Message boards :
Number crunching :
Titan V and GTX1060s
Message board moderation
Previous · 1 · 2 · 3 · 4 · Next
Author | Message |
---|---|
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
I've got a better idea, Stay away from Any New Release that refuses to load the Vendors Driver. Eventually they will fix it. One Year will be up in April, it usually works after about a Year or so. Look how long it took the AMD Drivers to work in 16.04, about two years wasn't it? Hmmm, I think I'll try the CUDA 10 Toolkit for 18.04 and see how that works. just for S & Gs. |
ML1 Send message Joined: 25 Nov 01 Posts: 20289 Credit: 7,508,002 RAC: 20 |
I jumped into AMD GPUs about a year ago after the Linus scorching that set the present amdgpu work in motion. nVidia meanwhile are still doing things 'their way'... Not looked at nVidia since, but the usual game to check is kernel version compatibility, CUDA toolkit in place, and whatever other 'quirks'. You also have the new joker of whatever it is that systemd expects/requires to be in place to play... Hopefully some pointers there to search on. And... Take a look at what the ppa does and where? Good luck. Happy super fast crunchin'! Martin PS: Also check your user/boinc has write access to the /dev/nvidia... See new freedom: Mageia Linux Take a look for yourself: Linux Format The Future is what We all make IT (GPLv3) |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
So, why did 18.04 remove the option to install gksu anyway? Seems to be just another move to annoy the user to me. Yes, that was an annoying development to me also. Something to do with Gnome desktop manager. There is a workaround. gedit admin:///etc/default/apportor if you just preface the gedit call for the file with admin:///you are able to edit the file as root in the file manager. 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: 13164 Credit: 1,160,866,277 RAC: 1,873 |
I've got a better idea, Stay away from Any New Release that refuses to load the Vendors Driver. Eventually they will fix it. One Year will be up in April, it usually works after about a Year or so. Just remember that the CUDA toolkit installs its own graphics driver. You can't mix and match the driver in the toolkit with the standalone driver apparently from multiple posts about the topic. 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: 13164 Credit: 1,160,866,277 RAC: 1,873 |
I got the official Nvidia .run installer to install the drivers properly in Ubuntu 18.04 AS LONG as I have purged all remnants of Nvidia ppa drivers AND done a autoremove. I also had to start from the Nouveau drivers. But they did install eventually. Too much work for me since the ppa drivers have always worked for me. The missing OpenCL driver on some versions is easily handled with a sudo apt-get install ocl-icd-libopencl1 Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
...I think I'll try the CUDA 10 Toolkit for 18.04 and see how that works. just for S & Gs.Same problem. All the nVidia software from the Toolkit says the driver isn't loaded. Additional drivers says a manually installed driver is in use. BOINC says No Usable GPUs found. Hmmm, perhaps that's why I'm the Only one that can seem to get a Mining board to work correctly in Ubuntu. I think I'm the only one using the nVidia Driver from nVidia in 16.04. Works nicely with 11 GPUs in 16.04. Oh, I never installed the repository driver in this system. Went straight from the nouveau driver to the Downloaded drivers installed from the Recovery mode. I didn't have any trouble enabling networking in recovery mode, it took about 3 seconds. No problems installing the drivers in the recovery mode. I took a hint from Lubuntu. I ran autoremove in Lubuntu and it seemed to remove half the OS...had to reinstall after that one. I think I'm going to give up for now, and keep using 16.04 for a while. |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
I think I'm going to give up for now, and keep using 16.04 for a while. Whatever works for you and causes the least amount of drama? Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
Yep, need to stay away from the drama. Else you might end up like these people, https://devtalk.nvidia.com/default/topic/1042520/driver/-when-will-the-nvidia-web-drivers-be-released-for-macos-mojave-10-14-/post/5293903/#5293903 From all that you'd think everything prior to Mojave suddenly stopped working. High Sierra still works as well as 16.04, which is more than you can say about nVidia drivers in 18.04 and Mojave. Someday they will get it working, until then everything else is still working fine. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
So, I decided to try installing the Repository 390 driver and see if that changed anything seeing as how nothing from nVidia would load. That worked as expected, missing OpenCL, and when I went to remove 390 autoremove only removed a few dozen files, somewhat better than the Lubuntu autoremove experience. Prior to 18.04 running autoremove after the uninstall command would only remove a handful of files, now it removes dozens of leftover driver related files. That seemed to work! I booted into Recovery Mode, ran the 410.78 installer, and after rebooting 410,78 loaded without any trouble. Something that happened during the Repository driver install/uninstall triggered the nVidia driver to load. Kinda weird. Anyway, I finally got the Downloaded driver to load and it seems to work normally. Right now everything is working in 18.04.1. Strange that after 7 months the Repository driver still can't install a driver with a working OpenCL. Also, having to jump through hoops to get the Downloaded driver to load after 7 months is worrisome. Hopefully they will get it working normally by April... |
Tod Send message Joined: 17 Apr 99 Posts: 27 Credit: 143,685,603 RAC: 0 |
@Ian&Steve No, its actually an old EVGA X99 system. |
Tod Send message Joined: 17 Apr 99 Posts: 27 Credit: 143,685,603 RAC: 0 |
@Keith I ended up installing 4.10, last night. Granted its only been running about 12 hours or so, but on the Titan V, the results with 4.15 are pretty much identical (about 38-39 seconds per work unit). I also increased the clocks of both memory and gpu (about 300 each), and neither really seemed to matter. |
Ian&Steve C. Send message Joined: 28 Sep 99 Posts: 4267 Credit: 1,282,604,591 RAC: 6,640 |
your host info for the Fedora system says the CPU is an i7-4820k, which is an LGA 2011 CPU and uses DDR3 memory X99 is LGA 2011-v3 and uses DDR4 memory. must be EVGA X79, no? your Windows host looks to be running an X99 platform though. in any case, it was just a curiosity if it was ASUS or not. most of the afflicted boards seem to be ASUS, but it can happen to any brand board depending on what components it has and how the BIOS is setup I suppose. Seti@Home classic workunits: 29,492 CPU time: 134,419 hours |
Tod Send message Joined: 17 Apr 99 Posts: 27 Credit: 143,685,603 RAC: 0 |
You're correct! I ran an lspci and sure enough, X79, but it is an EVGA board. I got the windows and fedora boards mixed up. My bad :-) Its getting close to being recycled. Likely will replace with an X299 and a core i9 |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
@Keith So what does Nvidia X Server settings app show for Graphics Clock and Memory Transfer Rate for the Titan V on the PowerMizer tab? I think you can get a lot more clock in the memory side than 300. I wonder are you actually seeing a +300 increase in the core clock from default P2 0 offset? I'm not at all familiar with the clocks on the Titan V so wonder if it does the GPU Boost3.0 thing or not. If I whack in a +100 core clock offset into any of my cards, they just laugh it off and maybe rise 30Mhz over the stock P2 0 offset clocks. The thermal headroom is not high enough to allow such a massive increase in clocks and the card just drags the clocks back to what it knows it can actually manage. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Tod Send message Joined: 17 Apr 99 Posts: 27 Credit: 143,685,603 RAC: 0 |
For the Titan, it has 3 power states, P0,P1,P2. I set the dropdown to Max Performance. Watching the course of a work unit, it mostly hovers on P1, with an occasional switch to P2. GPU Memory P0 135 Min - 135 Max 1700min-1700Max P1 135 Min - 1335 Max 1700min-1700Max P2 135 Min - 1912 Max 1700min-1700Max |
Ian&Steve C. Send message Joined: 28 Sep 99 Posts: 4267 Credit: 1,282,604,591 RAC: 6,640 |
looks like the memory doesnt have different clocks for your max power state, which is unlike the geforce cards. maybe it's a Titan V thing. does look like you're getting a heavy gimp on that core clock though if it's staying at P1 most of the time (the occasional switch to P2, or max power state is normal when BOINC switches to a new WU). Seti@Home classic workunits: 29,492 CPU time: 134,419 hours |
Tod Send message Joined: 17 Apr 99 Posts: 27 Credit: 143,685,603 RAC: 0 |
I've placed the GPU offset at 600 just to test for a few hours. Watching a few of the WUs, I dont see any change, but I'll give it time. All this leads me to believe the bottle neck might be the old hardware the card is installed on. But we'll see soon enough :-) |
Ian&Steve C. Send message Joined: 28 Sep 99 Posts: 4267 Credit: 1,282,604,591 RAC: 6,640 |
See here: https://devtalk.nvidia.com/default/topic/1036962/linux/titan-v-max-clock-speed-locked-to-1-335-mhz-and-underperforms-titan-xp-ubuntu-16-04-nvidia-390-amp-396-/ looks like the same thing me and Keith were talking about. but instead of gimping the memory clocks (like they do on Pascal), they have gimped your core clocks. one thing to look out for, when you apply that clock offset, it may be applying that on the brief switches to the higher power state, trying to add 600MHz to that 1900 value. several of us have seen issues and system crashes from this behavior. if you run into issue, check out the "keepP2" script that petri created. see here: NVIDIA P0, P2 states and overclocking 1080, 1080Ti and VOLTA in Linux Seti@Home classic workunits: 29,492 CPU time: 134,419 hours |
Keith Myers Send message Joined: 29 Apr 01 Posts: 13164 Credit: 1,160,866,277 RAC: 1,873 |
It looks like the suggested driver fix for Volta didn't happen in November like the Nvidia representative said it would. Still locked in 410.78 drivers. Did you see the core clock lock at 1335Mhz with the 415 drivers? It looks like it is pointless to try and use a core clock or memory clock offset on Volta as the driver is hard locking the clocks to defaults. You still get the advantage of much better FP16 and FP32 performance over consumer cards and the fastest performance on the special app of any card in play so far. Seti@Home classic workunits:20,676 CPU time:74,226 hours A proud member of the OFA (Old Farts Association) |
Tod Send message Joined: 17 Apr 99 Posts: 27 Credit: 143,685,603 RAC: 0 |
Thanks, I'll try installing that tonight :-) I'll revert back to 415 tonight as well and report back the clocks |
©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.