I would point anyone considering the 24/32C TR2 cpu to go read my earlier post in the thread. Message 1955118
and go view the link I posted by buildzoid. He explains the current requirements of the 24/32C cpus and why only one current board is capable of running them properly and at full capabilities.

The only current board that can do justice to the 24/32C cpus is the MSI MEG Creation X399 board.

https://www.newegg.com/Product/Product.aspx?Item=N82E16813144204 And its running $584 at Newegg as well as being an "extended" ATX mb which I believe means I can't fit it into my case.

Sounds like upgrading to a Threadripper2 is off the table for some 6 months.

Maybe I can get another project going :)

Tom

---

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

Depends on the board and the case. I have a SSI CEB form factor workstation X99 board and it is the same size as an EATX board. Fit the same standoffs in the case. Normally the only thing you can get in trouble with a EATX form factor motherboard is occluding the pass-through cable grommets in the case from the power supply. If that isn't an issue, most EATX motherboards fit a ATX case as long as there are EATX standoff positions you can use.

---

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

A proud member of the OFA (Old Farts Association)

Trying to balance the cpu and gpu loads for "faster production" overall.

A couple of different people running in the high end out here have said they appeared to be getting faster total processing by not using all the cpu cores.

What kind of load are the cpu's? Under a smaller Windows machine it isn't using 100% anymore.

When I set my single cpu e5-2670 box to 1 cpu / 1 gpu, it still isn't running "flat out" (I took -nobs off it after it locked up and erred out a lot of gpu tasks).

My dual cpu e5-2670 box when set to 1 cpu / 1 gpu for the CUDA91 tasks also isn't running flat out. It is running with -nobs.

So I have a two questions.
Is there general agreement that running with a couple of cpus idle makes the whole system "faster" for Seti processing?

Is there a preferred way to run a few idle cpu's? I can do it with app_config.xml by assigning 2 cpus to a single gpu task. I expect I can use the local config menu to idle a couple of cpus. Is there another way? Is one preferred over the other?

Am I missing something?

Thank you,
Tom

---

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

Just keep a 3 or 4 of cpu threads free for general desktop housekeeping. Or keep the overall system load around 60-75%. I think it is easiest just to use a max concurrent statement in an app_config file. So count up how many gpu tasks you want to run and add in 50-75% or your cpu threads for a total of how many max concurrent project tasks you want to run. I never have seen a system run very well with best system output with all cpu threads running work units.

---

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

A proud member of the OFA (Old Farts Association)

My gut feeling at this point Tom would be that to get this system to work at it's best, it'd probably be best to hang back a good 3-6 months, or at least till there is some competition in the TR2 optimized board market, and for them to be able to bake for a bit to work out the bugs. I went thru that fire long ago, had the latest-and-greatest-itis bug, though this was when I was doing networking back in the 90's, and I had heartburn for months and months with it. Hard lesson to learn, but for all the thrill of being out there walking on point, remember who it is that often gets picked off first. I'd let others be the heros, taking the slings and arrows, while you hang back till the smoke clears a bit. I know it sucks to wait, but that might let you acquire some more resources to make this thing be even an more of an uber-cruncher than it would be if you started putting it together today. Moar Korz! :-D

---

My gut feeling at this point Tom would be that to get this system to work at it's best, it'd probably be best to hang back a good 3-6 months, or at least till there is some competition in the TR2 optimized board market, and for them to be able to bake for a bit to work out the bugs. I went thru that fire long ago, had the latest-and-greatest-itis bug, though this was when I was doing networking back in the 90's, and I had heartburn for months and months with it. Hard lesson to learn, but for all the thrill of being out there walking on point, remember who it is that often gets picked off first. I'd let others be the heros, taking the slings and arrows, while you hang back till the smoke clears a bit. I know it sucks to wait, but that might let you acquire some more resources to make this thing be even an more of an uber-cruncher than it would be if you started putting it together today. Moar Korz! :-D

I ordered a matched pair of e5-2690 V2's today. Once I get done trying out all the changes I want to see with it on the e5-2670 v1's I will give it a little "boot" while I am waiting for the Threadripper2 project to jell.

