Forum

Author Topic: Version 1.3.1 100% CPU and 100% GPU whoaaa  (Read 4232 times)

Hans Keesom

  • Newbie
  • *
  • Posts: 15
    • View Profile
Version 1.3.1 100% CPU and 100% GPU whoaaa
« on: April 08, 2017, 07:24:18 PM »
So I installed 1.3.1. after having had bad experiences with the famous error at line 67 which forced me to go back to 1.2.6. Version 1.2.6 kept my GPU only 40-45% busy and saw a lot of perfcap due to vrel and thrm. Vrel means the GPU slows down because voltage goes up to much and thrm means the GPU reached it max-temp.

Version 1.3.1. seems to be so much better it just runs my CPU and GPU at a full 100%. Actually the GPU is running an average of 97%.
Now it is vrel and pwr that are the perfcap reasons, no longer thrm. So cooling is ok but maybe my power supply is being asked for more it can deliver? It is 800 W so I would think it can handle one Nvidia GTX Titan from about four years ago which has a TDP of 250W. It specs says you need to put it in a computer with at least 600 w powersupply

The motherboard is a asus saubertooth x79 with a intel i7 3930K and 64 GB. I have many external harddisk, but most of them have their own powersupply

I would like to know what you think of  the picture below together with the above specs, what should I invest in?   A larger the powersupply?  One with 1000 of 1200 watt (to get more out of the Titan or to support a new (second) GPU).


« Last Edit: April 08, 2017, 07:42:35 PM by Hans Keesom »

stephan

  • Full Member
  • ***
  • Posts: 129
    • View Profile
Re: Version 1.3.1 100% CPU and 100% GPU whoaaa
« Reply #1 on: May 03, 2017, 07:14:17 PM »
Vrel doesn't necessarily mean there is anything wrong, it's supposed to mean that the GPU is maxed out relative to the frequency it can reach at the current voltage. So if your CPU and GPU are runnning at 100% you're doing good :)

BobvdMeij

  • Full Member
  • ***
  • Posts: 157
    • View Profile
Re: Version 1.3.1 100% CPU and 100% GPU whoaaa
« Reply #2 on: May 04, 2017, 11:48:03 AM »
Hans,

Could you please clarify what settings you are running Agisoft on and how you have organized your Agisoft preferences and/ or additional system preferences relating to your CPU and GPU?

As I have just stated in another post in this General forum, my CPU is maxed out at 100% constantly whereas my GPUs (two GTX 1080's) average at about 20%. I have the feeling that there is still much processing time for me to gain, but whatever I try my GPUs are barely used and that is a bummer.

Thanks in advance!

Bob

Hans Keesom

  • Newbie
  • *
  • Posts: 15
    • View Profile
Re: Version 1.3.1 100% CPU and 100% GPU whoaaa
« Reply #3 on: May 05, 2017, 01:18:52 PM »
Hi Bob,

First of all, I am not aware of what I have done regarding settings in photoscan to reach this, I was sort of lucky I guess.

Also, many phases of the whole workflow still work without using the GPU, unfortunately.

Also, I have heard sometime GPU's can go into some sort of selfprotect mode where they will run at 324 mhz untill you restart the computer. Don't push the GPU to the limit I would think. Keep the voltage a bit low.

I backed up from trying to overclock things. Pushing the limits seems not to good for the stability and it is fine to me if a job runs 200 hours instead of 180 hours, if that means the machine will not crash.
At night, when the small room where the computer is located is closed, I even run it at a lower CPU multiplier. Does not seem to affect the time to finish.

My first question would be to start GPU-Z and look at the VREL at the second tab, that often tells a lot. If it is THRM the GPU's are just running hot.

I do know the pcie of my motherboard is still running at 2.0 speed according to GPU-Z, but it might be totally unrelated.
So look for the businterface speed in GPU-Z, it should not be at 1.1 when under load and maybe you can get 3.0.

BobvdMeij

  • Full Member
  • ***
  • Posts: 157
    • View Profile
Re: Version 1.3.1 100% CPU and 100% GPU whoaaa
« Reply #4 on: May 10, 2017, 10:41:46 AM »
Hi Bob,

First of all, I am not aware of what I have done regarding settings in photoscan to reach this, I was sort of lucky I guess.

Also, many phases of the whole workflow still work without using the GPU, unfortunately.

Also, I have heard sometime GPU's can go into some sort of selfprotect mode where they will run at 324 mhz untill you restart the computer. Don't push the GPU to the limit I would think. Keep the voltage a bit low.

I backed up from trying to overclock things. Pushing the limits seems not to good for the stability and it is fine to me if a job runs 200 hours instead of 180 hours, if that means the machine will not crash.
At night, when the small room where the computer is located is closed, I even run it at a lower CPU multiplier. Does not seem to affect the time to finish.

My first question would be to start GPU-Z and look at the VREL at the second tab, that often tells a lot. If it is THRM the GPU's are just running hot.

I do know the pcie of my motherboard is still running at 2.0 speed according to GPU-Z, but it might be totally unrelated.
So look for the businterface speed in GPU-Z, it should not be at 1.1 when under load and maybe you can get 3.0.

Hello Hans,

I have started using v1.3.1 as of yesterday I must say that indeed GPU acceleration has improved a lot. Compared to v1.3.0 the newer build obviously re-allocated some CPU calculations to the GPU. Depth Map generation processing time was lowered by 300% in one particular case, which I believe is impressive.

Also, I have turned CPU utilization off as I found this lowers processing time. When CPU utilization is on it is continously maxed at 100%, hereby further capping GPU acceleration. PerfCap in GPU-Z mostly displays Vrel and Util. During image alignment temperature is the major limitation.

I find it nonetheless strange, however, that your GPU load figure suggest a continous load on your GPU. In my case, on the other hand, the pattern is much more spikey (see attached screenshot)! Perhaps this is due to me utilizing two GPUs which Agisoft utilizes interchangably, although this is just a wild guess!

Hans Keesom

  • Newbie
  • *
  • Posts: 15
    • View Profile
Re: Version 1.3.1 100% CPU and 100% GPU whoaaa
« Reply #5 on: September 17, 2017, 02:02:50 PM »
Hi Bob,

My GPU is constantly hitting it's thermal limit of 80 degrees. So it could very well be that the GPU will work spiky like yours once I improve the cooling. At the moment a steady stream of hot air comes out of the backplate of the GPU, hot enough to make my hand hurt if I put it in front of it.

Experiment :
Placing an external vent to pull away the hot air seems to have an effect as that the gpu clock goes from 1019 to 1032 (even 1058)and the powerusage of the GPU rises with about 10% and finally that the card throttles not only because of THRM but also VREL and PWR.