Work Unit 18706- Too short deadline

[stratos412]

Hello all

I received a work unit 18706 (0, 1664, 126), which I currently finished, nevertheless the deadline was too short, so I probably missed the points (no worries).
The problem was that when I received the unit, the deadline was two days, however PC needed almost 2.2 days to complete it.
(Intel Core i5 3570, GPU: TI AMD Radeon HD 7700 Series).
The deadline have pasted and I was about 2 hours+30 minutes short to finish it (PC was running 24 hours and it was almost 95% complete)
Can you please check it and give an extra 1-2 days so we can have time to complete the next work units?

08:49:57:FS01:Initialized folding slot 01: gpu:1:0 R575A [Radeon R7 250X/HD 7700/8760]
14:27:10:17:127.0.0.1:New Web session
******************************* Date: 2022-10-22 *******************************
................
18:46:46:WU00:FS01:Received Unit: id:00 state:DOWNLOAD error:NO_ERROR project:18706 run:0 clone:1664 gen:126 core:0x22 unit:0x000006800000007e0000491200000000
...............
18:46:48:WU00:FS01:0x22:Project: 18706 (Run 0, Clone 1664, Gen 126)
...............
18:27:01:WU00:FS01:0x22:Completed 2375000 out of 2500000 steps (95%)
18:27:03:WU00:FS01:0x22:Checkpoint completed at step 2375000
******************************* Date: 2022-10-24 *******************************
18:46:26:WARNING:WU00:FS01:Past final deadline 2022-10-24T18:46:25Z, dumping
18:46:26:WU00:FS01:Shutting core down
18:46:26:WU00:FS01:0x22:WARNING:Console control signal 1 on PID 1776
18:46:26:WU00:FS01:0x22:Exiting, please wait. . .
18:46:26:WU00:FS01:0x22:Folding@home Core Shutdown: INTERRUPTED
18:47:26:WU00:FS01:Cleaning up

Best
Stratos

[Gary480six]

I reported this same issue back in July. You would think that the researcher would notice all the delayed and reissued work units is actually Slowing down their project. And adjust the deadline. But I guess not.

At the very least, they could restrict the project to fewer classes of video cards in the GPU.txt file.

Stop sending it out to 2:4 class cards (GTX 750Ti) and 1:5 cards (Radeon HD 7700) as those cards do not meet the deadlines - even running 24/7. And restrict this project to 2:7 (RTX 2060 Super) and 2:8 (RTX 3070) or 1:6 (Radeon RX 6700) cards.

[muziqaz]

Your GPU has been changed to different species, which means it was put to the different performance tier bracket. While you will not be receiving these type of projects anymore, please keep in mind your GPU will be getting much less work, since at the moment there are not too many small projects which would fit your GPU. While there might be some small atom count projects still in the wild which would fit your GPU, project owners have not been made aware of the changes. Any new project which is coming out of the gate from now, if suited, will be assigned to your type of cards.
For past month majority, if not all new projects are huge and are too big for your GPU.
On the side note, your GPU will also have a bit more correct name in the log file from now on. You can delete GPU.txt file in order to receive a new version.

This is in regards to AMD card you have. (In few days the whole range of AMD cards will be reassigned to different performance tiers. Up till now AMD only had Species 5 and 6, now it will have from 1 to 7 and with next next gen 8+.
nVidia cards might get performance tier revision too, if someone from FAH gets onto doing it

[stratos412]

Thank you guys.
So, if I am getting it right, there is not much I can do from my side. Let's hope I shall been receiving less demanding work units.
However, it is not a bad idea to expand the deadline in some short of automatic way, if work unit still running, but not meet the time to finish the project.
It is a kind of PC power waste, if you are running a project 24/7 and FaH dump it when deadline past. Better give a couple more days that abandon the whole work unit

[muziqaz]

Increasing deadlines will slow down the science more than having some slower GPUs miss the deadlines
Believe it or not, main power of FAH comes from high end GPUs, mainly nVidia. So if we expand the deadlines, and someone with slow GPU gets that WU instead of someone with much faster GPU. Each work unit series generates next set of work units, so if one is stuck with slow GPU, the whole project suffers. Since we accept all kinds of help, we try to differentiate and balance the assignments as much as possible that everyone could contribute as much as possible without slowing down science too much
Eventually your GPU will pick up new work, just at the moment everyone keeps releasing large projects
However I remember there might have been few small projects which might have had your GPU with current performance rating included.

