Folding Community Board

[mattheginger]

18 hours ago, GOTSpectrum said:

and yes, this is a two slot 4070ti

I'd be intrigued to know what temps you get. I've been eyeing up this card precisely because it's two slot  

[Gorgon]

On 5/15/2023 at 1:51 AM, GOTSpectrum said:

They say no crypto mining but say nothing about folding. Ryzen for scale, and yes, this is a two slot 4070ti

Yes, it's about the only one out there. It's not readily available in North America so the price is at a premium here.

18 hours ago, mattheginger said:

I'd be intrigued to know what temps you get. I've been eyeing up this card precisely because it's two slot  

Should be pretty decent as long if your power or clock-limiting it for better efficiency. My Zotac Trinity 4070ti sits around 51C Folding at 2205MHz but that's with a beefier cooler. Ada is, in my experience, twice as efficient as Ampere.

[GOTSpectrum]

@Gorgon am I right in the asumption that you can't power limit Nvidia GUPs with a GUI on ubuntu? 

[Gorgon]

1 hour ago, GOTSpectrum said:

@Gorgon am I right in the asumption that you can't power limit Nvidia GUPs with a GUI on ubuntu? 

There might be some tools to do that but

nvidia-smi -pm 1 (only needed once)
nvidia-smi -i <GPU  ID> -pl <Power>

or

nvidia-smi -i <GPIU ID> -lgc 0,<max Clock>

is trivial to type.

 

The find out the min and max limits either enter a nonsensical value

nvidia-smi -i <GPU ID> -pl 1000

or

nvidia-smi -i <GPU ID> -q | grep Power

[GOTSpectrum]

4 minutes ago, Gorgon said:

There might be some tools to do that but

nvidia-smi -pm 1 (only needed once)
nvidia-smi -i <GPU  ID> -pl <Power>

or

nvidia-smi -i <GPIU ID> -lgc 0,<max Clock>

is trivial to type.

 

The find out the min and max limits either enter a nonsensical value

nvidia-smi -i <GPU ID> -pl 1000

or

nvidia-smi -i <GPU ID> -q | grep Power

 

That's fair, I just like sliders and graphs lol 

[Gorgon]

15 minutes ago, GOTSpectrum said:

That's fair, I just like sliders and graphs lol 

I believe the psensors package has some pretty graphs for GPUs

 

I just use PuTTY to ssh into the systems and have a Session in PuTTY for each Slot.

 

Setup PKI to log in using PAgent

 

Window 29 Cols, 20 rows

Window Behaviour: Title (set to something useful like 4070)

SSH: Remote Command:

nvidia-smi -i <GPU ID> -l 1 --format=csv,noheader --query-gpu=temperature.gpu,power.draw,clocks.gr,fan.speed

The above works in a Windoze CMD Shell or PowerShell too

[Bitter]

https://gitlab.com/leinardi/gwe

 

Is that which you seek?

[BondiBlue]

I took the time this morning to replace the heatsink in my RTX 2070. I pulled an identical heatsink from a broken RTX 2080 (the 2080 works, but artifacts like crazy and won't run compute applications correctly), swapped it in and fired it up. It's back to working perfectly now, so I can start folding on it again. Right now I can consistently fold on the following cards:

• RTX 2070 8GB
• RTX 2070 Super 8GB
• GTX 1080 8GB
• Quadro P4000 8GB
• Quadro P4000 8GB
• GT 1030 2GB

There's one more card I'd like to add to the mix, and that's my trusty little GTX 1660 Ti 6GB. I've just got to find a system with room for it. 

[GOTSpectrum]

3 minutes ago, BondiBlue said:

I took the time this morning to replace the heatsink in my RTX 2070. I pulled an identical heatsink from a broken RTX 2080 (the 2080 works, but artifacts like crazy and won't run compute applications correctly), swapped it in and fired it up. It's back to working perfectly now, so I can start folding on it again. Right now I can consistently fold on the following cards:

• RTX 2070 8GB
• RTX 2070 Super 8GB
• GTX 1080 8GB
• Quadro P4000 8GB
• Quadro P4000 8GB
• GT 1030 2GB

There's one more card I'd like to add to the mix, and that's my trusty little GTX 1660 Ti 6GB. I've just got to find a system with room for it. 

I'm retiring my 10 series cards, they just aren't worth it anymore for me with the stupid high (0.47USD/KWh) power cost here

[BondiBlue]

21 minutes ago, GOTSpectrum said:

I'm retiring my 10 series cards, they just aren't worth it anymore for me with the stupid high (0.47USD/KWh) power cost here

Ouch! I'm in an area with nuclear power, and I only pay around $0.125/KWh right now. I feel for you and others in situations with high costs like that. 

[GOTSpectrum]

1 minute ago, BondiBlue said:

Ouch! I'm in an area with nuclear power, and I only pay around $0.125/KWh right now. I feel for you and others in situations with high costs like that. 

I remember only a couple years ago I was paying 0.14usd... oh how times have changed

[mattheginger]

1 hour ago, GOTSpectrum said:

...with the stupid high (0.47USD/KWh) power cost here

 

[justpoet]

Yeah, our costs went through the roof for power here too.  Joke's on them though, as now that it is nice and sunny, I produce more than I use with my solar...so they're giving me credit at these high rates for me to burn through come winter.  These stupid high rates are making my payback timeframe MUCH shorter for the solar.  It was originally going to be something like 10 years, since I way overbuilt the structure I put a bit more than half on, the other half are west facing with plenty of shade, and had to run underground conduit and all that jazz.  I'm guessing I'm close to 7 now.  Someday I'll have time to sit down and do the calculations right.

