My two gaming systems were running so fast this morning that the cooling fans were on in high gear. They were running a bit hot, so I will need to crank up the fans.
/ 
Reply
Quote
Erratum:
CUDA = nVidia
OpenCL = everything
I didn't want to be too specific therefore I used this short comparison. Thanks for clarification.
____________
/ Reply
Quote
Ugly performance on GTX10xx Cards. The new Cuda app doubles the calculation time on GTX10xx series Cards against Cuda55.
Even on new RTX cards is the performance not good, my GTX1080 with cuda 55 is faster than my RTX2070 with the new app. But the raw calculation Power of the RTX2070 is 25% higher.
And when you start comparing the Graphics app with actual cpu's then there is only one conclusion: stopp all graphics app's because of very low calculation speed.
Graphics app can do 4-5 WU/hour and and Ryzen 3900x with avx will do ~22 Wu/hour with lower power consumption. At the moment it's only waste of power which is paid by the volunteers.
Sorry for this hard comment but i think this is the reality.
First stop delivering the new app to non turing Cards,then try to get support from outside, perhaps the guys who optimized the Seti app.
/ Reply
Quote
Set "No new tasks" from "Project" tab for "Asteroids@Home" project.
Nice idea, but the project does not, in fact, respect the "No new tasks" flag.
This is very annoying if you want to clear running tasks before e.g. going off on holiday, decommissioning a PC, or just to let another task have a pop at the CPU for, say, emergency research on a newly discovered and dangerous virus currently causing a global pandemic.
Just sayin'.
/ Reply
Quote
"No new tasks" is a client-side BOINC setting. BOINC projects cannot forcibly push tasks to you. Start troubleshooting with the BOINC config and work up to reinstalling and re-adding projects.
I use "no new tasks" to limit my non-CoViD-related tasks, including Asteroids, to 1-2 batches of WUs per week and have not seen any issues doing so.
/ Reply
Quote
Speaking of new applications, are there any plans for FMA app? Or AVX2 for that matter?
/ Reply
Quote
I noticed something else that looks strange.
I am running the new application on a computer with 2 graphics cards (GTX 1070 Ti and GTX 2070).
In BOINC manager it shows 2 GPU tasks running (one on device0 and one on device1 as expected), but when I look at the GPU load using GPU-Z, I see that the GPU Load is 98% for the GTX 2070, but always 0% for the GTX 1070 Ti.
The workunits are completing successfully, but it doesn't look like the GTX 1070 Ti is actually being used.
I am using Windows 10 Professional with latest updates, BOINC Manager 7.14.2 and NVidia driver v445.75
I have tried resetting the project, but this didn't help.
Edit:
I just tried it on another computer with 2 GPUs (two identical GTX 760 cards) and the same thing is happening.
Even though BOINC manager shows 2 GPU tasks running, I only see load on 1 GPU.
Do you have a cc_config.xml file with <use_all_gpus> in it?
/ Reply
Quote
I noticed something else that looks strange.
I am running the new application on a computer with 2 graphics cards (GTX 1070 Ti and GTX 2070).
In BOINC manager it shows 2 GPU tasks running (one on device0 and one on device1 as expected), but when I look at the GPU load using GPU-Z, I see that the GPU Load is 98% for the GTX 2070, but always 0% for the GTX 1070 Ti.
The workunits are completing successfully, but it doesn't look like the GTX 1070 Ti is actually being used.
I am using Windows 10 Professional with latest updates, BOINC Manager 7.14.2 and NVidia driver v445.75
I have tried resetting the project, but this didn't help.
Edit:
I just tried it on another computer with 2 GPUs (two identical GTX 760 cards) and the same thing is happening.
Even though BOINC manager shows 2 GPU tasks running, I only see load on 1 GPU.
Same problem here with GTX 970 and 1660 Super.
I had the same issue with 2 Radeon at Moo! Wrapper some month ago and never found a solution.
