CPU lanes are almost never shared with other peripherals. They will be straight to PCIe slots. In the case of a consumer level CPU with 16 lanes. That means all 16 are run to the main 16x PCIe slot, and then 8 lanes run to a second 16x PCIe slot where applicable (in which both 16x slots will run at 8x if populated)

It’s the Chipset based lanes that are usually shared with other peripherals like the M.2 or SATA ports, etc.

---

Seti@Home classic workunits: 29,492 CPU time: 134,419 hours

Is it time to upgrade the CPU and Motherboard?

You might be able to use less power with a newer cpu/motherboard but the results generated by the gpu processing usually means it doesn't matter WHAT cpu/mb you are using as long as it is fast enough to push the gpus to full speed.

I have seen some real strange looking combos. Like an Intel Duo 2 core pushing something like a gtx 1080Ti. I will admit the CPU wasn't getting a whole lot else done but then anything it could do would have been minuscule compared to the GPU.

So I would hold off, get the new cards installed, tuned up etc. And then decide.

Tom

---

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

I like the specs on the i9-9900K Coffee Lake 8-Core.

Your present CPU is very old & power hungry for it's level of performance, but the difference in Run time & CPU time is purely due to it trying to do too much- reserving a CPU core for the GPU will sort that out at no expense.

I would wait to see the pricing and specs on the Ryzen 3000 series CPUs (end of this month, products arriving middle of next if the rumours are correct). High core counts, high clock speeds, good single threaded performance & way cheaper than Intel if the early noises are correct. Even if you go with Intel, these new AMD systems will hopefully make Intel lower their prices.
Saying that, i'm very happy with my i7-8700k systems with water cooling.

---

Grant
Darwin NT

Took a while, Newegg sent me a bad power supply then once that was sorted I found that one of the cards didn't work. So 3 months and I got the 2-2070's up an running power supply Corsair HX850.
I have a pre-order for a new AMD RYZEN 9 3900X 12-Core, plus MB and ram, but that is back ordered. So for the time being let's see what the 2070's do. My 750Ti's were taking about 15 mins. Looks like it might take the 2070 about 5 mins?

---

My 750Ti's were taking about 15 mins. Looks like it might take the 2070 about 5 mins?

If you were to make use of some command line values, their output would improve significantly (or are you running multiple WUs? My RTX 2060 is doing similar WUs a minute faster than your RTX 2070- Output may or may not improve with the values when running multiple WUs).

In your Seti project folder should be a file
mb_cmdline_win_x86_SSE3_OpenCL_NV_SoG.txt

If you copy & paste these values in to it, and save them (Notepad only- Not Word or Wordpad) the next WU that starts processing will use the new values & should complete considerably faster.

-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

To undo those changes, just delete everything in the file & save it, and the next WU to start processing will just use the default values as they are doing now.

---

Grant
Darwin NT

Done. I will see what it does today.
It seems that they took about 4.5 sec on average. Here are the last few before the change. Will compare this evening.

---

Well it looks like the command line values, helped save me an average of 1.1 minutes per work unit. Here are the 13 work units completed before the changed command line this morning vs this evening after the change had run about 12 hours. Thanks for the help Grant.

---

My 750Ti's were taking about 15 mins. Looks like it might take the 2070 about 5 mins?

If you were to make use of some command line values, their output would improve significantly (or are you running multiple WUs? My RTX 2060 is doing similar WUs a minute faster than your RTX 2070- Output may or may not improve with the values when running multiple WUs).

In your Seti project folder should be a file
mb_cmdline_win_x86_SSE3_OpenCL_NV_SoG.txt

If you copy & paste these values in to it, and save them (Notepad only- Not Word or Wordpad) the next WU that starts processing will use the new values & should complete considerably faster.

-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

To undo those changes, just delete everything in the file & save it, and the next WU to start processing will just use the default values as they are doing now.

I could be wrong but for that to take effect do you not to tell the Boinc to read config preference files? Option to achieve this located in the option menu. Carlos must of worked this out. I should have realised this before posting

---

... but for that to take effect do you not to tell the Boinc to read config preference files?

Nope.
While changes to app_info.xml require you to exit & restart BOINC, and changes to app_config.xml require you to "Read preference files" (or exit and restart).
The big advantage of using the command line file- as soon as the next WU starts, it uses those values.

---

Grant
Darwin NT

Thanks for the help Grant.

No worries.
Good to see your system putting it's capabilities to good use.

---

Grant
Darwin NT

... but for that to take effect do you not to tell the Boinc to read config preference files?

