Project 10496 (158,16,66)

[B.A.T]

Nope. I'm still getting this.
Had over 100 WU from this project the last 4 weeks.

[JonasTheMovie]

Yeah.... 640ppd instead of the 800ppd I got lately...

But hey, whatever really, as long as everyone put in the effort....

[ComputerGenie]

Dear Lord,

Please let John Chodera and the Chodera lab find whatever answers they are looking for, as it relates to MTOR kinase + FAT domain mutants. And please let it happen soon.

Sincerely yours,
Every folder that has lost 200k PPD, compared to their normal average, working on this project

[bruce]

Dear Lord,

Can you explain to me why ComputerGenie is so secretive? He doesn't tell us what his hardware he has or how it's configured. He should be aware that we can't really be helpful when he has neither filled out his hardware profile or posted it in this topic. Please let him know that we'd like to help both him and John Chodera to do better science research and to have a better understanding of Your World.

Sincerely yours,
Bruce.

[ComputerGenie]

Dear Lord,

Can you explain to Bruce that no one is being "secretive" about the hardware they have or how it's configured. He should be aware that every folder involved in Project 10496 has endured massive PPD losses from the majority of the Rs, Cs, and Gs involved in Project 10496, and that those losses are irrespective of hardware or configuration. Please let him know that we like helping both F@H and John Chodera to do better science research and to have a better understanding of Your World, but we wish that certain PRCGs didn't have such a dynamic impact on our bottom line when we get paid for each PPD that we produce (and that many of us depend on those payments to pay the bills in order to continue doing so).

Sincerely yours,
ComputerGenie

[bruce]

Recently I have learned something about another specific project (not 10496) that runs fine on some GPUs and poorly on others. Since you still have not provided data on your GPUs I will not be able to determine if your problem with 10496 is a similar issue.

On the previous page, others have mentioned a GTX1070 and a GTX1080. Several have been cooperative enough to fill out their hardware configuration in the User Control Panel. The alternative would be to post the top segment of FAH's log file.

I apologize, but my crystal ball isn't working very well tonight. ... and under the circumstances, it seems that the word "secretive" is appropriate.

BTW, we did solve the problem on the other project and the same fix might work here.

[B.A.T]

The project runs fine but the GTX 1080 get around 650K PPD compared to other projects where the 1080 does 800K PPD.

[bruce]

Thanks, B.A.T, for being cooperative. That's useful information.

One possible way of looking at p10496 is that maybe the GTX1080 is too fast for it. The GPU's compute engine is probably outrunning the ability of the CPU/PCIe bus to supply data to it. So my questions to everyone are (A) What GPU are you running and (B) What is the functional clock rate of the PCIe bus, and (C) have you allocated a full CPU thread to running FAHCore_21 on this project? (D) Are you running Window or Linux?

I'm particularly interested in reports from people who have lower performance GPUs -- like a mid-range GPU like say a GTX750Ti that runs at 1306 GFLOPS instead of 8873 GFLOPS. Does the lower performance GPU achieve its expected throughput?

[B.A.T]

A:
1: AMD FX 8150, 8 GB 1333MHZ, 256 GB Samsung evo 850, Radeon 280X
2: Intel i3 4170, 8GB ram, 120GB SSDNow V300, 2X KFA2 gtx 1080
3: AMD Phenom II x6 1100T, 8GB 1600MHZ, Radeon 280X
4: AMD FX 4300, 8GB 1600MHz, Crucial C300 128GB, 2X Radeon 290X
B:
1: PCI 2.0 16X
2: PCI 3.0 8X to both cards
3: PCI 2.0 16X
4: PCI 2.0 16X to both cards
C:
1: CPU at 13% (4 cores, 8 threads)
2: CPU at 53% (2 cores, 4 threads)
3: CPU at 15% (6 cores)
4: CPU at 17% (2 cores, 4 threads)
D:
1: W10 64bit
2: W10 64bit
3: W7 64bit
4: W7 64bit

The GTX 1080 get 600-650K PPD
The 290X get 280-310K PPD
The 280X get 135-150K PPD
The PPD is as expected for all the radeon cards, only my 1080 get lower.

