GPUPI 2.0 Launched in Cooperation With HWBOT (Full Integration + Points)

  • News, benchmarks
  • 78
  • HWBOT

GPUPI 2.0 Launched in Cooperation With HWBOT (Full Integration + Points)

HWBOT is today officially announcing the launch of GPUPI 2.0 from Overclockers.at, a new and updated benchmark app that calculates the number Pi using multiple OpenCL and CUDA graphics cards and multiple CPUs. GPUPI 2.0 will debut on the forthcoming Rookie Rumble at OC-Esports.io and is launched with an improved user interface that includes hardware detection and automatic result submission to HWBOT.org.

“I am glad how things worked out with GPUPI in the last few months, having overcome many technical hurdles since I started development back in 2008. I am more than proud that it’s officially added as the first CUDA/OpenCL benchmark to HWBOT.” commented Matthias “mat” Zronek, GPUPI 2.0 developer.

GPUPI 2.0: The Multi-GPU Pi Benchmark

Coded by Matthias “mat” Zronek from Overclockers.at, GPUPI 2.0 calculates the mathematical constant of Pi in parallel by using the BBP formula. GPUPI is optimized for OpenCL and CUDA capable graphics processors and is implemented with C++, STL and pure Win 32. GPUPI 2.0 supports the recently released CUDA 7.0 optimized to get optimized performance from NVIDIA cards.

GPUPI is the first General Purpose GPU benchmark officially listed on HWBOT and also the first that can simultaneously use different video cards. Only the same driver is required. This is a real game changer for overclockers and means that it is possible to bench several cards of different generation, model and architecture as long as they share the same driver.

Check out the screenshot below where Matthias is running GPUPI 2.0 using 3x GTX 980 plus one GTX 780Ti card:

Submission on HWBOT: http://hwbot.org/submission/2810820__mat__gpupi___1b_4x_geforce_gtx_980_8sec_361ms

Unfortunately AMD and NVIDIA cards cannot yet be used in tandem due to the different nature of the technologies involved and the absence of a mutually acceptable driver. But fear not, mat is currently working on the next version that removes this limitation at least for graphics cards.

GPUPI 2.0: Changelog

  • Multi-GPU and Multi-CPU support
  • New platform and device selection with tree view
  • HWBOT integration with automatic result submission including hardware detection (Graphics Card, CPU, OS) and screenshot upload
  • Improved detection of AMD graphics cards (HD 7xxx und Rx xxx only)
  • Encrypted kernel code using AES/CBC
  • Support for CUDA 7.0
  • Improvements for error logging (GPUPI.log in the working directory)
  • Compiled with Visual Studio 2013 Community (Update 4) and AMD OpenCL 2.0 headers
  • New about dialog including benchmark and support information

The new benchmark will make its competitive debut in the forthcoming Rookie Rumble, the perfect opportunity for hungry young overclockers gain kudos by helping to test debug the new benchmark.

To download GPUPI 2.0, click here: https://www.overclockers.at/news/gpupi-international-support-thread.

To learn more, including technical and developmental details: https://www.overclockers.at/news/gpupi-international-support-thread


78

Johni5 says:

Hi!
My first loop 21 sec i7 4790K@4800MHz but interesting i read first loop 2sec 4790K@4700MHz.Please Rules .

Austria _mat_ says:

Depending on the driver and the implementation it is possible, that certain memory buffers might be allocated during the run just before memory is written to it. Try reruns first. If that doesn't help, use smaller batch sizes (10M or even 1M). Be sure to have enough free RAM available, so it won't be swapped to your hard disk.

United States steponz says:

Is windows 8 and higher supported.. what about the timer issue?

Johni5 says:

gpupi old version is much faster charge than gpupi 2.0 1.4 version first loop 2 sec 2.0 is 21 sec. Please set up a rules. Please illustrate the settings. This is 6 min why? http://hwbot.org/benchmark/gpupi_for_cpu_-_1b/rankings?start=0&cores=4#start=0#interval=20

Strunkenbold says:

You have to download OpenCL Intel runtime. Right now you are using OpenCL driver from AMD.

Austria _mat_ says:

The AMD drivers are the fastest choice for CPUs, even for Intel.

In my tests GPUPI 2.0 was faster than 1.4. I will have to test some things today on Haswell/Ivy and check if there are any issues with the new version on these platforms. You could post a screenshot or score in the meantime.

Btw, your loop time depends heavily on the configured batch size. The smaller it is, the faster is your first loop.

