New TC stats system & progress

[zodac]

Hey all, just an update on some of the LARS automation work.

 

I've got most of the bits running locally now, and I think it should be ready to go live by the end of the week. The plan is to run it once a month, so after the stats are reset, and in the two days downtime before the stats start up again.

 

However, it does seem that there is some fluctuation with the PPD values which can look suspicious/out of sync. @BWG has already manually updated the multipliers for October, and me checking 8 hours later has shown some small differences.

 

To prevent any concerns, I'm going to update the list of hardware (available here) to include the average PPD and multiplier. The plan is to update all hardware values at once, so going forward you can consider the values here to be a snapshot of the PPDs for the upcoming month. It would be far too much effort to update more frequently than monthly, so I hope this is a fair compromise.

 

There are some PPD values up on that page now, but since we don't have the #1 hardware, there is nothing to compare it to. It will be there next month and you'll be able to compare yourselves if necessary.

 

I'll update here again once I have something running/tested locally, but you won't see any updated multipliers/PPD on the live stats site until November.

Edited October 5, 2021 by zodac

[damric]

It's too easy to poison the database, whether it's someone running a card for 2 minutes at stock or someone running something tweaked to weaken someone else's multiplier.

 

The painful solution is to freeze the competition for a month. Require eligible cards and backup cards to be fed to the LARS database at stock configuration for entirety of previous month (maybe 2 weeks would be long enough). Do away with the ALL OS since it's too easy to poison using linux due to far less samples.

 

You can compare FLOPS to have an idea how cards should perform. If it seems way off, then it probably is.

[BWG]

9 hours ago, damric said:

It's too easy to poison the database, whether it's someone running a card for 2 minutes at stock or someone running something tweaked to weaken someone else's multiplier.

 

The painful solution is to freeze the competition for a month. Require eligible cards and backup cards to be fed to the LARS database at stock configuration for entirety of previous month (maybe 2 weeks would be long enough). Do away with the ALL OS since it's too easy to poison using linux due to far less samples.

 

You can compare FLOPS to have an idea how cards should perform. If it seems way off, then it probably is.

 

I don't think that's going to be effective for cards that have ample submissions to LARS.

 

The cards that fit into that high submission count mold just happen to all be within close proximity to each other in ETF. The real solution would be restricting the submission count even further by increasing the submission count required, and then removing the ability for a user to feed LARS independently in order to enter into the event. That effectively removes a lot of the competitive advantages we've uncovered with LARS within the first few months.

 

Highly folded cards would saturate any attempt at manipulation. Less folded cards would then become disallowed as they probably should be. It's just not that cool to fold 30k ppd at 200 watts anyway, hypothetically.

 

Obviously some tweaks here and there, but I think that is the direction we should head into if we want to have tighter competition. There's all sorts of things to consider though, so keep the ideas flowing.

[damric]

23 minutes ago, BWG said:

 

I don't think that's going to be effective for cards that have ample submissions to LARS.

 

The cards that fit into that high submission count mold just happen to all be within close proximity to each other in ETF. The real solution would be restricting the submission count even further by increasing the submission count required, and then removing the ability for a user to feed LARS independently in order to enter into the event. That effectively removes a lot of the competitive advantages we've uncovered with LARS within the first few months.

 

Highly folded cards would saturate any attempt at manipulation. Less folded cards would then become disallowed as they probably should be. It's just not that cool to fold 30k ppd at 200 watts anyway, hypothetically.

 

Obviously some tweaks here and there, but I think that is the direction we should head into if we want to have tighter competition. There's all sorts of things to consider though, so keep the ideas flowing.

Well personally, I'm going to finally learn Linux just so my oem prebuilt can feed this 3200G at stock to LARS, which indeed is @35W configuration with a single stick of DDR4 @1866 so I can counter-espionage what the other team did to my multiplier. Yeah that whole ALL OS crap is so easily abused. In the meantime I might switch to something else for ETF. I'll let you know.

