Folding Community Board

[Gorgon]

6 hours ago, GOTSpectrum said:

@Gorgon these are running at stock settings

Yeah there might be an issue - just restarted my 4070ti at Stock (285W PL) and am seeing 80% in on p18448: TPF 00:37, PPD 7.7-7.9MPPD, Power ~195W, GPU Temp 65-67C, GClock 2820MHZ (pretty much locked there), Fans 39-40%

 

But you were observing a wide range on this WU:

MPPD TPF

2.234 01:24

5.287 00:47

5.778 00:44

6.707 00:40

6.738 00:40

7.027 00:39

7.040 00:39

 

Even throwing out the 2.2MPPD sample as an outlier your seeing 5.3-7.0MPPD

 

Where I was seeing at a locked 2250:

MPPD TPF    StdDev Samples

5,965 00:44 91,812   32

 

or 5.8-6.1MPPD with TPF 00:43-00:44

 

so a much tighter variance in my case so it doesn't appear to be excessive variability in the WU.

 

Perhaps that cooler is struggling under a 200W load and the GPU is dropping "bins" to meet the temperature target. It looks "beefy" enough for 200W though.

 

Is it Dual BIOS (Doesn't appear to be) and, if so, is it on "quiet" rather than "oc"/"performance" setting?

 

Is Prefer Max Performance State enabled for the GPU ("Boost Lock")

 

Is it choked for air in the chassis or getting baked by a lower card? Still wouldn't explain the wide variance as you'd expect "Spec" to running both GPUs

 

Tried reverting to 5.28 Nvidia Driver? That's what I'm running as 5.31 had issues reporting the correct Graphics Clock in the SMI tool (but not in HWinfo and presumably not in AfterBurner)

 

But it seems more like the system feeding it or the card is glitching out or if the card's PCIe slot is fed from the chipset could something else (m.2, SATA?) be saturating the DMI link to the CPU.

 

If this is your main system could it be another process interfering or is it the Display Out?

 

In my system the 4070Ti is in the upper slot of a Aorus x570 Master (R9 5900x) running at x8 with a Radeon Pro W4100 in the lower slot as Display out in a Lian-Li O11 Mini Air with 2 x 120mm Bottom Intake; 2 x 140mm front intake; 1 x 120mm Rear Exhaust and a EVGA 280mm CLC Top Exhaust

[GOTSpectrum]

19 minutes ago, Gorgon said:

Yeah there might be an issue - just restarted my 4070ti at Stock (285W PL) and am seeing 80% in on p18448: TPF 00:37, PPD 7.7-7.9MPPD, Power ~195W, GPU Temp 65-67C, GClock 2820MHZ (pretty much locked there), Fans 39-40%

 

But you were observing a wide range on this WU:

MPPD TPF

2.234 01:24

5.287 00:47

5.778 00:44

6.707 00:40

6.738 00:40

7.027 00:39

7.040 00:39

 

Even throwing out the 2.2MPPD sample as an outlier your seeing 5.3-7.0MPPD

 

Where I was seeing at a locked 2250:

MPPD TPF    StdDev Samples

5,965 00:44 91,812   32

 

or 5.8-6.1MPPD with TPF 00:43-00:44

 

so a much tighter variance in my case so it doesn't appear to be excessive variability in the WU.

 

Perhaps that cooler is struggling under a 200W load and the GPU is dropping "bins" to meet the temperature target. It looks "beefy" enough for 200W though.

 

Is it Dual BIOS (Doesn't appear to be) and, if so, is it on "quiet" rather than "oc"/"performance" setting?

 

Is Prefer Max Performance State enabled for the GPU ("Boost Lock")

 

Is it choked for air in the chassis or getting baked by a lower card? Still wouldn't explain the wide variance as you'd expect "Spec" to running both GPUs

 

Tried reverting to 5.28 Nvidia Driver? That's what I'm running as 5.31 had issues reporting the correct Graphics Clock in the SMI tool (but not in HWinfo and presumably not in AfterBurner)

 

But it seems more like the system feeding it or the card is glitching out or if the card's PCIe slot is fed from the chipset could something else (m.2, SATA?) be saturating the DMI link to the CPU.

 

If this is your main system could it be another process interfering or is it the Display Out?

