does mfakto work on the Intel HD Graphics 3000?

[ixfd64]

I'm getting conflicting information as to whether mfakto will run on the Intel HD Graphics 3000. I know the GPU supports OpenCL 1.1, which is the minimum version required for mfakto. Yet I've also seen reports that the HD Graphics 3000 is not supported.

In any case, I downloaded mfakto 0.12 as I have a couple of CPUs with an HD Graphics 3000. However, I'm getting an error message (with code 0xc000007b) when I run the application, even though I've installed the appropriate drivers. For the record, I'm using a 64-bit version of Windows 7.

The 32-bit version of mfakto does work, but it complains about being unable to find a GPU. As such, the program runs extremely slowly on the CPU and uses up almost all of the resources.

Any advice would be appreciated.

[kracker]

I thought only HD 4000 supported OpenCL 1.1, but I might be wrong.

[Dubslow]

http://www.mersenneforum.org/showthread.php?t=17538

The short answer is "sort of, but not really".

[Bdot]

Originally Posted by ixfd64

I'm getting conflicting information as to whether mfakto will run on the Intel HD Graphics 3000. I know the GPU supports OpenCL 1.1, which is the minimum version required for mfakto. Yet I've also seen reports that the HD Graphics 3000 is not supported.

In any case, I downloaded mfakto 0.12 as I have a couple of CPUs with an HD Graphics 3000. However, I'm getting an error message (with code 0xc000007b) when I run the application, even though I've installed the appropriate drivers. For the record, I'm using a 64-bit version of Windows 7.

The 32-bit version of mfakto does work, but it complains about being unable to find a GPU. As such, the program runs extremely slowly on the CPU and uses up almost all of the resources.

Any advice would be appreciated.

As mentioned in the mfakto thread earlier, Intel still seems to have limited use of their built-in OpenCL device to machine that do not have a discrete GPU. If you happen to have such a system, I'd be happy to help searching what would be needed to make it work. But so far, nobody has seen the HD3000 (or HD4000) listed in the clinfo output. And that would be the requirement (as far as I understood this so far):

• install AMD Catalyst driver
• install Intel OpenCL SDK
• run clinfo and make sure the Intel GPU is listed

If clinfo does not find a device, then mfakto won't either.

[LaurV]

Originally Posted by Bdot

But so far, nobody has seen the HD3000 (or HD4000) listed in the clinfo output

I do! (i7 2600k, asus maximus extreme z mobo supporting both nvidia/sli/cuda/ati/crossfire/opencl, win7-64).
GPU-Z shows me, beside of the "list" of gtx580's which I have, also one HD3000, but I never knew what to do with it. I am a "nvidia" guy. Is there any chances I could use this for something, without destroying the gtx setup (which is running mfactc/cudalucas currently)? BTW, when I start GPU-Z, it always showed an error related to opencl drivers being wrong or absent, which I always ignored it, because I don't know how those would interact with installed nvidia drivers.

[Bdot]

Originally Posted by LaurV

I do! (i7 2600k, asus maximus extreme z mobo supporting both nvidia/sli/cuda/ati/crossfire/opencl, win7-64).
GPU-Z shows me, beside of the "list" of gtx580's which I have, also one HD3000, but I never knew what to do with it. I am a "nvidia" guy. Is there any chances I could use this for something, without destroying the gtx setup (which is running mfactc/cudalucas currently)? BTW, when I start GPU-Z, it always showed an error related to opencl drivers being wrong or absent, which I always ignored it, because I don't know how those would interact with installed nvidia drivers.

Hey LaurV, now that's gonna be interesting. Not that it could add a lot to your bottom line, but the educational benefits could be worth it ...

How about you downloading mfakto and try to run it straight away. I assume it will complain about some missing libraries, libopencl.so/opencl.dll maybe.
This should be solvable by installing the windows or linux catalyst driver. I just hope it will install on a box without any AMD H/W.

But maybe that is not even needed, as you say clinfo lists it already.
In case mfakto does start, you can play around with the -d option in order to make it use the HD3000. Take care about a message "GPU not found, fallback to CPU." This is not what you want. The -d option for mfakto can take a 2-digit number. The first digit specifies the platform, i.e. the software stack provider. In your case, there can potentially be 3 of them: NVIDIA, Intel and AMD. The second digit is the device number, starting with 1. You may have a total of <number of discrete GPUs> plus 2 devices. So the minimum of the options is "-d [1-3][1-3]" (9 combinations) that you can try. Some will try to select your GTXes (known to fail mfakto's kernel compilation), but maybe you'll find the HD3000.

If any error occurs, let me know the message, and maybe also include the clinfo output here.

What will be the influence to the rest of your system? Well, nobody tried that before
I think the drivers should not interfere with each other if you're running a reasonably modern OS (I'd not try it with XP). However, a restore point/backup would not hurt ...
When running mfakto on the iGPU, then it will consume some CPU (which is certainly expected). But using the iGPU will also add towards the TDP of the chip, possibly downclocking the CPU. I'd be interested in the results ...

