7640 & 7643 PPD

[HaloJones]

One of the few things I hate about FAH is when a new type of project delivers massively less ppd for the same energy output.

[P5-133XL]

One of the few things I hate about FAH is when a new type of project delivers massively less ppd for the same energy output.

While I understand your issue, I also understand the problem.

The fact is that GPU's capabilities have a very big range and are either under-utilized or over-subscribed. They always start out with small WU's and they get bigger with time. So, folding starts out drastically under utilizing almost all the GPU's and thereby virtually all GPU's under-perform. As the GPU's get bigger, all the PPD's increase till the lower-end ones start getting over-subscribed and thus their PPD's start dropping. Over time, more and more GPU's are over subscribed and for the bottom end ones they are drasticly overworked. To help compensate, there are always new GPU's that are faster and are under utilized again. So we get a moving curve and PPD's for older cards keep dropping and new top-end cards get getting higher PPD's.

The 460 GTX is really a very low end card, so I'm not at all surprised at the PPD drop over time. The same thing has happened to SMP. I used to be able to get 10K PPD on over-clocked AMD x2 3800's but now I'm lucky to get 4-6K PPD on q6600 which are much faster processors. But there are quad socket processors that can easily get 200K PPD's. I agree that it isn't particularly fair but it is what it is.

One can get annoyed, and vocalize your displeasure but in the ending analysis It's just the way it is. One has to find some other motivation other than pure points because if that's the only reason you fold, you'll eventually get disgusted and quit like so many others before you.

P.S. This thread isn't really the place for this discussion. May I suggest that if you wish to continue try: Suggested change to the PPD system

[Zagen30]

The 460 GTX is really a very low end card

Not to get this too off-topic, but the 460 is not "very low-end." A GT 520 or HD 6450 is very low-end; the 460 is solidly mid-range, maybe sliding to lower end of the mid-range spectrum now that it's over a generation old. The Pande Group seems to have deliberately picked mid-range hardware for the last few benchmark purchases they've made (e.g. an i5 for CPU WUs).

[bruce]

The first Fermi that came out was GTX 470 and 480 which eventually became the top of the 4xx line as lower and lower priced 4xx's appeared, including the 420 / 430 / 440 / 450 and eventually the 460SE. Today that puts the 460 near the middle of the 4xx series -- in fact, not all that low in the 5xx series.

[sirmonkey]

One of the few things I hate about FAH is when a new type of project delivers massively less ppd for the same energy output.

While I understand your issue, I also understand the problem.

The fact is that GPU's capabilities have a very big range and are either under-utilized or over-subscribed. They always start out with small WU's and they get bigger with time. So, folding starts out drastically under utilizing almost all the GPU's and thereby virtually all GPU's under-perform. As the GPU's get bigger, all the PPD's increase till the lower-end ones start getting over-subscribed and thus their PPD's start dropping. Over time, more and more GPU's are over subscribed and for the bottom end ones they are drasticly overworked. To help compensate, there are always new GPU's that are faster and are under utilized again. So we get a moving curve and PPD's for older cards keep dropping and new top-end cards get getting higher PPD's.

The 460 GTX is really a very low end card, so I'm not at all surprised at the PPD drop over time. The same thing has happened to SMP. I used to be able to get 10K PPD on over-clocked AMD x2 3800's but now I'm lucky to get 4-6K PPD on q6600 which are much faster processors. But there are quad socket processors that can easily get 200K PPD's. I agree that it isn't particularly fair but it is what it is.

One can get annoyed, and vocalize your displeasure but in the ending analysis It's just the way it is. One has to find some other motivation other than pure points because if that's the only reason you fold, you'll eventually get disgusted and quit like so many others before you.

P.S. This thread isn't really the place for this discussion. May I suggest that if you wish to continue try: Suggested change to the PPD system

then maybe its time PG finally comes up with a way to organize the WU distribution so that it can send the appropriate WU's to the appropriate GPU's. it's not that difficult for the program to detect the GPU, send that data back to the assignment server so it knows what WU's to send for that gpu. it removes the whole under utilization and over saturation of GPU's that are still being used for F@H. it keeps everyone happy and all the WU's get done instead of the assignment server mass spamming 1 type of WU for weeks on end because not enough of them got done. other then that honestly i think the PPD systems fine for GPU's, its the only system that hasn't been destroyed by the broken bonus point system that plagues the cpu client.

[P5-133XL]

Again, this is just stealing the thread and there are better threads to put this into. The problem with right-sizing the WU's is that over time the projects get bigger, and the old projects disappear but the cards don't. You can't right-size the projects to the cards when the only projects that remain over-subscribe the older GPU's.

I personally believe there is no true solution to the point system being unfair other than to have a different motivation to folding such as folding for science. That way the points are not so important. I no longer believe that it is possible to construct a perfect point system that people won't object to. Too many conflicting goals; too many conflicting opinions, too much self interest and greed. You can't produce a competition where no one loses.

