GPUPI - SuperPI on the GPU

[_mat_]

It should be listed as OpenCL CPU device then. Select it and start the bench.

 

If it's not listed neither ignored, you have not installed the Intel OpenCL drivers yet. Have a look at the FAQ, you'll find a link for the download there.

Edited December 11, 2014 by _mat_

[GENiEBEN]

Will try to update the drivers. Here's a screenshot: https://www.dropbox.com/s/r21xvr8x5o1oc4l/2014-12-11%2016_26_40-Greenshot.png?dl=0

 

EDIT: Added the OCL drivers from the FAQ section, still not working https://www.dropbox.com/s/f6kagqtt6knaxs8/2014-12-11%2016_50_37-Greenshot.png?dl=0

Edited December 11, 2014 by GENiEBEN

[_mat_]

API: OpenCL CPU with 1 devices

Seems to work as excepted. What's the problem exactly? Is the CPU not listed in the settings dialog? What happens when you start the calculation?

[GENiEBEN]

Seems to work as excepted. What's the problem exactly? Is the CPU not listed in the settings dialog? What happens when you start the calculation?

 

That's my problem I guess, there is no settings dialog, starts as shown in the screenshot(s).

[_mat_]

That's my problem I guess, there is no settings dialog, starts as shown in the screenshot(s).

Press Alt+C, it should always come up.

[GENiEBEN]

Press Alt+C, it should always come up.

 

No luck.

 

[tiborrr]

Pushed Pitcairn below 1 minute for 1B: http://hwbot.org/submission/2696798_tiborrr_gpupi___1b_radeon_hd_7870_%28pitcairn_core%29_59sec_231ms?recalculate=true

 

Also, there's an issue - you can open multiple instances of GPUPI, which is problematic:

- you get end result whether the calculation is valid or not. it's easy to open another instance and run short 1M bench, hide it from screen and apply "Pi calculation is done" window to the invalid run.

 

My recommendations would be:

- locked main computing window while dialog window (Pi calculation is done) is opened.

- only one instance of GPUPI can be run.

- no result should be print to screen if there is an error in calculation.

- if possible, an invalid run should halt the calculation right after it encounters rounding error or at least after each loop, like in original SuperPI (if possible by the algorithm).

- Hwbot should add a valid checksum textbox, which should be mandatory for each submission.

 

Regards,

Niko

[Moonman]

Here's an example:

 

Last 'legit' score:

http://hwbot.org/submission/2696936_moonman_gpupi___1b_radeon_r9_290x_18sec_243ms

 

With higher GPU overclocks I got better score but with Invalid result. It's too easy to come around this, just like tiborrr mentioned.

Here's an example:

 

mat, keep up the good work!

[GENiEBEN]

We can see the 3 icons in the taskbar

[Moonman]

Now you can't

[GENiEBEN]

^ Actually that will not be a valid SS soon, so yeah.

[Moonman]

Good to know, will have to abandon the hipro space saving screenshots then

[_mat_]

It's easy to distinguish an invalid result from a valid one. Whatever the dialog says, the hexadecimal digits for 1B will always be 5895585A0. If it's not then something went wrong. It might be difficult to moderate, so we could add something to help it. For example set the font of the run red or add a watermark.

 

We could also go back to the way SuperPI does it an make the result dialog modal.

 

Nico, there is no easy way to check for a valid result during the run. The errors are normally very small, no mathematical boundaries are hit (like division by zero) and the results of each batch are influenced by multiple input parameters like the batch size.

[tiborrr]

We can see the 3 icons in the taskbar

Fixed in less than 20 seconds.

 

 

@mat

I may have a solution for this screenshot issue.

 

The solution would be to save the result in form of a file. Once the calc is done the background info is encrypted and saved into a binary file, which is then uploaded to hwbot. This way the user would have to upload both valid file (encrypted) as well as screenshot. This way the photoshop overclockers would have more issues bypassing the results.

 

This way the benchmark submission routine would consist of:

- valid file

- screenshot per hwbot rules

 

You could add a "Save" button next to OK button to "Pi Calculation is done" splash window.

 

What do you (Genie, mat, massbo) think?