[ixfd64]

I think I now know why mfakto won't run.

From what I've heard, the integrated GPU will only become active if the display port is connected. I use my discrete GPU for the display, so the integrated CPU does not turn on. Does anyone know if it's possible to make the computer think the integrated GPU port is connected, say, with a program like ZoneScreen?

[Ralf Recker]

Originally Posted by ixfd64

I think I now know why mfakto won't run.

From what I've heard, the integrated GPU will only become active if the display port is connected. I use my discrete GPU for the display, so the integrated CPU does not turn on. Does anyone know if it's possible to make the computer think the integrated GPU port is connected, say, with a program like ZoneScreen?

I don't know if it's possible using software but some people use dummy VGA adapters for similar purposes (Folding@home, BOINC, etc.).

[LaurV]

Originally Posted by Bdot

Hey LaurV, now that's gonna be interesting. Not that it could add a lot to your bottom line, but the educational benefits could be worth it ...

Ok, back home, this is what I see, I had to cut it a bit because the forum won't let me attach big photos, I hope is still readable.

Downloaded mfakto 0.12 win 64. It first complained about vc runtime, I thought I have that, I don't know how it disappeared. I reinstalled it from M$ site. Now it says:

e:\-99-Prime\mfakto\mfakto-0.12win>mfakto-x64
mfakto 0.12-Win (64bit build)


Runtime options
  Inifile                   mfakto.ini
  SievePrimesMin            5000
  SievePrimesMax            200000
  SievePrimes               25000
  SievePrimesAdjust         1
  NumStreams                3
  GridSize                  4
  WorkFile                  worktodo.txt
  ResultsFile               results.txt
  Checkpoints               enabled
  CheckpointDelay           300s
  Stages                    enabled
  StopAfterFactor           class
  PrintMode                 full
  V5UserID                  none
  ComputerID                none
  AllowSleep                yes
  TimeStampInResults        no
  VectorSize                4
  GPUType                   AUTO
  SieveOnGPU                no
  SmallExp                  no
  SieveCPUMask              0
Compiletime options
  SIEVE_SIZE_LIMIT          36kiB
  SIEVE_SIZE                289731bits
  SIEVE_SPLIT               250
  MORE_CLASSES              enabled
Select device - Error: No platform found
init_CL(3, 0) failed

What's next? (drivers? openCL? something which will not kill my gtx farm setup, of course) Any idea?

(edit: just to repeat what I said before, every time when I restart the computer it complains about OpenCL driver being absent or buggy, but I never knew what to do with that and being afraid to install new drivers which may influence nvidia setup, so I ignored it every time).

[Bdot]

Originally Posted by LaurV

Ok, back home, this is what I see, I had to cut it a bit because the forum won't let me attach big photos, I hope is still readable.

It is. From what you post it is hard to believe that you really see the HD3000 in the clinfo output. Can you please post it as well?

As the next step I'd say installing the Cayatlyst driver (see my prev post) would be next. But if you don't have that, I wonder where you have the clinfo binary from ? Is the nvidia driver shipping that as well?

[LaurV]

Ok, it took me a while to realize that we were talking about different things, and as I got some free time, I decided to give a try to mfakto, to understand how the things are going on first, then talk after.

So, I dug through a pile of scrap which I have in some cartoon boxes here and took out a Radeon HD 4650. There are few more there, but they do not really worth the power they consume (see below), and also there aren't any more free slots on this mobo. Anyhow, this will help me to understand how this toy works.

Mfakto runs without any problem on this card, it took me (no prior knowledges) mostly half hour to dld and install everything, maybe one hour including hardware instalation, and from the former discussion on the forum, which I vaguely remember, I was able to solve all issues related to missing dll's and other stuff.

Good, we passed (very slowly) the -st and -st2 tests, played a bit with --CLtest but got billions of errors, so we gave up, then decided to do something useful, so we took 10 LLTF assignments from GPU72, but we are afraid that AMD will release HD9990 until we will finish these assignments.

If someone is interested, HD4650 is about 40 times slower then modern cards when it comes to mfakto. We also found out that this card is so old that mfakto has no idea how to call it by her name... We also had to tweak part of the parameters to make the graphic card more responsive (with default parameters it looked like stoned for minutes!). We don't know exactly what we did, but now it is working. Slow, but working... And the computer is quite responsive too, at least when we are writing this post...

c:\mfakto\mfakto-0.12win>mfakto-x64
mfakto 0.12-Win (64bit build)