Edit: Here is a perfectly nice score using GPUPI 2.0 on Haswell-E: http://hwbot.org/submission/2820569_pulse88_gpupi_for_cpu___1b_core_i7_5960x_5min_56sec_448ms

Austria _mat_ says:

steponz said: Is windows 8 and higher supported.. what about the timer issue?
Windows 8 will be supported by checking for an existing HPET timer with enough resolution. We are currently discussing this internally, an updated version of the bench will be available very soon.

Strunkenbold says:

_mat_ said: The AMD drivers are the fastest choice for CPUs, even for Intel.


I just had time to compare 100M scores and there was Intel OpenCL way fasten than AMD. (32sec vs 48sec) Catalyst 14.12 installed

4770k @ 4Ghz used

France Taloken says:

Rules have to be set before you unleash the submit-tsunami.

About OS, versions of the benchs, heterogenous multi-GPU ... Or don't complain about filtering thousands of wrong scores in few weeks ...

United Kingdom ObscureParadox says:

Can we make a rule saying no win8 for now then before the system gets flooded with win 8 results? Might have been an idea to wait until the win 8 version is ready before giving points, or just disallow win 8 and clearly state that for now.

Austria _mat_ says:

Already implemented the fix, currently testing it on multiple platforms. :)

Greece sofos1990 says:

Nice to see a new benchmark but the way it came out it's sadly too weird. So many things need clarification. The options the benchmark has, reminds me UCbench. Can we use version 1.4 as well or only 2.0? Some submissions are with 1.4.

Austria _mat_ says:

sofos, the benchmark options (batch size, reduction size) are there to optimize the calculation for your hardware. The better you tweak it, the more calculations will run in parallel. Use them for your own good, they are not weird at all. ;) The moderated results for version 1.4 are okay, but it makes no sense to use this version right now. It should be slightly slower (especially for CUDA) and won't support multiple graphics cards. The official downloads are no longer available too, so I would recommend to set up a rule that it's no longer allowed to use 1.4 after this release. Regarding the timing issue again: We will release a version 2.1 very soon (maybe tomorrow). This should be the only version that is allowed to use, because it will block the possibility to cheat by downclocking the system bus. That is the only thing that is weird right now, but not for much longer.

United Kingdom ObscureParadox says:

_mat_ said: Already implemented the fix, currently testing it on multiple platforms. :)


Glad to hear you got the fix done so quick, many thanks :)

Greece TASOS says:

Why rush and make this bench a point awarded one , when there are some "issues" open ?

Two words.

Make the bench bullet-proof for cheaters.
Set the rules.

Australia zeropluszero says:

TASOS said: Why rush and make this bench a point awarded one , when there are some "issues" open ?

Two words.

Make the bench bullet-proof for cheaters.
Set the rules.


This.
Honestly, this happens every time, how is this so bunnying hard?
Don't release it while it still beta
And certainally don't give it points.

UCBench 2.0.

Enjoy cleaning up the mess mods.

Jesus, there isnt even screenshot rules.
What are you people thinking?

GENiEBEN says:

Thanks Obama

United States steponz says:

im glad to see more benchmarks.. but I think it should could come out a different way... maybe get a bunch of top guys to test it more to find all the little gotchas... I noticed a bunch of things...
But overall very happy to see something new... changes things up a bit... if ya guys need any help testing.. let me know ;)

United States steponz says:

Guys.... just for a thought.. if it had no points... many people wouldn't even bench it... also.. we should start letting points for other benches like 3d11 extreme.. other fm benches...

Australia zeropluszero says:

steponz said: Guys.... just for a thought.. if it had no points... many people wouldn't even bench it

It should be tested beforehand.
steponz said:
we should start letting points for other benches like 3d11 extreme.. other fm benches...

We definitely should not.
GENiEBEN said: Thanks Obama

[img]https://www.statusquo.org/aru_flagitfolder/images/flag%20austflag.jpg[/img]

France Taloken says:

If 2.1 is released soon, and will be mandatory, no need to let people submit now, it just polluate the rankings.

Doc.Brown says:

Strange benchmark,I have the same score with +100mhz overclock on my AMD CPU,only tested on 100M but I guess it will be the same with 1B...

