Then where do I get V3.1+? The link only has V3.3.2. Is anybody going to do anything about this? Or are we just not doing GPUPI anymore? I'm so lost. Please explain.
Thanks!
PS, V3.3.2 doesn't even work...AT ALL!!! So you don't need to worry about that POS.
The problem that's confusing everybody is the fork of GPUPI (3.2 and 3.1, gets points) and GPUPI 3.3 (no points). I can only accept the fact, but I am definitely not a fan of this.
I would advise to not bench GPUPI 3.3 currently. I am working on GPUPI 4 that will improve a lot of things, especially benchmark security and timer precision/reliability. It will also get rid of the plagued OpenCL implemtations for CPUs, so it will work out of the box in the future. No need to install any OpenCL driver.
Got an AMD/ATi graphics card? I've found the best way to get OpenCL for CPU working is installing the Radeon Drivers prior to 18.8.1 (I think) Lately I've been using 18.5.1. Otherwise you can try your luck with the various OpenCL SDK packages that can be found on the net. Check out the GPUPI page on Overclockers.at. Alternatively you can try Intels OpenCL package. Links at the page mentioned.
@cbjaust
Yes I have an Radeon 570, but i don't plan to install older VGA drivers just to run this bench.Thanks anyway
@macsbeach98
I just need the Amd opencl.
All i find online is the whole SDK at almost 300mb.
There is not a simple driver file or something?
@_mat_
I have downloaded the "AMD APP SDK 3.0 (OpenCL 2.0)" file and during the installation, I have selected the first option "OpenCL Runtime libraries", but still no option to select my CPU at the benchmark.
Maybe I have to install something more?
Submit mode doesn't work for me at Win7.
2 times I tried to submit the results through the integrated process and both times I have received an Application error message something about instruction at memory 0x00000000 and that the memory could not be written. And I lost the results.
Submit mode doesn't work for me at Win7. 2 times I tried to submit the results through the integrated process and both times I have received an Application error message something about instruction at memory 0x00000000 and that the memory could not be written. And I lost the results.
Suggest you just save the submit file instead and upload that separately. It usually works.
I have unticked the boxes for the CPU and GPU at the results screen and it allowed me to save my results normally..
Go figure. I thought that it's recommended to have them selected.
nope depends on hardware, OS, SDK driver used… sometimes even running HWINFO in save mode helps to save... you just have to check if it saves properly before any serious benching
Unfortunately lately it doesn't allow me to save or submit anything any more on win7.
It just gives the message that the app stopped working and it's closing down. I have tried any possible combination with ticking or not the boxes and the result remains the same.
I believe my problem is straight Intel's Fault. I'm using a EVGA x299 Dark w/DDR4-4133. I have been able to run GPUPI 3.2 & 3.3 on the CPU (Using INTEL'S OpenCL) with a i9-7900x & i9-9820x. The only change has been the update of a i9-10940x. The program crashes before it even starts the test. It has completed the Bench using AMD's OpenCL Driver. I have not tried this on Win-10.
I've tied 18.1 and just updated to DL version of 2020. It could also be... From where I have aggressively pushed my timings to 4000-CL 12-11-11-28/220 1T. I have also tried to make it do 12-11-10-28/220 from another sub. I may have damaged my Win-7 install from way tooo many BSOD's . I have made many runs with 4000-12-11-28/220 1T on GPU benches, I have to credit this to Teamgroup DDR4-4133 and m||rk||r88 on helping me fine tune the memory ? I'm using A2 B-Die memory moduals with a set of A0's reserved for my EVGA z390 ?
The error occurs in Intel's threading library. If this happens on air/stock as well, it's pretty safe to say that this is a bug in Intel's OpenCL implementation. A bug report on the Intel forums would help them to have a look at it.
There is going to be a major BenchMate release first. All my time and effort is going into that.
GPUPI 4 will still arrive this year though.
I have no clue what that means. Is this about 3.2 and 3.3?
Well there are so many previous posts about this matter.
3.2 definitely has the problem of not saving the data file, as I have tried again with a new cooling system and it's always breaking the saving process as it force closes the app.
