New TC stats system & progress

[Bastiaan_NL]

17 hours ago, Avacado said:

P.S. We really do need to have the conversation about updating stats monthly instead of quarterly (I think you said something about quarterly  @BWG) Disregard the rest if it is monthly. If you actually look at just how drastic some of the changes are already, you will see what I mean. What is the point of having  a handicapped system if it isn't on par. It's like a professional athlete playing in the special Olympics.  

 

For instance that 960, would finally come down a notch, by todays rankings, it would only be a 21x multiplier. That is huge and goes to show just how much a niche card with low default LAR's values and high overclockability can achieve over the rest of us losers. 

I agree with this, though I don't know how we could improve the system. The handicap is a good thing, but it doesn't even out the cards. Instead it's a game of finding the card with the worst PPD compared to what it can actually do, and that's not what I'm able/willing to do. The hardware I fold on needs to be efficient power to PPD wise(or Units Per Day), so I'm not switching to anything old. 

 

56 samples on a 7850 vs 117547 on a 2080Ti, just saying

Edited August 11, 2021 by Bastiaan_NL

[Avacado]

5 minutes ago, Bastiaan_NL said:

I agree with this, though I don't know how we could improve the system. The handicap is a good thing, but it doesn't even out the cards. Instead it's a game of finding the card with the worst PPD compared to what it can actually do, and that's not what I'm able/willing to do. The hardware I fold on needs to be efficient power to PPD wise(or Units Per Day), so I'm not switching to anything old. 

 

56 samples on a 7850 vs 117547 on a 2080Ti, just saying

Yep and Iv'e already found my next card to add for next month. 

[Bastiaan_NL]

I bet you won't tell me...

[damric]

Considering the rules, I'm not sure why anyone would use a modern expensive card with low OC headroom for ETF. Pre-boost era cards with voltage control, particularly ones based on cut-down dies are the way to go for overclocking and getting the most handicap.

 

There is a particular Fermi card I'm thinking of if anyone can guess.

[BWG]

I think I have enough of a GPU variety to switch the stats monthly, but the initial decision was quarterly. I seriously don't think my work units are in the last 500 lars submissions unlike poor @damric who it is probably likely for. But, we were working toward automating the stats to automatically update the multipliers. 

 

Right now, it's a bit trivial to achieve with the current state of the stats site. I can't seem to update the multipliers without creating a new hardware name. I really don't want to bother zodac with that right now. He just moved and started his job. He did a ton of work to get us here and may need a little time.

 

Asking me to change a bunch of hardware, multipliers, and stuff on the last day of the month every month isn't really in the cards for me. The last day of the month in the mortgage business gets very crazy. But, if things get real out of hand in September, maybe we could in October. 

 

I don't think any moonshot wu crunching card will do very good next month. Could be a real crapshoot either way. I will say this though, Michael Hertz is looking down and laughing at his GPU kicking all of your @$$3$!  

 

[damric]

At least we're all folding. I was watching the news tonight and it's depressing but we are at least doing something good.

[zodac]

Just an FYI to everyone, I did have a little proof of concept that loaded all the GPU entries from LARS at the start of each month, and updated all the multipliers automatically. I didn't put it in the live system because wasn't sure if we wanted multipliers changing every month. Also, it gives no flexibility to allow for non-LARS GPUs to be added (wouldn't be hard to add, just needs more time).

 

If it's a big hindrance, it's definitely something I could try and make time for.

 

1 hour ago, BWG said:

Right now, it's a bit trivial to achieve with the current state of the stats site. I can't seem to update the multipliers without creating a new hardware name. I really don't want to bother zodac with that right now. He It just moved and started his its job. He It did a ton of work to get us here and may need a little time.

 

 

Hey, I just tested some updates on axihub myself. From what I can see, the updates are taking effect on the backend, but not updating on the UI (I'm new to website caching, so probably haven't set things up properly). If you do a SHIFT+F5 to refresh the page, the values should be updated.

Let me know if that works. If so, I'll disable the caching for now so at least you can modify things a bit easier.