 

In my system the 4070Ti is in the upper slot of a Aorus x570 Master (R9 5900x) running at x8 with a Radeon Pro W4100 in the lower slot as Display out in a Lian-Li O11 Mini Air with 2 x 120mm Bottom Intake; 2 x 140mm front intake; 1 x 120mm Rear Exhaust and a EVGA 280mm CLC Top Exhaust

The card is in the top slot of an X370 board, with a quadro in the lowest slot, which is why it is running at x8 and afaik it is running direct from the CPU. 

 

It is my main system, temps are perfectly fine on it currently seeing 61C on the card while folding, it was 5C hotter in here mid day than it is not but Can't see that causing a massive issue. 

 

The quadro is coming out of this system anyway to be able to use it on the jellyfin server so I will soon get to that and see if things improve or if I need to look into something else being the problem. 

 

I do pause the GPU while gaming so that shouldn't affect the results we see here. 

 

It holds the clocks great so I'm thinking it is either:

 

PCIe limitations(doesn't seem likely though)

 

Drivers(Will DDU and downgrade to test that)

 

Something in the background(not really easy to test for this as I use this machine for 12 hours a day)

 

Something I'm not thinking off(thinking to try the above and see if I see any differences, if not, when I install my new SSD will do a clean install of windows and see if that corrects the issue, but that would be a massive PITA)

[GOTSpectrum]

This is the card running rn

[GOTSpectrum]

Right, so one card now, running at x16, 528.24 driver version, all else remains the same, will leave 'er running overnight and see what we get 

[Gorgon]

1 hour ago, GOTSpectrum said:

The card is in the top slot of an X370 board, with a quadro in the lowest slot, which is why it is running at x8 and afaik it is running direct from the CPU.

Yup, that would be the case.

 

I do pause the GPU while gaming so that shouldn't affect the results we see here.

Yes, pausing a WU in mid-stride will affect the points. The longer it's paused the lower the Quick Return Bonus.

 

57 minutes ago, GOTSpectrum said:

This is the card running rn

Whoo hoo! I'm 1 bin faster than you. NM - Current WU (p16569) is at 2805 also

31 minutes ago, GOTSpectrum said:

Right, so one card now, running at x16, 528.24 driver version, all else remains the same, will leave 'er running overnight and see what we get

That would be the same BW as my card in a PCIe4 x8 then

[GOTSpectrum]

1 hour ago, Gorgon said:

Yup, that would be the case.

 

 

 

Yes, pausing a WU in mid-stride will affect the points. The longer it's paused the lower the Quick Return Bonus.

 

Whoo hoo! I'm 1 bin faster than you. NM - Current WU (p16569) is at 2805 also

That would be the same BW as my card in a PCIe4 x8 then

It's crazy to me that the fans are just at the lowest for this fan and everything is happy

[RollinLower]

11 hours ago, GOTSpectrum said:

It's crazy to me that the fans are just at the lowest for this fan and everything is happy

that's pretty sick indeed!

[Gorgon]

15 hours ago, GOTSpectrum said:

It's crazy to me that the fans are just at the lowest for this fan and everything is happy

I really want to get hold of a 4060 or 4060ti as I suspect that at 2205-2250MHz that these GPUs could likely run with absolutely no problem with the Fans and Shroud removed in a chassis with adequate airflow.

 

Given their number of CUDA cores in relation to a 4070ti:

    GPU Yield Pwr Eff.
4070 ti 7,170 108 66.5
   4070 5,497  83 66.5
4060 ti 4,063  61 66.5
   4060 3,585  54 66.5

So it would appear that though these models are definite "Do NOT Buys" for gaming they are the best values for Folding so far in the Ada line-up and are more than twice as efficient at their Base Clocks than Ampere.

[Gorgon]

On 5/25/2023 at 10:08 AM, GOTSpectrum said:

@Gorgon these are running at stock settings

 

Here is my data at stock:

Spoiler

DatabaseID,ProjectID,ProjectRun,ProjectClone,ProjectGen,Name,Path,Username,Team,CoreVersion,FramesCompleted,FrameTime,Result,Assigned,Finished,WorkUnitName,KFactor,Core,Frames,Atoms,PreferredDays,MaximumDays,SlotType,PPD,Credit,BaseCredit
113492,18448,7,56,911,bash Slot 00,localhost:36330,gordonbb,223518,0.2,100,00:00:36,FINISHED_UNIT,05/25/2023 20:59:30,05/25/2023 22:01:29,p18448,0.75,0x22,100,126828,1.5,2,Unknown,7913964.58968,329748.52457,55859
113493,17636,2,12,101,bash Slot 00,localhost:36330,gordonbb,223518,0.2,100,00:00:28,FINISHED_UNIT,05/25/2023 22:00:51,05/25/2023 22:49:32,p17636,0.75,0x22,100,94172,2,5,Unknown,8778477.06926,284487.6828,27012
113494,16569,2,98,60,bash Slot 00,localhost:36330,gordonbb,223518,0.2,100,00:00:34,FINISHED_UNIT,05/25/2023 22:49:01,05/25/2023 23:47:50,p16569,0.75,0x22,100,225658,2,4,Unknown,10889208,428510.5,50000
113498,16577,1,159,26,bash Slot 00,localhost:36330,gordonbb,223518,0.2,100,00:00:43,FINISHED_UNIT,05/25/2023 23:47:08,05/26/2023 01:00:47,p16577,0.75,0x22,100,231032,2,4,Unknown,10156518.11721,505474.86,66000
113504,17636,4,67,54,bash Slot 00,localhost:36330,gordonbb,223518,0.2,100,00:00:28,FINISHED_UNIT,05/26/2023 00:59:56,05/26/2023 01:48:51,p17636,0.75,0x22,100,94172,2,5,Unknown,8757514.21371,283808.331,27012
113507,16569,3,130,101,bash Slot 00,localhost:36330,gordonbb,223518,0.2,100,00:00:34,FINISHED_UNIT,05/26/2023 01:48:21,05/26/2023 02:47:32,p16569,0.75,0x22,100,225658,2,4,Unknown,10855423.05882,427181,50000
113516,16570,1,88,106,bash Slot 00,localhost:36330,gordonbb,223518,0.2,100,00:01:10,FINISHED_UNIT,05/26/2023 02:46:52,05/26/2023 04:47:34,p16570,0.75,0x22,100,469796,3,7,Unknown,13187318.4,1068417,135000
113520,16569,0,154,65,bash Slot 00,localhost:36330,gordonbb,223518,0.2,100,00:00:34,FINISHED_UNIT,05/26/2023 04:46:08,05/26/2023 05:45:52,p16569,0.75,0x22,100,225658,2,4,Unknown,10805336.47059,425210,50000
113525,16568,4,52,107,bash Slot 00,localhost:36330,gordonbb,223518,0.2,100,00:00:34,FINISHED_UNIT,05/26/2023 05:45:12,05/26/2023 06:44:09,p16568,0.75,0x22,100,225706,2,4,Unknown,10876896,428026,50000
113530,18449,3,0,1101,bash Slot 00,localhost:36330,gordonbb,223518,0.2,100,00:00:39,FINISHED_UNIT,05/26/2023 06:43:28,05/26/2023 07:50:54,p18449,0.75,0x22,100,137998,1.5,2,Unknown,7865401.93809,355035.50415,62731
113537,18717,45,4,367,bash Slot 00,localhost:36330,gordonbb,223518,0.2,100,00:01:12,FINISHED_UNIT,05/26/2023 07:50:13,05/26/2023 09:51:53,p18717,0.75,0x22,100,77809,1.5,2,Unknown,6067411.2,505617.6,120000
113540,17638,0,85,32,bash Slot 00,localhost:36330,gordonbb,223518,0.2,100,00:00:36,FINISHED_UNIT,05/26/2023 09:50:38,05/26/2023 10:52:21,p17638,0.75,0x22,100,114671,2,5,Unknown,8409584.1984,350399.3416,37460
113544,18448,10,9,1223,bash Slot 00,localhost:36330,gordonbb,223518,0.2,100,00:00:36,FINISHED_UNIT,05/26/2023 10:51:44,05/26/2023 11:53:51,p18448,0.75,0x22,100,126828,1.5,2,Unknown,7905465.08424,329394.37851,55859
113549,16566,0,60,105,bash Slot 00,localhost:36330,gordonbb,223518,0.2,100,00:01:23,FINISHED_UNIT,05/26/2023 11:53:14,05/26/2023 14:14:27,p16566,0.75,0x22,100,441800,3,7,Unknown,12186355.66265,1170680,160000

 

I'm moving it back to Graphics Clock locked to the Base Clock Frequency (2250MHz)

[RollinLower]

Upgrades done! I rebuilt the cooling system in my F@H box just in time for the summer sprint. Now with double the pumps and higher flow QDCs.

Lets fold some proteins!

[justpoet]

1 hour ago, RollinLower said:

Upgrades done! I rebuilt the cooling system in my F@H box just in time for the summer sprint. Now with double the pumps and higher flow QDCs.