The concept of having result saving, encryption and uploading inside each benchmark is flawed. Especially if latest benchmark versions are not allowed, like GPUPI 3.3.
Why can't we just upload the full screenshot as we do for so many other benchmarks?
To quote myself from the mission statement on benchmate.org:
Quote
Benchmarking is in a horrible spot right now. Even on HWBOT, the world's biggest result database, most results have to be submitted as screenshots and system data is entered manually. Due to undetectable time measurement bugs and easily abusable vulnerabilities of benchmarks, results can not be correctly judged by the inconclusive data available. Additionally, manual submission is error-prone, inconsistent, time consuming and hard to learn while still being far from trustworthy. Although there are better submission methods and online rankings available like 3DMark's Hall of Fame or the Geekbench Browser, they all set their own opaque standards for validation and are only available for their (latest) products.
Christian Ney says:
GPUPI: International support thread
leeghoofd says:
Version 3.3 or beyond will not be allowed due to changed algorythms and other improvements!
mr.wulten says:
Then where do I get V3.1+? The link only has V3.3.2. Is anybody going to do anything about this? Or are we just not doing GPUPI anymore? I'm so lost. Please explain. Thanks! PS, V3.3.2 doesn't even work...AT ALL!!! So you don't need to worry about that POS.
_mat_ says:
If you search for GPUPI on the web you will find my site with all versions to download: https://www.overclockers.at/news/gpupi-international-support-thread
The problem that's confusing everybody is the fork of GPUPI (3.2 and 3.1, gets points) and GPUPI 3.3 (no points). I can only accept the fact, but I am definitely not a fan of this.
I would advise to not bench GPUPI 3.3 currently. I am working on GPUPI 4 that will improve a lot of things, especially benchmark security and timer precision/reliability. It will also get rid of the plagued OpenCL implemtations for CPUs, so it will work out of the box in the future. No need to install any OpenCL driver.
_mat_ says:
I have updated the download page on overclockers.at to reflect the situation on HWBOT better. Let me know if that helps.
viper says:
How to run this bench for the CPU? When I click the Calculate menu, it only allows me to select my Graphics card, so I can't run it for the CPU.
cbjaust says:
Got an AMD/ATi graphics card? I've found the best way to get OpenCL for CPU working is installing the Radeon Drivers prior to 18.8.1 (I think) Lately I've been using 18.5.1. Otherwise you can try your luck with the various OpenCL SDK packages that can be found on the net. Check out the GPUPI page on Overclockers.at. Alternatively you can try Intels OpenCL package. Links at the page mentioned.
macsbeach98 says:
Install the intel openCL and whatever AMD openCL you prefer and GPUPI will give you the option to run either Intel or AMD OpenCL.
And it works everytime whatever AMD openCL you desire to use
viper says:
@cbjaust Yes I have an Radeon 570, but i don't plan to install older VGA drivers just to run this bench.Thanks anyway @macsbeach98 I just need the Amd opencl. All i find online is the whole SDK at almost 300mb. There is not a simple driver file or something?
_mat_ says:
You have to download the whole AMD APP SDK from here, see "OpenCL drivers for CPUs": https://www.overclockers.at/news/gpupi-international-support-thread
But you will only need to install the OpenCL driver (first entry if I remember), uncheck the rest in the installer.
viper says:
@_mat_ I have downloaded the "AMD APP SDK 3.0 (OpenCL 2.0)" file and during the installation, I have selected the first option "OpenCL Runtime libraries", but still no option to select my CPU at the benchmark. Maybe I have to install something more?
_mat_ says:
As stated above your AMD GPU driver is overwriting the OpenCL entry for the CPU. In the posts above you are good countermeasures.
leeghoofd says:
New 2020 scores: Require CPU-Z 1.91 or newer version New 2020 scores: Full desktop screenshot required ( no clipping)
viper says:
Any updates for 2020? New adrenaline drivers are out. Is it still required to run old graphics drivers to be able to run it?
keeph8n says:
2020 drivers have been able to be used since their launch.
leeghoofd says:
Matt is working hard on GPUPI 4.0, no need to redesign this version.
viper says:
Hello everyone,
For win7 is the procedure the same or maybe it's a bit more easy?