Edited August 12, 2021 by zodac

[Avacado]

15 hours ago, Bastiaan_NL said:

I bet you won't tell me...

Not a chance. I actually snagged 2 different cards. We'll see which is better.

[BWG]

9 hours ago, zodac said:

Just an FYI to everyone, I did have a little proof of concept that loaded all the GPU entries from LARS at the start of each month, and updated all the multipliers automatically. I didn't put it in the live system because wasn't sure if we wanted multipliers changing every month. Also, it gives no flexibility to allow for non-LARS GPUs to be added (wouldn't be hard to add, just needs more time).

 

If it's a big hindrance, it's definitely something I could try and make time for.

 

 

Hey, I just tested some updates on axihub myself. From what I can see, the updates are taking effect on the backend, but not updating on the UI (I'm new to website caching, so probably haven't set things up properly). If you do a SHIFT+F5 to refresh the page, the values should be updated.

Let me know if that works. If so, I'll disable the caching for now so at least you can modify things a bit easier.

 

It's not really something I can take on at month end, but the automation sounds interesting. I didn't realize you already worked on it. We moved away from using Windows/Linux values though, and it's all o/s now. I don't really want to change it for September, but possibly October and after.

 

 What does it think? 

 

18 minutes ago, Avacado said:

Not a chance. I actually snagged 2 different cards. We'll see which is better.

 

I could make a testing team and put 2 of you on it with your new gpus to test what they really do if you'd like. I wanted to eventually have something like that anyway for people wanting to test hardware.

Edited August 12, 2021 by BWG

[Avacado]

12 minutes ago, BWG said:

 

It's not really something I can take on at month end, but the automation sounds interesting. I didn't realize you already worked on it. We move away from using Windows/Linux values though, and it's all o/s now. I don't really want to change it for September, but possibly October and after.

 

 What does it think? 

 

 

I could make a testing team and put 2 of you on it with your new gpus to test what they really do if you'd like. I wanted to eventually have something like that anyway for people wanting to test hardware.

Kind of seems like a moot point. I bought the cards to combat unfair advantages in the current system (960, 7850). Though I would be more than happy to do testing towards ETF improvements, it would be a slap in the face to keep getting dominated by you 3 whilst doing so. Like you mentioned, when you let me add my new card of choice, I will be happy to do a detailed write up of how flawed the LARs database can be if one knows how to look. It will be about a week for me to get both cards, change TIM/pads and do preliminary testing. 

 

Seems to me the only real fair way to add cards would be to make users run them for a month with LAR's chrome browser open to track said cards and give a more accurate value for the averages. One could argue on cards like the 2080Ti, like bastiaan said, that there are already so many entries and it is very close to baseline that this testing wouldn't be needed. 

 

Lastly, we would all need to have some integrity, it would be very easy to run your cards gimped to obtain low stats, before slamming a 50% OC on them. I trust everyone here wouldn't be doing something like this. 

Edited August 12, 2021 by Avacado

[Avacado]

This is an approved stickied response from @Ducrider's PM.

 

"I agree with you 100 percent. Picking a card that can overclock and make a bit more ppd is the goal. To pick a card that has a stupid handicap with a ppd gain over the Lars stats (which change daily) is not playing fair. We all should be able to make more ppd with our cards over the listed ppd. That also means we should all be in a close race for teams and category leaders. We should not see 2 people not to say names that can literally blow past everyone and make twice the adjusted points daily. To be honest the rules state we can't end low ppd work units etc. Well isn't only having the covid moonshot work units about the same? There needs to be some changes. Changes need to be made soon. Like you fold on that card for 30 days with said passkey and you get a daily average to be based off of. You have hardware problems well it is what it is. Can Nascar swap in a new motor during a race? Same principle.  It's hard to base a fair game on outdated hardware and outdated stats. If you think your hardware won;t make 30 days then you need to use something else. Just my thoughts."

 

 

 

I agree with everything you said @Ducrider, however selecting COVID-19 as a project priority in the F@H client does NOT guarantee you will get those units. End users have literally 0 control over what wu's are being fed to them. If a person is purposefully ending low WU's to remain competitive, then they have 100% lost all of my respect and it shouldn't be happening at all.  