Tom

---

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

Just keep a 3 or 4 of cpu threads free for general desktop housekeeping. Or keep the overall system load around 60-75%. I think it is easiest just to use a max concurrent statement in an app_config file. So count up how many gpu tasks you want to run and add in 50-75% or your cpu threads for a total of how many max concurrent project tasks you want to run. I never have seen a system run very well with best system output with all cpu threads running work units.

Using the project_max when you are running 4-5 different projects is going to require me to set a hard limit on each project that totals up to the maximum number of threads I want to actually run.

Tom

---

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

I would like to announce that one of my experiments is coming closer to "success".

The system I set out to replicate is here: https://setiathome.berkeley.edu/show_host_detail.php?hostid=8432395 It is running 26th on the LeaderBoard at a 99,859 RAC. When I started this project the machine was running 101,000 or so and parked on the bottom of the first page.

I set out to replicate Ian&SteveC's 2 Gtx 1060 3GB system. In another post about potential upgrades from the e5-2670 that I had, he suggested a couple of e5-2690 v2's.

These cpus have 10c/20t @3GHz baseline compared to the 2670's 8c/15t @2.6GHz.

My machine appeared to get "stuck" at slightly above 80,000 RAC. After some consultation with Ian&SteveC it is moving again. It is perched at the next step down on the board and is only 7,000 RAC short of the target machine.

Since the graph is still pointed upwards, I am hopeful it "snuggle" right up to 8432395 before it peters out.

I have had no luck with the daily RAC tools that everyone else seems to be able to get to work. I am fully registered but something isn't quite right. Haven't gotten enough support from that forum to figure out what I am doing wrong.

This system is https://setiathome.berkeley.edu/show_host_detail.php?hostid=8571144

Tom

---

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

I would like to announce that one of my experiments is coming closer to "success".

The system I set out to replicate is here: https://setiathome.berkeley.edu/show_host_detail.php?hostid=8432395 It is running 26th on the LeaderBoard at a 99,859 RAC. When I started this project the machine was running 101,000 or so and parked on the bottom of the first page.

I set out to replicate Ian&SteveC's 2 Gtx 1060 3GB system. In another post about potential upgrades from the e5-2670 that I had, he suggested a couple of e5-2690 v2's.

These cpus have 10c/20t @3GHz baseline compared to the 2670's 8c/15t @2.6GHz.

My machine appeared to get "stuck" at slightly above 80,000 RAC. After some consultation with Ian&SteveC it is moving again. It is perched at the next step down on the board and is only 7,000 RAC short of the target machine.

Since the graph is still pointed upwards, I am hopeful it "snuggle" right up to 8432395 before it peters out.

I have had no luck with the daily RAC tools that everyone else seems to be able to get to work. I am fully registered but something isn't quite right. Haven't gotten enough support from that forum to figure out what I am doing wrong.

This system is https://setiathome.berkeley.edu/show_host_detail.php?hostid=8571144

Tom

RAC is an AVERAGE of your daily production over the course of 6 weeks or so. "Daily RAC" is not a thing.

you can get an idea of where your RAC will settle out, by looking at your daily production.

you can see your daily production by clicking either of the "FreeDC" or "BoincStats" buttons on the system page you linked next to the line called "Cross project credit"

heres the link to your Free DC page for that system: https://stats.free-dc.org/stats.php?page=hostbycpid&cpid=36afce90b05ddfaace54771692b8ada2

your total SETI RAC is a bit low becauser it looks like you split up your resources to other projects. I put 100% of mine on SETI

---

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

you can get an idea of where your RAC will settle out, by looking at your daily production.

you can see your daily production by clicking either of the "FreeDC" or "BoincStats" buttons on the system page you linked next to the line called "Cross project credit"

Or better yet just looking at your WU runtimes.

RAC is dependant on whatever Credit New is paying out at the time. Sometimes it pays more, but generally it pays less, and then less, and then less...

---

Grant
Darwin NT

I guess I just never seem to figure out how to use that website properly. Every time I ask the question you or someone else finds the information that apparently was right in front of my nose, but I can't find it :(

Tom

---

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

you can get an idea of where your RAC will settle out, by looking at your daily production.

you can see your daily production by clicking either of the "FreeDC" or "BoincStats" buttons on the system page you linked next to the line called "Cross project credit"

Or better yet just looking at your WU runtimes.

