Message boards :
Number crunching :
RTX 2080 Ti - Odd performance
Message board moderation
Author | Message |
---|---|
CaptainHer0 Send message Joined: 17 Jan 18 Posts: 38 Credit: 502,222 RAC: 0 |
Hello all, I'm getting odd readings and poor performance with the RTX 2080 Ti compared to my old AMD. The GPU usage average is 26%. ??? It bounces all over the place from 20's to 90's. Wattage is 32% TDP. I'm only using 107 watts. Avg is 87 Watts over a 2 hour period. (CUDA 50 & CUDA 42) Should be closer to 300W. Readings are from GPU-z and match readings from MSI Dragon center. My AMD seemed to be more consistent. Is this normal for an RTX? Are there optimizations I can do that I may be ignorant to? Maybe my settings are incorrect? I have done other benchmarks, (FF15, PC Mark, et al) and card performs fantastic @ 100% GPU usage and 300W power. Card model is MSI RTX 2080 Ti Sea Hawk Ek X - @ 2010 MHz / 7140Mhz Memory. - default one click OC in MSI Dragon center. (~30% OC from ref.) Card performance is similar with OC disabled. This card should be crushing work units like Rambo. Thanks for any insight. Much appreciated. |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13161 Credit: 1,160,866,277 RAC: 1,873 |
What you are seeing is normal for a CUDA42 or CUDA50 application which is what you are running. Those applications perform very poorly on modern Nvidia devices. You should be running the latest and fastest application which is the SoG application. Since I don't see any sign of the SoG application in any of your reported tasks, I can assume you are unable to use that application because you do not have the correct graphics drivers for the application. The SoG application is an OpenCL application. If you don't see any reporting of OpenCL capabilities in your Event Log after first startup and only CUDA drivers loaded, I have to assume you let Microsoft load your video drivers which does not seem to ever install the OpenCL drivers. You need to go to Nvidia's website and download the current drivers for your card and install the drivers directly from Nvidia. That will install both the CUDA and OpenCL components of the drivers and you should soon start seeing the scheduler send you OpenCL SoG tasks. After the scheduler realizes that the SoG is fastest on your system, you should received nothing but SoG tasks. SoG tasks on that card should be running in the 3 minute range. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Chris Oliver Send message Joined: 4 Jul 99 Posts: 72 Credit: 134,288,250 RAC: 15 |
It might not be a bad idea to run the driver uninstaller found here:- https://www.guru3d.com/files-details/display-driver-uninstaller-download.html This will remove the mess that Windows has installed before you install the official Nvidia drivers giving you the best chance for success !! |
CaptainHer0 Send message Joined: 17 Jan 18 Posts: 38 Credit: 502,222 RAC: 0 |
Awesome, thanks! I'll look at that. I did download the drivers from the NVIDIA site, and also checked again via the GeForce Experience software this morning, Says I have the latest driver: 419.35 I have activated "experimental features" in the driver settings area. GPU-Z does not show 'OPEN CL' as an option. Do you have a link for the Open CL Drivers? I see CUDA Development Kits and Direct X libraries. |
Oddbjornik Send message Joined: 15 May 99 Posts: 220 Credit: 349,610,548 RAC: 1,728 |
Somehow you still don't have the right driver. Your computer should say something along the lines of "NVIDIA RTX 2080Ti(4095MB) driver: 419.35 OpenCL: 1.2" in the GPU column. Feel free to take a peek at my computer list to see ... |
juan BFP Send message Joined: 16 Mar 07 Posts: 9786 Credit: 572,710,851 RAC: 3,799 |
What you are seeing is normal for a CUDA42 or CUDA50 application which is what you are running. Those applications perform very poorly on modern Nvidia devices. You should be running the latest and fastest application which is the SoG application. Since I don't see any sign of the SoG application in any of your reported tasks, I can assume you are unable to use that application because you do not have the correct graphics drivers for the application. The SoG application is an OpenCL application. If you don't see any reporting of OpenCL capabilities in your Event Log after first startup and only CUDA drivers loaded, I have to assume you let Microsoft load your video drivers which does not seem to ever install the OpenCL drivers. You need to go to Nvidia's website and download the current drivers for your card and install the drivers directly from Nvidia. That will install both the CUDA and OpenCL components of the drivers and you should soon start seeing the scheduler send you OpenCL SoG tasks. After the scheduler realizes that the SoG is fastest on your system, you should received nothing but SoG tasks. Adding to that, did you run 1 WU at a time? If yes with a top GPU like the one you have and running SoG builds you should try with 2 or even 3. And not forget, this hungry GPU's need a GPU core just to feed their hungry. |