_mat_ says:
No clue. Try to avoid Windows 7 for Compute benchmarks though.
viper says:
Submit mode doesn't work for me at Win7. 2 times I tried to submit the results through the integrated process and both times I have received an Application error message something about instruction at memory 0x00000000 and that the memory could not be written. And I lost the results.
Matsglobetrotter says:
Suggest you just save the submit file instead and upload that separately. It usually works.
viper says:
Small update.
I have unticked the boxes for the CPU and GPU at the results screen and it allowed me to save my results normally..
Go figure. I thought that it's recommended to have them selected.
leeghoofd says:
nope depends on hardware, OS, SDK driver used… sometimes even running HWINFO in save mode helps to save... you just have to check if it saves properly before any serious benching
viper says:
Happy Easter to everyone.
Unfortunately lately it doesn't allow me to save or submit anything any more on win7.
It just gives the message that the app stopped working and it's closing down. I have tried any possible combination with ticking or not the boxes and the result remains the same.
Is this ever going to be fixed? ?
Mr.Scott says:
leeghoofd says:
So what did you change from when it was working? That's the main question you need to ask yourself.
viper says:
The OS remains the same. Only RAM or CPU tweaking.
And it was never 100% stable before, so I don't think that we can ever be sure of what is causing these problems.
It can even be the priority (Real time or high) that is affecting it. Who knows except the people who are related to this app?
MaddMutt says:
I believe my problem is straight Intel's Fault. I'm using a EVGA x299 Dark w/DDR4-4133. I have been able to run GPUPI 3.2 & 3.3 on the CPU (Using INTEL'S OpenCL) with a i9-7900x & i9-9820x. The only change has been the update of a i9-10940x. The program crashes before it even starts the test. It has completed the Bench using AMD's OpenCL Driver. I have not tried this on Win-10.
Thank You
_mat_ says:
What Intel OpenCL version are you using? I have tried 18.1 on an i9-10980XE and it's working fine.
MaddMutt says:
I've tied 18.1 and just updated to DL version of 2020. It could also be... From where I have aggressively pushed my timings to 4000-CL 12-11-11-28/220 1T. I have also tried to make it do 12-11-10-28/220 from another sub. I may have damaged my Win-7 install from way tooo many BSOD's . I have made many runs with 4000-12-11-28/220 1T on GPU benches, I have to credit this to Teamgroup DDR4-4133 and m||rk||r88 on helping me fine tune the memory ? I'm using A2 B-Die memory moduals with a set of A0's reserved for my EVGA z390 ?
Thank for the help
leeghoofd says:
its probably not the OC, just run stock and you maybe can rule that out
_mat_ says:
The error occurs in Intel's threading library. If this happens on air/stock as well, it's pretty safe to say that this is a bug in Intel's OpenCL implementation. A bug report on the Intel forums would help them to have a look at it.
viper says:
Hey is there going to be any update soon?
If not,.maybe it's better to close totally the section.
It's not fair not to be able to save any newer scores. It's a waste of time and effort for better results.
_mat_ says:
There is going to be a major BenchMate release first. All my time and effort is going into that.
GPUPI 4 will still arrive this year though.
I have no clue what that means. Is this about 3.2 and 3.3?
viper says:
Well there are so many previous posts about this matter.
3.2 definitely has the problem of not saving the data file, as I have tried again with a new cooling system and it's always breaking the saving process as it force closes the app.
_mat_ says:
Use BenchMate, it fixes all problems with result saving: https://benchmate.org
The concept of having result saving, encryption and uploading inside each benchmark is flawed. Especially if latest benchmark versions are not allowed, like GPUPI 3.3.
viper says:
Why can't we just upload the full screenshot as we do for so many other benchmarks?
_mat_ says:
To quote myself from the mission statement on benchmate.org:
Please log in or register to comment.