GTX 460 superclocked

[TouchuvGrey]

i currently am running two video cards   a GT220 and a GTS 250, i have a GTX 460 ( superclocked ) on the way.
i intend to replace the GT220 with the GTX460 and put the 220 in my #2 machine. i will report back with
the results.

Current Configuration:

7/27/2010 8:33:51 PM      Starting BOINC client version 6.10.56 for windows_x86_64
7/27/2010 8:33:51 PM      Config: use all coprocessors
7/27/2010 8:33:51 PM      log flags: file_xfer, sched_ops, task, sched_op_debug
7/27/2010 8:33:51 PM      Libraries: libcurl/7.19.7 OpenSSL/0.9.8l zlib/1.2.3
7/27/2010 8:33:51 PM      Data directory: C:\ProgramData\BOINC
7/27/2010 8:33:51 PM      Running under account Mike
7/27/2010 8:33:51 PM      Processor: 8 GenuineIntel Intel(R) Core(TM) i7 CPU         920  @ 2.67GHz [Family 6 Model 26 Stepping 5]
7/27/2010 8:33:51 PM      Processor: 256.00 KB cache
7/27/2010 8:33:51 PM      Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss htt tm pni ssse3 cx16 sse4_1 sse4_2 syscall nx lm vmx tm2 popcnt pbe
7/27/2010 8:33:51 PM      OS: Microsoft Windows 7: Home Premium x64 Edition, (06.01.7600.00)
7/27/2010 8:33:51 PM      Memory: 11.99 GB physical, 23.98 GB virtual
7/27/2010 8:33:51 PM      Disk: 419.93 GB total, 332.79 GB free
7/27/2010 8:33:51 PM      Local time is UTC -5 hours
7/27/2010 8:33:51 PM      NVIDIA GPU 0: GeForce GTS 250 (driver version 25896, CUDA version 3010, compute capability 1.1, 998MB, 470 GFLOPS peak)
7/27/2010 8:33:51 PM      NVIDIA GPU 1: GeForce GT 220 (driver version 25896, CUDA version 3010, compute capability 1.2, 987MB, 131 GFLOPS peak)
7/27/2010 8:33:51 PM   SETI@home   Found app_info.xml; using anonymous platform
7/27/2010 8:33:52 PM   SETI@home   URL http://setiathome.berkeley.edu/; Computer ID 5241862; resource share 1000
7/27/2010 8:33:52 PM   SETI@home   General prefs: from SETI@home (last modified 21-Feb-2010 19:16:08)
7/27/2010 8:33:52 PM   SETI@home   Computer location: home
7/27/2010 8:33:52 PM   SETI@home   General prefs: no separate prefs for home; using your defaults
7/27/2010 8:33:52 PM      Reading preferences override file
7/27/2010 8:33:52 PM      Preferences:
7/27/2010 8:33:52 PM         max memory usage when active: 6139.56MB
7/27/2010 8:33:52 PM         max memory usage when idle: 11665.16MB
7/27/2010 8:33:52 PM         max disk usage: 2.00GB
7/27/2010 8:33:52 PM         don't use GPU while active
7/27/2010 8:33:52 PM         suspend work if non-BOINC CPU load exceeds 60 %


Architecture:     GenuineIntel Intel(R) Core(TM) i7 CPU 920 @ 2.67GHz [Family 6 Model 26 Stepping 5]
OS Details:     Microsoft Windows 7 Home Premium x64 Edition, (06.01.7600.00)
Number of CPU's:     8
Created:     Sat, 26 Dec 09 17:11:17 -0700
Timezone:     GMT -5
Floating Point Speed:     2,260.79 million ops/sec
Integer Speed:     8,033.70 million ops/sec
Memory Bandwidth:     125Mbit/sec
Ram:     11.99Gb
Cache:     256.00Kb
Swap:     23.98Gb
Disk Total:     419.93Gb
Disk Free:     356.30Gb
Ave Upload Rate:     10.80 Kb/sec
Ave Download Rate:     52.21 Kb/sec
Ave Turnaround:(tbf)     365,377.22
cpid:    a23269cdfd792a32f4cdad5ec3412d55
Rank:     2,320
Last Update:     12,874
7 Day Average:     12,625
Last 7 Days:     101,245
Last 28 Days:     286,597
RAC:     11,973
Coprocessor:     0 x 6.10.56CUDA with 2 (BOINC)

[TouchuvGrey]

1/1/2002 12:15:58 AM      NVIDIA GPU 0: GeForce GTX 460 (driver version 25856, CUDA version 3010, compute capability 2.1, 739MB, 363 GFLOPS peak)
1/1/2002 12:15:58 AM      NVIDIA GPU 1: GeForce GTS 250 (driver version 25856, CUDA version 3010, compute capability 1.1, 998MB, 470 GFLOPS peak)

