PrimeGrid Challenges 2021

[StefanR5R]

it looked like I got some tasks that were very short early on,

llrSOB, like almost all of the other LLR based PrimeGrid subprojects, uses the new validation scheme with one normal task and one short verification task.

it keeps locking up.

LLR exercises the vector units, and RAM. The hardware settings (clocks, voltage, whatnot) for both need to be set conservatively. (That's what I find convenient about using server-type hardware.)

If it could ever stay up, its says 14 hours each, how is that ?

That's super quick for doing two at once. My 22-core E5-2696 v4's are among the rather few Intel processors which are big enough to run two tasks at once, and they need ~22 hours for that, pulling about 210...220 W for the CPU alone. (Estimated based on whole system consumption.)

 

[Markfw]

I appreciate the info @crashtech . I think the short ones are the double checker tasks. I'll keep an eye on mine, I did get a few hours late start so maybe my tasks will start rolling in. It just had me a bit alarmed that I'm barely turning in any tasks.


Can you try 2x8 ? Primegrid doesn't generally respond well to hyperthreading or SMT, so it will probably run better at 2 tasks with 8 threads each.

I paused F@H. The next time it locks up, I will turn off smt and run 2 x 8

 

[StefanR5R]

You can keep SMT enabled, just make sure that BOINC doesn't use more logical CPUs than intended.

 

[crashtech]

oops...

 

[StefanR5R]

I have no doubt my ranking will deteriorate as more work rolls in.

The cloud users may have run times at the order of 2 or 4 days perhaps.

Also I would imagine you are bunkering a bit...

Nope, I just have some computers at TN-Grid still. Will put some more to PrimeGrid later. IMO, 10 days is a long time for a contest at a number theory project, even considering the length of the tasks here.

PS, for effect, I configured my computers for best run times initially, but switched them to best throughput today.

 

[Markfw]

OK, it locked up again. I went into bios. I can NOT find the option to disable SMT. I found the user manual, it says there is a menu "CPU features", but I can NOT find it. MSI MAG x570 Tomahawk wifi.

Until I can find it and turn it off, I am back to WCG and F@H

Oh, and the memory is gskill 4000 running at 3600 and 1800 bus. CPU and voltages are all stock/default.

 

[Justinus]

OK, it locked up again. I went into bios. I can NOT find the option to disable SMT. I found the user manual, it says there is a menu "CPU features", but I can NOT find it. MSI MAG x570 Tomahawk wifi.

Until I can find it and turn it off, I am back to WCG and F@H

Oh, and the memory is gskill 4000 running at 3600 and 1800 bus. CPU and voltages are all stock/default.

Not sure about MSI but on my Asus board the core count, CCD, and SMT control is under the AMD overclocking menu.

 

[StefanR5R]

@Markfw, in the long run it would IMO be worthwhile to get this machine stable under PrimeGrid load (or similar, like Prime95). If it is in a state within which it frequently crashes in PrimeGrid, it may as well randomly miscalculate in other projects for all we know. Though since everything is at defaults, there is nothing sticking out what to try next. Maybe the BIOS or AGESA has still some rough edges.

You can keep SMT enabled, just make sure that BOINC doesn't use more logical CPUs than intended.

Note, I suggested to set program thread count = core count only as a means for throughput optimization, not for stabilization. It is supposed to work stable either way, obviously.

 

[Markfw]

@Markfw, in the long run it would IMO be worthwhile to get this machine stable under PrimeGrid load (or similar, like Prime95). If it is in a state within which it frequently crashes in PrimeGrid, it may as well randomly miscalculate in other projects for all we know. Though since everything is at defaults, there is nothing sticking out what to try next. Maybe the BIOS or AGESA has still some rough edges.


Note, I suggested to set program thread count = core count only as a means for throughput optimization, not for stabilization. It is supposed to work stable either way, obviously.

I have the best power supply with both 8 pin CPU power plugged in. I have memory running UNDER rated speed, and I did not even tighten timings. the bus is under rated speed. The bios is the newest that is not a beta.(7C84V15), and all bios settings are default except the memory at 3600 (instead of rated 4000).

There is no reason for it to be unstable that I can think of, and on WCG its fine. And F@H its fine, both at the same time.

Sorry, I just don't trust that primegrid software, if it can't run on a "under-stock" setup.

