New TC stats system & progress

[BWG]

Here's my notes. @Avacado

 

Looks like the PPD estimates change on LARS, so that's another wrench in the plan. Edit: We'll likely do a quarterly lock on LARS. I'll end up grabbing a copy of the page, posting it, and then doing the math before the event starts a month. Adjust all of the multipliers for everyone. Lock it in for 3 months. Redo after 3rd month. Repeat. 

 

I'm going to take the existing stats, plug in some new multipliers for basing everything on a 3090, and then show everyone the difference between existing, basing each categories multipliers on the top card actively folding in the category, vs everything being based on the 3090. We will discuss with the Team Captains, and then decide which method to use.  

 

Edited April 22, 2021 by BWG

[damric]

If this is indeed going to be monthly, I would like to experiment some CPU folding in the future since I have an interesting overclocking chip on my test rig, a 1600 AF.

[Avacado]

47 minutes ago, BWG said:

Here's my notes. @Avacado

 

Looks like the PPD estimates change on LARS, so that's another wrench in the plan.

 

TY. Ok, so the math is computated against the highest performing card average in the category vs LARS averages, got it.

 

Now the real problem, assuming that the above is true, there are 2 different revisions of the Ti with vastly different averages. Is Bastiaans Ti a non-rev A for certain? I'm sure you validated this, so i'll assume it is. 

 

Problem #1- We have no way of knowing if the averages on LARS are with Or without OC's. I would assume that on newer cards they most likely will be and to a lesser extent the older the card becomes.

 

Problem #2- The disparity in overclockability between newer and older cards. I'll use my best Ti as an example. Currently folding at 94% load, stock it's at 1830 core, OC it gets +200 core to 2030. This fluctuates a lot, but for arguments sake, bear with me just to get a percentage.  That is an 11% increase in perf. I don't think there will be any way to squeeze that kind of OC out of a Tesla, but I have yet to experiment. Will be cool to see. 

 

I fully expect the advantage to lie with newer gen cards. I WILL test this and get back to you. 

Edited April 22, 2021 by Avacado

[zodac]

Hey guys.

 

I'm running into trouble reworking the DB tables and still keeping the existing stats running. I'm going to have to shut down for a little bit and do some testing. I do have the initial stats for all users for this month, so I should be able to get a valid monthly total for everyone (I can do it manually if I have to), and hope that I'll be back up over the weekend. But I underestimated how much stuff would need to move around.

 

I'm probably gonna bring it down in the next hour or two.

[BWG]

35 minutes ago, Avacado said:

TY. Ok, so the math is computated against the highest performing card average in the category vs LARS averages, got it.

 

Now the real problem, assuming that the above is true, there are 2 different revisions of the Ti with vastly different averages. Is Bastiaans Ti a non-rev A for certain? I'm sure you validated this, so i'll assume it is. 

 

Problem #1- We have no way of knowing if the averages on LARS are with Or without OC's. I would assume that on newer cards they most likely will be and to a lesser extent the older the card becomes.

 

Problem #2- The disparity in overclockability between newer and older cards. I'll use my best Ti as an example. Currently folding at 94% load, stock it's at 1830 core, OC it gets +200 core to 2030. This fluctuates a lot, but for arguments sake, bear with me just to get a percentage.  That is an 11% increase in perf. I don't think there will be any way to squeeze that kind of OC out of a Tesla, but I have yet to experiment. Will be cool to see. 

 

I fully expect the advantage to lie with newer gen cards. I WILL test this and get back to you. 

 

Yep, everything we ever do will have all sorts of issues, but I think the issues above a far less vs the old way we used to run the event where your m40 would be against gpu's it stood no chance at all of beating, and the gap in points was insurmountable. For the people who used to compete in the old TC, this system is so refreshing! 

 

I will say this though, every piece of hardware I have folding is overclocked and is beating the LARS DB by a good amount. It's a decently wide range, 9xx, 1xxx and 3xxx. Yeah, your Tesla's not going to be great since it can't OC easily, but you are competitive against a 2080ti right now, and the points are so very close between the teams.