GTX 460 installed, i had expected that it would have been more like 900-950 GFLOPS peak.
Unless i am missing something it seems as if the GTX 460 will do less crunching than the
GTS 250 ( of course i miss a lot )

[Ghost0210]

hi,
not sure if Boinc is reasing the Gflops correctly for my 465 either. It can go from 570->855->2553 Gflops in a day.
855 is the correct (nVidia's) rating for the 465, so I would expect your 460 to be around that mark as well
Check out http://en.wikipedia.org/wiki/GeForce_400_Series and you can see Wiki's rating there

[Richard Haselgrove]

hi,
not sure if Boinc is reasing the Gflops correctly for my 465 either. It can go from 570->855->2553 Gflops in a day.
855 is the correct (nVidia's) rating for the 465, so I would expect your 460 to be around that mark as well
Check out http://en.wikipedia.org/wiki/GeForce_400_Series and you can see Wiki's rating there

If you're using BOINC v6.10.45 or later, it should calculate peak GFlops correctly for GTX 480 and GTX 470 - using the original Fermi (GF100) chip with 32 shaders per multiprocessor.

But the GTX 460 uses the GF 104 chip, with 48 shaders per MP. No version of BOINC calculates that correctly yet (although DA was notified on 17 July): You should add 50% extra to those reported GFlops values.

Edit - the GTX 465 is a GF100 card, so the GFlops should be right.

[Ghost0210]

Thanks Richard,
I saw the post you made to DA on the Boinc_Alpha list, from what he said it sounds like nVidia, didn't write this in to their API correctly?
Also any idea's why Boinc would report my Gflops at 2253 Gflops?
31/07/2010 17:50:48 |  | NVIDIA GPU 0: GeForce GTX 465 (driver version unknown, CUDA version 3010, compute capability 2.0, 994MB, 2253 GFLOPS peak)
31/07/2010 17:50:48 |  | ATI GPU 0: ATI Radeon HD5x00 series (Redwood) (CAL version 1.4.737, 1024MB, 620 GFLOPS peak)

 

[Richard Haselgrove]

Thanks Richard,
I saw the post you made to DA on the Boinc_Alpha list, from what he said it sounds like nVidia, didn't write this in to their API correctly?

It looks as if they didn't even put it into the API at all, although that's the obvious place. Instead, you have to read the printed developer guide, and hard-code the shader counts into the application. Anybody got a hot-line to the NVidia dev team?

Also any idea's why Boinc would report my Gflops at 2253 Gflops?
31/07/2010 17:50:48 |  | NVIDIA GPU 0: GeForce GTX 465 (driver version unknown, CUDA version 3010, compute capability 2.0, 994MB, 2253 GFLOPS peak)
31/07/2010 17:50:48 |  | ATI GPU 0: ATI Radeon HD5x00 series (Redwood) (CAL version 1.4.737, 1024MB, 620 GFLOPS peak)

No idea on that one, sorry.

[Ghost0210]

Ok,
Because having to hard-code something like that is the way to go obviously Only nVidia........

No idea on that one, sorry.

No worries, it reports correctly occasionally so its not a big issue, just a shame I don't actually get that amount

[Jason G]

Hmmm, I was reading this stuff not long ago... looking

[Later:] Here's how boinc 6.10.58 sources do the estimation, which matches the technical specifications of the compute capabilities in the Cuda 3.1 reference material.

    // Estimate of peak FLOPS.
    // FLOPS for a given app may be much less;
    // e.g. for SETI@home it's about 0.18 of the peak
    //
    inline double peak_flops() {
        // clock rate is scaled down by 1000;
        // each processor has 8 or 32 cores;
        // each core can do 2 ops per clock
        //
        int cores_per_proc = (prop.major>=2)?32:8;
        double x = (1000.*prop.clockRate) * prop.multiProcessorCount * cores_per_proc * 2.;
        return x?x:5e10;
    }

I would suggest they are GigaFlops on the 'cold day in hell' benchmark to start with.  For that outrageous Multi-TeraFop figure I suspect the clock rate or multiProcessorCount may be being misreported internally somewhere. 

[Richard Haselgrove]

Hmmm, I was reading this stuff not long ago... looking

http://boinc.berkeley.edu/trac/changeset/21034
http://boinc.berkeley.edu/trac/changeset/21036

No change to http://boinc.berkeley.edu/trac/browser/trunk/boinc/lib/coproc.h since then.