Sixkid Send message Joined: 10 Jan 12 Posts: 17 Credit: 8,248,305 RAC: 21 |
Ahh now i understand why my GPU is now crunching as a Roadrunner :) SoG ( what ever the acronym means ) in the task list :) |
Chris Oliver Send message Joined: 4 Jul 99 Posts: 72 Credit: 134,288,250 RAC: 15 |
Awesome, thanks! If I was you I would run the uninstaller program and select the reboot when finished option. Then after restarting I would install a fresh download of the latest drivers from Nvidias site and when the driver installer asks choose the Express install option. Then everything should be good to go. |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13161 Credit: 1,160,866,277 RAC: 1,873 |
There is a comment in the 419.35 drivers document that says it has experimental OpenCL 2.0 capability but that is it not compatible with anything but OpenCL 2.0 applications. I believe the SoG application is not one of them. It is a OpenCL 1.2 application and that is what the driver normally reports for capability. I would turn off the experimental settings in the driver and just run with stock driver settings. I don't know if this is the issue or whether you had a MS update of your driver after you installed the Nvidia driver directly from Nvidia. Microsoft has a tendency to install whatever it wants even if you just installed a manufacturer driver. The DDU driver purge might be the best avenue for fixing things. And then install the standard Nvidia driver directly from Nvidia and not select any of the experimental features. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13161 Credit: 1,160,866,277 RAC: 1,873 |
Awesome, thanks! There is no link for Windows OpenCL drivers. The OpenCL component is always bundled with the main graphics drivers for Intel, AMD and Nvidia. At least for Windows. And MS graphics drivers constantly reinstalled over manufacturer drivers always LACKS the OpenCL component. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13161 Credit: 1,160,866,277 RAC: 1,873 |
Ahh now i understand why my GPU is now crunching as a Roadrunner :) SignalsonGpu Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
You've been waiting to try that haven't you....hahaha |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13161 Credit: 1,160,866,277 RAC: 1,873 |
You're darn tootin' I think my new favorite highlight color is aqua. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
CaptainHer0 Send message Joined: 17 Jan 18 Posts: 38 Credit: 502,222 RAC: 0 |
[SOLVED] WHAT>>> The issue was Win 10 Opencl.dll became corrupted at some point in the last 2 months. (Fairly fresh Win10 build) I figured this was the case after driver reinstalling in Safe mode via DDU didn't fix it. I put the drivers on a USB, and disconnected the LAN so windows couldn't do "funny stuff" and "help" me out by installing a 6 month old version of the NVIDIA drivers. :) After some Googling it appears it's not uncommon. There was a Win10 update in Oct/Nov 2018 that seemed to have caused the OpenCL.DLL to be rewritten, and this caused it to malfunction. HOW TO FIX>>> 1: Download a fresh copy of Opencl.dll (approx 146Kb) 2: Place file into: C:\Windows\System32 3 :If using 64bit Windows, place an additional copy of the file into: C:\Windows\SysWOW64 Great, now let's run the DISM (Deployment Image Servicing and Management) 1: Open/Run CMD (Command Prompt - run as administrator or Admin Powershell via windows icon) and perform the following commands in sequence: Dism /Online /Cleanup-Image /StartComponentCleanup -this may take several minutes and may appear to stop at times, great time to make lunch ;) Dism /Online /Cleanup-Image /RestoreHealth 2: After these have completed, DONOT run SFC/Scannow, use the following: Dism /Online /Cleanup-Image /CheckHealth If done properly, no corruption(s) should be found. 3: Open/relaunch/refresh or install GPU-Z and now "OpenCL" should have a check-mark next to it. YAY - https://www.techpowerup.com/download/gpu-z/ 4.Reboot your PC to save changes. 5: Launch GPU-Z to make sure OpenCL is still working. If the issue persists, :( - please read, and follow the additional steps in the articles below. References: https://windowsreport.com/opencl-problems-windows-10/ https://troubleshooter.xyz/wiki/fix-corrupt-opencl-dll-windows-10/ Thanks! |
CaptainHer0 Send message Joined: 17 Jan 18 Posts: 38 Credit: 502,222 RAC: 0 |
You had mentioned running more than one WU at a time with an 'SoG' build, so I am interested to see if I can do that and how it performs. Now that OpenCL is working my GPU load is 80-98%, BUT the odd thing is my GPU is only using about 108Watts. PC total power running BOINC and 4k monitor is 397Watts total. I have poked around on the forums and see command lines people have come up with, but am hesitant to grab a random one and use it. Do you know of one I should try for the RTX 2080Ti? Any other settings I should tweak like CPU usage? etc? I'm not a SETI Jedi Master like ya'll. :) |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13161 Credit: 1,160,866,277 RAC: 1,873 |