[BWG]

Yeah, K5000 is a bit out of scope too, so I'll probably switch it. The 960 wasn't bad except during moonshot. But, I started having issues with it again, and I thought it would be fun to directly compete with Avacado. I still crack up over that video. There's more tweaking still to be had with the event, but I think it's been fun. Each month has gradual changes on popular GPUs.

[tictoc]

1 hour ago, damric said:

Well personally, I'm going to finally learn Linux just so my oem prebuilt can feed this 3200G at stock to LARS, which indeed is @35W configuration with a single stick of DDR4 @1866 so I can counter-espionage what the other team did to my multiplier. Yeah that whole ALL OS crap is so easily abused. In the meantime I might switch to something else for ETF. I'll let you know.

 

The only reason why I ran my 3200g was because the average on LARS looked quite a bit lower than what a stock 3200g would be.  Linux really has nothing to do with it.  Just going off of the ppd for the first unit I ran this is what it looks like:

 

You're 3200g

Core 1725MHz - 38% OC

p18201 - tpf 16:53

 

My 3200g

Core 1240MHz - 0% OC

p18201 - tpf 23:34

 

The tpf difference is pretty much an exact match to the clock speed difference.  I just noticed that the LARS number was way lower than it should have been (23k ppd at that time), so I thought I'd run it for a bit to get an idea of what the stock ppd numbers should be.  My guess before running it was that 50k ppd was probably closer to what the real stock number was.  I definitely wasn't trying to game the handicap or sabotage your numbers.  It just looked like a funky outlier compared to what the chip should have been, and I had the hardware on hand to do a little comparison.

Edited October 6, 2021 by tictoc

[damric]

5 hours ago, BWG said:

Yeah, K5000 is a bit out of scope too, so I'll probably switch it. The 960 wasn't bad except during moonshot. But, I started having issues with it again, and I thought it would be fun to directly compete with Avacado. I still crack up over that video. There's more tweaking still to be had with the event, but I think it's been fun. Each month has gradual changes on popular GPUs.

K5000 is interesting to me. It's similar to what a GTX 660 or something? It's really close spec-wise to my HD7850 as far as FLOPS but it has more compute cores so I'm guessing the stock clock speed is low which makes it a really good overclocker? I just want to see a better baseline that, well...looks like a line and not a dot. Then I'd like to do battle with that with my HD7850. At one point I was considering getting the MXRT version of Pitcairn for ETF since they are pretty cheap on ebay, similar gimped clocks for reliability in medical work.

[damric]

4 hours ago, tictoc said:

 

The only reason why I ran my 3200g was because the average on LARS looked quite a bit lower than what a stock 3200g would be.  Linux really has nothing to do with it.  Just going off of the ppd for the first unit I ran this is what it looks like:

 

You're 3200g

Core 1725MHz - 38% OC

p18201 - tpf 16:53

 

My 3200g

Core 1240MHz - 0% OC

p18201 - tpf 23:34

 

The tpf difference is pretty much an exact match to the clock speed difference.  I just noticed that the LARS number was way lower than it should have been (23k ppd at that time), so I thought I'd run it for a bit to get an idea of what the stock ppd numbers should be.  My guess before running it was that 50k ppd was probably closer to what the real stock number was.  I definitely wasn't trying to game the handicap or sabotage your numbers.  It just looked like a funky outlier compared to what the chip should have been, and I had the hardware on hand to do a little comparison.

The part Linux had to do with it is the fact that there were no Linux submissions in months, thus the weight of your single submission pulled the performance line way up. You got to remember that 99% of those chips were in some junk HP Pavilions with single channel RAM and the 35w config for 1000MHz GPU clock. It's why the performance line was pretty consistently low for Windows. When properly configured those 500 vega cores do pretty well as long as it's dual channel RAM pushed as far as the IMC will let it so that the bandwidth and infinity fabric isn't choking it. Of course they don't sell pre-built machines with 3200G and Ubuntu or whatever so yeah that Linux submission really screwed me since the Windows PPD is only like 20K on LARS. It's fair game what you did unless the rules are changed. It's still not as bad as what I accidentally did to myself back in July when I accidentally turned on the dark mode for almost a week at full overclock on my HD 7850.

