I'm getting 75% GPU usage, 50c temp max, 70% frame buffer, 20% bus interface, memory 12% in HWmonitor doing 3wu per 1070ti. On both of my machines using 1070ti's its pretty much the same. This seems actually pretty low as my 1080ti machine hovers around 90% gpu and 90% frame buffer with 50c average 40c max (the 1080ti is watercooled). From my perspective it seems the 1070ti's have a lot more to go.

Isnt 75% max GPU seem awfully low? All the 1070ti's are staying awfully cool no less on air. However they are in a controlled temp server room. Most of the 1070tis are different brands too, with the tri setup as all blowers and the 2 1070ti's are 3 fan direct's.

75% does seem low. Hard for me to remember the utilization on the SoG app. I thought it used to be 89-92% range running singles. What PCIe bus width and slot speed is the 1070 running at? My 1070 has to run in a X4 slot because of the motherboard PCIe lane layout. It runs with less utilization than the other cards in X8 slots.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

What is the Ram on the 1070s? He can only utilize 27% for the work units. If his commandline is higher than 1028 then it's not going to do much else. 3 seem pretty aggressive for a 1070.

---

Ahh yes, the old OpenCL memory limit. I always forget about that. 3 tasks on a 1070 is just too many. Two max.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

2x 1070ti's direct plugged directly into 2 x16 slots and one with extension cable (not a riser) in an x8 slot on the xeon machine. 8350k has one direct into the x16 slot and the other is in a thunderbolt 3 enclosure.

The xeon machine itself is pretty hodgepodged together built from a failed dell 730 server on a HP Z series mobo. Had lots of problems with drivers and RAM issues at first as it has ECC.

The 8350k is just a run of the mill asrock z370 with a thunderbolt box for the second 1070ti.

I am not sure which host you are having the issues with. Just looked at all your hosts and noticed you are running them stock with no command line. If you used a command line to increase their performance, the utilization would go up.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

I'm getting 75% GPU usage, 50c temp max, 70% frame buffer, 20% bus interface, memory 12% in HWmonitor doing 3wu per 1070ti. On both of my machines using 1070ti's its pretty much the same. This seems actually pretty low as my 1080ti machine hovers around 90% gpu and 90% frame buffer with 50c average 40c max (the 1080ti is watercooled). From my perspective it seems the 1070ti's have a lot more to go.

Isnt 75% max GPU seem awfully low? All the 1070ti's are staying awfully cool no less on air. However they are in a controlled temp server room. Most of the 1070tis are different brands too, with the tri setup as all blowers and the 2 1070ti's are 3 fan direct's.

Take a look at the wall clock for processing an SOG task. It is twice as high as it "oughta" be based on my current gtx 1060 3GB results under Windows 10. So yes, there is something not quite up to speed.

Could you post your mb*sog.txt command line, please?
And your app_config.xml file?

Thank you.
Tom

---

A proud member of the OFA (Old Farts Association).

Take a look at the wall clock for processing an SOG task. It is twice as high as it "oughta" be based on my current gtx 1060 3GB results under Windows 10.

If he's running 3 at a time on the 1070Ti then he's running about right. 3 at a time on 1080TI was between 9-12 minutes depending on where the data came from. It will also depend on the timing speed of his RAM on the Mobo.

---

Take a look at the wall clock for processing an SOG task. It is twice as high as it "oughta" be based on my current gtx 1060 3GB results under Windows 10.

If he's running 3 at a time on the 1070Ti then he's running about right. 3 at a time on 1080TI was between 9-12 minutes depending on where the data came from. It will also depend on the timing speed of his RAM on the Mobo.

Wasn't thinking of that but if he did post his app_config.xml it would have been obvious.

So the question is, can he drive up the gpu load level to 90+% with either a better command line or another gpu task?

Tom

---

A proud member of the OFA (Old Farts Association).

So the question is, can he drive up the gpu load level to 90+% with either a better command line or another gpu task?

Tom

Commandline will give us a clue. He's running 2048

Maximum single buffer size set to:2048MB

Fftlength=1024,pass=3:Tune: sum=27959.3(ms); min=4.525(ms); max=344.3(ms); mean=15.52(ms); s_mean=16.04; sleep=15(ms); delta=1970; N=1801; high_perf
Fftlength=2048,pass=3:Tune: sum=20660.4(ms); min=2.273(ms); max=141.2(ms); mean=5.737(ms); s_mean=5.282; sleep=0(ms); delta=1; N=3601; usual

He should lower the value to 1024. That might increase the usage and decrease the time.

---

I'm getting 75% GPU usage, 50c temp max, 70% frame buffer, 20% bus interface, memory 12% in HWmonitor doing 3wu per 1070ti.

Too many WUs.
Running 1 WU at a time my GTX 1070s generally have around 95% utilisation, although power usage is generally only around 65% Of course different WU types will result in different figures.


Edit- as has been mentioned many times in the past- GPU usage is not a good indicator of performance. All that matters is the number of WUs processed per hour.
What good is 100% GPU Utilisation if you're only producing half the work compared to when Utilisation was at 60%?

---

Grant
Darwin NT

Putting

-tt 1500 -hp -period_iterations_num 1 -high_perf -high_prec_timer -cpu_lock -sbs 2048 -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

in the
mb_cmdline-8.22_windows_intel__opencl_nvidia_SoG.txt
file in the
C:\ProgramData\BOINC\projects\setiathome.berkeley.edu
folder will provide a big boost to your output on cards using the SoG application (even suitable for the GTX 1050Ti, however -period_iterations_num should be set to around 30 or so if you use that computer for things other than crunching).

Once the file has been saved, the next WU that starts processing will use the new values- no need to exit & restart BOINC, or read config files.

Dear All, I did the above on a Windows10 System (First time, I try to run BOINC on Win10). But somehow BOINC still downloads CUDA42 and CUDA50 Apps. It is this computer:
http://setiathome.berkeley.edu/show_host_detail.php?hostid=8695664

Did I miss something?

thanks,
klepel

Your Windows 10 computer is still very new, so it will be trying the various application types until it finds the optimum one, which can take several weeks if the wrong mix of tasks is being sent out. Once can force the selection by using the "Lunatics installer" so you only have the application you want to run, but one has to be careful with its use as it is possible to select a completely wrong application and end up with a very poor performance, or loads of tasks being dumped.
The other computer is not running anywhere near its best - three (SoG) tasks is way to much for anything below a GTX1080, and even then is marginal. Drop it down to one, and see what happens.
Another thing to consider is "freeing up" a CPU core for each concurrent GPU task you are running - it is quite spectacular how much difference that makes.

---

Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

It is normal for a brand new computer seen by BOINC to have BOINC try all possible science applications. It is trying to determine which is the fastest. BOINC should land on the SoG application eventually as the fastest and from then on send only SoG tasks to your host. The only way to prevent getting the CUDA42 and CUDA50 tasks is to run an anonymous platform and specifically list only the SoG application. You can do that by either creating your own app_info file or to use the Lunatics Installer and select the SoG application in the gpu chooser menu.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)