Runtime options
  Inifile                   mfakto.ini
  SievePrimesMin            5000
  SievePrimesMax            200000
  SievePrimes               25000
  SievePrimesAdjust         1
  NumStreams                2
  GridSize                  3
  WorkFile                  worktodo.txt
  ResultsFile               results.txt
  Checkpoints               enabled
  CheckpointDelay           300s
  Stages                    enabled
  StopAfterFactor           class
  PrintMode                 full
  V5UserID                  none
  ComputerID                none
  AllowSleep                yes
  TimeStampInResults        no
  VectorSize                2
  GPUType                   AUTO
  SieveOnGPU                no
  SmallExp                  no
  SieveCPUMask              0
Compiletime options
  SIEVE_SIZE_LIMIT          36kiB
  SIEVE_SIZE                289731bits
  SIEVE_SPLIT               250
  MORE_CLASSES              enabled
Select device - Get device info -
WARNING: Device does not support atomic operations. This may lead to errors
         when multiple factors are found in the same block. Possible errors
         include reporting just one of the factors, or (less likely) scrambled
         factors. If the reported factor(s) are not accepted by primenet,
         please re-run this test on the CPU, or on a GPU with atomics.
Compiling kernels ..........
WARNING: Unknown GPU name, assuming VLIW5 type. Please post the device name "ATI RV730 (Advanced Micro Devices, Inc.)" t
o http://www.mersenneforum.org/showthread.php?t=15646 to have it added to mfakto. Set GPUType in mfakto.ini to select a
GPU type yourself and avoid this warning.

OpenCL device info
  name                      ATI RV730 (Advanced Micro Devices, Inc.)
  device (driver) version   OpenCL 1.0 AMD-APP (937.2) (CAL 1.4.1734)
  maximum threads per block 128
  maximum threads per grid  2097152
  number of multiprocessors 8 (640 compute elements)
  clock rate                450MHz

Automatic parameters
  threads per grid          1048576
  optimizing kernels for    VLIW5

running a simple selftest ...
Selftest statistics
  number of tests           50
  successful tests          50

selftest PASSED!

got assignment: exp=61516901 bit_min=71 bit_max=73
Starting trial factoring M61516901 from 2^71 to 2^72 (7.77GHz-days)
  k_min = 19191337685520 - k_max = 38382675379483
Using GPU kernel "barrett15_75"
No checkpoint file "M61516901.ckp" found.
  done |    ETA |     GHz |time/class|    #FCs | avg. rate | SieveP. |CPU idle
  0.1% | 15h13m |   12.24 |  57.180s | 893.39M |  15.62M/s |   25000 |  79.79%
  0.2% | 15h03m |   12.36 |  56.587s | 883.95M |  15.62M/s |   28125 |  79.43%
  0.3% | 14h53m |   12.50 |  55.993s | 875.56M |  15.64M/s |   31640 |  79.10%
  0.4% | 14h48m |   12.55 |  55.745s | 866.12M |  15.54M/s |   35595 |  78.93%
  0.5% | 14h28m |   12.82 |  54.570s | 857.74M |  15.72M/s |   40044 |  78.24%
  0.6% | 14h19m |   12.95 |  54.028s | 849.35M |  15.72M/s |   45049 |  77.31%
  0.7% | 14h09m |   13.08 |  53.473s | 840.96M |  15.73M/s |   50680 |  76.40%
  0.8% | 14h00m |   13.21 |  52.969s | 833.62M |  15.74M/s |   57015 |  75.53%
  0.9% | 13h51m |   13.34 |  52.456s | 825.23M |  15.73M/s |   64141 |  74.51%
  1.0% | 13h42m |   13.47 |  51.938s | 817.89M |  15.75M/s |   72158 |  73.21%
  1.1% | 13h32m |   13.61 |  51.400s | 809.50M |  15.75M/s |   81177 |  71.75%
  1.3% | 13h23m |   13.76 |  50.837s | 802.16M |  15.78M/s |   91324 |  70.32%
  1.4% | 13h17m |   13.85 |  50.519s | 794.82M |  15.73M/s |  102739 |  69.28%
  1.5% | 13h06m |   14.03 |  49.861s | 788.53M |  15.81M/s |  115581 |  67.65%
  1.6% | 12h57m |   14.17 |  49.371s | 781.19M |  15.82M/s |  130028 |  66.32%
  1.7% | 12h48m |   14.33 |  48.814s | 773.85M |  15.85M/s |  146281 |  64.84%
  1.8% | 12h40m |   14.47 |  48.361s | 767.56M |  15.87M/s |  164566 |  63.25%
  1.9% | 12h33m |   14.58 |  47.997s | 761.27M |  15.86M/s |  185136 |  61.58%
  2.0% | 12h27m |   14.68 |  47.661s | 757.07M |  15.88M/s |  200000 |  60.35%
  2.1% | 12h28m |   14.65 |  47.754s | 757.07M |  15.85M/s |  200000 |  60.30%
  done |    ETA |     GHz |time/class|    #FCs | avg. rate | SieveP. |CPU idle

Soon we will move to better things, like tickling the intel chip again, we are still missing something, but we are going to find out what (maybe).