Folding Community Board

[rkv_2401]

Finally hit 100 million points for the team.

[CT854]

Got a 3070 (@3.0x16) and a 1070 (@3.0x1) running along with my 3080 on my desktop. Just hit 10M PPD yesterday.

 

Once I get my x8 to x16 risers, I can add my 2080 Super. Playing with GPUs like this is literally so much fun.

[RollinLower]

3 hours ago, rkv_2401 said:

Finally hit 100 million points for the team.

awesome!

[Gorgon]

On 5/20/2022 at 12:43 PM, Gorgon said:

GPU	Yield		Eff.	Value	Gen.
1070ti	1,439,799	 9.6	 3.0	 0.0%		
2070s	2,667,237	17.8	 4.7	86.5%
3070ti	3,887,798	25.9	 6.1	45.7%

Efficiencies (kPPD/W); Value (kPPD/Inflation adjusted MSRP US$) and Generational Improvement (PPD %).

Above are the final values over a week. 14 different projects were sampled with at least one WU for the 1070ti and 3070ti and 2-8 samples for the 2070 Supers.

 

Of interest is the "back-of-napkin" initial estimate was fairly accurate. The 2070 super has a 87% improvement over the 1070Ti in terms of Yield (PPD) and the 3070Ti 46% again over the 2070 Super. There has been a $100US price bump for each successive generation so though the Value (kPPD/$) has improved just not as much as it could have.

[CT854]

I feel like this is running on borrowed time...

 

Edit: Looks to be on track for 13M-14M PPD!

[RollinLower]

10 hours ago, CT854 said:

I feel like this is running on borrowed time...

Edit: Looks to be on track for 13M-14M PPD!

looks fine to me! what could possibly go wrong here?

[CT854]

3 minutes ago, RollinLower said:

looks fine to me! what could possibly go wrong here?

I left it alone to go to work... please don't jinx it

[Shlouski]

10 hours ago, CT854 said:

I feel like this is running on borrowed time...

 

Edit: Looks to be on track for 13M-14M PPD!

 

This picture is giving me anxiety.

 

For those like me now suffering mental anguish, please enjoy this picture to regain some sense of normality, trust me it will get easier with time.

 

 

Much better.

[RollinLower]

15 hours ago, CT854 said:

I left it alone to go to work... please don't jinx it

Haven't heard back from you, did the rig catch on fire/did you die?

[CT854]

3 minutes ago, RollinLower said:

Haven't heard back from you, did the rig catch on fire/did you die?

Still going strong... I've just decided to perpetually tempt fate.

[rkv_2401]

21 minutes ago, CT854 said:

Still going strong... I've just decided to perpetually tempt fate.

I would risk it all for a rig that gave me 14 mill PPD, too

[CT854]

4 minutes ago, rkv_2401 said:

I would risk it all for a rig that gave me 14 mill PPD, too

That on its own is pushing around 8.2M. Add that with the ~6.2M from my desktop and it's gonna be around there if it stays consistent.

[RollinLower]

PSA to check your drivers when folding on linux. 

between the downtime in this graph where no hardware changes, just a driver update!

[Shlouski]

No pc fans, all the heat is pumped straight outside and the room actually stays cooler than without it.

 

 

 

 

@Gorgon

 

If you are not too busy can you tell me how to configure hfm to collect the data, and if possible average it out?

 

[RollinLower]

2 minutes ago, Shlouski said:

No pc fans, all the heat is pumped straight outside and the room actually stays cooler than without it.

 

mental setup, you're a madman!

[rkv_2401]

4 hours ago, Shlouski said:

No pc fans, all the heat is pumped straight outside and the room actually stays cooler than without it.

beautiful

[Gorgon]

9 hours ago, Shlouski said:

...

@Gorgon

 

If you are not too busy can you tell me how to configure hfm to collect the data, and if possible average it out?

...

Working on it ...

 

[Gorgon]

13 hours ago, RollinLower said:

PSA to check your drivers when folding on linux. 

between the downtime in this graph where no hardware changes, just a driver update!

... or disable automatic updates Ah - OK - Thanks @Kilrah - So @RollinLower which version are you talking about (Old & New?) I'm running 510.60.02 currently but I see 510.73.05 is the most current in the PPA for 22.04 and was released on the 17th.

 

But we've also seen a bunch of WUs that are really low in PPD these days.

