadsheets, I was wondering if there was a way to add who the highest climber is on the side of the sheet (posted here since I cannot comment on blog posts )?

that would be valid for some but the problem is if people didn't start on day 1 then they will be super inflated, and the sheet would get super complicated quickly.  

[Baha]

4 hours ago, Favebook said:

Okay guys, I pulled the trigger and bought 4090 ASUS TUF. I gotta admit, I am ashamed and happy at the same time.

I have to say that I am not really up to date with 4090 power connector/adapter shenanigans, can anyone tell me if there were different versions of 12VHPWR cable or are they all the same coming from the manufacturer?

To my knowledge, the main thing to look out for is that there are two versions of the socket.

On the original version of the socket, the four sense pins that sit above the primary 12 power pins are the same length. In the revised version of the socket, the sense pins are shortened so that they are recessed into the socket. If the plug is not securely inserted with the newer socket type, the card is less likely to receive power. This should help prevent some of the burning problems. For a comparison, here's the original:



And here's the revised, shorter length:



These started shipping back in July, so as long as you bought new, you're likely to receive the newer version. There has been some revision on the cable end as well, but my understanding is that it's just requirements on the width and tension of the cable clip -- nothing warranting an actual revision for existing high-quality cables. It would be better to avoid introducing more points of failure with adapters, as long as you're using a good PSU vendor with a good cable (either included in box, first party from your PSU manufacturer, or known good third party) you should be fine on that front.

[GOTSpectrum]

1 hour ago, soccermomisqueen said:

@GOTSpectrum I was looking at your blog posts and the spreadsheets, I was wondering if there was a way to add who the highest climber is on the side of the sheet (posted here since I cannot comment on blog posts )?

Hey sorry for not getting back to you before now, I'm actually out with people. I'm unlikely to do this as it really isn't that important of a metric 

[GOTSpectrum]

45 minutes ago, SansVarnic said:

Wait this doesn't work on F@H though does it?

Yes it does

[SansVarnic]

1 minute ago, GOTSpectrum said:

Yes it does

I have never used the service before ... hmm might be worth looking into on the occasion. Skimming through that setup instruction for boinc... I may need help with setting up for f@h as linux is not a strong point for me.

Maybe is a good time for me to finally learn me some linux. <- inner laugh at self.

[GOTSpectrum]

13 minutes ago, SansVarnic said:

I have never used the service before ... hmm might be worth looking into on the occasion. Skimming through that setup instruction for boinc... I may need help with setting up for f@h as linux is not a strong point for me.

Maybe is a good time for me to finally learn me some linux. <- inner laugh at self.

I believe there is a folding Google cloud how to somewhere on the forum

[RollinLower]

 

42 minutes ago, SansVarnic said:

I have never used the service before ... hmm might be worth looking into on the occasion. Skimming through that setup instruction for boinc... I may need help with setting up for f@h as linux is not a strong point for me.

Maybe is a good time for me to finally learn me some linux. <- inner laugh at self.

 

28 minutes ago, GOTSpectrum said:

I believe there is a folding Google cloud how to somewhere on the forum

 

[Alex Atkin UK]

2 hours ago, Gorgon said:

I bough a boatload of 3000-series B-stock EVGA Cards when they were still on Fire Sales at their web site and though I love EVGA Build Quality I quickly regretted this as the 4000-series, if run at the sweet spot (2200-2500MHz) are almost twice as efficient as 3000-series (1440MHz). For gaming they don't make fiscal sense but for folding, if you care about your power bill they are the best.

I have 4 Zotac Trinity OC 4070tis in 2 rigs so 1/2 slot of space between the lower and upper cards. IF you run them clock or power limited they're fine with that little space (48-53C on lower and 66-70C on the upper) but I am considering removing the backplate on the lower cards to get more room to breathe. And, combined, you can feed these two cards from 3 dual 8-ping PCIe PEG Power Cables if your running them with a clock or power-limit (likely a bad idea if you want to run them full-bore).

The easiest way I've found on Linux to automatically apply the limits is via a shell script run at reboot from the root crontab:

cat /root/setpwr.sh

#!/bin/bash
nvidia-smi -pm 1
nvidia-smi -i 0 -pl 175
nvidia-smi -i 0 -lgc 0,2205
nvidia-smi -i 1 -pl 175
nvidia-smi -i 1 -lgc 0,2205

crontab -l

# use /bin/bash to run commands, instead of the default /bin/sh
SHELL=/bin/bash
@reboot /root/setpwr.sh > /dev/null 2>&1

I have it set in /etc/sddm/Xsetup mainly because I couldn't figure out how else to run nvidia-settings with the right display access to set the fan controls.  Drawback there is it requires you to use Xorg not Wayland, well its a drawback if Fedora kill off Xorg.

