Page 6 of 13

Re: FAHBench (OpenMM 5.1)

Posted: Sun Jul 28, 2013 6:15 pm
by foldy
FAHbench on my HD 7950 @ 1000Mhz
Driver 13.4

Explicit 30.1
Implicit error

Re: FAHBench results (Most Powerful GPU)

Posted: Tue Jul 30, 2013 3:24 am
by edo75
[I7-3820 3.7Ghz 16GB DDR3 @ 1600Mhz + Win7Pro 64bit FAHBench 1.2.0 w/OpenMM 5.1]
[ASUS GTX TITAN 314.22 - GPU @990 MHz during tests]

OpenCL SinglePrecision
Explicit:
46.5601 ns/day
Implicit:
185.798 ns/day

OpenCL DoublePrecision
Explicit:
17.7465 ns/day
Implicit:
27.9688 ns/day

Re: FAHBench (OpenMM 5.1)

Posted: Sun Aug 25, 2013 10:59 am
by casuta
i5-2500k @ 4 GHz
GTX 560Ti @ 823 Mhz

OpenCL SinglePrecision
Explicit:
16.1452 ns/day
Implicit:
72.0165 ns/day

Image

OpenCL DoublePrecision
Explicit:
4.54758 ns/day
Implicit:
6.96494 ns/day

Image

Re: FAHBench (OpenMM 5.1)

Posted: Mon Aug 26, 2013 3:31 am
by Spongebob25
EVGA 780 GTX Classified@1201 MHz
Folding@Home Benchmark v.1.2.0 OpenCL-Single

Explicit: 49.678 ns/s
Implicit: 202.33 ns/s

OpenCL-Double

Explicit: 10.0926 ns/s
Implicit: 14.8229 ns/s

Re: FAHBench (OpenMM 5.1)

Posted: Mon Aug 26, 2013 5:19 am
by Jesse_V
FAHBench 1.2.0

Asus Nvidia GTX 480:
Single precision OpenCL explicit: 20.6915 ns/day
Single precision OpenCL implicit: 83.349 ns/day
Double precision OpenCL explicit: 7.37991 ns/day
Double precision OpenCL implicit: 11.1071 ns/day

Gigabyte GTX 560TI factory-overclocked edition (display driver, GPU clock 900 MHz, Memory clock 2004 MHz):
Single precision OpenCL explicit: 16.281 ns/day
Single precision OpenCL implicit: 72.4887 ns/day
Double precision OpenCL explicit: 4.55196 ns/day
Double precision OpenCL implicit: 6.9429 ns/day

Re: FAHBench (OpenMM 5.1)

Posted: Tue Aug 27, 2013 5:25 pm
by planetclown
Radeon HD 7950 @925 MHz
Intel i7-2600K @4.0 GHz
Win7 64-bit

Using AMD Catalyst 13.4 drivers:

Single Precision Explicit Solvent: 32.8755 ns/day
Single Precision Implicit Solvent: error
Double Precision Explicit Solvent: 7.37084 ns/day
Double Precision Implicit Solvent: 15.3652 ns/day

[updated on 04-OCT-2013]
Using AMD Catalyst 13.9 drivers:

Single Precision Explicit Solvent: 32.7459 ns/day
Single Precision Implicit Solvent: 102.478 ns/day
Double Precision Explicit Solvent: 7.60427 ns/day
Double Precision Implicit Solvent: 15.3795 ns/day

Re: FAHBench (OpenMM 5.1)

Posted: Tue Aug 27, 2013 11:24 pm
by folding_hoomer
Q9550@2975MHz, WinXP Pro, Forceware 320.49

MSI GTX670 Power Edition, Factory-OC´d to 1019/1176MHz (with boost), Mem 6008MHz, measured while folding (no values for gaming yet)
Manualy OC´d:
GPU: +67MHz, resulting in max. 1241MHz GPU
Mem: +100MHz, resuling in 6412MHz Mem
Fan: manualy set to 50%, Temps between 49°-51°C while calculating
TDP: 100%, no additional voltage for GPU, Mem or Aux