My 1080 runs at 2012-2038 MHz depending on project and temp. and they get 750-900K PPD on other projects.

[bruce]

CPU at 13% (4 cores, 8 threads)

If FAJCpre_21 uses one thread, Windows reports that as 13%. Nevertheless, I have to ask:
How much of the CPU is reported to be idle?

Do you mean that the CPU is ~87% idle or is the other 87% in use by other tasks?

[B.A.T]

87% of the cpu is idle.

[Ricky]

Bruce,

I find this project PPD is normal with Maxwell, but low with Pascal GTX1080.

Code: Select all

 Project ID: 10496

Titan XP, Win 8.1, PCI 3.0 x16 
 Number of Frames Observed: 300
 Min. Time / Frame : 00:01:26 - 1,062,448.2 PPD
 Avg. Time / Frame : 00:01:36 - 900,842.2 PPD <= This is actually higher than 800k average with all projects. Peak PPD is 1.36M for the best project. 

GTX980SC, Win 8.1, PCI 3.0 x16
 Number of Frames Observed: 300
 Min. Time / Frame : 00:02:40 - 418,673.5 PPD
 Avg. Time / Frame : 00:02:50 - 382,280.5 PPD <= PPD is almost exactly perceived average.

GTX1080FTW, Win 8.1, PCI 3.0 x16 (slot 1)
 Number of Frames Observed: 300
 Min. Time / Frame : 00:01:56 - 678,216.5 PPD
 Avg. Time / Frame : 00:02:02 - 628,804.9 PPD <= PPD is below 800K perceived average.

GTX980tiFTW, Win 8.1, PCI 3.0 x8
 Number of Frames Observed: 300
 Min. Time / Frame : 00:01:56 - 678,216.5 PPD
 Avg. Time / Frame : 00:02:02 - 628,804.9 PPD <= PPD is just a bit below 650K perceived average.

GTX1080FTW, Win 8.1, PCI 3.0 x16 (slot 2)
 Number of Frames Observed: 300
 Min. Time / Frame : 00:01:56 - 678,216.5 PPD
 Avg. Time / Frame : 00:01:59 - 652,732.2 PPD <= PPD is below 800K perceived average.

[ComputerGenie]

...So my questions to everyone are (A) What GPU are you running and (B) What is the functional clock rate of the PCIe bus, and (C) have you allocated a full CPU thread to running FAHCore_21 on this project? (D) Are you running Window or Linux?...

Rig 1:
A 2x GeForce GTX 1080
B PCI 2.0 16X -- 5.0 GT/s
C Yes, nothing runs on this PC except F@H and no CPU folding
D Linux 3.16.0-4-amd64

Rig 2:
A 1x GeForce GTX 1080
B PCI 3.0 16X -- 8.0 GT/s
C Yes, and no CPU folding on this PC either
D Windows 7 Ultimate

Both rigs experience the same results; some RCGs in this project will run in the normal 800-850k PPD range, while most RCGs run in the 600-650k PPD range.
Can you guess which days I hit the latter?
Day--Points
05.12.17--2,495,128
05.11.17--2,156,558
05.10.17--2,136,510
05.09.17--2,447,404
05.08.17--2,308,872
05.07.17--2,587,470
05.06.17--2,536,278
05.05.17--2,147,935

Edit:As a more specific example...
RCG 71,71,19 (TPF 2:14) --- Est PPD (@ 30% completed) = 550006
RCG 27,39,23 (TPF 1:47) --- Est PPD (@ 37% completed) = 765490
Both WUs currently running on Rig 1, both slots are identical on board and have identical cards with identical configurations

[Nathan_P]

As already posted earlier, GTX 1080 & GTX 1070, both on PCIe 3.0 x16, There is only F@H running so the 2 gpu's have access to all 8 threads, Linux - Specifically Zorin 9

[bruce]

The OpenCL driver for Linux is significantly different than the Windows OpenCL driver. Each GPU uses at most one CPU thread. Many projects do significantly better if you have a faster CPU (Central Processing Unit). The GPU can't get data fast enough to stay busy but it does depend on Windows/Linux, on the particular Project, and also on the speed of both the CPU and the GPU.

FAH Development is aware of these issues and is exploring options to improve the situation.