Folding Forum

Fah only uses SP. DP numbers posted just for reference.

Ah ok thx

Couldn't run DP on my 260 anyway , I'm sure it supports it though.......
I've pasted the error messages in my previous post.

Assimilator1 wrote:Well I can't find out if my GTX 260 c216 does support O CL 1.1 or not *shrug*, so seeing as no one's given a benchmark for this oldie I will .
...

Thanks for your report, I have updated it. FYI, I did provide the data for an earlier version of FAHBench (viewtopic.php?p=239647#p239647).

Please report any errors to me via a PM or post in this thread.

No probs, & thx for the link , I see the numbers are fairly close, was your 260 a c216 version or 192? The numbers of cores makes quite a difference, worth adding that in the listing I reckon. Btw you never added your 260 to the list in that old thread.

Is there much difference in the stats between v1.0 & v1.2?

Btw do the 260s not do DP?

Assimilator1 wrote:Btw do the 260s not do DP?

They do, or at least should. They have CC (Compute Capability) 1.3, which is the first one to support DP. See http://en.wikipedia.org/wiki/CUDA#Versi ... ifications for additional details. Having said that, CC 2.x (read: Fermi and newer GPUs) introduced a whole lot of additional features.

Assimilator1 wrote:...I see the numbers are fairly close, was your 260 a c216 version or 192?...

It was a 216 SP which was factory overclocked by Gigabyte.

Assimilator1 wrote:...Btw you never added your 260 to the list in that old thread...

The thread was originally maintained by k1wi but later, k1wi was busy so it wasn't updated. Finally, a new version of FAHBench was released so I decided to start fresh.

Assimilator1 wrote:...Is there much difference in the stats between v1.0 & v1.2?...

Yep, a major one:
-Updated to use OpenMM 5.1 with significantly faster speed (http://fahbench.com/)

That translates to a huge improvement for GPUs folding (http://folding.stanford.edu/home/sneak- ... u-core-17/).

I guess fahbench needs CC 2.0 then seeing as I got the previously mentioned error?

Re speed up v1.0 vs v1.2, I guess that my 260 numbers vs yours were close because 260s can't use core 17?
I just discovered the 260 I have is a 55nm version ('only' had it a year or so, lol), so I'm guessing that should o/c quite a bit
Awesome speed up on MM 5.1 anyway!

Re old thread, didn't notice someone else had posted it, doh!

Assimilator1 wrote:I guess fahbench needs CC 2.0 then seeing as I got the previously mentioned error?...

AFAIK, OpenCL 1.1 is used by FahCore_17 which is based on FAHBench. Not sure how that relates to Compute Capability.

Assimilator1 wrote:...Re speed up v1.0 vs v1.2, I guess that my 260 numbers vs yours were close because 260s can't use core 17?...

Assuming that the overclock was the same and the drivers used were too, it's possible that FahCore_17 uses different hardware that isn't present in Pre-Fermi GPUs thus, the difference isn't there as the code is skipped since it can't run.

Assimilator1 wrote:...Re old thread, didn't notice someone else had posted it, doh!

I did mention it at the end of the list but I don't think that anyone would be interested in those results since they aren't useful with the latest FahCore_17 version.

Let me remind everyone that when Drivers "support" OpenCL version X.x, that does not tell you whether it is limited to hardware that has feature F or if the drivers contain emulator code providing support that simulates feature F in older hardware at decidedly inferior performance.

PantherX
Re old thread, that's not what I meant, I meant I hadn't realised someone else had started it.
My 260 isn't o/ced (yet) btw.

New benchmark.
ATI HD 5850 1GB, stock clocks, Cat 13.1, Win 7 64 bit, FaH Bench v1.2 - verify accuracy on

Open CL, SP
Explicit 11.1046 ns/day
Implicit 45.1101 ns/day

Open CL, DP
Explicit 3.66112 ns/day
Implicit 6.30671 ns/day

Going to run it on my C2Q @3.6 GHz too, think that might take a while so it'll be SP only!

Gigabyte R9 290X Windforce 3X OC@1047MHz
C2D6300@2600MHz, Win7 Pro 64Bit, Catalyst 14.1 BetaV1.6

FAHBench1.2

OpenCL
Settings: Single Precision
Verify Accuracy

Explicit Solvent: 46.7955 ns/day
Implicit Solvent: 151.498 ns/day

Settings: Double Precision
Verify Accuracy

Explicit Solvent: 10.2063 ns/day
Implicit Solvent: 18.9532 ns/day

Few more for the list...

FAHBench1.2

Settings: Single Precision

GTX660 (327.23)

Explicit: 22.6831
Implicit: 91.9702

GTX660Ti (327.23)

Explicit: 29.4018
Implicit: 137.002

GTX750Ti (334.89)

Explicit: 20.5086
Implicit: 92.9817

Rig details:-
* Sager NP9170/Clevo P170EM chassis
*Intel i7 Ivy Bridge 3610QM
*AMD 7970M
*16GB 1600MHz RAM
*256GB Crucial M4 SSD
*Windows 7 Ultimate
*F@H bench 1.2.0

With the last WHQL AMD drivers (13.12):

Implicit single precision: 74.8 ns/day.
Explicit single precision: 21.3 ns/day.

With the last Beta drivers (14.2 Beta 1.3):

Implicit single precision: 86.2 ns/day.
Explicit single precision: 25.1 ns/day.

FahCore_17.

Which places the mobility 7970M near the desktop HD 7870 on explicit and the GTX 570. Not too shabby

update for OpenMM 6 in the works?