Nope.
While changes to app_info.xml require you to exit & restart BOINC, and changes to app_config.xml require you to "Read preference files" (or exit and restart).
The big advantage of using the command line file- as soon as the next WU starts, it uses those values.

Thanks Grant. You learn something new every day

---

Well I installed the 3rd RTX 2070 but Bonic is only using two of them. What do I need to change to get the third one recognized?

---

Well I installed the 3rd RTX 2070 but Bonic is only using two of them. What do I need to change to get the third one recognized?

In the BOINC Event log startup messages, does it recognise all 3 (probably not as it's still only showing 2 on your computer details).
If not, in Windows Task Manager, Performance tab, does it show GPU 0, GPU 1, GPU 2?
If not, in Device Manager, do all 3 cards show up?
If not, time to start pulling & swapping cards, just to make sure the new card is OK.

---

Grant
Darwin NT

Add

<use_all_gpus>1</use_all_gpus>

on you cc_config file

Not all 2070 are exactly equals

---

Add

<use_all_gpus>1</use_all_gpus>

on you cc_config file

Not all 2070 are exactly equals

+1

---

OK, I had to reinstall the Nvidia driver. Now I have all 7,680 cores running. Let's crunch.
Oh and this is really putting out a lot more heat with 3 cards.

OK, I had to reinstall the Nvidia driver. Now I have all 7,680 cores running. Let's crunch.
Oh and this is really putting out a lot more heat with 3 cards.

Glad it was a simple fix.

---

Grant
Darwin NT

Edit- I just notices there's one more thing you can do now to give your output a big boost and that is reserve 1 CPU core to support the GPU.
Your CPU WUs are showing this-
Run time 3 hours 44 min 58 sec
CPU time 2 hours 26 min 17 sec
Which indicates the CPU is really, really struggling. Run time & CPU time should have less than a minutes difference, even on a heavily loaded system.

Today I started trying out my new 2080 Ti on SETI. I noticed this comment about the delta between run time & CPU time. So I was watching mine and CPU time seems to run about 2 minutes less than run time. Does this mean that hardly any compute time occurs on the GPU???
Skimmed the threads and found some suggestions for my app_config. I tried 2, 4, 8, 10 & 16 WUs at once. At 16 (1.0c/0.0625g) it started throwing off a lot of errors so I backed off and at 4 it seems reliable. This is the app_config I'm using:

<app_config>
<app>
    <name>astropulse_v7</name>
    <gpu_versions>
        <cpu_usage>1.0</cpu_usage>
        <gpu_usage>0.5</gpu_usage>
    </gpu_versions>
    <cmdline>-unroll 68 -oclFFT_plan 256 16 256 -ffa_block 12288 -ffa_block_fetch 6144 -tune 1 64 4 1 -tune 2 64 4 1</cmdline>
</app>
<app>
    <name>setiathome_v8</name>
    <gpu_versions>
        <cpu_usage>1.0</cpu_usage>
        <gpu_usage>0.25</gpu_usage>
    </gpu_versions>
   <cmdline>-sbs 1024 -period_iterations_num 1 -high_perf -hp -high_prec_timer -tt 1500</cmdline> 
</app>
</app_config>

I have 40 CPUs so I set Use 90% so that left 4 CPUs uncommitted. I was also running WCG CPU WUs & no other GPU WUs.
I don't understand anything about the <cmdline> syntax, nothing. I just copied it from Keith I think.
I also saw that Zalster uses a different version with app_classes. Is there an advantage to specifying that level of detail???
Would appreciate any help in finding the sweet spot for a 2080 Ti (and also 1080 Ti, 1080, 1070 Ti & 1070). TIA

---

I have 40 CPUs so I set Use 90% so that left 4 CPUs uncommitted.

As long as you reserve a CPU core for each GPU WU being processed, that isn't necessary.
I am using all of my cores & threads & my CPU WU Runtime & CPU time are within 2 minutes or less of each other.
If you use your system for significant work (eg rendering, transcoding etc) while crunching, then leaving a core or 2 free may be of benefit.

I'd suggest using all cores & threads, with a CPU core reserved for GPU WU support and see how things go. Then reserve a core for non-BOINC use if necessary.

So I was watching mine and CPU time seems to run about 2 minutes less than run time. Does this mean that hardly any compute time occurs on the GPU???

No, it means the CPU is supporting the GPU, keeping it fed so to speak.
That is why it's best to reserve a CPU core for each GPU WU that is being processed, as it stops your CPU processing times from being much longer than they should be because it's trying to process a WU on the CPU, and keep the GPU fed as well.
The CPU time & Runtime for GPU WUs can vary considerably, depending on the WU being processed and is not an issue. For CPU WUs it should be as close as possible; the bigger the difference, the more the system is struggling.