 

[Justinus]

I have the best power supply with both 8 pin CPU power plugged in. I have memory running UNDER rated speed, and I did not even tighten timings. the bus is under rated speed. The bios is the newest that is not a beta.(7C84V15), and all bios settings are default except the memory at 3600 (instead of rated 4000).

There is no reason for it to be unstable that I can think of, and on WCG its fine. And F@H its fine, both at the same time.

Sorry, I just don't trust that primegrid software, if it can't run on a "under-stock" setup.

Anything over 3200MHz ram/1600MHz FCLK is technically an overclock on X570. You may not be able to run it on stock voltages for the memory controller, I/O die, or SOC.

 

[crashtech]

There's nothing wrong with Primegrid's software. It is just a bit hard on things and will reveal any weakness.

 

[Justinus]

I'll admit I know nothing about primegrid or boinc but I installed and started it up on my 5950X. Whatever it did, it did automatically - It's running PPS SR2 Sieve in 32 threads, getting around 1% per 10 minutes.

 

[Markfw]

Anything over 3200MHz ram/1600MHz FCLK is technically an overclock on X570. You may not be able to run it on stock voltages for the memory controller, I/O die, or SOC.

OK, I went to the manual. it says this:
▪3rd Gen AMD Ryzen™ Processors support DDR4 1866/
2133/ 2400/ 2667/ 2800/ 2933/ 3000/ 3066/ 3200 Mhz by
JEDEC, and 2667/ 2800/ 2933/ 3000/ 3066/ 3200/ 3466/
3600/ 3733/ 3866/ 4000/ 4133/ 4266/ 4400/ 4533/ 4600+
Mhz by A-XMP OC MODE

I did NOT have A-XMP enabled. So I enabled it. Then I changed to OC advanced, and found SMP and disabled it. Then I reconfigured for 2 x 8 and its running.

Lets see what happens NOW.

 

[Justinus]

OK, I went to the manual. it says this:
▪3rd Gen AMD Ryzen™ Processors support DDR4 1866/
2133/ 2400/ 2667/ 2800/ 2933/ 3000/ 3066/ 3200 Mhz by
JEDEC, and 2667/ 2800/ 2933/ 3000/ 3066/ 3200/ 3466/
3600/ 3733/ 3866/ 4000/ 4133/ 4266/ 4400/ 4533/ 4600+
Mhz by A-XMP OC MODE

I did NOT have A-XMP enabled. So I enabled it. Then I changed to OC advanced, and found SMP and disabled it. Then I reconfigured for 2 x 8 and its running.

Lets see what happens NOW.

Enabling the XMP for voltages is a bit of a hammer but it should work and it's fast to try. Testing and tuning the voltages yourself is a little time consuming but usually you can get by with quite a lot lower voltages than XMP will set. For example on my board the XMP will set 1.2VSOC but I am fully stable at 1.05VSOC.

 

[Icecold]

I'll admit I know nothing about primegrid or boinc but I installed and started it up on my 5950X. Whatever it did, it did automatically - It's running PPS SR2 Sieve in 32 threads, getting around 1% per 10 minutes.

You can of course run any subproject you want to at any time, but for the competition that's going on now if you wanted to participate in it, you would want to go under your Primegrid preferences on the Primegrid website(My Account --> Primegrid Preferences), and only select the Seventeen or Bust LLR project. In that same section you can select how many threads you want use to use, and how many simultaneous tasks. Running 2 tasks at 8 threads per task seems like a good starting point on a 5950x. The Seventeen or Bust tasks seem to take quite a bit of time(12 hours is the lowest I've seen mentioned, most PC's are completing them in much less time than that). LLR tasks(unlike Sieve tasks like what you were running in your post) also have the possibility of finding a very large prime number and getting your name 'in the record book' which is pretty cool.

 

[Markfw]

OK, I went to the manual. it says this:
▪3rd Gen AMD Ryzen™ Processors support DDR4 1866/
2133/ 2400/ 2667/ 2800/ 2933/ 3000/ 3066/ 3200 Mhz by
JEDEC, and 2667/ 2800/ 2933/ 3000/ 3066/ 3200/ 3466/
3600/ 3733/ 3866/ 4000/ 4133/ 4266/ 4400/ 4533/ 4600+
Mhz by A-XMP OC MODE

I did NOT have A-XMP enabled. So I enabled it. Then I changed to OC advanced, and found SMP and disabled it. Then I reconfigured for 2 x 8 and its running.

Lets see what happens NOW.

Nope. Locked up in 9 minutes,.

