Folding Community Board

[Gorgon]

16 hours ago, Skiiwee29 said:

Folding makes my space heater go BRRRRRRRRRRRRRRR...

 

 

Green is GPU white CPU. I have my office AC running too

 

I feel your pain:

Unlike gaming folding is a slow and steady thing. Try sliding the power limit slider in Afterburner all the way to the left and you should see a massive decrease in Power from the GPU and only a slight decrease in Points per Day (PPD).

 

Likewise down clocking the CPU or adjusting PowerPlay Tables/BIOS limits can tame the heat from the CPU.

 

Both the CPU and GPU have exponential increases in Power required for small increases in performance.

 

[LazyDev]

Just an update to my system, overall.

 

RTX 2070

I've successfully replaced one of the memory modules, but was greeted with code 43 in Windows. Another mats test shows another failed module.

 

GTX 1080

I've recently bought (More like, on the 1st of April) a Gigabyte GTX 1080 G1 Gaming, that did not detect. It was narrowed down to a faulty fuse. 

 

This has been running 24/7 without issues. The only issue to note is that touching the memory clock, causes a BSOD (0x00000119). 

 

In the next folding event, The 1080 will be running, and hopefully, the 2070. I really do hope that I'm not chasing down a rabbit hole with it.

[marknd59]

Anyone else having this problem with F@H in the dark ?

 

[RollinLower]

51 minutes ago, marknd59 said:

Anyone else having this problem with F@H in the dark ?

 

yep, same here. it never seems to load total points.

[marknd59]

4 minutes ago, RollinLower said:

yep, same here. it never seems to load total points.

Well EOC stats are working so the problem must be somewhere else. The Donor Statistics are not working so I guess that is the problem.

[LAR_Systems]

Looks like F@H is in the middle of a server and API upgrade and may have broken a little bit of everything?

 

Is F@H working for anyone else or no?

[LAR_Systems]

6 hours ago, marknd59 said:

Well EOC stats are working so the problem must be somewhere else. The Donor Statistics are not working so I guess that is the problem.

It's not on the F@H In the Dark end... F@H appears to have broken a number of APIs and possibly the work servers.

[Starman57]

42 minutes ago, LAR_Systems said:

Looks like F@H is in the middle of a server and API upgrade and may have broken a little bit of everything?

 

Is F@H working for anyone else or no?

Nope.  EOC stats not working here either.

 

Are you getting the returned Folding in the Dark database updates from GPU and CPU?  I trust those are still working.

[piratemonkey]

somewhat related, I believe my client is stuck on downloading. unsure of how long

[marknd59]

16 hours ago, LAR_Systems said:

Looks like F@H is in the middle of a server and API upgrade and may have broken a little bit of everything?

 

Is F@H working for anyone else or no?

For me EOC stats are working. F@H Donor Statistics for my user name are not working. F@H stats for team 223518 are working. As you said it looks like F@H have update the site and have broken a few things.

[Captainmarino]

3 hours ago, marknd59 said:

For me EOC stats are working. F@H Donor Statistics for my user name are not working. F@H stats for team 223518 are working. As you said it looks like F@H have update the site and have broken a few things.

Same here on all fronts. Seems like may some folders' data may be on different servers that are getting updated or something.

[LAR_Systems]

21 minutes ago, Captainmarino said:

Same here on all fronts. Seems like may some folders' data may be on different servers that are getting updated or something.

@marknd59 @RollinLower I have pushed a work around for F@H in the dark to point to legacy APIs at F@H to restore the functionality of points, project etc. in my client.

They have not yet fixed redirects or replacement of these APIs with their new page launches so the default web client they ship is broken.   I'm also seeing the issue with little to no work on the WS or long waits for it.   Hopefully they are working through the issues today.

Anyways... if you force update of F@H INTD to v 1.1.4 you get most of the functionality back.    Wait and see until / how they are going to fix their own clients.

[Starman57]

5 hours ago, LAR_Systems said:

Wait and see until / how they are going to fix their own clients.

Would this fix to the clients also fix the under reporting and under allocation of points for the AMD 6900XT (quite possibly the entire 6xxx series)?

[LAR_Systems]

2 hours ago, Starman57 said:

Would this fix to the clients also fix the under reporting and under allocation of points for the AMD 6900XT (quite possibly the entire 6xxx series)?

Don't know what you are referring to regarding points.

But I'm gonna say no, what I'm talking about is F@H orphaned some APIs their web clients use to show your total points and project descriptions.  Has nothing to do with the folding engine, card performance or folding cores.   

[Starman57]

3 hours ago, LAR_Systems said:

Don't know what you are referring to regarding points.

Here's an example.  WU 13446.  From your GPU database.