Austria _mat_ says:

It actually scales pretty well with clock frequency. Haven't tested too much with AMD CPUs, but there shouldn't be a difference. Don't forget it's a multi threaded benchmark and needs all CPU resources it can get. Your daily platform with lots of background processes might not be the best system to test efficiency and tweaks on. ;) If any you finds a bug or some kind of inconsistency I am happy to have a look it at. Please post a screenshot though. ;)

Doc.Brown says:

It's not under my daily use PC but on a Seven 64b tweaked for bench ;) Anyway problem solved,it was the hpet ^^

Austria _mat_ says:

I have a version 2.1 ready that detects the OS and adapts the time measurement to avoid any issues. I don't want to just throw out another version that might introduce a bug. Therefor I will take a few days to think about my approach and do some more testing. Please be patient. [img]https://www.overclockers.at/attachment.php?attachmentid=202689[/img]

Belgium Massman says:

Rules were added.

V2.1 will add support for Win8/8.1/10 :-)

steponz said: im glad to see more benchmarks.. but I think it should could come out a different way... maybe get a bunch of top guys to test it more to find all the little gotchas... I noticed a bunch of things...
But overall very happy to see something new... changes things up a bit... if ya guys need any help testing.. let me know ;)


Actually, the UCBench story taught us that only when points are enabled the benchmark gets really tested. Competitions beforehand are okay to get the major bugs out.

United States Mr.Scott says:

*sigh* Cannot get 2.0 working on XP. 1.4 worked fine. What changed? Specifically, AMD CPU. As stated, 1.4 worked fine. All I did was upgrade to 2.0 and now I get "cannot find entry point" error.

Belgium Massman says:

Did you update the vga drivers? I had a similar issue with my Nvidia graphics card. Updating the drivers fixed everything :)

Italy delly says:

Now is required a datafile to load the results? I did not see the option to save it...:confused:

United States MetalRacer says:

So no more submissions until the new 2.1 is released?

NVM I figured it out.

Colombia saint19 says:

Mr.Scott said: *sigh* Cannot get 2.0 working on XP. 1.4 worked fine. What changed?
Specifically, AMD CPU.
As stated, 1.4 worked fine. All I did was upgrade to 2.0 and now I get "cannot find entry point" error.


Try to use a never AMD APP SDK version, or run it on Win Vista with APP SDK 2.7, it worked for me.

United States Mr.Scott says:

saint19 said: Try to use a never AMD APP SDK version, or run it on Win Vista with APP SDK 2.7, it worked for me.

I'll give it a go. 1.4 worked with SDK 2.5.
I'll try newer ones.

Germany Agent_Fresh says:

Hey, can anyone pls tell me how to save a file, which i can upload later?!?
I just wanna upload a screen, and now it says it needs a file...

United States Mr.Scott says:

saint19 said: Try to use a never AMD APP SDK version, or run it on Win Vista with APP SDK 2.7, it worked for me.

XP is only supported to SDK 2.5.
2.0 does not work on XP. 1.4 still works fine. What changed?
I can only assume that 2.0 is Vista/W7 only.
Maybe we can address this before 2.1 is released? :o

Australia FatBoyNotSoSlim says:

delly said: Now is required a datafile to load the results?
I did not see the option to save it...:confused:

Agent_Fresh said: Hey, can anyone pls tell me how to save a file, which i can upload later?!?
I just wanna upload a screen, and now it says it needs a file...


Anyone? I have all these screens with results that are useless right now since theres a need for a datafile?

United Kingdom TheMadDutchDude says:

Using 2.8 on Win 7 doesn't work for AMD chips. I get an invalid result. It's rather annoying after 40 mins of benching to get an invalid result thrown in your face. I'm going to try 3.0 tomorrow.

Austria _mat_ says:

I am currently setting up an XP system to test 2.1 on it. I will report back my findings here. Be sure to use GPUPI_OpenCL.exe on XP though, because all new CUDA versions won't support XP anymore. Btw, there is no reason to use XP in the first place with other benches like wPrime and SuperPI. Normally the newest OpenCL/CUDA drivers will get you the best performance, so Windows 7, 8 and 10 is the way to go. Another heads up for submissions of data files: I will implement the possibility to save it to your hard disk and submit it later. Regarding "Invalid result", this is mostly due to instability or old drivers. Precision is everything in this benchmark, because billions of partial results will be accumulated to a final sum giving the digits of Pi. The smallest error will change the final output. Especially double precision was really buggy on old driver implementations and therefor always return an invalid result. So be sure to take the newest if that happens.

Australia SniperOZ says:

nice work on this _MAT_, but where do we find the data file? (as there is no way of generating it manually...that I can see! )..... and as it is now a requirement for submission!!!.... do we need to wait for next version before being able to sub scores??