 

[Justinus]

Nope. Locked up in 9 minutes,.

Have you checked in event viewer for critical/error/warning WHEA messages?

 

[Justinus]

You can of course run any subproject you want to at any time, but for the competition that's going on now if you wanted to participate in it, you would want to go under your Primegrid preferences on the Primegrid website(My Account --> Primegrid Preferences), and only select the Seventeen or Bust LLR project. In that same section you can select how many threads you want use to use, and how many simultaneous tasks. Running 2 tasks at 8 threads per task seems like a good starting point on a 5950x. The Seventeen or Bust tasks seem to take quite a bit of time(12 hours is the lowest I've seen mentioned, most PC's are completing them in much less time than that). LLR tasks(unlike Sieve tasks like what you were running in your post) also have the possibility of finding a very large prime number and getting your name 'in the record book' which is pretty cool.

I got that all setup, 2x8 thread tasks and I set the affinity of each task's threads to the 8 physical cores on each CCD.

It seems to be hitting the CPU really darn hard - it's only running 3.2-3.3 GHz hitting stock current limits.

Considering how hard this particular project looks to be on zen3, @Markfw are you sure there's no auto-PBO/OC funny business happening on your motherboard upping power limits/allowing higher clocks?

 

[Markfw]

I got that all setup, 2x8 thread tasks and I set the affinity of each task's threads to the 8 physical cores on each CCD.

It seems to be hitting the CPU really darn hard - it's only running 3.2-3.3 GHz hitting stock current limits.

Considering how hard this particular project looks to be on zen3, @Markfw are you sure there's no auto-PBO/OC funny business happening on your motherboard upping power limits/allowing higher clocks?

First, I downed the memory to 3200. Second, at 2 x 8 it says it will be 24 hours, and has not locked up in 15 minutes, so that may be it. Even 4000 cas 18 will only run 3200 on this app. I will check the ETA later, as all the locked up time is messing with the estimates. 2% in 30 minutes, so it seems right, but very slowwwwww

Edit: mine is running 4.1 ghz, but it could be that I have excellent cooling (NH-D15)

 

[Justinus]

First, I downed the memory to 3200. Second, at 2 x 8 it says it will be 24 hours, and has not locked up in 15 minutes, so that may be it. Even 4000 cas 18 will only run 3200 on this app. I will check the ETA later, as all the locked up time is messing with the estimates. 2% in 30 minutes, so it seems right, but very slowwwwww

I'm running 3800CL14 tuned timings and it's stable so far for me. Still pretty slow, about 0.09%/minute which should end up around 18-19 hours. For some reason the estimated remaining time is over 6 days, so I don't know what that's about.

I have made sure my memory config is stable though, with a 60-pass memtest86

 

[Markfw]

I'm running 3800CL14 tuned timings and it's stable so far for me. Still pretty slow, about 0.09%/minute which should end up around 18-19 hours. For some reason the estimated remaining time is over 6 days, so I don't know what that's about.

I have made sure my memory config is stable though, with a 60-pass memtest86

Now I am curious... I can't run over 3200 cl18, but you are at 3800 cl14. What motherboard ? We should compare our times on this as the day goes on.

 

[Justinus]

Now I am curious... I can't run over 3200 cl18, but you are at 3800 cl14. What motherboard ? We should compare our times on this as the day goes on.

For sure, it looks like we started this batch of work around the same time - as of 3:47 PST I am 23 minutes elapsed and 2.040%/2.020% complete.

I'm on an Asus Dark Hero X570. I've got a kit of 2x16GB dual rank b-die 4000CL16 G.Skill Trident-Z RGB. I applied the timings from Ryzen Clock Tuner for 3800 fast b-die and then tweaked CL, secondary, and subtimings from there. It's taking 1.5Vmem so I got a memory cooler. The cooler keeps the memory at 35C at the sensors. Memtest will throw errors without it and the memory will heat up to 50C at the sensors.

Before this I was running 3800CL16 from the Ryzen Clock Tuner profile with no tweaking at 1.35 vmem and no cooler stable.

 

[Markfw]

At one hour, I am at 4.5%

 

[Justinus]

At one hour, I am at 4.5%

5.55% at 1 hour.

What actual and effective clockspeeds are you seeing?

 

[Markfw]

6.6 at 1 hour 30 minutes. 4.1 ghz (linux, lscpu is all I have)

Edit: my memory is slower, but its not locking up, so I will let it run !