[Kilrah]

15 minutes ago, Gorgon said:

... or disable automatic updates

I think he's pointing out how much higher his PPD is with the new driver, not the downtime

[CT854]

 

Spoiler

 

[RollinLower]

10 hours ago, Gorgon said:

... or disable automatic updates Ah - OK - Thanks @Kilrah - So @RollinLower which version are you talking about (Old & New?) I'm running 510.60.02 currently but I see 510.73.05 is the most current in the PPA for 22.04 and was released on the 17th.

 

But we've also seen a bunch of WUs that are really low in PPD these days.

jup, currently on 510.73.05. this change seems too drastic to be just a few bad WU's though. nearly 3M PPD up from before, and i've been stable on that low PPD number for a long time too. Seems like i would get a 'normal' WU atleast once in that time? 

 

Also, if anyone can point me in the right direction to control the fan on this quadro without a desktop environment that would be great. Nvidia settings on Ubuntu need atleast some form of desktop to function (Xorg?), and without direct fan control this quadro will just aim for 80C 24/7, even if the fan spins at like 50%...

 

tried multiple guides from google without success, maybe someone here has the magic touch?

[rkv_2401]

I read some posts about varying PPDs with different WUs and wanted to add in my two cents, I've seen anywhere between 1.05 million to 1.4+ million recently, but most of the time it's around 1.3 million on my 1660Ti mobile. I expect that ~40% PPD difference I've observed would be even larger on larger GPUs, as QRB is exponential. I use 'High Priority' as a research cause.

[Gorgon]

8 hours ago, RollinLower said:

...

Also, if anyone can point me in the right direction to control the fan on this quadro without a desktop environment that would be great. Nvidia settings on Ubuntu need atleast some form of desktop to function (Xorg?), and without direct fan control this quadro will just aim for 80C 24/7, even if the fan spins at like 50%...

 

tried multiple guides from google without success, maybe someone here has the magic touch?

Ahh - the headless conundrum. Short of installing HDMI Dummy Plugs to spoof a monitor you need to pass nvidia-setting the location of the Xauthority file (buried deep in nvidia documentation)

 

You must have a minimal X config running and be using Gnome (gdm3). I usually install the Midnight Commander which installs enough of X to get things going.

 

Disable Wayland:

sudo nano /etc/gdm3/custom.conf
uncomment: #WaylandEnable=false

Configure X-server for Coolbits, enable all GPUs and Allow X-server to start with no connected displays:

nvidia-xconfig -a --cool-bits=12 --allow-empty-initial-configuration

Disable nvidia-modeset:

sudo nano /usr/lib/modprobe.d/nvidia-kms.conf
change:
options nvidia-drm modeset=1 to options nvidia-drm modeset=0

Note the id used for Xauth:

root@dcn03:~# htop
root@dcn03:~# ps -ef | grep auth
root        1192    1190  0 May22 tty1     00:05:06 /usr/lib/xorg/Xorg vt1 -displayfd 3 -auth /run/user/127/gdm/Xauthority -nolisten tcp -background none -noreset -keeptty -novtswitch -verbose 3
root     3601489 2298207  0 11:45 pts/1    00:00:00 grep --color=auto auth

Then use that id for nvidia-settings:

DISPLAY=:0 XAUTHORITY=/run/user/127/gdm/Xauthority nvidia-settings \
 -a [gpu:1]/GPUGraphicsClockOffsetAllPerformanceLevels=50 \
 -a [gpu:1]/GPUFanControlState=1 \
 -a [fan:2]/GPUTargetFanSpeed=75

The fanID can be tricky, only 1 ID controllable for a Pascal GPU, 2 or more for each Turing & Ampere GPUs and they start at 0 and work upwards across all cards with nvidia-settings supporting a max ID of 7)

[Gorgon]

8 hours ago, RollinLower said:

jup, currently on 510.73.05. this change seems too drastic to be just a few bad WU's though. nearly 3M PPD up from before, and i've been stable on that low PPD number for a long time too. Seems like i would get a 'normal' WU atleast once in that time? ...

Thanks - I'll try updating to 510.73.05 tomorrow when the Time-of-Use electricity pricing gets expensive again during the day and my systems stop folding. Do you remember which version you were at previously?

[RollinLower]

15 hours ago, Gorgon said:

Thanks - I'll try updating to 510.73.05 tomorrow when the Time-of-Use electricity pricing gets expensive again during the day and my systems stop folding. Do you remember which version you were at previously?

don't remember sadly, though i believe it wasn't that old, maybe 2/3 versions. I did reinstall to 22.04 recently and that also meant reinstalling the drivers. Maybe i just installed the wrong ones? It's Linux so i ran into some issues doing the reinstall ofcourse, and i finished that at like 2AM anything can happen if you work on a pc untill 2 AM.