To add to that, older GPUs are not widely available to buy to test for more precise assignments

[BobWilliams757]

Your GPU has been changed to different species, which means it was put to the different performance tier bracket. While you will not be receiving these type of projects anymore, please keep in mind your GPU will be getting much less work, since at the moment there are not too many small projects which would fit your GPU. While there might be some small atom count projects still in the wild which would fit your GPU, project owners have not been made aware of the changes. Any new project which is coming out of the gate from now, if suited, will be assigned to your type of cards.
For past month majority, if not all new projects are huge and are too big for your GPU.
On the side note, your GPU will also have a bit more correct name in the log file from now on. You can delete GPU.txt file in order to receive a new version.

This is in regards to AMD card you have. (In few days the whole range of AMD cards will be reassigned to different performance tiers. Up till now AMD only had Species 5 and 6, now it will have from 1 to 7 and with next next gen 8+.
nVidia cards might get performance tier revision too, if someone from FAH gets onto doing it

I think having a better performance breakdown of the AMD GPU's will be a really good thing. Having folded with a slow iGPU for a while, it often got assignments that were too much for onboard graphics. I think a lot of time was wasted with users folding in hopes to meet the timeout or at least beat the deadline, and often they were reassigned and finished before then. That being said, some of the older and/or slower stuff still has a place when assigned correctly, so I'm glad F@H is taking this approach to keeping more hardware available for use.

Doing these things now makes sense, as if another (world goes sideways) comes along there will already be more potential for resources to be in place.

As for such assignment designations/tier revisions, who actually does such things? Is it actual researchers (or their staff) or volunteers? Just a question, as I would gladly dedicate some time here and there if it helped make F@H more capable and possibly allow a wider base of people folding.

To add to that, older GPUs are not widely available to buy to test for more precise assignments

Understanding that we can't expect to find a bunch of various hardware to get exacts, I was thinking about this while frustrated one day after getting yet another assignment to my iGPU with no hopes of completing it. Couldn't F@H use the data at LAR systems (along with possible user input when needed) to help break down GPU species? Though the data obviously isn't perfect it's raw data that at a shows averages, though probably at a sometimes slightly low number due to non 24/7 or people running power limited and such. Here and there some kinks might be found, but overall I think looking towards the higher side numbers it could be a good resource to use.

[Gary480six]

Your GPU has been changed to different species, which means it was put to the different performance tier bracket. While you will not be receiving these type of projects anymore, please keep in mind your GPU will be getting much less work, since at the moment there are not too many small projects which would fit your GPU. While there might be some small atom count projects still in the wild which would fit your GPU, project owners have not been made aware of the changes. Any new project which is coming out of the gate from now, if suited, will be assigned to your type of cards.
For past month majority, if not all new projects are huge and are too big for your GPU.
On the side note, your GPU will also have a bit more correct name in the log file from now on. You can delete GPU.txt file in order to receive a new version.

This is in regards to AMD card you have. (In few days the whole range of AMD cards will be reassigned to different performance tiers. Up till now AMD only had Species 5 and 6, now it will have from 1 to 7 and with next next gen 8+.
nVidia cards might get performance tier revision too, if someone from FAH gets onto doing it

When I posted my original comment back in July, I had not thought about restructuring the GPU.txt species bracket to 'solve' the problem. But it works for me! I would rather have an older Folding video card sit idle... than working, missing deadlines and slowing down the research. Right now, this does not affect the Nvidia cards - but hopefully those performance tiers get revised soon too.
Meanwhile, Thanks to whoever was working on the AMD tiers!