[zodac]

25 minutes ago, zodac said:

Hey guys.

 

I'm running into trouble reworking the DB tables and still keeping the existing stats running. I'm going to have to shut down for a little bit and do some testing. I do have the initial stats for all users for this month, so I should be able to get a valid monthly total for everyone (I can do it manually if I have to), and hope that I'll be back up over the weekend. But I underestimated how much stuff would need to move around.

 

I'm probably gonna bring it down in the next hour or two.

 

Last stats for a few days. We can compare against this when things are running again:

 

[BWG]

34 minutes ago, zodac said:

 

Last stats for a few days. We can compare against this when things are running again:

 

 

Cool, Damric got one into verify he's working!

[tictoc]

Liking all of the above.  There will always be some outliers with any handicapping system, but I dig the overall plan. 

 

For what it's worth the db baseline numbers are almost certainly mostly cards running at stock with whatever version of GPU boost the card has.  Everything that I've ran recently and in the past always exceeded whatever averages were out there, whether it was LARS or the old wu database on OCN.

 

@Avacado I've got some tricks for your Teslas that we can test out over the next month, and it should be smoking by the time the comp starts on June 1st.

Edited April 22, 2021 by tictoc

[zodac]

Just now, tictoc said:

Liking all of the above.  There will always be some outliers with any handicapping system, but I dig the overall plan. 

 

For what it's worth the db baseline numbers are almost certainly mostly cards running at stock with whatever version of GPU boost the card has.  Everything that I've ran recently and in the past always exceeded whatever averages were out there, whether it was LARS or the old wu database on OCN.

 

@Avacado I've got some tricks for your Teslas that we can test out over the next month, and it should be smoking by the time the comp starts on June 1st.

 

You know, while it's not exposed on the UI, I do also keep the non-multiplied points for each user. You could always keep going with the non-multiplied points for a month or so until you learn to tweak the multipliers to be more valid? Just spitballing, don't know if it's worth bringing in manual multipliers, or just trying to leverage the LARS DB.

[BWG]

51 minutes ago, zodac said:

 

You know, while it's not exposed on the UI, I do also keep the non-multiplied points for each user. You could always keep going with the non-multiplied points for a month or so until you learn to tweak the multipliers to be more valid? Just spitballing, don't know if it's worth bringing in manual multipliers, or just trying to leverage the LARS DB.

 

I think tictoc already had that run through his mind when he asked me if he should be folding at stock. lol

[Bastiaan_NL]

  

2 hours ago, Avacado said:

Now the real problem, assuming that the above is true, there are 2 different revisions of the Ti with vastly different averages. Is Bastiaans Ti a non-rev A for certain? I'm sure you validated this, so i'll assume it is. 

My card is a Rev A. 

 

45 minutes ago, tictoc said:

For what it's worth the db baseline numbers are almost certainly mostly cards running at stock with whatever version of GPU boost the card has.  Everything that I've ran recently and in the past always exceeded whatever averages were out there, whether it was LARS or the old wu database on OCN.

 

@Avacado I've got some tricks for your Teslas that we can test out over the next month, and it should be smoking by the time the comp starts on June 1st.

Same for me, it's not hard to beat the database averages with any of the cards I own. 
Can I share in the M40 knowledge? I'm not going to share it with anyone, just interested.

[BWG]

Oh I'm sure it involves extracting the bios, hitting it with the editor, and then flashing it.?

[Avacado]

1 hour ago, tictoc said:

Liking all of the above.  There will always be some outliers with any handicapping system, but I dig the overall plan. 

 

For what it's worth the db baseline numbers are almost certainly mostly cards running at stock with whatever version of GPU boost the card has.  Everything that I've ran recently and in the past always exceeded whatever averages were out there, whether it was LARS or the old wu database on OCN.

 

@Avacado I've got some tricks for your Teslas that we can test out over the next month, and it should be smoking by the time the comp starts on June 1st.

Aye Aye captain! Look forward to it!