[tictoc]

4 hours ago, damric said:

The part Linux had to do with it is the fact that there were no Linux submissions in months, thus the weight of your single submission pulled the performance line way up. You got to remember that 99% of those chips were in some junk HP Pavilions with single channel RAM and the 35w config for 1000MHz GPU clock. It's why the performance line was pretty consistently low for Windows. When properly configured those 500 vega cores do pretty well as long as it's dual channel RAM pushed as far as the IMC will let it so that the bandwidth and infinity fabric isn't choking it. Of course they don't sell pre-built machines with 3200G and Ubuntu or whatever so yeah that Linux submission really screwed me since the Windows PPD is only like 20K on LARS. It's fair game what you did unless the rules are changed. It's still not as bad as what I accidentally did to myself back in July when I accidentally turned on the dark mode for almost a week at full overclock on my HD 7850.

 

I see what you mean.  In hindsight I probably should not have even enabled the LARS extension, since I really was just trying to see if the listed average was anything close to what a stock chip could do when just running F@H.

 

This really is an outlier of a chip. It's almost impossible to get a true average ppd number, since the use case of this chip is not really conducive to getting a realistic average ppd.  The other wrench in the system is that I'm sure LARS is also counting the 2200g, which has a base clock of 1000MHz compared to the 3200g, and it's 1250MHz base clock.  I'm not sure what the best solution is to deal with hardware that has artificially low avg ppd numbers.  Most likely we'll just need to look a bit closer at uncommon hardware, before giving it the go ahead for ETF.          

 

For what it's worth the recent Windows numbers (53k this week) are also bringing the the average up.

 

Back on topic for this thread. @BWG The stats have been down since sometime yesterday afternoon.

[tictoc]

5 hours ago, damric said:

K5000 is interesting to me. It's similar to what a GTX 660 or something? It's really close spec-wise to my HD7850 as far as FLOPS but it has more compute cores so I'm guessing the stock clock speed is low which makes it a really good overclocker? I just want to see a better baseline that, well...looks like a line and not a dot. Then I'd like to do battle with that with my HD7850. At one point I was considering getting the MXRT version of Pitcairn for ETF since they are pretty cheap on ebay, similar gimped clocks for reliability in medical work.

 

The K5000 is the same GPU as the GTX 680, but it runs a much lower stock clock to fit in a smaller TDP budget (120W).  Upping the power limit and giving it proper cooling allows for pretty nuts OCs that are in the range of 40% over stock.

Edited October 6, 2021 by tictoc

[zodac]

9 hours ago, tictoc said:

Back on topic for this thread. @BWG The stats have been down since sometime yesterday afternoon.

Yeah, axihub was down for a few hours and inaccessible. Seems to be back online now though.

[axipher]

13 hours ago, zodac said:

Yeah, axihub was down for a few hours and inaccessible. Seems to be back online now though.

 

Yep, the joys of moving houses, slowly building up new rack now:

[zodac]

Hey all. As promised, the LARS automation work is basically done and will be turned on for November. Here's the output of the multiplier changes if I was to run it right now on the existing hardware, for reference. @BWG