Edited August 12, 2021 by Avacado

[BWG]

There will always be someone who won't like the way things are working. 

 

I ran my 960 at it's max oc for the last 4 months testing in this events beta. I didn't change to the 3080ti for nothing at the start of the month. 960 started failing units, so I knew something was up. 

 

There will always be issues. But, what will be the next complaint from you after we go to monthly LARS?

[Avacado]

5 minutes ago, BWG said:

There will always be someone who won't like the way things are working. 

 

I ran my 960 at it's max oc for the last 4 months testing in this events beta. I didn't change to the 3080ti for nothing at the start of the month. 960 started failing units, so I knew something was up. 

 

There will always be issues. But, what will be the next complaint from you after we go to monthly LARS?

Wow, ok. You act like I am the only one voicing concerns here. Do you even want anyone to help improve this or does it upset you enough that we should just leave it be?

 

I bet when you ran your 960 for 4 months, you weren't tracking those stats with LARs.

 

Don't even worry about it, I guess its fine the way it is. Its good to be king right? If you and I are working together on this and you think my suggested improvements are complaints, I won't have anything else to add. 

Edited August 12, 2021 by Avacado

[firedfly]

20 minutes ago, Avacado said:

Seems to me the only real fair way to add cards would be to make users run them for a month with LAR's chrome browser open to track said cards and give a more accurate value for the averages. One could argue on cards like the 2080Ti, like bastiaan said, that there are already so many entries and it is very close to baseline that this testing wouldn't be needed. 

 

LAR's only uses the most recent 500 ppd samples to calculate averages.  We could limit cards to the ones with at least 500 samples in LARs.  That may mean someone needs to fold on a card at stock settings for about two weeks with the LAR's chrome extension running to fill in the required samples.

 

 

16 minutes ago, Avacado said:

From Ducrider:

 

Changes need to be made soon. Like you fold on that card for 30 days with said passkey and you get a daily average to be based off of.

 

This is also an interesting idea.  I would make a small twist to it and say that the ETF stats need to have a month of stock clocked folding for the card to be eligible.  I think the same card from anyone would be fine.  e.g. if I fold at stock clocks on a 960, then others can use those stats as well--we don't need to benchmark every card.

 

 

I think that both of these ideas have merit to make the competition more competitive.   However, both of these options also make it more difficult to find new members to join the competition due to the extra overhead.  It would also be more difficult to replace failed hardware.  These would also both require more work by @BWG or others to keep up with the new restrictions.

 

I'm undecided on what would be best for the competition and look forward to more discussion about this.

[Avacado]

8 minutes ago, firedfly said:

 

LAR's only uses the most recent 500 ppd samples to calculate averages.  We could limit cards to the ones with at least 500 samples in LARs.  That may mean someone needs to fold on a card at stock settings for about two weeks with the LAR's chrome extension running to fill in the required samples.

This is interesting, I can see numerous cards that have averages assigned with way less than 500 sample sizes. Do you think they just assign the first value they get and then modify after 500?

[axipher]

I know I'm not providing much for my team since I'm on a 5700 at stock clocks hitting 24/7 folding.  All I can adjust is the power limit and kind of adjust the clocks a little, but nothing that has consistent results on Folding@Home without random lock-ups.

 

I know that I could throw in some older hardware and OC the piss out of it to make the most of the larger multipliers on older hardware but I don't have anything else aside from a Radeon HD 8400 that I use in my emulation rig for S-Video output to my old Tube TV for retro games.

 

 

I know right now the GPU shortages means that most people are likely running hardware that is at least 4 years old, but in an ideal world, I would think that maybe the main AMD and Nvidia categories could have a requirement to be within the last 2 generations of GPU's then the wildcard can be anything goes.  This could at least bring two of the categories a little closer and more reliant on uptime as most of the newer cards will be a little closer.

[firedfly]

14 minutes ago, Avacado said:

This is interesting, I can see numerous cards that have averages assigned with way less than 500 sample sizes. Do you think they just assign the first value they get and then modify after 500?