Edited December 11, 2014 by tiborrr

[_mat_]

No luck.

 

Thanks for the video, with your help I could find a possible solution for the problem. Please test this version and see if it works now: Link removed

 

It's just a temporary link, if this fixes the problem, I will update the download links.

 

EDIT: GPUPI 1.4.1 is now available for download.

 

https://www.overclockers.at/news/gpupi-international-support-thread

Edited December 12, 2014 by _mat_

[GENiEBEN]

Awesome mat, it works now

 

@tiborr @mat

 

Yeah, I think using the HWBOT API to create a datafile like the one used by Aquamark would be great. Let me know if you need help.

[tiborrr]

@Genie: Excellent! This is what I wanted to hear!

[_mat_]

I may have a solution for this screenshot issue.

 

The solution would be to save the result in form of a file. Once the calc is done the background info is encrypted and saved into a binary file, which is then uploaded to hwbot. This way the user would have to upload both valid file (encrypted) as well as screenshot. This way the photoshop overclockers would have more issues bypassing the results.

 

This way the benchmark submission routine would consist of:

- valid file

- screenshot per hwbot rules

 

You could add a "Save" button next to OK button to "Pi Calculation is done" splash window.

 

What do you (Genie, mat, massbo) think?

The hwbot API can automate the uploading of submission results and already supports encryption. I don't know if some kind of validation is possible, but I guess not because the logic would have to be on the server's side.

 

Well, if I can add something to the bench to actively prevent photoshop overclocking, I'll do it. But these are clearly bigger issues, that have to be decided by the hwbot staff.

 

Edit: Owned. Well, if it's possible, I'll see to it. Genieben, do you know if there are C++ examples for hwbot automation?

Edited December 11, 2014 by _mat_

[tiborrr]

In my opinion, hwbot API would be perfect solution. Actually there's no validation required if you're able to save the result only when the checksum is correct.

 

Thanks again, Matthias. You are making Karl proud!

[_mat_]

In my opinion, hwbot API would be perfect solution. Actually there's no validation required if you're able to save the result only when the checksum is correct.

 

Thanks again, Matthias. You are making Karl proud!

[Massman]

The API supports uploading of encrypted datafiles. I'm not a developer, so my input will be limited on this but I'm sure Geni can help with this.

 

There should be some information on the encryption in the API documentation: http://static.hwbot.org/api/Generic_API_1.1.pdf

 

You can always adjust the benchmark application settings via "My Account > Your Benchmarks"

[GENiEBEN]

Genieben, do you know if there are C++ examples for hwbot automation?

 

Yes, I'm the process of rewriting that though so it's pure without 3rd party libraries, will keep you posted.

[_mat_]

GPUPI 1.4.1 is now available for download and fixes a minor bug with a certain OpenCL CPU configuration. Thanks again to Genieben for his bug report!

 

https://www.overclockers.at/news/gpupi-international-support-thread

[buildzoid]

Is it normal that the VRAM frequency of my R9 290 keeps bouncing up and down from 150 to my 1575 or am I doing something wrong. Also how straining is this for a GPU because this would make a great GPU stability test if it didn't run 15C cooler than Unigine.

[_mat_]

I have no access to Radeon graphics cards at the moment, so I can't test it myself. Seems like between the kernel calls the memory goes back to idle mode. Definitely strange behaviour, that should cost some efficiency.

 

Regarding stability testing, GPUPI is mainly using 64 bit integer calculations and some double precision math. Those parts of the hardware will be tested for stability of course, even the slightest miscalculation ends up in an invalid run. That's very different from game benchmarks, where only a big hiccup will stop the run (hardware lockup and the driver watchdog resets everything in about 5 seconds), small errors might be seen in certain shaders and different image quality, but not necessarily. But game benchmarks also test different parts of the hardware, because they mostly use shaders for floating point math, which is the fastest, most common use of graphics hardware. My guess for the additionally produced heat is the usage of more parts of the graphics pipeline and things like the geometry shaders (for tesselation) and so on.

 

So I would recommend GPUPI as an additional GPU stability test that is focused on math precision and not on full utilization of all hardware features.