Belgium Massman says:

For now, submitting is only possible via online submission

Australia SniperOZ says:

lol......!!!..... yeah that doesn't work on my end either,... I get a message saying ..... need cpu for 3d benchmark...... or something like that..... even when I run the cpu bench.

Belgium Massman says:

Can you make a screenshot of the error ?

United Kingdom topdog says:

SniperOZ said: lol......!!!..... yeah that doesn't work on my end either,... I get a message saying ..... need cpu for 3d benchmark...... or something like that..... even when I run the cpu bench.


I also get the same error

Austria _mat_ says:

SniperOZ said: lol......!!!..... yeah that doesn't work on my end either,... I get a message saying ..... need cpu for 3d benchmark...... or something like that..... even when I run the cpu bench.
I guess you have a Xeon CPU, right? This bug is already fixed and will be available with 2.1 as well.

Australia SniperOZ says:

ok I just tested it again......... and it submitted ( the cpu test)....lol....don't you hate that.... but the GPU score didn't work

cpu is intel 4930k

http://www.overclock.net/g/i/2421256/bench-screens/sort/display_order/

United Kingdom topdog says:


Australia SniperOZ says:

Ok so I have tried a different OS and I got it to work..... for the GPU test... but I'm still unsure why its not working on my fully updated 24/7 rig.

Belgium Massman says:

I think it's because the datafile doesn't including a processor name in the submission, so the score gets rejected.

Australia SniperOZ says:

will that be an ongoing problem or it will be sorted out?

Belgium Massman says:

It will be sorted out; I don't think it's a major issue.

Australia SniperOZ says:

sweet ;)

Australia zeropluszero says:

beta benchmarks with points ... nice one.

Australia newlife says:

so wants to share some tweaks bro?

i'll begin with Pentium g3258 likes 4m batch and 64 reduction

zeropluszero said: beta benchmarks with points ... nice one.


did you miss the comments about how a bench doesn't get tested properly till it gets points

Austria _mat_ says:

zeropluszero said: beta benchmarks with points ... nice one.
It was a beta benchmark since the 11/11/2014. Even though we held a competition for the Turrican charity thing the bench had barely 300 submission in the main category. After a few days with points the entrys have already doubled. It's like Massman says: No points, nobody cares. :(

France COMIAS says:

Hello,
This is true since the addition of the points it is to members of the BOT which carry loop, always with the same reproach!
- No good version ...
- Can not Score ...
- Etc etc ...

Most of downloads GPUPI like sheep without taking the time to read the FAQ.
Still not giving a lot of information.
Notament the SDK and downloads for OpenCL.

By cons _Mat_ I have noticed that when using the Intel OpenCL SDK 1.2, timeshares have the bench by nearly 2, compared to the AMD OpenCL 2.0.

I also enjoyed a few test when I place the OpenCL.dll / AMD_OpenCL32.dll and OPENCL64.dll / AMD_OpenCL64.dll files in the folder GPUPI, I earn a few hundredths of a second, every bench.

For my part I download the following files for GPU and GPU bench Intel / AMD:

- AMD-SDKInstaller v1.3.158-Beta-windows-x64-W
- AMDSDKInstallManager-1.3
- Intel_sdk_for_ocl_applications_2014_x64_setup
- Opencl_runtime_14.2_x64_setup

This significantly improves the results.

I think he put more faudrais Advent need to download the SDK from Intel or AMD. because his carry-loop starts typing on certain system.

Anyway great job _Mat_! :)

Sorry for my english it's Google that translates for me.

Uruguay andressergio says:

Thanks Mat good job !

Any problem with Older Chipsets and GPUs ?
I can run it perfect on x99 etc but i cant't on P45 with my E8400 / Q6600 / Gtx 285 / 295 / 260 none works i re installed win 7 2 times already and put the OpenCL and visual c++ 2013 redist. But same all...CPU is a no GO says "cannot start..." GPU on stock setting runs 2 loops and makes NVIDIA driver fail. All that hard i mentioned same error.

Thanks !
Kind Regards for All
Sergio

Colombia saint19 says:

andressergio said: Thanks Mat good job !

Any problem with Older Chipsets and GPUs ?
I can run it perfect on x99 etc but i cant't on P45 with my E8400 / Q6600 / Gtx 285 / 295 / 260 none works i re installed win 7 2 times already and put the OpenCL and visual c++ 2013 redist. But same all...CPU is a no GO says "cannot start..." GPU on stock setting runs 2 loops and makes NVIDIA driver fail. All that hard i mentioned same error.