1 hour ago, zodac said:

 

You know, while it's not exposed on the UI, I do also keep the non-multiplied points for each user. You could always keep going with the non-multiplied points for a month or so until you learn to tweak the multipliers to be more valid? Just spitballing, don't know if it's worth bringing in manual multipliers, or just trying to leverage the LARS DB.

Thank you for all the hard work on the back end. Nothing I say is a critique of anything you or BWG are doing, just digging for a better understanding. 

7 minutes ago, Bastiaan_NL said:

  

My card is a Rev A. 

 

Same for me, it's not hard to beat the database averages with any of the cards I own. 
Can I share in the M40 knowledge? I'm not going to share it with anyone, just interested.

M40 knowledge is open source Teddy Broosevelt. You can have all you can handle. 

Edited April 22, 2021 by Avacado

[Avacado]

2 hours ago, BWG said:

 

Yep, everything we ever do will have all sorts of issues, but I think the issues above a far less vs the old way we used to run the event where your m40 would be against gpu's it stood no chance at all of beating, and the gap in points was insurmountable. For the people who used to compete in the old TC, this system is so refreshing! 

 

I will say this though, every piece of hardware I have folding is overclocked and is beating the LARS DB by a good amount. It's a decently wide range, 9xx, 1xxx and 3xxx. Yeah, your Tesla's not going to be great since it can't OC easily, but you are competitive against a 2080ti right now, and the points are so very close between the teams.

 

Fair points, If you looks at the figures, I would wager that the "Best PPD" stats are literally coming from my Teslas. That is the exact PPD I am getting with them, beating the average by 200k PPD, they are also submitting WU's <4 Hours making them great for the 4 hour update window. I guarantee that the only reason I am getting the best PPD stats are solely due to the cooling solution. The stock passive with high flow CFM still see the cards thermal throttle. They rarely get above 60c now. 

 

With 266 "Windows Samples" returned, I would also wager that I am the only person in the world folding M40's right now. Just a hunch  Hurry up with the build @ENTERPRISE  

Edited April 22, 2021 by Avacado

[BWG]

Yeah, that's the variable you get to take advantage of, as do I. You're exceeding, and it showed in your stats. 

[zodac]

Stats are back up. Let me know if anything looks wrong, I already caught one place I mixed up multiplied and non-multiplied points.

[ENTERPRISE]

I just wanted to say a huge well done to everyone:)

[tictoc]

Other than the very obvious 1.2M mixup in FPN and TFA's point totals everything looks good to me.

[zodac]

3 hours ago, tictoc said:

Other than the very obvious 1.2M mixup in FPN and TFA's point totals everything looks good to me.

I assume this is a joke, but if not, I am prepared to give @BWG -2mil points, just to be sure.

[tictoc]

3 minutes ago, zodac said:

I assume this is a joke, but if not, I am prepared to give @BWG -2mil points, just to be sure.

 

The -2M should probably be given, but for now I'm prepared to let him be happy at the top of the leaderboard as a reward for getting things fired up.

 

All jokes aside, I don't see any issues, and thanks for taking the time to help. 

[Avacado]

1 hour ago, tictoc said:

 

The -2M should probably be given, but for now I'm prepared to let him be happy at the top of the leaderboard as a reward for getting things fired up.

 

All jokes aside, I don't see any issues, and thanks for taking the time to help. 

For real, let him have his fun, Lets get that Titan X BIOS going for the teslas and turn up the heat. Will be switching to a WB'd card when I get that rig going. 

Edited April 23, 2021 by Avacado

[BWG]

@zodacwhere's your team thread?  

[zodac]

6 minutes ago, BWG said:

@zodacwhere's your team thread?  

 

Private. Team members only.

[Avacado]

13 minutes ago, zodac said:

 

Private. Team members only.

No Sex Folding in zodac's room, NONE! 

 

 

Edited April 23, 2021 by Avacado

[zodac]

Hey everyone. I've updated the system to no longer filter on the Stanford team you Fold on, after we discussed it probably wasn't necessary. So as long as you continue to abide by the 1 H/W for 1 user+passkey, you should be able to switch between OCN, EHW or anywhere else whenever you want.