Lets fold some proteins!

Your thick rad....excites me.

[GOTSpectrum]

Need more WUs like this one 

 

[GOTSpectrum]

@Shlouski what do you think? 

 

[Gorgon]

10 hours ago, GOTSpectrum said:

Need more WUs like this one 

 

I get 10,115,160, TPF 01:24 for only 100W power.

 

Wait until you get a 16566 or a 16570

[dogwitch]

a updated.

1 card clean.

fixing a ssd atm.

 

a stupid question. with got or other might know.

how many llt groups are their on boinc projects?

am trying to clean up abandon leadership of teams.

[RollinLower]

On 5/28/2023 at 5:43 AM, Gorgon said:

I get 10,115,160, TPF 01:24 for only 100W power.

 

Wait until you get a 16566 or a 16570

i didn't read your whole convo with GOT, but that's on linux with a 4080, right?
There's no Linux numbers in the LAR PPD database and i'm seeing how the top 3 40-series cards stack up to see which one has the best value as a potential upgrade

 

i need to watercool them, so that also plays a part. For reference:
4090 would be about 1900 euro's

4080 is about 1300 euro's

4070Ti is about 1000 euro's

Those prices all include waterblocks.

[Gorgon]

38 minutes ago, RollinLower said:

i didn't read your whole convo with GOT, but that's on linux with a 4080, right?
There's no Linux numbers in the LAR PPD database and i'm seeing how the top 3 40-series cards stack up to see which one has the best value as a potential upgrade

 

i need to watercool them, so that also plays a part. For reference:
4090 would be about 1900 euro's

4080 is about 1300 euro's

4070Ti is about 1000 euro's

Those prices all include waterblocks.

No, we are discussing a 4070 ti.

 

So far, in terms of value, the lineup goes like this:

	Price	Shaders	Clock	TDP	Value	Value
Model	($US)	(#)	(MHz)	(W)	(shClk/$)	(sh/$)
RTX 4090	$1,599	16,384	2,235	450	22,901	10.2
RTX 4080	$1,199	9,728	2,205	320	17,890	8.1
RTX 4070ti	$799	7,680	2,310	285	22,204	9.6
RTX 4070	$599	5,888	1,920	200	18,873	9.8
RTX 4060ti	$399	4,352	2,310	160	25,196	10.9
RTX 4060	$299	3,840	1,830	115	23,502	12.8

So purely in terms of Shaders/$ the order is:

RTX4060, 4060ti. 4090, 4070, 4070ti then 4080.

 

In terms of (BaseClock*Shaders)/$ the order is:

4060, 4060ti. 4090, 4070, 4070ti, 4080.

 

So, if you were to buy just one card and want to put a waterblock on it I'd suggest a 4090.

 

If money was not a consideration though I'd likely suggest 2 x 4070ti. The reason being is there are still a lot of WUs that can't seem to take full advantage of all the Shaders in a 4090 and for around the same price (excluding waterblocks) 2 x 4070ti may very well out-perform a 4090 in the long term.

 

At the very least, in terms of the high-end cards, unless NVidia "blinks" and adjusts prices to more closely reflect reality the 4080 is the absolute worst value and should be avoided at all costs.

 

Where Ada really shines, however, is efficiency. Hands down Ada beats Turing across the lineup by nearly double. Meaning for the same amount of power Ada will double the PPD compared to Turing if one ignores cost.

[RollinLower]

1 hour ago, Gorgon said:

Where Ada really shines, however, is efficiency. Hands down Ada beats Turing across the lineup by nearly double. Meaning for the same amount of power Ada will double the PPD compared to Turing if one ignores cost

That's a big selling point for me, right now I'm doing 350W on my 3090 and 250W on my 2080 24/7. Going to ADA might be a big cost up front, but in the long term the savings add up!

 

Thanks for the explanation, seeing 2x 4070Tis would be only marginally more expensive than a single 4090 it seems like the way to go.

[GOTSpectrum]

highest I've seen so far 

[Gorgon]

2 hours ago, RollinLower said:

That's a big selling point for me, right now I'm doing 350W on my 3090 and 250W on my 2080 24/7. Going to ADA might be a big cost up front, but in the long term the savings add up!

 

Thanks for the explanation, seeing 2x 4070Tis would be only marginally more expensive than a single 4090 it seems like the way to go.

The difficulty might be in getting blocks for 4070ti as they're not very popular - go figure. But I have mine capped to 2250MHz and it draws only an average of 107W.