 

nvidia-settings -a "[gpu:0]/GPUFanControlState=1" -a "[fan:0]/GPUTargetFanSpeed=65"
nvidia-smi -pm 1
nvidia-smi -gtt 70 # Aim for this target operating temperature
nvidia-smi -pl 200 # Power Limit

 

Target temperature doesn't seem to really work though which is why I have the power limit there too, as this is the 2080 config where the thermal pads are not in great condition so hot spot runs about 20-25C higher than core temp.  As you can't monitor hot spot in Linux, I just have to try to keep the core temp down and hope hot spot tracks linearly.

 

Both 4090s are regrettably on Windows as I'm really curious if they score higher on Linux, but I'm not completely insane, both those boxes have actual uses that require Windows.  I'm not about to spend 4090 money and ONLY use it for Folding.

 

The 3080, INNO3D 4070 Ti and 2080 are dedicated Foldboxes, the rest are in PCs where I actually need them for stuff.

[Lightwreather]

38 minutes ago, GOTSpectrum said:

I believe there is a folding Google cloud how to somewhere on the forum

 

11 minutes ago, RollinLower said:

 

 

Just an fyi, you might be better off selecting a regular ubuntu image and setting up the drivers on your own instead of using the pre-installed driver image. idk, It's never really worked for me

[Favebook]

2 hours ago, Baha said:

-snip-

This is exactly the information I was looking for. Thank you very much kind man!

[Justaphf]

1 hour ago, Alex Atkin UK said:

Both 4090s are regrettably on Windows as I'm really curious if they score higher on Linux, but I'm not completely insane, both those boxes have actual uses that require Windows.  I'm not about to spend 4090 money and ONLY use it for Folding.

I broke down and finally got myself a 4090 Friday evening, hence my recent PPD bump.  Will be my daily driver in Windows in Dec, but for now doing an "extended burn in test" under Linux.  Using stock settings, according to my HFM.NET history, the 40 WUs it's done so far are showing an average PPD of 26.7 M.  Not sure how accurate the PPD estimates per WU in HFM.NET history viewer are though compared to actual awarded values.  Power seems to be staying between 240 to 315 W, depending on WU.

[Justaphf]

On 11/1/2023 at 10:26 PM, Schnoz said:

Please consider filling out this form to see how your hardware stacks up against the competition! Of those who've filled out the form, there's a pretty decent variety of hardware this year!

 

https://forms.gle/z44vthgbc9CYAvReA

 

Link to view the responses

@Schnoz Is the survey/sheet smart enough that if I enter a second time to edit my original list (under the same google account), will it replace the original entry, or will it cause a double entry in the sheet?

[GOTSpectrum]

5 minutes ago, Justaphf said:

@Schnoz Is the survey/sheet smart enough that if I enter a second time to edit my original list (under the same google account), will it replace the original entry, or will it cause a double entry in the sheet?

It will cause a double entry 

[Shlouski]

13 hours ago, GOTSpectrum said:

We need more power! 

 

Looks to be leveling out , still really impressive stuff guys

 

I was thinking about all that stock Linus bought from that computer shop, then hardware prizes crossed my mind, anyone need random hardware from the early 2000's ?

[Shlouski]

23 minutes ago, Justaphf said:

broke down and finally got myself a 4090 Friday evening, hence my recent PPD bump

I'm watching you

[Alex Atkin UK]

10 minutes ago, Shlouski said:

Looks to be leveling out , still really impressive stuff guys

 

I was thinking about all that stock Linus bought from that computer shop, then hardware prizes crossed my mind, anyone need random hardware from the early 2000's ?

Oh please no, anything but that......

[HoldSquat]

38 minutes ago, Alex Atkin UK said:

Oh please no, anything but that......

I suggested the Christmas album before, you sure you wouldn't rather have early 2000s stuff?

[Sequence211]

7 hours ago, LAR_Systems said:

4080 is currently best for power usage per PPD.

4070ti is the sweet spot for cost of card & power usage per PPD as @GOTSpectrum mentioned.

 

4090 is a beast for gaming / machine learning etc. but for folding you can buy two 4070tis for the same price as a 4090, get more PPD overall while completing 2x the work units.

 

https://folding.lar.systems/gpu_ppd/overall_ranks_power_to_ppd

Thank you. I did not realise you had a table for that. But the thing is that the table only considers TDP and not card draw. My 4090  has a tdp of 450w, but on,y draws 300w while folding, with stock settings. 
It would be nice to have a column for typical folding power draw for each model.  But i know that is not easy to collect data on.

[Justaphf]

52 minutes ago, Shlouski said:

I'm watching you

I'm only trying to out pace Jake, honest 

[GOTSpectrum]

Not folding related, but I sent off a form to adopt a 17-week-old kitten that was abandoned at a cattery today!