Moderator could we move this conversation out of this thread and into a PPD discussion thread?

[bruce]

then maybe its time PG finally comes up with a way to organize the WU distribution so that it can send the appropriate WU's to the appropriate GPU's. it's not that difficult for the program to detect the GPU, send that data back to the assignment server so it knows what WU's to send for that gpu. it removes the whole under utilization and over saturation of GPU's that are still being used for F@H. it keeps everyone happy and all the WU's get done instead of the assignment server mass spamming 1 type of WU for weeks on end because not enough of them got done.

Better WU assignments is certainly a reasonable request, but I don't think you have any basis for what you're saying.

Maybe it's easy and maybe it's not. I don't know enough about what either V6 or V7 currently sends to the servers, and if it has to be changed, that is a big deal because not only does the client software have to change, but the server code probably has to change on all the servers so they don't choke on the unexpected data. (and during the transition, everything has to keep working.) That would most definitely make it a big deal -- but I don't know enough about that to really comment -- other than it MIGHT be a big deal.

And which GPUs are "appropriate" for these WUs? I don't see a lot of people who are demanding access to these projects? How do you know it's even possible to remove under utilization or under saturation that you're complaining about. Is there some better way to change the project internally so that it works better for you? If so, how?

I'm interested in anything that can be done to keep everyone happy but in my experience, everybody is always griping about something, and when Stanford changes ANYTHING, somebody benefits (silently) while somebody else starts griping (loudly).

other then that honestly i think the PPD systems fine for GPU's, its the only system that hasn't been destroyed by the broken bonus point system that plagues the cpu client.

The QRB for CPUs has most definitely improved the scientific throughput by completing WUs more rapidly, partly by allowing a reduced number of active WUs needed. If it were ever applied to GPUs I'm not sure how much that would actually change the GPU projects but if it makes for better science, it might happen someday.

Yes, I'll move it.

[HaloJones]

My two 8800GT's have never varied their PPD. GROGPU2 is consistent and keeps them running at around 5500ppd each. OPENMMGPU has been consistent for ages and now suddenly drops 30% and my 450GTS is barely producing more than the 8800GT. That can't be right and if I knew how to make them avoid these 76xx WU I would.

[P5-133XL]

Change your client-type. I did and I'm getting p803x's now using advanced.

[artoar_11]

Today I got the first WU - p7642. OK, reduction of points for me is not fatal. I want to ask another.

bruce, if you can explain, why the load on the GTX 460 dropped? With p803x temperature was ~ 63*C, now ~ 58*C.
These WUs have more atoms. PCI-E2 x16 bus limit?

[bruce]

At this point, I can only guess, but I've decided to see if we can find the pattern. I think you're on the right track, though.

viewtopic.php?f=74&t=21223

[apeman556]

Hello please help. since the new 7640 wu came out, my gpu has only completed about 2 unit, but failed over 8 times saying Nans detected on gpu unstable machine.
My gpu is a gtx 560 ti, I have 8 GB ram and an overclocked 2500k, (overclocked to 4.3, which also folds reliably) i've been folding for many years and this paticular rig has been getting over 35000PPd for the last year, and now it won't even finsh a GPU work unit. Its not oveclocked by me, but comes overclocked to 900MHz fromthe factory (Gigabyte). Ant ideas?

[derrickmcc]

Hello please help. since the new 7640 wu came out, my gpu has only completed about 2 unit, but failed over 8 times saying Nans detected on gpu unstable machine.
My gpu is a gtx 560 ti, I have 8 GB ram and an overclocked 2500k, (overclocked to 4.3, which also folds reliably) i've been folding for many years and this paticular rig has been getting over 35000PPd for the last year, and now it won't even finsh a GPU work unit. Its not oveclocked by me, but comes overclocked to 900MHz fromthe factory (Gigabyte). Ant ideas?

The factory overclock may be ok for gaming, but folding really stresses a GPU. Try reducing the overclock on your GPU.

I have 2 x GTX460 which where over clocked to 825/1600/1650, but just recently a got an error folding p8035 on one of them, reduced the overclock on that GPU to 800/1600/1600 and now all is fine (for the moment)

[bruce]

Try reducing the overclock on your GPU.

Right.

NVidia suggests that the gtx 560 ti should run at 822 Core 1645 Shader 4008 Memory I recommend you try that. If it does work, then you can try complaining to Gigabyte who sold you a GPU that was overclocked to 900MHz but won't fold at that rate.

You may also try increasing the fan speed. Sometimes that works when nothing else does.

[apeman556]

ok thanks for response and suggestions.
I did try reducing overclock to 850 last night but again failed at 60%, I'll ry today at the 822 Core, hopefully this will cure the problem.