 

They update every 3 hours (about 30 minutes after EOC).  They recalculate the averages then, but only use up to the most recent 500 submissions.  If a card has fewer submissions, then all submissions will be used to calculate the average.

 

Additionally, the Chrome extension submits PPD samples about every 30 minutes--not when a WU is completed.  At least that was the case when I investigated this a month or two ago.

 

These details are all based on my own observations and some comments on the LAR's site.  I don't have any official knowledge.

 

This is from the GTX 570 that I'm populating in LARs right now.  It has fewer than 500 samples and notes that in the average calculation:

 

This is from the RTX 3080TI that has more than 500 samples:

Edited August 12, 2021 by firedfly

[BWG]

23 minutes ago, Avacado said:

Wow, ok. You act like I am the only one voicing concerns here. Do you even want anyone to help improve this or does it upset you enough that we should just leave it be?

 

I bet when you ran your 960 for 4 months, you weren't tracking those stats with LARs.

 

Don't even worry about it, I guess its fine the way it is. Its good to be king right? If you and I are working together on this and you think my suggested improvements are complaints, I won't have anything else to add. 

 

Hey, I do want to improve things, but we're 12 days into the event. The ideas are all being taken into account for discussion with the Team Captains during the last week of this month. From there, we'll all decide what to work on and implement. 

 

It's the 12th, and your first full tracked day of folding has you pretty near the top of daily production. I don't think it's as broken as you lead us onto believe. I have squeezed nearly every bit of oc out of the 960 as I could. Does a K4000 overclock? Is it under water? It's not that far away from my 960.

 

Edited August 12, 2021 by BWG

[Avacado]

1 minute ago, BWG said:

 

Hey, I do want to improve things, but we're 12 days into the event. The ideas are all being taken into account for discussion with the Team Captains during the last week of this month. From there, we'll all decide what to work on and implement. 

 

It's the 12th, and your first full tracked day of folding has you pretty near the top of daily production. I don't think it's as broken as you lead us onto believe. I have squeezed nearly every bit of oc out of the 960 as I could. Does a K4000 overclock? Is it under water? It's that far away from my 960.

 

Hey, relax,

 

I know we are 12 days in, I don't expect changes overnight, I am just passionate about the things I choose to invest my time in. I thank you for all of your handwork every chance I get. I know how labor intensive this project is and you have done amazing things so far. I am not saying the system is broken, I see room for manipulation which can lead to massive changes in outcome. You are showing my stats, this is not by chance, it's after calculated failure and research in the current system (Plus I had it lying around). 

 

I could care less about winning or loosing, compared to you and Fired in the foldathons and the ETF, I have been doing my fair share of loosing. It's not a personal competition. Again, I am not the only person voicing these concerns. 

[BWG]

12 minutes ago, axipher said:

I know I'm not providing much for my team since I'm on a 5700 at stock clocks hitting 24/7 folding.  All I can adjust is the power limit and kind of adjust the clocks a little, but nothing that has consistent results on Folding@Home without random lock-ups.

 

I know that I could throw in some older hardware and OC the piss out of it to make the most of the larger multipliers on older hardware but I don't have anything else aside from a Radeon HD 8400 that I use in my emulation rig for S-Video output to my old Tube TV for retro games.

 

 

I know right now the GPU shortages means that most people are likely running hardware that is at least 4 years old, but in an ideal world, I would think that maybe the main AMD and Nvidia categories could have a requirement to be within the last 2 generations of GPU's then the wildcard can be anything goes.  This could at least bring two of the categories a little closer and more reliant on uptime as most of the newer cards will be a little closer.

 

You know I did think about that, but it's still hard to even get the last 2 generations. I'd probably have 50 GPUS by now if it weren't. It's still a good idea though. Probably one of the easiest to implement too. We could take a tally of what GPUS people have in the event to see if it would cause someone to drop out, and possibly re-align which Category people are in to make it work.