Plug in a screen to the 2nd gpu when you boot the machine or Windows can disable a gpu that's 'not being used'. There is a plug you can make to simulate a monitor if you'd like:
https://www.geeks3d.com/20091230/vga-hack-how-to-make-a-vga-dummy-plug/
/ Reply
Quote
Georgi VidinskiVolunteer moderator
Project administrator
Project developer
Project tester
Send message
Joined: 22 Nov 17
Posts: 36
Credit: 12,934,560
RAC: 11,359
Hi guys,
Yes, there was an issue with addressing multiple CUDA devices. This has been fixed at the beginning of April.
I talked to Daniel, who already confirmed that the application now is distributed correctly on multiple CUDA devices.
Thanks.
Georgi
/ Reply
Quote
Why are there not listed "minimum requirements" for the GPU CUDA or OpenCL tasks? Yes, we do have some modern devices, but also would want to run them on some older (if possible).
Can somebody write min. requirements for cuda55 & cuda102?
____________
non-profit org. Play4Life in Zagreb, Croatia, EU
/ Reply
Quote
Georgi VidinskiVolunteer moderator
Project administrator
Project developer
Project tester
Send message
Joined: 22 Nov 17
Posts: 36
Credit: 12,934,560
RAC: 11,359
Hi KLiK,
You are absolutely right and we will add such information as soon as possible.
Meanwhile here are some more info about the difference and the requirements used by the Asteroids@Home server and when it will send "cuda55" or "cuda102" applications to the clients.
The "cuda55" applications (x86 & x64 both) are build using the CUDA v5.5 library which addresses GPUs with CC of 2.0 by 5.3. More details HERE. As these applications were here for quite long time, there were many new NVIDIA GPUs coming out which we could not use because of the restrictions of the CUDA v5.5 library. It seams that at some point in time someone managed to build successfully a newer version using more recent CUDA library but he has kept the old name "cuda55" of the applications, as many of us were able to run that application on GPUs with CC of 6.x which was not supported by the previous CUDA v5.5.
By that time the project was suffering already from lack of support by the developers, who have made the application in first place. So when I joined the team NVIDIA has been released their CUDA v10.2 library already. More information HERE. And it has its whole new list of requirements along with providing suport for the latest (by that time) GPUs with CC of 7.x. Unfortunately with the new version coming some restrictions. For instance in CUDA v10.2 NVIDIA has deprecated their support for the devices with CC 35, 37 (Kepler) and CC 50 (Maxwell) while these devices can still run apps using this library. Along with it NVIDIA has deprecated their support for a list of OSes as well. And last but not least the CUDA v10.2 library supports only x64 projects.
By the time I'm writing this post there is a new version released by NVIDIA - CUDA v11.0. It adds support for the NVIDIA Ampere GPU micro-architecture with CC 8.0 (like Nvidia GeForce RTX 3080). In the same time they finally dropped the support of devices, declared as deprecated in v10.2 for good. More information HERE. And for those who are reading this post in a year from now HERE is a link to the Archive with all Releases.
Now. As to the drivers. Every new version of CUDA libraries comes with a minimum requirements for the drivers, that will support them. In the links above you will find tables where that is described in details.
So what we have done was to add rules which will ensure that server will send to the clients the "cuda102" app only to computers running x64 OSes and only to those of them, which has installed certen minimum required version of the NVIDIA driver. Which in THIS case (CUDA 10.2) was as follows:
Linux x86_64 Driver Version >= 440.33
Windows x86_64 Driver Version >= 441.22
This means that every other machine which falls outside this requirements will receive the older application "cuda55".
Here is the time to explain why we are unable to support the older applications and keep their code base up-to-date. For instance, in order to rebuild the code using the CUDA v5.5 SDK, the developer team will have to use quite old development environment which includes the need of old and already not supported Operating systems. As all we have limited time resources to contribute to the project we cannot afford to keep such outdated systems up and running no maters physically or using Virtual machines.
I hope this was helpful.
Regards,
Georgi
/ Reply
Quote