Nvidia 3080: Average WU time (h:mm) 1:23, average WU PPD 4,670,868, points/WU average 268,642

Nvidia 2080ti (TU102): WU time (h:mm) 2:36), average WU PPD 3,821,898, points/WU average 255,328

 

My 6900XT: WU time (h:mm) 1:45, WU PPD 3,492,525, points 244,743 (shouldn't this be around 260,000+???)

 

The 6900XT computes this WU in 105 minutes compared to the 2080ti at 156 minutes, but credit PPD number is ~400,000 lower and the WU points is ~11,000 lower.  Under reporting/under allocated???

 

Note that my 6900xt doesn't compute this WU in your database average of 2:55.  Are all the rest of the 6xxx series that much slower?  I think not.

[Captainmarino]

So THIS is what F@H has been working on... 

 

[RollinLower]

5 hours ago, Captainmarino said:

So THIS is what F@H has been working on... 

well dang, fancy!

[Captainmarino]

4 hours ago, RollinLower said:

well dang, fancy!

Well, it seems like that was short-lived. Everything seems like its back to boring again. Boo!

[LAR_Systems]

11 hours ago, Starman57 said:

Here's an example.  WU 13446.  From your GPU database.

Nvidia 3080: Average WU time (h:mm) 1:23, average WU PPD 4,670,868, points/WU average 268,642

Nvidia 2080ti (TU102): WU time (h:mm) 2:36), average WU PPD 3,821,898, points/WU average 255,328

 

My 6900XT: WU time (h:mm) 1:45, WU PPD 3,492,525, points 244,743 (shouldn't this be around 260,000+???)

 

The 6900XT computes this WU in 105 minutes compared to the 2080ti at 156 minutes, but credit PPD number is ~400,000 lower and the WU points is ~11,000 lower.  Under reporting/under allocated???

 

Note that my 6900xt doesn't compute this WU in your database average of 2:55.  Are all the rest of the 6xxx series that much slower?  I think not.

Points scores etc. all come out of the F@H core client I have no say in their calculation, reward etc.   

 

I capture samples of the data reported by F@H during folds to be averaged and they fluctuate up and down while folding, they are never 100% consistent, you can see that in the charting view in the client. 

 

This means you can not really look at how your card is folding right now on a specific Work Unit PRCG and have it be 100% match of the average for the project/gpu as each WU PRCG will fold differently day to day or within the folds inside the work unit, which is why averages are used and update over time as F@H change the work units.    

That said.. adding to the issue I assume your AMD 6900XT gets identified as the following https://folding.lar.systems/gpu_ppd/brands/amd/folding_profile/navi_21_radeon_rx_68006800_xt_6900_xt

NAVI 21 [RADEON RX 6800/6800 XT / 6900 XT]

 

That means your card, as far as F@H core is concerned is the same as the RX 6000/6800 XT and all results captured and averaged in the database will be for all 3 models of cards.

 

I don't know if this an AMD bios, driver or F@H thing but F@H does not identify AMD cards as distinctly as they do Nvidia cards (may not optimize for them as well I'm guessing).

But it means your averages might be a bit off because the DBs number for your card will be a blend of all 3 AMD models due to the way F@H does not split them.   

This also can be compounded when comparing AMD to Nvidia work unit results as it's possible F@H rewards PPD differently between Nvidia and AMD due to hardware, driver or folding architecture supported by the different vendors.
 

Sorry can't be of more help here, but there are a lot of unknowns as to the how/why they do things in the F@H engine, the folding DB just reflects the output of what they are providing.

[LAR_Systems]

16 minutes ago, Captainmarino said:

Well, it seems like that was short-lived. Everything seems like its back to boring again. Boo!

They botched the rollout... it broke the APIs of the web clinets and other integrations... they need to account for the legacy systems then try again would be my guess.

[RollinLower]

14 minutes ago, LAR_Systems said:

They botched the rollout... it broke the APIs of the web clinets and other integrations... they need to account for the legacy systems then try again would be my guess.

knowing the F@H dev team (or lack there of) this might be done in the year 2256!

[Starman57]

59 minutes ago, LAR_Systems said:

Points scores etc. all come out of the F@H core client I have no say in their calculation, reward etc.   

 

I capture samples of the data reported by F@H during folds to be averaged and they fluctuate up and down while folding, they are never 100% consistent, you can see that in the charting view in the client. 

 

This means you can not really look at how your card is folding right now on a specific Work Unit PRCG and have it be 100% match of the average for the project/gpu as each WU PRCG will fold differently day to day or within the folds inside the work unit, which is why averages are used and update over time as F@H change the work units.    

That said.. adding to the issue I assume your AMD 6900XT gets identified as the following https://folding.lar.systems/gpu_ppd/brands/amd/folding_profile/navi_21_radeon_rx_68006800_xt_6900_xt

NAVI 21 [RADEON RX 6800/6800 XT / 6900 XT]

 

That means your card, as far as F@H core is concerned is the same as the RX 6000/6800 XT and all results captured and averaged in the database will be for all 3 models of cards.

