Some Fermi projects not yet re-benchmarked

[Mactin]

My biggest irritant with fah : (big) PPD variability.
Always had it, probably always will.
I have to live with it.

[7im]

The benchmark computer always gets the same PPD. Buy one like that if you can't understand why the PPD on different configs produce different PPD.

[artoar_11]

7im, now I remember why PPD in p8018 is so low.

Postby tjlane » 26 Oct 2012, 00:32

Folding Forum,

Please be aware that we are rolling out a new version of GPU Core15, v2.25. This core includes support for the new generation of nVidia boards, the Kepler series.

While this core has been extensively beta-tested and not shown issues, we are doing a controlled release to the general FAH community. Over the next few days, the p762x series will run on advmethods and require core15 v2.25. Your client should upgrade automatically to the new core.

Please let us know by posting here what your experience with the new core is -- if all goes well, we will release to the entire FAH community shortly, so we can support Kepler boards.

Thanks!

TJ

Postby 7im » 26 Oct 2012, 02:56

k1wi wrote:
I haven't noticed any increase in performance from v2.25 on my 500 series GPU, are you referring to Kepler 7im?


Yes.

tjlane wrote:
... This core includes support for the new generation of nVidia boards, the Kepler series.

Also concerned that v2.25 caused slightly longer frame times on a few projects using Fermi boards.

viewtopic.php?f=74&t=22793

On the old core_2.22 PPD on p8018 was ~ 15,000 points (my GTX 460 @ 775 MHz faster than the benchmark machine - GTX 460 @ 720MHz).

[Mactin]

The benchmark computer always gets the same PPD. Buy one like that if you can't understand why the PPD on different configs produce different PPD.

I perfectly understand the benchmarking process, I simply dont have any faith in it. For me, it's part of folding.
There are sometimes quite large variations within the same project, so the benchmark itself will see variation.

The only two other choises are to complain here or stop folding.

[7im]

And those variations within one project are rare and well explained. What's to complain about?

[artoar_11]

The only two other choises are to complain here or stop folding.

Not a good choice to give up just due to poor PPD. PPD gives us the satisfaction for the work done, but these are figures only (statistics)

[PantherX]

...I perfectly understand the benchmarking process, I simply dont have any faith in it. For me, it's part of folding.
There are sometimes quite large variations within the same project, so the benchmark itself will see variation.

The only two other choises are to complain here or stop folding.

Since there is a huge variation of GPUs that are supported, the PPD will fluctuate rather significantly depending on the GPU model. You can see how much the variation occurs with FahCore_17 by reading this blog post (http://folding.stanford.edu/home/a-peek ... marking-2/).

I don't really think that any resources will be put into FahCore_15 since FahCore_17 is the future of GPU folding. Moreover, if there is enough data provided by donors to indicate that a particular project is out of line PPD-wise, action might be taken to rectify that. Just stating that PPD is too low/high without any supporting data makes it impossible to take any action, if it needs to be taken.

[DutchForce]

I think that a good way to see if a Project (released before Dec. 2012) was re-benchmarked is to compare the current Project Stats (Points, Timeout and Deadline) with the Project Stats when it was first released, which usually can be found on the Beta forum.
*Project 8018* Stats when it was first released was - Point: 5757 ; Timeout: 13.77 ; Deadline: 19.65
This is the same as the current Stats, which is a good reason to presume that this Project is not yet re-benchmarked, because other re-benchmarked Projects not only had their Points increased, but also the Timeout and Deadline changed.

For example:
*Projects 7624-7625-7626* changed from:
Points : 5187 to 14093
Timeout : 12.6 to 37.94
Deadline: 17.9 to 48.78

I hope this helps...

*I can not post links to the Projects because this is my first post, but you can find the threads on the Beta forum*

sorry for spelling or grammar errors, english is not my native language

[7im]

*Project 8018* Stats when it was first released was - Point: 5757 ; Timeout: 13.77 ; Deadline: 19.65
This is the same as the current Stats, which is a good reason to presume that this Project is not yet re-benchmarked, because other re-benchmarked Projects not only had their Points increased, but also the Timeout and Deadline changed.

Not a good reason to assume that. You didn't account for WU sizes, like P7620 and P7621, which more closely match P8018 in size. You also didn't account for the fact that the original WUs had points at the ~2700 level, while projects in this series released several months later were at the ~5700 points level (the same as the corrected values).

One could assume the correction was made between when P8010 first came out (May 2012, ~2700 points) and when P8018 came out (July 2012, ~5700 points).

Let's not assume anything. I contacted the research to investigate.

[HayesK]

My HFM history has 41 p8018 completed in 2013, all using core_15 v2.25 and having ppd about 20% below typical for those GPU. Completed several hundred p8018 in 2012, all using using core_15 v2.22 and having ppd about middle of the range for those GPU. The HFM benchmark data below shows the same gpus on both core_15 versions, with frame times for v2.25 about 20% longer than v2.22.

p8018 on core_15 v2.25

Code: Select all

 Project ID: 8018
 Core: OPENMMGPU
 Credit: 5757
 Frames: 100

 Name: Hayes-P5Kpro-Q9450-3.6+2xGTX550Ti-XP32-V729 Slot 01
 Number of Frames Observed: 300
 Avg. Time / Frame : 00:09:50 - 8,431 PPD

 Name: Hayes-P5Kpro-Q9450-3.6+2xGTX550Ti-XP32-V729 Slot 02
 Number of Frames Observed: 300
 Avg. Time / Frame : 00:09:51 - 8,416 PPD

 Name: Pam-P5Kpro-Q9450-3.4+2xGTS450-XP32-V729 Slot 00
 Number of Frames Observed: 300
 Avg. Time / Frame : 00:10:48 - 7,676 PPD

 Name: Pam-P5Kpro-Q9450-3.4+2xGTS450-XP32-V729 Slot 02
 Number of Frames Observed: 300
 Avg. Time / Frame : 00:10:46 - 7,700 PPD

p8018 on core_15 v2.22

Code: Select all

 Project ID: 8018
 Core: OPENMMGPU
 Credit: 5757
 Frames: 100

 Name: Hayes-P5Kpro-Q9450-X32--V7 Slot 00
 Number of Frames Observed: 300
 Avg. Time / Frame : 00:07:24 - 11,203 PPD

 Name: Hayes-P5Kpro-Q9450-X32--V7 Slot 02
 Number of Frames Observed: 300
 Avg. Time / Frame : 00:07:25 - 11,178 PPD

 Name: Pam-P5Kpro-Q9450-3.4+2xGTS450-XP32-V729 Slot 00
 Number of Frames Observed: 300
 Avg. Time / Frame : 00:08:10 - 10,151 PPD

 Name: Pam-P5Kpro-Q9450-3.4+2xGTS450-XP32-V729 Slot 02
 Number of Frames Observed: 300
 Avg. Time / Frame : 00:07:55 - 10,472 PPD

[artoar_11]

Today, I notice that there is a change of base points of p7620 and p7621. I guess it's not a mistake My GTX 460 works only on FahCore_15 projects.

7620 171.64.65.105 p7620 1980 37.94 48.78 14093.00 100 OPENMMGPU Description jadeshi 0.00
7621 171.64.65.105 p7621 1980 37.94 48.78 14093.00 100 OPENMMGPU Description jadeshi 0.00

[Calibrator]

Something is amiss with projects 7620 and 7621. I seem to be stuck folding them and my PPD have severely dropped off. I used to fold at around 400,000 PPD but now with projects 7620 and 7621 I'm folding at around 80,000 PPD! I'm using Folding at Home version 7.4.4 with two GTX 780 Ti's. I have changed nothing on my end. I even tried deleting the work folder and config files to get fresh work units. The server keeps giving me projects 7620 and 7621. I've been stuck with them for about a week now! What gives?
http://folding.extremeoverclocking.com/ ... =&u=351730

http://folding.extremeoverclocking.com/ ... or&t=11108

[7im]

If they need more 7620s folded, that's they send and that's what we fold. This older project doesn't have a QRB. We're all in the same boat points-wise.

[Calibrator]

But my fellow "competitors" on my team have not seen a drop off and don't appear to be folding 7620 and 7621. Why am I only folding 7620 and 7621? Is anyone else? It is my understanding that 7620 and 7621 are 3 year old projects... If they need to be completed, I'm all in for that but I would just like to know if I'm not alone on this!

[P5-133XL]

All I can say is that getting specific projects seem to come in streaks. You are in the same boat as everyone else with similar HW and client-type. If necessary, you can change the set of projects you pull from by changing the client-type to something different like from general release (no client-type defined) to advanced.

When you change the client-type you will also change the risk of failure for a specific project. All projects go through a process of testing and changing the client-type changes where you are in the testing curve. However, as a side benefit, each client-type has a different set of projects that you can get. So if you don't like these project, changing the client-type will likely change the project you are getting.