Additional Info: While Explicit Solvent was calculated using GPU at 1228MHz, Implicit Solvent got one extra boost step of +13MHz, resuling in 1241MHz (Single- and Double Precision)

OpenCL
Settings: Single Precision
Verify Accuracy

Explicit Solvent: 36.9868 ns/day
Implicit Solvent: 138.674 ns/day

Settings: Double Precision
Verify Accuracy

Explicit Solvent: 5.54031 ns/day
Implicit Solvent: 7.37772 ns/day

Re: FAHBench (OpenMM 5.1)

Posted: Tue Sep 03, 2013 7:48 pm
by proteneer
FAHBench open sourced on github: github.com/simtk/fahbench

Re: FAHBench (OpenMM 5.1)

Posted: Fri Sep 20, 2013 5:13 am
by n_w95482
I ran this on all of my currently-running PCs. It took a lot longer than I anticipated. I also have results for an i5-3470 and a Radeon HD 7470, but I left them at work. I'll add them tomorrow.
Edit: I apparently forgot to run the double precision bit on the i5, so I'll be posting that PC's results next week.


Main PC:
GTX 680 - 1163 MHz boost, 314.22 driver
  • OpenCL Explicit: 35.4297 ns/day
  • OpenCL Implicit: 135.925 ns/day
  • OpenCL Explicit Double Precision: 5.73037 ns/day
  • OpenCL Implicit Double Precision: 7.983 ns/day
i7-980X @ 4.24 GHz, HT on - Intel OpenCL 2013 runtime
  • OpenCL Explicit: 4.7256 ns/day
  • OpenCL Implicit: 5.07258 ns/day
  • OpenCL Explicit Double Precision: 3.36264 ns/day
  • OpenCL Implicit Double Precision: 3.88878 ns/day
i7-980X @ 4.24 GHz, HT on - AMD APP
  • OpenCL Explicit: 3.82007 ns/day
  • OpenCL Implicit: 5.52605 ns/day
  • OpenCL Explicit Double Precision: 2.14357 ns/day
  • OpenCL Implicit Double Precision: 3.28687 ns/day
---------------------------------------------------------------------------
Spare PC:
GTX 570 @ 797/1594/975 MHz core/shader/memory - 314.22 driver
  • OpenCL Explicit: 20.2203 ns/day
  • OpenCL Implicit: 91.8993 ns/day
  • OpenCL Explicit Double Precision: 7.7988 ns/day
  • OpenCL Implicit Double Precision: 11.6211 ns/day
  • [i](note - this one hit the card HARD, hehe)[/i]
Opteron 185 @ 2.8 GHz - AMD APP
  • OpenCL Explicit: 0.422991 ns/day
  • OpenCL Implicit: 0.628385 ns/day
  • OpenCL Explicit Double Precision: 0.243642 ns/day
  • OpenCL Implicit Double Precision: 0.423059 ns/day
---------------------------------------------------------------------------
HTPC:
Radeon HD 6550D @ 900 MHz - Catalyst 13.6
  • OpenCL Explicit: 3.60741 ns/day
  • OpenCL Implicit: 19.145 ns/day
  • OpenCL Explicit/Implicit DP: GPU not capable, fell back on CPU - see below results
A8-3870K @ 3.1 GHz - AMD APP (used Remote Desktop to prevent the GPU from assisting :))
  • OpenCL Explicit: 1.22301 ns/day
  • OpenCL Implicit: 1.61156 ns/day
  • OpenCL Explicit Double Precision: 0.615837 ns/day
  • OpenCL Implicit Double Precision: 1.04417 ns/day
---------------------------------------------------------------------------
Laptop:
C2D T7600 @ 2.33 GHz - AMD APP (can't run Intel OpenCL 2013 runtime)
  • OpenCL Explicit: 0.522528 ns/day
  • OpenCL Implicit: 0.559855 ns/day
  • OpenCL Explicit Double Precision: 0.26674 ns/day
  • OpenCL Implicit Double Precision: 0.347506 ns/day
---------------------------------------------------------------------------
Secondary work PC:
GTX 460 1 GB @ 763/1526/950 MHz core/shader/memory - 314.22 driver
  • OpenCL Explicit: 13.6397 ns/day
  • OpenCL Implicit: 61.402 ns/day
  • OpenCL Explicit Double Precision: 3.71182 ns/day
  • OpenCL Implicit Double Precision: 5.7906 ns/day
C2D E8400 @ 3.6 GHz - AMD APP (errors w/Intel OpenCL 2013 runtime)
  • OpenCL Explicit: 0.993747 ns/day
  • OpenCL Implicit: 1.16912 ns/day
  • OpenCL Explicit Double Precision: 0.498737 ns/day
  • OpenCL Implicit Double Precision: 0.775832 ns/day

