mfakto: an OpenCL program for Mersenne prefactoring

[Bdot]

Originally Posted by kracker, private message

Hi, was wondering (maybe you already posted somewhere) is it possible/planned for gpu sieving on mfakto in the future? I think many of us including me would really appreciate it ....

Thanks,
kracker

Well, I sure plan for that, have been for a long time.

Unfortunately, my first attempt to "simply port it" did not work. My GPU sieve does not work, it seems to randomly kill FCs ... I have not yet been able to debug it to see why. I had put this back a little and instead played around with 2 montgomery kernels (which are ~10% slower than the barretts as of now). And I made the kernels that are based on 15-bit math work with George's improvements to the barrett reduction. A better plan to make the best out of all the different AMD GPU generations is also on the way.

So yes, I plan for GPU sieving also on OpenCL, but I can't tell when this will come.

Bdot

[kracker]

Originally Posted by Bdot

Well, I sure plan for that, have been for a long time.

Unfortunately, my first attempt to "simply port it" did not work. My GPU sieve does not work, it seems to randomly kill FCs ... I have not yet been able to debug it to see why. I had put this back a little and instead played around with 2 montgomery kernels (which are ~10% slower than the barretts as of now). And I made the kernels that are based on 15-bit math work with George's improvements to the barrett reduction. A better plan to make the best out of all the different AMD GPU generations is also on the way.

So yes, I plan for GPU sieving also on OpenCL, but I can't tell when this will come.

Bdot

I see. Thanks, nice to know

[bcp19]

With the breakthroughs on GPU sieving on the Nvidia models, is there similar work being done for the AMD GPUs?

[kracker]

Originally Posted by bcp19

With the breakthroughs on GPU sieving on the Nvidia models, is there similar work being done for the AMD GPUs?

I think Bdot's last post will explain

[bcp19]

Originally Posted by kracker

I think Bdot's last post will explain

Thanks, I should have seen that.

[Rodrigo]

A few weeks ago, I posted the following:

Originally Posted by Rodrigo

This is a brand-new HP system and the iGPU video-out ports are covered with a bracket that reads, "Do Not Remove." In spite of the oddball screws, I wouldn't have an objection to removing the brackets, except for the possibility that doing that might void the warranty. I will have to look into that before proceeding.

Life and work have been busy since then, but I did look into this issue and found out that it was OK with the vendor to remove that bracket -- so I did, with the help of a specialized PC repair toolkit (thanks, Radio Shack). I now have both the onboard and the add-on graphics connected, each via their own DVI cables, to the same monitor. (The iGPU's DVI cable is hooked up via an adapter to the monitor's VGA input.)

Now, while GPU-Z did like that and is providing additional info on the Intel HD 4000, and Device Manager says that that's working properly, still I can't get anybody to make use of it. mfakto is still giving error messages if I test using the switch -d 2, -d 02, -d 21, etc. It only ever finds two platforms and one device.

So I have not yet been able to make use of the onboard graphics. Next, I'll try connecting it to a different monitor and see what happens.

Rodrigo

[tului]

Originally Posted by Rodrigo

A few weeks ago, I posted the following:


Life and work have been busy since then, but I did look into this issue and found out that it was OK with the vendor to remove that bracket -- so I did, with the help of a specialized PC repair toolkit (thanks, Radio Shack). I now have both the onboard and the add-on graphics connected, each via their own DVI cables, to the same monitor. (The iGPU's DVI cable is hooked up via an adapter to the monitor's VGA input.)

Now, while GPU-Z did like that and is providing additional info on the Intel HD 4000, and Device Manager says that that's working properly, still I can't get anybody to make use of it. mfakto is still giving error messages if I test using the switch -d 2, -d 02, -d 21, etc. It only ever finds two platforms and one device.

So I have not yet been able to make use of the onboard graphics. Next, I'll try connecting it to a different monitor and see what happens.

Rodrigo

I'll keep watching. I've got an HD4000 doing nothing at the moment, but my board supports Lucid Virtu or whatever it's called. I know it's not gonna keep up with my 7970, but it might help a bit.

[Bdot]

Originally Posted by Rodrigo

A few weeks ago, I posted the following:


Life and work have been busy since then, but I did look into this issue and found out that it was OK with the vendor to remove that bracket -- so I did, with the help of a specialized PC repair toolkit (thanks, Radio Shack). I now have both the onboard and the add-on graphics connected, each via their own DVI cables, to the same monitor. (The iGPU's DVI cable is hooked up via an adapter to the monitor's VGA input.)

Now, while GPU-Z did like that and is providing additional info on the Intel HD 4000, and Device Manager says that that's working properly, still I can't get anybody to make use of it. mfakto is still giving error messages if I test using the switch -d 2, -d 02, -d 21, etc. It only ever finds two platforms and one device.

So I have not yet been able to make use of the onboard graphics. Next, I'll try connecting it to a different monitor and see what happens.

Rodrigo

Hi Rodrigo,

does clinfo now include the HD 4000?

Do you have a picture on the HD4000 (when switching the monitor to the VGA input)?

Do things change, if you disconnect the cable from the real GPU before booting up? BTW, the AMD drivers should still allow you to use the HD7770, even if it is not connected.

If nothing helps, you need to extract the real GPU from the system to test it ...