[Gorgon]

1 hour ago, GOTSpectrum said:

I'm retiring my 10 series cards, they just aren't worth it anymore for me with the stupid high (0.47USD/KWh) power cost here

I'm selling my 1000-series and 2000-series which will fund a couple more 4070(ti). Giving folks here a good deal on a used card and getting much more power efficient cards.

 

1070ti - 14kPPD/W

2070S - 22kPPD/W

3070ti - 29kPPD/W

4070ti - 60kPPD/W

 

I might even start selling the 3000-series as the 4000-series are so much more efficient. 60kPPD/W is likely low and will only improve as more complex WUs start appearing as it currently averages 6MPPD but can hit over 10MPPD even clock-limited down to 2250MHz and drawing 80-140W (103W average).

 

So yes, the 4000-series cards are waay over-priced for gaming but for compute they're worth it.

[Captainmarino]

1 hour ago, Gorgon said:

I'm selling my 1000-series and 2000-series which will fund a couple more 4070(ti). Giving folks here a good deal on a used card and getting much more power efficient cards.

 

1070ti - 14kPPD/W

2070S - 22kPPD/W

3070ti - 29kPPD/W

4070ti - 60kPPD/W

 

I might even start selling the 3000-series as the 4000-series are so much more efficient. 60kPPD/W is likely low and will only improve as more complex WUs start appearing as it currently averages 6MPPD but can hit over 10MPPD even clock-limited down to 2250MHz and drawing 80-140W (103W average).

 

So yes, the 4000-series cards are waay over-priced for gaming but for compute they're worth it.

Wish I could justify buying one to the wife. I think I'll have to continue folding on the ol' 1660S for a bit longer

[Gorgon]

19 hours ago, Captainmarino said:

Wish I could justify buying one to the wife. I think I'll have to continue folding on the ol' 1660S for a bit longer

Once the 4060(ti) comes out selling the 1660s for a couple of hundred bucks might help fund an upgrade, would be offset by energy savings and be a nice birthday present.

[Gorgon]

Just a heads-up ...

 

I upgraded my Windows NVidia driver to 531 and noticed that my GPU monitor

nvidia-smi -i 0 -l 1 --format=csv,noheader --query-gpu=temperature.gpu,power.draw,clocks.gr,fan.speed

that I run on my desktop was not reporting the actual graphics clock speed that I set using:

nvidia-smi -i 0 -lgc 0,2250

HWinfo64, however, does report the correct speed. I downgraded to Driver 528.49 and can confirm the the SMI utility is once again reporting the correct graphic clock frequency.

[mattheginger]

All these 18717's are killing me...   

[BondiBlue]

I finally hit 2 billion points! I'm so close to that last badge!

[GOTSpectrum]

So question time, x8 3.0, will it bottleneck a 4070ti? 

 

Seeing lower than expected perf on the new card and wondering if that will be the cause 

[mattheginger]

42 minutes ago, GOTSpectrum said:

So question time, x8 3.0, will it bottleneck a 4070ti? 

 

Seeing lower than expected perf on the new card and wondering if that will be the cause 

I'm not sure regarding the pcie. But as an anecdote, I installed my 4070 (non-Ti) the other night and proceeded to get mostly low-atom WU's for the evening, giving seriously lackluster PPD.

 

The difference in points between low and high atom WU's for these 40's series cards is wild. Sometimes a full 50% drop.   

[GOTSpectrum]

14 minutes ago, mattheginger said:

I'm not sure regarding the pcie. But as an anecdote, I installed my 4070 (non-Ti) the other night and proceeded to get mostly low-atom WU's for the evening, giving seriously lackluster PPD.

 

The difference in points between low and high atom WU's for these 40's series cards is wild. Sometimes a full 50% drop.   

I'm using the Extension from @LAR_Systems that tells me how my card compares against other 4070TIs 

[Gorgon]

2 hours ago, GOTSpectrum said:

So question time, x8 3.0, will it bottleneck a 4070ti? 

 

Seeing lower than expected perf on the new card and wondering if that will be the cause 

No - still more than enough bandwidth - run HfM for a week or so and take a look at it. I'm running mine at 2250MHz as that seems to be the sweet spot but a lot of the WUs out there really don't let this card stretch it's legs. That's why I suspect the 4060 and 4070ti will be much more consistent performers as with fewer CUDA cores they'll be able to work at they're full potential more often.

[GOTSpectrum]

5 minutes ago, Gorgon said:

No - still more than enough bandwidth - run HfM for a week or so and take a look at it. I'm running mine at 2250MHz as that seems to be the sweet spot but a lot of the WUs out there really don't let this card stretch it's legs. That's why I suspect the 4060 and 4070ti will be much more consistent performers as with fewer CUDA cores they'll be able to work at they're full potential more often.

I wonder why my card is underperforming when compared to other cards of it's class then... sitting at 2970 on the core and 10251 on the mem. Seeing like 25-30% below other 4070tis across multiple WUs of all sizes 

[Gorgon]

13 minutes ago, GOTSpectrum said:

I wonder why my card is underperforming when compared to other cards of it's class then... sitting at 2970 on the core and 10251 on the mem. Seeing like 25-30% below other 4070tis across multiple WUs of all sizes 

OS? Drivers? You could try x16 for a while but it really shouldnt hit you that hard even under windows. My 4070ti is under windows currently bu still manages to get above 10MPPD on some WUs even at 2250MHz with no mem overclock. Try removing the memory overclock.