[BWG]

7 minutes ago, Avacado said:

Hey, relax,

 

I know we are 12 days in, I don't expect changes overnight, I am just passionate about the things I choose to invest my time in. I thank you for all of your handwork every chance I get. I know how labor intensive this project is and you have done amazing things so far. I am not saying the system is broken, I see room for manipulation which can lead to massive changes in outcome. You are showing my stats, this is not by chance, it's after calculated failure and research in the current system (Plus I had it lying around). 

 

I could care less about winning or loosing, compared to you and Fired in the foldathons and the ETF, I have been doing my fair share of loosing. It's not a personal competition. Again, I am not the only person voicing these concerns. 

 

Yeah, I've already had someone mention benching a card for a month and running it stock to manipulate LARS, and then re-entering the event. But, has it actually happened? I really couldn't tell you, but I'd say it has not because it's just now coming to light (I wish it did during testing), and no one has really changed GPUS for reasons other than hardware issues, except for tictoc and firedfly. Is it maybe enough for me to just add a rule against it, and just take the angle that everyone will be honest? I think it should be. We're all folding for a good cause. There's no real advantage to cheating after this month when the prizing goes away.

 

As much as the folding baseline option seems like the perfect cure, it's just not going to happen for the reasons firedfly stated. That one is a hard no from me, dogg. But, if we go to monthly LARS, and we can implement a 2 gen cap at some point in 2 categories, and make the rule against lars manipulation, how happy would that make everyone?

 

 

 

 

[damric]

No one complained about my HD 7850 over the last 4 months while running the same 40% OC, and that was with a x73 multiplier. Now it's x56.

 

My points should really be 30% higher if the handicap is based on the PPD from the stock clock shipped with an HD 7850, which is 860MHz, not the 1200MHz I run it now.

 

I have higher PPD cards, but I read the rules and saw that the handicap favors overclocking and that's why I chose my HD 7850. Using one of the newer generation cards that have hardly any headroom would best be used for the 3 day foldathons.

[BWG]

4 minutes ago, damric said:

No one complained about my HD 7850 over the last 4 months while running the same 40% OC, and that was with a x73 multiplier. Now it's x56.

 

My points should really be 30% higher if the handicap is based on the PPD from the stock clock shipped with an HD 7850, which is 860MHz, not the 1200MHz I run it now.

 

I have higher PPD cards, but I read the rules and saw that the handicap favors overclocking and that's why I chose my HD 7850. Using one of the newer generation cards that have hardly any headroom would best be used for the 3 day foldathons.

 

That's why I really did not say anything myself during testing. It's pretty obvious that you have a highly overclocked GPU that should have been dominant. That level of effort is supposed to be a way everyone in the event can do better. But, what I'm hearing is your units shouldn't have affected LARS unless you ran the Chrome LARS Extension? @firedfly I gotta get with you on that whole process and at least see how it all works. 

 

 

[Avacado]

28 minutes ago, damric said:

No one complained about my HD 7850 over the last 4 months while running the same 40% OC, and that was with a x73 multiplier. Now it's x56.

 

My points should really be 30% higher if the handicap is based on the PPD from the stock clock shipped with an HD 7850, which is 860MHz, not the 1200MHz I run it now.

 

I have higher PPD cards, but I read the rules and saw that the handicap favors overclocking and that's why I chose my HD 7850. Using one of the newer generation cards that have hardly any headroom would best be used for the 3 day foldathons.

Of course no one said anything, because we want your card to succeed, also, I spent my time trying to find a similar situation like yours. Find a card that can OC and bring it back to life. I love that idea. 

[damric]

16 minutes ago, BWG said:

 

That's why I really did not say anything myself during testing. It's pretty obvious that you have a highly overclocked GPU that should have been dominant. That level of effort is supposed to be a way everyone in the event can do better. But, what I'm hearing is your units shouldn't have affected LARS unless you ran the Chrome LARS Extension? @firedfly I gotta get with you on that whole process and at least see how it all works. 

 

 

Yeah I ran the dark mode extension. That's how it got messed up and pushed my GPU ranking so high. I was like oooh dark mode what's this looks cool...doht!