 

wildfly_dev | Best PPD is '6,643,151'
wildfly_dev | 212 from LARS, 15 in DB, 15 to update
wildfly_dev | Updated hardware 'Navi 10 [Radeon RX 5600 OEM/5600 XT/5700/5700 XT]' (ID: 46), multiplier: 5.07 -> 4.7, average PPD: 1301320 -> 1413961
wildfly_dev | Updated hardware 'TU104 [GeForce RTX 2080 Super]' (ID: 55), multiplier: 2.48 -> 2.6, average PPD: 2693818 -> 2552709
wildfly_dev | Updated hardware 'GP107GL [Quadro P620]' (ID: 40), multiplier: 48.13 -> 49.5, average PPD: 138084 -> 134216
wildfly_dev | Updated hardware 'GM204 [GeForce GTX 980] 4612' (ID: 24), multiplier: 8.76 -> 7.86, average PPD: 771118 -> 845108
wildfly_dev | Updated hardware 'GM107 [GeForce GTX 750] 1111' (ID: 41), multiplier: 97.13 -> 96.45, average PPD: 68879 -> 68879
wildfly_dev | Updated hardware 'Vega 20 [Radeon VII]' (ID: 56), multiplier: 3.33 -> 3.31, average PPD: 2008233 -> 2008233
wildfly_dev | Updated hardware 'GM206 [GeForce GTX 960] 2308' (ID: 57), multiplier: 21.3 -> 20.64, average PPD: 314056 -> 321821
wildfly_dev | Updated hardware 'GK104 [Quadro K5000]' (ID: 45), multiplier: 61.27 -> 60.84, average PPD: 109196 -> 109196
wildfly_dev | Updated hardware 'TU102 [GeForce RTX 2080 Ti Rev. A] M 13448' (ID: 48), multiplier: 1.32 -> 1.5, average PPD: 4960313 -> 4430496
wildfly_dev | Updated hardware 'TU106 [GeForce RTX 2070 Rev. A] M 7465' (ID: 50), multiplier: 2.71 -> 2.7, average PPD: 2448739 -> 2463563
wildfly_dev | Updated hardware 'GM204 [GeForce GTX 970] 3494' (ID: 21), multiplier: 12.97 -> 13.44, average PPD: 511994 -> 494421
wildfly_dev | Updated hardware 'Ellesmere XT [Radeon RX 470/480/570/580/590]' (ID: 42), multiplier: 12.91 -> 13.49, average PPD: 519263 -> 492376
wildfly_dev | Updated hardware 'GP104 [GeForce GTX 1070 Ti] 8186' (ID: 51), multiplier: 4.78 -> 4.74, average PPD: 1399391 -> 1400077
wildfly_dev | Updated hardware 'GM107GL [GRID M40]' (ID: 54), multiplier: 14.51 -> 57.64, average PPD: 115244 -> 115244
wildfly_dev | Updated hardware 'raven [Radeon RX Vega gfx902]' (ID: 53), multiplier: 153.94 -> 115.05, average PPD: 44057 -> 57743

 

[zodac]

Hi all,

 

The stats site may be unavailable/broken for a few hours, just changing some internal links around and it might take some time to propagate out.

[axipher]

14 hours ago, zodac said:

Hi all,

 

The stats site may be unavailable/broken for a few hours, just changing some internal links around and it might take some time to propagate out.

 

Should be all up now with HTTPS implemented by a Z and his Canadian side-kick...

 

Accessible via https://teamcomp.axihub.ca/ or https://etf.axihub.ca/

[zodac]

10 hours ago, axipher said:

 

Should be all up now with HTTPS implemented by a Z and his Canadian side-kick...

 

Accessible via https://teamcomp.axihub.ca/ or https://etf.axihub.ca/

 

Yup, nice and secure now. If anyone is updating bookmarks, use the https://etf.axihub.ca link, might get rid of the other one at some point and only use ETF going forward.

Edited November 3, 2021 by zodac

[zodac]

Hi everyone,

 

Been a while since I've made an update, but there's a new feature coming to the site later today (live now!). I've added the ability for users to request some common changes directly through the site, rather than needing to go through @BWG/ @damric.

 

You'll see a new option on the navigation bar, at System> Request A User Change

 

From here, you can request a change for:

• Your Folding@Home username
• Your passkey
• Your live stats link (if you have one)
• Your hardware

Note: You will need to enter the existing passkey first, to avoid anyone requesting changes for someone else.  

 

 

You can also request the change to made ASAP, or schedule it for the start of next month.