 

I don't know if this an AMD bios, driver or F@H thing but F@H does not identify AMD cards as distinctly as they do Nvidia cards (may not optimize for them as well I'm guessing).

But it means your averages might be a bit off because the DBs number for your card will be a blend of all 3 AMD models due to the way F@H does not split them.   

This also can be compounded when comparing AMD to Nvidia work unit results as it's possible F@H rewards PPD differently between Nvidia and AMD due to hardware, driver or folding architecture supported by the different vendors.
 

Sorry can't be of more help here, but there are a lot of unknowns as to the how/why they do things in the F@H engine, the folding DB just reflects the output of what they are providing.

Yes, I am aware that you just record what the FAH client reports, and have no say in the calculation.  Thanks for that.  Yes, WU calculations are not 100% consistent, but on each individual WU, they are close to being consistent assuming that the underlying computing loads are the same.  That's where a problem crops up, because if you are also using the computer at the same time, this will extend FAH compute times as it's fighting for computing resources.

 

Yes, my 6900XT falls into the "bucket" you specified.  I found out from FAH is that they "categorize" cards based loosely on the PCI ID, however they don't fully parse it hence the "bucket" approach.  As for the blend, yes of course, but I wonder what it would take in terms of performance from the "slower" members of the group to account for an "average" that rises from 1 hour 45 minutes to 2 hours 55 minutes?  It must mean that the slower 6xxx cards must be incredibly slow at FAH loads to account for an almost doubling of the time!  I looked at the FP32 performance of them, and they aren't that much slower, as the 6800 (the slowest) is only 30% slower than the 6900XT (fastest) in the group.

 

Quoted from the FAH forum where I raised this issue "FAH will continue to award the same baseline points for both AMD and nVidia."  As for folding architecture, FAH defines that.  The Nvidia and AMD GPUs aren't specifically designed folding cards.  They aren't ASICs.  As such, it should be a pretty easy correlation between cards based on the predominant compute instructions used, namely FP32 (as reported by FAH - "In fact, FP64 performance is quite different, too, favoring nV by a large factor but I don't use that number since it makes a much smaller contribution to overall performance than FP32 does.")

 

Looking at my FAH numbers as reported by Folding in the Dark, I'm surmising that a LOT of people are somehow running the FAH client (both GPU and CPU) really badly.  Correction, inefficiently.  Could it be they just install the program and that's it?  Windows does a lousy job of scheduling a specific heavy load, or at least the FAH load, on modern multicore CPU's which also impacts GPU's as well.  Witness the differences in performance with Linux which either has a smarter scheduler or less bloat to deal with.  I can only speak to my personal setup, and my contribution of Linux WUs on 5700XT GPUs for the last couple of weeks has changed the trend line quite noticeably.  As the days progress, I'm guessing the Linux line will deflect even more based on the PPD numbers my 2 machines are locally reporting.  Take a look.

 

https://folding.lar.systems/gpu_ppd/brands/amd/folding_profile/navi_10_radeon_rx_5600_oem5600_xt57005700_xt

 

As for help, you are already doing monumental work by providing Folding in the Dark and it's website database.  Kudos.   I just asked the basic question of client "rebalancing" just in case you were somehow plugged into the goings on behind the scenes (pending change lists) as to client changes and API calls.  Since you were blind-sided by the latest API difficulties, I'm guessing not.  Never hurts to ask.

 

Anyway, I'll still plod along with the 6900XT at 15-25% higher than the average.  Now that I've finally got my 5950x, I'm getting 75-100+% higher PPD:PLP number.  Example?  WU 16959.  5950x CPU average: 23,442.  My current: 49,124.  I've had the same effect on my 3950x's which possibly contributed to the skewed results comparing the 3950x to the 5950x.  Ever notice the 3950x is quite a bit higher than the 5950x?  That should change over time as I know based on local results that the 5950x is quite a bit faster than the 3950x (my guess based on FAH is about 20%).  Whether it will ever show up in your database after being watered down by inefficient results and averaged remains to be seen.

 

Again, thanks for the work you do.

[Captainmarino]

1 hour ago, LAR_Systems said:

They botched the rollout... it broke the APIs of the web clinets and other integrations... they need to account for the legacy systems then try again would be my guess.

That had been my thoughts exactly. Too bad. I was enjoying my smiley face.

[RollinLower]

12 minutes ago, Captainmarino said:

That had been my thoughts exactly. Too bad. I was enjoying my smiley face.

don't forget they handed you an award as well!

[WhitetailAni]

F@H in the dark is reporting points for me now, though it wasn't before.

The EOC stats were working while F@H in the dark wasn't, but both work now.

Side note, 22 million!

Spoiler