Bdot

[Rodrigo]

The research I did last night doesn't offer a lot of hope that we can get the HD4000 to work. People are having trouble getting OpenCL to work with that iGPU:

http://communities.intel.com/message/173795#173795

http://communities.intel.com/thread/29246?tstart=0

http://communities.intel.com/message/179163#179163

http://communities.intel.com/message/176221#176221 (I don't even have a VGA port, only DVI's through either the GPU or the iGPU)

http://software.intel.com/en-us/forums/topic/342698

http://software.intel.com/en-us/forums/topic/292292

That last one is from August, but according to an Intel employee --

I found out more about the issue. Theoritically this is supposed to work _if_ you are in a hybrid mode - meaning you have Intel's Processor graphics and have a discrete card - with both the drivers installed and using the discrete as your primary adapter. Currently this is not working and we are working on a fix in the driver.

In my Web travels, I didn't find a single instance of someone reporting that they successfully ran the HD4000 alongside an add-on graphics card, let alone a description of how they did it. Someone else may have better luck finding an example, and I would welcome that (with a link!).

As a further test, I just downloaded and tried to run mfakto on the HD4000 in a new i5-3210M laptop. It doesn't find the HD4000 (the ONLY graphics installed in that computer). In fact, it didn't even find the CPU, so all -d combinations failed.

We may have to give up on this idea of using both the GPU and the iGPU until the thing becomes more mature.

Rodrigo

[Rodrigo]

Originally Posted by Bdot

Hi Rodrigo,

does clinfo now include the HD 4000?

Do you have a picture on the HD4000 (when switching the monitor to the VGA input)?

Do things change, if you disconnect the cable from the real GPU before booting up? BTW, the AMD drivers should still allow you to use the HD7770, even if it is not connected.

If nothing helps, you need to extract the real GPU from the system to test it ...

Bdot

Hi Bdot, (I meant to post this one before the one above)

Nope, clinfo still isn't listing the iGPU, and switching the source to the VGA input merely gives a blank screen.

Here are the next two steps I tried, and a third one:

1. Power off the computer, connect the iGPU output to an input on a separate monitor, and reboot.

Result: There was no display to the separate monitor (blank screen). Not really sure if it booted into Windows.

2. Power off the computer, reconnect the iGPU to the original monitor, then disconnect the output from the add-on GPU and reboot.

Result: There was no display on the monitor (blank screen). I think that it did boot into Windows, though, because after reconnecting the addo-n GPU and rebooting, my Gadgets were in the middle of the screen instead of off to the right, which suggests that there was an invisible display at some lower resolution. But maybe that was simply some kind of generic low-res output, rather than actual output from the HD4000.

3. Remove the add-on GPU from inside the PC case and reboot.

Steps 2 and 3 are the ones you suggested. I'm hoping to avoid Step 3. (It may be rendered moot by the information in the previous post.)

Rodrigo

[Bdot]

Originally Posted by Rodrigo

The research I did last night doesn't offer a lot of hope that we can get the HD4000 to work. People are having trouble getting OpenCL to work with that iGPU:

...

As a further test, I just downloaded and tried to run mfakto on the HD4000 in a new i5-3210M laptop. It doesn't find the HD4000 (the ONLY graphics installed in that computer). In fact, it didn't even find the CPU, so all -d combinations failed.

If it did not even find the CPU, then you most likely did not install the AMD Catalyst driver. This machine probably has the highest chance to see mfakto running on the HD4000 as its display is already working ...

I just hope Catalyst would at least install the OpenCL runtime, even if it does not find any AMD hardware ... Would you give it a try?

Originally Posted by Rodrigo

We may have to give up on this idea of using both the GPU and the iGPU until the thing becomes more mature.

Rodrigo

I can understand you ... you've put a lot of time into this already - thanks for your efforts. At least, this thread is now some documentation of what does not work.

Originally Posted by Rodrigo

Hi Bdot, (I meant to post this one before the one above)

Nope, clinfo still isn't listing the iGPU, and switching the source to the VGA input merely gives a blank screen.

As already mentioned, clinfo listing some GPU is required for mfakto running on it. It may not be sufficient though.

When you have connected both GPUs to some monitor (can be the same), could you try pressing the Windows-P combination and select "Duplicate"? I think by default, windows does not automatically enable additional screens.

Also, if you go the the "Screen resolution" control panel applet, does it show both GPUs?

Originally Posted by Rodrigo

Here are the next two steps I tried, and a third one:

1. Power off the computer, connect the iGPU output to an input on a separate monitor, and reboot.

Result: There was no display to the separate monitor (blank screen). Not really sure if it booted into Windows.

2. Power off the computer, reconnect the iGPU to the original monitor, then disconnect the output from the add-on GPU and reboot.

Result: There was no display on the monitor (blank screen). I think that it did boot into Windows, though, because after reconnecting the addo-n GPU and rebooting, my Gadgets were in the middle of the screen instead of off to the right, which suggests that there was an invisible display at some lower resolution. But maybe that was simply some kind of generic low-res output, rather than actual output from the HD4000.

3. Remove the add-on GPU from inside the PC case and reboot.

Steps 2 and 3 are the ones you suggested. I'm hoping to avoid Step 3. (It may be rendered moot by the information in the previous post.)

Rodrigo

This is all not very encouraging. I think you can skip the 3rd test if Catalyst installs on your i5 laptop.

Bdot