Whew! :D

I tried three different nVidia drivers

Posted: Sat Sep 28, 2013 2:23 pm
by user123
Want to share something.
Tested versions 301.42, 320.49 and 327.23 (all are WHQL drivers) with Folding@Home benchmark 1.2.0.
Ran the tests twice and took screenshots (but without CPU-Z nor CPU-Z screenshots).

Image
301.42 (older version) appears to be faster than 320.49 (later version) and 327.23 (latest version). I take the results to mean folding on my GTS450 will be faster if I use 301.42.

Re: I tried three different nVidia drivers

Posted: Sat Sep 28, 2013 2:35 pm
by 7im
The differences are small. It's possible other factors could also come in to play, like WU size.

Re: FAHBench (OpenMM 5.1)

Posted: Mon Sep 30, 2013 11:46 am
by Napoleon
Yep, rather small differences, less than 3% drop at a glance. In any case, FahBench starts folding the exact same WU every time unless you do something about it manually - viewtopic.php?f=38&t=24225&start=75#p248778 - so variation between WUs shouldn't be the cause. The slight differences between passes even with the same driver (less than 1%) are most likely caused by CPU background tasks. You need the CPU to feed the GPU, and there's always some background "noise" from the OS, AV utility, younameit. One can try to minize that by raising FahBench priority before starting the test.

Actually, I'm not terribly surprised by seeing slightly lower results for a Fermi card with a newer driver. Most likely the driver performance improvement efforts are focused on newer GPUs, and the previous generations might suffer a little in the process. I suppose NVidia weighs customer satisfaction pretty carefully when such choices have to be made.

Re: FAHBench (OpenMM 5.1)

Posted: Mon Sep 30, 2013 3:01 pm
by 7im
Why does everyone assume past optimizations are tossed aside for newer ones? NV not smart enough to maintain 2 code branches?

Re: FAHBench (OpenMM 5.1)

Posted: Mon Sep 30, 2013 3:46 pm
by Napoleon
Smart enough and probably has to maintain a whole lot more branches than just two... but if you have to make runtime choices, the number of branches to choose from just might affect things. Depends on how and when you have to choose, doesn't it?

Editing... a hypothetical switch case:

Code: Select all

switch( code_path ) {
  case Kepler1:
    // some specific stuff
    break;
  case Kepler2:
    // ...
    break;
  case Kepler3:
    // ...
    break;
  // 
  default:
    // some legacy stuff
}
That's a crude example, but I hope it clarifies my speculation. If you absolutely had to do something as clumsy as that, not very likely that you'd put legacy stuff on top of the stack...

Re: FAHBench (OpenMM 5.1)

Posted: Tue Oct 08, 2013 3:54 pm
by mdk777
Well, FYI for anyone interested:

The last few releases of AMD CAT Beta have run the Implicit Solvent test without issue.
My current result with AMD 7970

33.7636 ns/day Explicit
126.75 ns/day Implicit

These numbers are not unusual. However, testing was run on AMD CAT 13.11 Beta which supports the next generation of cards.

Hopefully this indicates a continuum of FOLDING support. :mrgreen: