I have a dedicated Nvidia 750 TI (no monitor attached) running one task and the temperature of the GPU is 50C. According to the SETI Task Page a typical AP Run Time of 2100 seconds results in a CPU (GPU) Time of around 80 seconds. There is one virtual CPU core free to support the GPU and "top" always shows the load on it at less than 15%. The question is, why would the GPU only show that it is working less than 5% of the time?

I have compared the server times using “ps ax” and the server logs to the times reported by the task page and they seem to agree. I suspect it is what it is, but it would be nice to know why it is. Perhaps the Linux algorithm for reporting the CPU time is flawed, but you would think the GPU temperature would be higher if that were the case. The reported temperature of 50C would certainly indicate that the GPU is not being overtaxed

I have a dedicated Nvidia 750 TI (no monitor attached) running one task and the temperature of the GPU is 50C. According to the SETI Task Page a typical AP Run Time of 2100 seconds results in a CPU (GPU) Time of around 80 seconds. There is one virtual CPU core free to support the GPU and "top" always shows the load on it at less than 15%. The question is, why would the GPU only show that it is working less than 5% of the time?

I have compared the server times using “ps ax” and the server logs to the times reported by the task page and they seem to agree. I suspect it is what it is, but it would be nice to know why it is. Perhaps the Linux algorithm for reporting the CPU time is flawed, but you would think the GPU temperature would be higher if that were the case. The reported temperature of 50C would certainly indicate that the GPU is not being overtaxed

I have a dedicated Nvidia 750 TI (no monitor attached) running one task and the temperature of the GPU is 50C. According to the SETI Task Page a typical AP Run Time of 2100 seconds results in a CPU (GPU) Time of around 80 seconds. There is one virtual CPU core free to support the GPU and "top" always shows the load on it at less than 15%. The question is, why would the GPU only show that it is working less than 5% of the time?

I have compared the server times using “ps ax” and the server logs to the times reported by the task page and they seem to agree. I suspect it is what it is, but it would be nice to know why it is. Perhaps the Linux algorithm for reporting the CPU time is flawed, but you would think the GPU temperature would be higher if that were the case. The reported temperature of 50C would certainly indicate that the GPU is not being overtaxed

Maybe the -use_sleep function is causing a weird issue with the monitoring utility? Perhaps it is polling at an interval when things are simply not active.
I looked at a few other hosts that also have 750 Ti's & the AP processing times seems very similar to yours. Some were a bit higher. So they may be running more than 1 task at a time or some other configuration that is less optimized. However your CPU times are the lowest I found.

I have a dedicated Nvidia 750 TI (no monitor attached) running one task and the temperature of the GPU is 50C. According to the SETI Task Page a typical AP Run Time of 2100 seconds results in a CPU (GPU) Time of around 80 seconds. There is one virtual CPU core free to support the GPU and "top" always shows the load on it at less than 15%. The question is, why would the GPU only show that it is working less than 5% of the time?

I have compared the server times using “ps ax” and the server logs to the times reported by the task page and they seem to agree. I suspect it is what it is, but it would be nice to know why it is. Perhaps the Linux algorithm for reporting the CPU time is flawed, but you would think the GPU temperature would be higher if that were the case. The reported temperature of 50C would certainly indicate that the GPU is not being overtaxed

Perhaps you are confusing GPU with CPU -- "top" knows nothing about GPUs, only CPUs. In fact, I know nothing that will give you an estimate of GPU usage under Linux -- nvidia_smi will tell you the GPU temperature, the percentage of the fan speed and memory usage. But it won't tell you the things that GPUz (purports to) tell(s) you about the GPU on Windows systems.

50 C is a doddle for a 750 Ti -- it will easily run two instances, where the one or the other takes over computing in the "slack" periods that the other job is transferring data to the host PC. Set it to run two instances and watch your RAC climb a bit -- and your GPU run 10 or so (Celsius) degrees warmer. (I've got a 660 Ti in this Linux machine, it's showing 63 C running two Astropulse WUs with an ambient of about 25 C; I also have a 750 Ti in a Windows machine at work, it handles two SETI jobs with ease, though I do notice a bit of mouse lag when one of its jobs is an Astropulse WU).

According to the SETI Task Page a typical AP Run Time of 2100 seconds results in a CPU (GPU) Time of around 80 seconds.

E: I get temps around 58C on one of my 750Tis under 99% GPU load. Perhaps your's just has a very efficient cooler.

#dsh:
I'm using Nvidia driver 344.22 in Linux-64. I run "nvidia-settings" which is part of the driver package utilities to get GPU % utilization. When the nvidia-settings window opens, click on GPU0 to see lots of info including utilization and GPU memory used, etc.
Even though you have a different driver version I would be surprised if it did not have the "settings" utility.

Gene;

According to the SETI Task Page a typical AP Run Time of 2100 seconds results in a CPU (GPU) Time of around 80 seconds.

What you take as "CPU (GPU) Time" is actually CPU time only.
The real "GPU time" will be (close to) that shown as "Run Time".

E: I get temps around 58C on one of my 750Tis under 99% GPU load. Perhaps your's just has a very efficient cooler.

E: I get temps around 58C on one of my 750Tis under 99% GPU load. Perhaps your's just has a very efficient cooler.

Depends on the cooler & ambient temperature.

My Gigabyte GTX 750Tis generally sit at around 60-65°c with around 90-95% load on them with the ambient temperatures around 36°c.
At present (coldest days so far this year) ambient temperature is down to 22°c, card temperatures are mid 50°s, but fan speed is down to 65% of maximum. Usually they're at around 75-80%.

I have a dedicated Nvidia 750 TI (no monitor attached) running one task and the temperature of the GPU is 50C. According to the SETI Task Page a typical AP Run Time of 2100 seconds results in a CPU (GPU) Time of around 80 seconds. There is one virtual CPU core free to support the GPU and "top" always shows the load on it at less than 15%. The question is, why would the GPU only show that it is working less than 5% of the time?

I have compared the server times using “ps ax” and the server logs to the times reported by the task page and they seem to agree. I suspect it is what it is, but it would be nice to know why it is. Perhaps the Linux algorithm for reporting the CPU time is flawed, but you would think the GPU temperature would be higher if that were the case. The reported temperature of 50C would certainly indicate that the GPU is not being overtaxed

Perhaps you are confusing GPU with CPU -- "top" knows nothing about GPUs, only CPUs. In fact, I know nothing that will give you an estimate of GPU usage under Linux -- nvidia_smi will tell you the GPU temperature, the percentage of the fan speed and memory usage. But it won't tell you the things that GPUz (purports to) tell(s) you about the GPU on Windows systems.

I was not basing the utilization of the GPU on “top”. I used “top” only to judge the degree to which the AP application was utilizing the free core which I indicated at no more than 15%. I did that only to show that I did not think the lack of CPU cycles in support of the GPU was slowing down the GPU.

I used “ps –ax” to list the AP application and show how much time it was actually running. In the example below it shows 69 seconds and when a WU completes, the last time “ps ax” displays matches the CPU (GPU) time shown on the SETI Results Page which is nominally 80 seconds

12286 RNl 1:09. ./../projects/setiathome.berkeley.edu/astropulse_7.08…….

The actual elapsed real time is typically 2100 seconds which is what is shown in the Results Page as Run Time.

I was basing the low usage on the disparity between the reported “CPU [GPU] time” and the “Run Time” which seems to indicate that the GPU is only working less than 5% of the time. That is probably not the case and I was just curious as to why that might be. My apologies for not being clearer in my first post.

Edit: To anyone reading the above, it is in error and Ivan is absolutely correct. I realized "top" was reporting the CPU time used by the GPU app but I was not making the connection that same was true for "ps -ax" reported time. My mistake. Thanks to Ivan and Castor for pointing out the error of my ways.

50 C is a doddle for a 750 Ti -- it will easily run two instances, where the one or the other takes over computing in the "slack" periods that the other job is transferring data to the host PC. Set it to run two instances and watch your RAC climb a bit -- and your GPU run 10 or so (Celsius) degrees warmer. (I've got a 660 Ti in this Linux machine, it's showing 63 C running two Astropulse WUs with an ambient of about 25 C; I also have a 750 Ti in a Windows machine at work, it handles two SETI jobs with ease, though I do notice a bit of mouse lag when one of its jobs is an Astropulse WU).

My results have been different. If I try to run two tasks, the tasks take twice as long and if the temperature of the GPU is any indication, I suspect I actually lose ground a bit since the temperature reported by “nvidia-smi” is about 2 degrees cooler than when running one WU.

My ambient temperature is currently at 27C and the 750 TI is at 51C. That is a lot less heat generated than yours and Castor's but I have no idea what I can contribute that to.