RAC is dependant on whatever Credit New is paying out at the time. Sometimes it pays more, but generally it pays less, and then less, and then less...

I have been comparing current RAC's because I do remember when the machine I am trying to replicate was running a RAC in the 100,000+ range.

Unless I am confused, everyone has been getting tasks that take more time to execute on both the gpu and cpu sides.

Tom

---

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

your total SETI RAC is a bit low becauser it looks like you split up your resources to other projects. I put 100% of mine on SETI

I NNT's the Seti Beta. The Einstein@Home(sp) will only run if I run out of tasks. Its one of the "0" resource setups. It also only runs on the gpu's.

Sometimes I will run out of CPU tasks and still be cheerfully crunching Seti gpu tasks all day long ;) [on tuesday]

Tom

---

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

Unless I am confused, everyone has been getting tasks that take more time to execute on both the gpu and cpu sides.

And the amount of Credit you get for them is less than processing the other WUs on a per hour basis, which is why everyone's RAC has been falling until recently, when more Arecibo work started coming through.

---

Grant
Darwin NT

I guess I just never seem to figure out how to use that website properly. Every time I ask the question you or someone else finds the information that apparently was right in front of my nose, but I can't find it :(

Tom

Your host did 108.2K credits yesterday from FreeDC Stats.

---

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

A proud member of the OFA (Old Farts Association)

I am comparing my 2x1060 system to both of yours from time to time, even tho i run 6gb version with 1 more CU per card, it is not enough to compensate for you guys huge number of cpu threads..

I mostly do a tad below 100k daily, peak into the 100k from time to time..

I am curious how you two run your system? nobs or not? how many threads are used for cpu work?
Right now i am running 14cpu 2gpu 2free , but i will probably experiment

---

I am comparing my 2x1060 system to both of yours from time to time, even tho i run 6gb version with 1 more CU per card, it is not enough to compensate for you guys huge number of cpu threads..

I mostly do a tad below 100k daily, peak into the 100k from time to time..

I am curious how you two run your system? nobs or not? how many threads are used for cpu work?
Right now i am running 14cpu 2gpu 2free , but i will probably experiment

I run that system with the -nobs flag. and have it set to use 90% of the CPU. So i think it runs 4 threads free all the time.

that leaves 28 threads in use, 4 free
2 threads for the GPU tasks (2 GPUs)
26 threads for CPU jobs running

---

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

Woops i meant 12cpu 2gpu 2 free, so it is pretty much similar to yours but scaled down to 16 threads.

---

I am comparing my 2x1060 system to both of yours from time to time, even tho i run 6gb version with 1 more CU per card, it is not enough to compensate for you guys huge number of cpu threads..

I mostly do a tad below 100k daily, peak into the 100k from time to time..

I am curious how you two run your system? nobs or not? how many threads are used for cpu work?
Right now i am running 14cpu 2gpu 2free , but i will probably experiment

I run that system with the -nobs flag. and have it set to use 90% of the CPU. So i think it runs 4 threads free all the time.

that leaves 28 threads in use, 4 free
2 threads for the GPU tasks (2 GPUs)
26 threads for CPU jobs running

I have also gone back to the no -nobs after the conversation elsewhere that basically said it doesn't really make the gpu process much faster. I tried the 90% CPU but it seems to be running better with all cores. Once I get it stable (more or less) on my RAC I will play with the "free core" theory. If it is right for my system, my RAC should go up again.

Tom

---

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

Woops i meant 12cpu 2gpu 2 free, so it is pretty much similar to yours but scaled down to 16 threads.

It would be interesting to see if running 100% cpu changes your results. Also would be interesting to see if running 1 core free changes the results.

Then you could change the available cpu to gpu ratio and see if it changes your results.

I think I remember trying the old standby of 0.33 cpu per 1 gpu and seeing the gpu load drop from the high 90%-100% down to the 30%-40% range so I scrapped that experiment fast. I used the "nvidia-smi" on the command line to find that out.

Tom

---

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