Would appreciate any help in finding the sweet spot for a 2080 Ti (and also 1080 Ti, 1080, 1070 Ti & 1070). TIA

I would suggest running just 1 WU at a time to get a rough idea of what the base performance is (the current mix of work does make that more difficult. There was a month or two where we had just the 1 type of WU for days (even weeks) at a time which made comparisons very easy).
With the current SoG application many others have found with high end hardware that 2 or 3 WUs at a time gives the best results- the most work crunched each hour. Myself, I've only ever found 1 WU at a time gives the best results, regardless of what command line values I have used.

I personally like to keep things as simple as possible.
My app_config.xml file

<app_config>
 <app>
  <name>setiathome_v8</name>
  <gpu_versions>
  <gpu_usage>1.0</gpu_usage>
  <cpu_usage>1.0</cpu_usage>
  </gpu_versions>
 </app>
 <app>
  <name>astropulse_v7</name>
  <gpu_versions>
  <gpu_usage>1.0</gpu_usage>
  <cpu_usage>1.0</cpu_usage>
  </gpu_versions>
 </app>
</app_config>

The command line values I use go in the command line file itself- the advantage of this is that as soon as the next WU starts, it uses those values. With anything in app_config.xml you have to use Options, Read config data for it to take effect.

Running stock, the file name is

mb_cmdline-8.22_windows_intel__opencl_nvidia_SoG.txt

and located in the Seti project data folder.

and the values I use for all my cards are

-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

That is for your Windows systems.
For your Linux systems I strongly advise you make use of the optimised application developed by Petrie and refined & packaged by TBar for easier implementation as it will provide roughly 3 times the performance of the SoG application, and it is self-optimising.
I hope your systems are well cooled.
:-)

Tbar's all-in-one

---

Grant
Darwin NT

Today I started trying out my new 2080 Ti on SETI. I noticed this comment about the delta between run time & CPU time. So I was watching mine and CPU time seems to run about 2 minutes less than run time. Does this mean that hardly any compute time occurs on the GPU???
Skimmed the threads and found some suggestions for my app_config. I tried 2, 4, 8, 10 & 16 WUs at once. At 16 (1.0c/0.0625g) it started throwing off a lot of errors so I backed off and at 4 it seems reliable. This is the app_config I'm using:

<app_config>
<app>
    <name>astropulse_v7</name>
    <gpu_versions>
        <cpu_usage>1.0</cpu_usage>
        <gpu_usage>0.5</gpu_usage>
    </gpu_versions>
    <cmdline>-unroll 68 -oclFFT_plan 256 16 256 -ffa_block 12288 -ffa_block_fetch 6144 -tune 1 64 4 1 -tune 2 64 4 1</cmdline>
</app>
<app>
    <name>setiathome_v8</name>
    <gpu_versions>
        <cpu_usage>1.0</cpu_usage>
        <gpu_usage>0.25</gpu_usage>
    </gpu_versions>
   <cmdline>-sbs 1024 -period_iterations_num 1 -high_perf -hp -high_prec_timer -tt 1500</cmdline> 
</app>
</app_config>

I have 40 CPUs so I set Use 90% so that left 4 CPUs uncommitted. I was also running WCG CPU WUs & no other GPU WUs.
I don't understand anything about the <cmdline> syntax, nothing. I just copied it from Keith I think.
I also saw that Zalster uses a different version with app_classes. Is there an advantage to specifying that level of detail???
Would appreciate any help in finding the sweet spot for a 2080 Ti (and also 1080 Ti, 1080, 1070 Ti & 1070). TIA

There is no need to go into detail like my app_config. It's just a throw back to something I've used in the past. Keith's slimmed down version is fine.
The sweet spot for SoG is 5 work units on the 2080Ti. Besure to leave 1 CPU free to support the OS and help feed the GPU.
The commandline you are using is fine. Grant's add more complicated items which you really don't need. He also has a -cpu lock in there which you don't need as that is a throw back to assigning a work unit to a single CPU rather than have it float between thread while crunching. We did that back in the beginning when we crunched more work units then there were actual cores. A work unit would crunch and when a core became free, it would jump to that free core and start again from zero. I believe it was supposed to be coupled with number in instance (referring to how many work units were being crunched on each gpu). You can still use it to make sure that a work unit stays with a specific thread and doesn't migrate but not really necessary as we tell people don't crunch more work units than the number of threads on your CPU.

---