No he was speaking of running more than one task with the CUDA42 and CUDA50 applications. As you found yourself with your first post, those applications have little need for any cpu support. Witness your low cpu utilization and low clocks. The SoG app on the other hand expects to use and receive a full cpu thread to support the gpu task. So it runs best singly. You may be able to run 2 tasks on the 2080 Ti but that is probably the limit. The reason the SoG is faster than the old CUDA apps is that is uses all of a card's resources all at once. Thanks for the update on why OpenCL disappeared on your system. You can also check with clinfo from the command line for installation verification of OpenCL drivers support. I would try my old command line. -sbs 1024 -period_iterations_num 1 -tt 1500 -high_perf -high_prec_timer -spike_fft_thresh 4096 -tune 1 64 1 4 -oclfft_tune_gr 256 -oclfft_tune_lr 16 -oclfft_tune_wg 256 -oclfft_tune_ls 512 -oclfft_tune_bn 64 -oclfft_tune_cw 64 Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
In order to process more than 1 work unit per GPU, probably should use a app_config.xml. SoG requires 1 Thread per work unit for best performance. If you can find the last lunatics installer and use it to install the SoG application and Astropulse app, it will create your app_info.xml Then you would just have to create the app_config.xml and install it into the Seti@home project folder and restart your machine. BUT I would run out all current work units so that you don't "ghost" your current work units. Set Boinc to No New Tasks and run them all out. Then use the installer. I can't find my old app_config.xml but threw this one together. Someone want to check it for syntax errors? The X in project max concurrent will allow you to decide how many work units, both CPU and GPU will run on your machine. Example, I have a 8 thread CPU and 1 GPU and want only 4 thread of the 8 to run then running 2 GPU and 2 CPU would mean I am only running 4 of the 8 threads of my CPU. If I wanted 4 CPU and 2 GPU then the X would be 6. <app_config> <app_version> <app_name>setiathome_v8</app_name> <plan_class>opencl_nvidia_SoG</plan_class> <avg_ncpus>1</avg_ncpus> <ngpus>0.5</ngpus> <cmdline>-sbs 1024 -period_iterations_num 1 -high_perf -hp -high_prec_timer -tt 1500</cmdline> </app_version> <app_version> <app_name>astropulse_v7</app_name> <plan_class>opencl_nvidia_100</plan_class> <avg_ncpus>1</avg_ncpus> <ngpus>1</ngpus> <cmdline>-unroll 28 -oclFFT_plan 256 16 256 -ffa_block 12288 -ffa_block_fetch 6144 -tune 1 64 4 1 -tune 2 64 4 1</cmdline> </app_version> <project_max_concurrent>x</project_max_concurrent> </app_config> |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13161 Credit: 1,160,866,277 RAC: 1,873 |
In order to process more than 1 work unit per GPU, probably should use a app_config.xml. SoG requires 1 Thread per work unit for best performance. If you can find the last lunatics installer and use it to install the SoG application and Astropulse app, it will create your app_info.xml Then you would just have to create the app_config.xml and install it into the Seti@home project folder and restart your machine. BUT I would run out all current work units so that you don't "ghost" your current work units. Set Boinc to No New Tasks and run them all out. Then use the installer. I can't find my old app_config.xml but threw this one together. Someone want to check it for syntax errors? The X in project max concurrent will allow you to decide how many work units, both CPU and GPU will run on your machine. Example, I have a 8 thread CPU and 1 GPU and want only 4 thread of the 8 to run then running 2 GPU and 2 CPU would mean I am only running 4 of the 8 threads of my CPU. If I wanted 4 CPU and 2 GPU then the X would be 6. Change the unroll count to 68 for the number of SM's in the RTX 2080 Ti. The rest is fine. You could add my workgroup[ tunings also to the MB app. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
CaptainHer0 Send message Joined: 17 Jan 18 Posts: 38 Credit: 502,222 RAC: 0 |
Thanks for the explanation. I see that on a CUDA 50/42 the task says (0.748 CPUs + 1 NVIDIA GPU). Like you said, not a whole lot of CPU. P.S. Nice scope in your bio pic. Speaking of scopes, Wife wants to see the supermoon tonight so I'll get my Meade back on the pier while I let my work units run out then configure it for SoG app as Zalster explained. |
juan BFP Send message Joined: 16 Mar 07 Posts: 9786 Credit: 572,710,851 RAC: 3,799 |
Thanks for the explanation. I see that on a CUDA 50/42 the task says (0.748 CPUs + 1 NVIDIA GPU). Like you said, not a whole lot of CPU. If you not have you could DL the Lunatics Installer here: https://arkayn.us/forum/index.php?action=downloads;cat=1 Just choose the v45.b6 32 or 64 bits depending of you OS They are easy to install, just not forget to choose the SoG builds on the installation process. And don't worry, it will crunch all the WU you already have in your host cache with the new builds. |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.