Thanks !
Kind Regards for All
Sergio


Verify that you only have C++ 2013, if you have more that one version the benchmark doesn't run.

Doc.Brown says:

v2.1 is allowed here or not yet?
If yes is there an updated rule which allow 8/8.1/10 ?

_Mat_ downloads links for v2.1 at overclockers.at are broken ;)

GENiEBEN says:

-allowed -yes

Belgium Massman says:

We'll publish the guidelines for V2.1 shortly. Stay tuned! (and yes, it will be allowed for Win8 and up)

Australia zeropluszero says:

Offline score upload?

Belgium Massman says:

V2.1 will support creating data files to upload later.

United States techjesse says:

That's Good News Massman :) this WiFi is driving me crazy lol

Doc.Brown says:

Ok,thanks for answering ;)

United States Mr.Scott says:

Question: Does Lucid MVP affect the score?, and is there a way to check to see if it was used?, and will it be allowed or not? I guess I should have said questions. lol

GENiEBEN says:

What does MVP have to do with GPGPU?

United States Mr.Scott says:

Virtu Universal MVP combines the CPU's on die GPU (if it has one) with your regular GPU card. It's called HyperFormance. It is similar to what occurs with Nvidias Hybrid SLI and AMD's hybrid tech. I assume that off loading to the on die GPU will increase performance and lower the time. Maybe somebody would like to check it out?
If not, I maybe just gave somebody leverage on their scores.
If nobody cares...........then, as you were.

EDIT- I have a couple people checking this out now. ;)

Australia newlife says:

Doesn't Virtu Universal MVP only work with directx? which would mean no opencl support

United States Mr.Scott says:

It says it works with Opencl 1.2 or better.

Germany Agent_Fresh says:

Now we have the function to save a file, but no more need for uploading it when we submit...?!?

United Kingdom TheMadDutchDude says:

It's a manual submission because some of the CPUs are not recognised in our database. For example: if I bench my Phenom II 555 BE as an unlocked chip to a Phenom II X4 B55, that chip is not in the database as it is actually a Phenom II 555 but with three cores active instead of two. Manual submission allows the user to enter such details on their own accord. It is a little slower, sure, but that's why we have the option to use the previous submission hardware if you've made some before. All you need to do is enter the result, and the relevant things such as CPU and RAM frequency (if it has changed).

Austria _mat_ says:

GPUPI 2.1 already takes care of those cases where the OpenCL driver device name does not match the entry in HWBOT's database. Before uploading or saving the result you can skip the detection of CPU or GPU with two checkboxes on the submissions dialog. Uncheck whatever couldn't be detected. The score will be uploaded but with an error like "Processor is needed for 2D benchmarks". Just follow the submission link and fill in the mandatory fields on the HWBOT site manually. Regarding the uploading of saved result files, this will start to work after May the 3rd, as the announcement states. We will disable the pure manual submissions then, which automatically enables an upload option for result files.

France COMIAS says:

Hello Mat, With GPUPI 2.1, impossible to submit directly with the app, I get an error each time, (http://img15.hostingpics.net/pics/916389exemple.png). and when I pass by the site daily data.result submit the file, same error. ( http://img11.hostingpics.net/pics/442566gpupisubmit.png) ""HAWOT did not accept this result [error:Sorry, an error occured. The datafile is possibly corrupt or we don't support the file.]"" I spend a whole afternoon yesterday after testing different possibility, but it changes nothing. have you any idea? Merci

Austria _mat_ says:

Hi, I answered your question in this thread. Special characters in your detected OS version are the problem. This will be fixed in GPUPI 2.1.1 very soon!

United States TraceFallen says:

i have no option to save my file and submit to hwbot is greyed out.. . been reading for hours and nothing.. .everyone here just says they fixed it then never say how they did. . .. .

United States TraceFallen says:

and the only option i get for manual is upload file. . . very confusing indeed and not set up well at all

Austria _mat_ says:

If the menu item for HWBOT submission is disabled, there is currently nothing to submit. Be sure to select only your CPU for 100M and only your GPU for 32B. The 1B results will work with both. It's also necessary that your run is valid, so no red background should appear with the result. If you still can't submit, please upload a screenshot of your run including CPU-Z/GPU-Z.

Poland barti27 says:

when you can will yield the results from this, correct it in the end

Brazil IanMarcelo says:

Iam trying ro submit,but all times it says that I need a newer version. I already downloaded this one million times in the correct link and it dont work. What I need to do ?????

Please log in or register to comment.