 

 

Once the request is made, please send a PM to @damric, and they'll review it and approve (assuming they're happy with the change).

 

Pinging the captains so they're aware.

@BWG @CharliesTheMan @damric @firedfly @Supercrumpet@tictoc

 

Edited March 7, 2022 by zodac

[Avacado]

57 minutes ago, zodac said:

Hi everyone,

 

Been a while since I've made an update, but there's a new feature coming to the site later today. I've added the ability for users to request some common changes directly through the site, rather than needing to go through @BWG/ @damric.

 

You'll see a new option on the navigation bar, at System> Request A User Change

 

From here, you can request a change for:

• Your Folding@Home username
• Your passkey
• Your live stats link (if you have one)
• Your hardware

Note: You will need to enter the existing passkey first, to avoid anyone requesting changes for someone else.  

 

 

You can also request the change to made ASAP, or schedule it for the start of next month.

 

 

Once the request is made, please send a PM to @damric, and they'll review it and approve (assuming they're happy with the change).

 

Pinging the captains so they're aware.

@BWG @CharliesTheMan @damric @firedfly @Supercrumpet@tictoc

 

This is really good work, thank you zodac! Outstanding addition. 

[ENTERPRISE]

9 hours ago, zodac said:

Hi everyone,

 

Been a while since I've made an update, but there's a new feature coming to the site later today (live now!). I've added the ability for users to request some common changes directly through the site, rather than needing to go through @BWG/ @damric.

 

You'll see a new option on the navigation bar, at System> Request A User Change

 

From here, you can request a change for:

• Your Folding@Home username
• Your passkey
• Your live stats link (if you have one)
• Your hardware

Note: You will need to enter the existing passkey first, to avoid anyone requesting changes for someone else.  

 

 

You can also request the change to made ASAP, or schedule it for the start of next month.

 

 

Once the request is made, please send a PM to @damric, and they'll review it and approve (assuming they're happy with the change).

 

Pinging the captains so they're aware.

@BWG @CharliesTheMan @damric @firedfly @Supercrumpet@tictoc

 

That is some mighty fine work bud!

[damric]

14 hours ago, zodac said:

Hi everyone,

 

Been a while since I've made an update, but there's a new feature coming to the site later today (live now!). I've added the ability for users to request some common changes directly through the site, rather than needing to go through @BWG/ @damric.

 

You'll see a new option on the navigation bar, at System> Request A User Change

 

From here, you can request a change for:

• Your Folding@Home username
• Your passkey
• Your live stats link (if you have one)
• Your hardware

Note: You will need to enter the existing passkey first, to avoid anyone requesting changes for someone else.  

 

 

You can also request the change to made ASAP, or schedule it for the start of next month.

 

 

Once the request is made, please send a PM to @damric, and they'll review it and approve (assuming they're happy with the change).

 

Pinging the captains so they're aware.

@BWG @CharliesTheMan @damric @firedfly @Supercrumpet@tictoc

 

 

I love it. Thanks for doing this.

[Supercrumpet]

On 07/03/2022 at 15:55, zodac said:

Hi everyone,

 

Been a while since I've made an update, but there's a new feature coming to the site later today (live now!). I've added the ability for users to request some common changes directly through the site, rather than needing to go through @BWG/ @damric.

 

You'll see a new option on the navigation bar, at System> Request A User Change

 

From here, you can request a change for:

• Your Folding@Home username
• Your passkey
• Your live stats link (if you have one)
• Your hardware

Note: You will need to enter the existing passkey first, to avoid anyone requesting changes for someone else.  

 

 

You can also request the change to made ASAP, or schedule it for the start of next month.

 

 

Once the request is made, please send a PM to @damric, and they'll review it and approve (assuming they're happy with the change).

 

Pinging the captains so they're aware.

@BWG @CharliesTheMan @damric @firedfly @Supercrumpet@tictoc

 

 

This looks great, thank so much!