[muziqaz]

LAR systems is very unreliable, as different users run their GPUs at different speeds, and conditions. Then there is this thing AMD use with their GPus, is using one PCI ID for several different gen GPUs, thus LAR reports those under same name, so results are all over the place. For a moment we (well, I) have 6900xt, Radeon 7 and 5700xt which are doing AMD side of testing, there was Vega64, but it died in this glorious death and was honored in massive multinational parade a month or so ago Soon that list will have 7900xt and probably will byte the bullet with some older GPU too.

As for who is doing these things, it's volunteers. In regards to rebalancing performance brackets - people who enjoy point and click very much

regarding nVidia, it has 7 or 8 species, so while all of them are not perfect, they still have broader reach when enabling disabling certain performance brackets from assignments

[BobWilliams757]

Muziqaz,

I folded for a couple years with an AMD iGPU, and no doubt the species matter needed a vast overhaul, so I'm glad efforts have been made. My Vega 11 iGPU was trying to keep up with the species designation of much more powerful GPU's due to the device ID issue.

Just my personal opinion here, but I think folding could get quite a bit of info from actual users if it was an option for them. The ultimate would be some type of in depth F@H bench work units or something, but just finding a few common work units could help quite a bit to break down what GPU capabilities are. Proper species designations and assignment help everyone really, and if folding efficiency goes up the science is done quicker. It's a no lose situation really.

Likewise I think that there are people who will help if they know how to contact the people seeking help. With Beta all but invisible these days, those paths to being part of the solution seem cryptic to many average or casual folders. If those paths in remain hidden, people will quit looking for them.

[muziqaz]

The way we set points is we run each new project through different GPUs, which we have. We have a lot of various tier nVidia GPUs, however AMD only is represented by me, and with UKs energy crisis, I'm starting to scale down.
Bench WUs will not help, because every new project is very different. We know how GPU perform in general, we don't know how they perform per project basis
Give it some time, once we have some small projects you will see things improve

[BobWilliams757]

I ended up biting the bullet and got an Nvidia card about a month or so ago. But at least AMD will get some more attention now. I know there is still a lot of AMD stuff out there that can fold given the chance, even those little iGPU's are not bad with smaller atom count work units.

I thought the points were more or less set unless people in Beta brought up PPD issues beyond normal.

[muziqaz]

Points are set before beta, and very rarely are adjusted during beta or after that

[BobWilliams757]

With the species changes I may have to see if my iGPU will coexist properly with my GPU on my motherboard. Even if it just picks up work units now and then it will be a little boost.

[muziqaz]

At the given moment there is no projects which are suitable for anything bellow species 4. For reference AMD RX 460 is lowest species 4 GPU.
For reference:
Species0 – Not Supported/Disabled
Species1 – TeraScale (Still supported)
Species2 – up to 0.999TFLOPS
Species3 – 1-1.999TFLOPS
Species4 – 2-4.999TFLOPS
Species5 – 5-9.999TFLOPS
Species6 – 10-23.999TFLOPS
Species7 – 24TFLOPS+ (yes, AMD has such GPU )

[BobWilliams757]

Good scoop Muziqaz, and thanks.

I will probably see if my iGPU wants to configure and play properly, then check now and then to see if there are any work units available for it. This now has me wondering if any of the newer iGPU's will make it to species 4, and I would guess yes, but will have to dig some when I have time.

Long term this should help F@H and speed up the science, especially if we ever go through another COVID type situation where projects of all different sizes and shapes were becoming available. If nothing else having some of the slower and narrow GPU's around keeps the people with the power GPU's from raging when the small atom count projects eat into their PPD ability.


As for the Species 7 AMD GPU, I think someone is trying to squeeze in a half a TFLOP or something!