 

LARs database shows an interesting trend between the the 4070ti and 4090:

1	GeForce RTX 4090	14,606,150	20,352,008	13,763,766
3	GeForce RTX 4070 Ti	10,198,178	10,557,612	9,538,461

[Gorgon]

On 5/29/2023 at 10:38 AM, RollinLower said:

That's a big selling point for me, right now I'm doing 350W on my 3090 and 250W on my 2080 24/7. Going to ADA might be a big cost up front, but in the long term the savings add up!

 

Thanks for the explanation, seeing 2x 4070Tis would be only marginally more expensive than a single 4090 it seems like the way to go.

At the very least I'd cap the GPU Clock on both the 3090 and 2080 to a maximum of 1440-1500MHz. You should see a 40-45% reduction in power for a 12-17% reduction in PPD and the cards will run even cooler and the fans on that beast of a system will be even quieter.

[Shlouski]

On 5/25/2023 at 8:38 PM, Gorgon said:

Sorry - I don't have long term data at Stock settings and as HfM does not display the clock frequency It's a little hard to compare apples to oranges.

 

I'm also not running stock settings as I'm limited to one pcie 8pin connector per gpu.

 

On 5/23/2023 at 11:55 PM, Gorgon said:

See the Std. Deviation in this post to give an idea of how much variance I'm seeing on my 4070ti:

This is interesting, thanks for sharing it

 

What sort of sample size are we talking here?

 

Have you seen any data to suggest projects with larger atom counts may have larger deviations?

 

A standard deviation of 777,503 in project 16569 is high IMO, are there outliers with millions of points between them?

[Gorgon]

7 hours ago, Shlouski said:

What sort of sample size are we talking here?

186 WUs

7 hours ago, Shlouski said:

Have you seen any data to suggest projects with larger atom counts may have larger deviations?

 

Project	Atoms	#	PPD	StdDev	TPF
14941	6,882,735	2	6,821,344	61,391	00:18
14942	6,759,952	1	7,020,028	0	00:18
14943	7,020,028	1	6,914,918	0	00:18
14946	6,914,918	3	6,547,256	239,254	00:18
16566	6,709,364	1	9,096,274	0	01:41
16567	6,208,996	1	8,159,711	0	00:41
16568	6,723,408	6	8,102,304	247,925	00:42
16569	9,096,274	8	8,040,503	102,963	00:41
16570	8,159,711	3	10,033,744	60,781	01:27
16572	8,602,902	3	10,018,967	9,305	01:25
16573	7,971,891	6	10,491,434	315,154	01:23
16574	7,922,479	4	8,880,609	159,179	01:40
16575	8,233,383	4	8,695,038	89,681	01:39
16576	7,938,843	8	8,612,129	53,983	01:39
16577	7,944,326	5	7,665,770	81,395	00:52
16706	7,975,594	4	7,685,652	132,361	01:13
17636	7,933,382	4	6,445,595	219,397	00:35
17637	7,996,022	2	6,066,748	97,481	00:44
17638	7,907,112	3	6,458,192	8,314	00:43
18448	10,495,144	32	5,964,680	91,812	00:44
18449	8,176,759	31	5,933,731	331,915	00:47
18717	8,156,887	42	4,742,522	79,433	01:25
18909	8,168,224	4	7,489,059	142,362	00:29
19001	8,010,041	5	6,685,900	17,491	00:21

Not really - ^here's the data set summary with counts - all the WUs had fairly large atom size and if you sort by Atom Size there doesn't seem to be any correlation but the project sample sizes were likely too low in most cases to see any wide variances.

 

Project 18448 & 9 and 18717 had large sample sizes but variances were fairly tight.

7 hours ago, Shlouski said:

A standard deviation of 777,503 in project 16569 is high IMO, are there outliers with millions of points between them?

Yes, one that I should have discarded - I may have reverted to stock accidentally - same with 16570 - Corrected above.

 

I _MIGHT_ have just added another 4070ti today so in a few weeks may have a larger dataset to look at.

[Mxyzptlk]

Anyone seen a 3 slot 2060 Super!  Yikes 

[Gorgon]

7 hours ago, Mxyzptlk said:

Anyone seen a 3 slot 2060 Super!  Yikes 

For a 175W TDP GPU a 3-slot cooler is way overkill. I’m just finishing getting rid of all my 1000 and 2000-series GPUs so I’m seeing much better efficiencies. Dropped from 1500 to 1000W of GPUs but keeping around the same yield so easier on the electricity bill.