LAR_Systems

https://folding.lar.systems/projects/folding_profile/19004

It's not the spikes as we take multiple samples across work units. What I commonly see with new GPUs is that F@H will toss work units that can be too easy or more less not optimized for the new GPU and it results is crazy results for the first few weeks. Then things seem to settle (average) down in almost all cases which suggests they benchmark then put the GPUs to work on the kind of work where they can get the most out of the hardware. The chart below showing the first two weeks vs this week so far suggest this is the case and it's normalizing down as they appear to push far more Influenza and pull back on Alzheimer's / remove membrane transport projects. This kind of thing is why I show the chart and I felt the PPD DB was important enough to build because "the ppd a GPU gets" changes a lot over time with F@H, Drivers, OS etc. evolving what they are doing.

I'm working on it

I manually manage my own database and fill in the unspecified projects using their descriptions

Congratulations to everyone getting down to the final hours of the event. To keep things exciting my scheduled network outage starts soon, so my machines are wrapping up their final WUs for the event, which may result in a rank change in the top 10 tomorrow. keep on folding to the bitter end, it’s a bummer my place is going to start getting cold.

I'm in a similar boat, about half my rigs cards are due for a re-paste, that will be a fun afternoon.

Fix for this will hopefully make it's way in to the public in the next 12 hours or so. Thanks for letting me know.

Get some rest, sleep well.

Thanks for the update, been doing a bunch of rack updates here and apparently after a power cycling the script did not restart as it should. Will been hardening that along with other upgrades shortly. Cheers,

Hey I'm as shocked as you are, or am I? Frankly, not sure what happened to @Oromit but It might have saved my bacon as I might have a scheduled network shutdown near the last day of the event.

If I had to guess based on the number of work unit samples... only a couple. 444 Samples over the past 4-5 days... avg is about 23 WU a day on that card so like 4 people maybe.

Don't make me buy a spite GPU this late in the event.

You have been seen https://folding.lar.systems/gpu_ppd/brands/nvidia/folding_profile/ad103_geforce_rtx_4080

Our rigs go cold until late Monday until the research/ F@h staff refill our supply of work units. It’s happened before where we have cleaned them out over a weekend.

I expect pain.

TIL many members of this forum are well versed in being horny.

It looks like it's made it's way on to the list now. https://folding.lar.systems/projects/folding_profile/16984 More less, it can take 6 hours to a day of folding a WU to get enough samples for a device and to allow the time for some of the backend data processing scripts on my side to process that info, add it to the DB then re-publish the DB pages / API data for the client.

I can't tell them apart is the issue, because F@H does not tell them apart. So the PPD is all blended average as is the power avg as they are all considered ONE GPU with that magic multiple model name by F@H. So no, there is not really a way for me to more accurately do that calculation as I don't know what specific model is providing the PPD report to then calculate and pair with the wattage etc. AMD GPUs get the second citizen treatment by the F@H team and I can't override that as if you have the best or worst model of that series they are all reported the same in client, F@H and therefore the DB.

It's not a flaw it's a reality of providing numbers on a list that users can use to see how different cards stack up against each other in general, on average by GPU model and most importantly at a glance as it's always changing. The general design and purpose of the DB as stated on the pages is to provide an average idea of performance and ranking of cards as they change regularly due to changes in work units being provided to fold over time. As the folding client does not track or provide real time wattage use data by WU, the database does not either as baseline power spec numbers for a GPU is all I can draw from when calculating averages and if I use the average TDP spec it's a static variable forever when calculating those averages. The idea of doing this this way is fine as it does not change the general outcome of the list rank for the "AVERAGE user" either as those new to folding or just trying to get their hands on a card at a decent price / cost of ownership etc. can make an informed average decision with the info provided. For the more savvy users out there, yes... of course you can overclock, undervolt, power limit, power shunt, hack drivers, change OS and BIOS settings etc. etc. etc. on top of there being multiple different card power / clock specs for each model and across different coolers / water cooling and room temps and thermal limits etc. and manufactures all changing the parameters for board partner cards. If I spent the time to create a compiled program to record all that sensor data on all OS and GPU types etc. the information would be pretty much useless reported raw by all those specifics as there are too many variables for people to digest and compare the information at that level of detail. So... knowing this from my day job... I would have to... wait for it... average it all out or group it and average it broadly so users can get the info they need at a glance. Much like hardware reviewers provide doing comparison in percentages at stock settings. Which brings us back to how it works now lol, if someone can hot rod the card and that data is logged it brings up the averages... then others stop being noobs and wonder how... start digging and next thing you know they are tinkeing, which brings up their averages so on and so on. How they handle the power settings, clocks and if they run a 200w box fan in the side of their case is not something I can account for, nor should I providing overview baseline data. You can look at the data on most Nvidia cards and see the moment in the performance curves the driver that removed LHR and added 4000 series updates including performance gains for 3000 series... yet another factor that can't really be captured easily. Then there will also just be the people that go 4090 GO BURRR and optimize nothing though brute force because they just buy the top of the list... it's not what I would do... but for those new to folding, and as long as they are folding I don't really care.. it's all still for the cause.

It's on the high side to reflect the highest card in the grouping as F@H does not isolate Ellesmere cards and no matter what of the following models you are folding on all get blended together in the DB as that's how F@H has decided to do it for some reason.

It won't be in my list unless someone is folding with it using the extension.

The M1 chips are impressive for what they are, but I would not be buying them as folding machines... Like M1 Max all cores might get 153k PPD down to the M1 getting around 44k PPD https://folding.lar.systems/cpu_ppd/brands/apple/folding_profile/apple_m1_max https://folding.lar.systems/cpu_ppd/brands/apple/folding_profile/apple_m1_pro https://folding.lar.systems/cpu_ppd/brands/apple/folding_profile/apple_m1

Reading this page of posts, I simply have to accept I will not be in the 10 ten by the end of the event pretty much regardless of what I do. I salute you those with sick hardware hookups.