[Josef W. Segur]

Another issue involved is that when BOINC reads the properties, Fermi cards are usually downclocked. It's a good feature of the newer GPUs, but leads to lower GFLOPS display. BOINC isn't really using the value much now, instead relying on trying to measure actual performance, so the developers aren't highly motivated to fix it.
                                                                                             Joe

[Ghost0210]

Thanks all
From what I've got above it does seem to make sense.
This fits in niceley with what I'm seeing, when I first boot the box, Boinc gives a Gflops of 570, add 50% on to this estimate from Richards post and this gives the correct Gflop rating of 855 but this is a GF100 chip so not sure why you would have to do this? Then, using nVidia's control panel - change the fan speed to manual and up it to 65% and restart Boinc it now gives the 2.5 Tflop rating.
I have a feeling that this may have to do with the way that Boinc reports the clock rate, EVGA Precsion and GPU-Z all rate the clock at 810 Mhz and if I don't use nVidia's toll to increase the fan speed, Boinc will report that 810Mhz clock rate in the stderr of any tasks.
But, when the Gflops is being reported as 2.5Tflops the clock rate in the stderr is being reported as 3200Mhz.
Since this figure has really no effect on the speed of the tasks or the performance of the card, like I said I'm not really bothered what Boinc reports it's more out of personel interest than anything

[Richard Haselgrove]

If we can track down what's happening, I can add it to my _alpha post about the shader count - that needs a bump, in any event.

Remember that CUDA cards have three different clocks. The only one that matters for BOINC is the shader clock, which would be something like 1.215 GHz for the 465. If you're seeing 3.2 GHz, then it looks as if both BOINC and the app are picking up the memory clock instead - that would be a bug. But whether it's a bug in the app(s) coding, the NVidia API, or EVGA Precsion, I couldn't begin to tell. Jason?

[Ghost0210]

EVGA Precision and GPU-Z are both picking up 1215 Mhz for the shader clock which is correct, but if you look at the stderr for a task completed on the 465 when it is reporting a Gflop of 2253 then the clock rate is being picked up as 3.2Ghz

<core_client_version>6.11.4</core_client_version>
<![CDATA[
<stderr_txt>
Before cudaAcc_initializeDevice(): Boinc passed gCUDADevPref 1
setiathome_CUDA: Found 1 CUDA device(s):
  Device 1: GeForce GTX 465, 993 MiB, regsPerBlock 32768
     computeCap 2.0, multiProcs 11
     clockRate = 3200000

I've attached the full stderr for you, I'm assuming that the clock rate in the stderr is the shader clock?
If you look at a task that has been completed with a gflop of 570, then it picks up the 810Mhz clock rate, which according to GPU-Z/EVGA is the Memory clock

But whether it's a bug in the app(s) coding, the NVidia API, or EVGA Precsion, I couldn't begin to tell.

Don't think its a bug in EVGA as GPU-z is reporting this as well as nVidia's Performance tool

[Jason G]

Most likely a driver API or Cuda Runtime library bug/omission.  Not surprising since it's a new card & there are almost always some teething issues with drivers & new card models. By my understanding GTX 460 is a new die (GF104), rather than a cut down GF100, correct me if I am wrong.  That could mean the clock circuitry &/or strategies are entirely different.  If that's the case, it's probably due to refinement of the memory controller, which was proving to be the limit for GF100.  Reports I've had from a Gamer/Benchmarker  friend seems to indicate that something like that has indeed been 'fixed' and achieves 'Unexpectedly high performance' for the money.

To clarify in case of any confusion: Boinc, for its peak flops estimates should be using the 'Core Clock' specification, which on my 480 is currently at 782MHz, but shows in my task stderr output as '810000', so I feel it's a generic driver or Cuda library encoded figure rather than necessarily read from the current rate.

Fermi's have the shaders locked at 2x the core clock (i.e. 1564MHz on my 480) ... but we don't use shaders, we use 'Cuda Cores' @ 32 per Multprocessor ( 15 * 32 = 480 cores )

- Boinc shows mine as 778GFlops Peak
- Using actual figures: (1000.*782000) * 15 * 32 * 2 = 750.72 GFlops peak
- Using figures from stderr: (1000.*810000) * 15 * 32 * 2 = 777.6 GFlops peak

So Boinc is using the 'slightly dodgy' figure reported by the Cuda Runtime, via the cudaGetDeviceProperties() call as the app does. Since mine is reporting a generic figure, slightly overstated even though it's a reference card, we may not see 'real; figures on newer cards until a new Cuda SDK & library version comes out.

[Richard Haselgrove]

But Ghost's 465, although a new-ish card (released 31 May), has the older GF100 chip. Should have got the driver bugs out of that by now.

And he reported earlier in this thread that "[BOINC GFlops] can go from 570->855->2553 Gflops in a day". That doesn't sound like a static library value: more like idle/downclocked: then running normally: then a borked figure from EVGA Precision. In general, I don't buy the library idea (although Jord seems to like it too): people who over/underclock these cards report BOINC reported speeds which vary in proportion, so I think the runtime is genuinely getting some representation of a clock speed, even if not always the right one.

And again, this is quite different from the problem with the 460 / GF104 that TouchuvGrey reported at the start of this thread: that one stems from the

'Cuda Cores' @ 48 per Multprocessor ( 7 * 48  = 336 cores

for Compute Capability 2.1