"Like a breath of fresh air" - my new Ryzen R9 5900X CPU in my main rig.

[VirtualLarry]

It's pretty amazing, seeing 24 CPU thread-charts in Task Manager. Tbat's the most that I've ever personally seen on rigs that I have owned or currently own.

I'm CPU mining (XMR through NH) right now, at 100% CPU load, and my temps have only reached 70C, 75C peak, with a 3 or 4 year old CoolerMaster MasterLiquid Lite 240mm AIO.

(I didn't even change the thermal paste when I installed the new CPU! Shh, Don't tell anyone....)

I have yet to try some PrimeGrid, maybe I will soon, and see if the PC reboots on it's own again.

Memory is 4x8GB Trident RGB DDR4-3800 CAS18-22-22-38 (39?). But it beeps when I cold-boot it, not sure if the RAM is syncing properly, I have it set to XMP.

I have a couple of kits of Team Vulcan DDR4-3200 2x16GB (so 4x16GB total, for 64GB) that I can throw in, and I just might. Might need the extra RAM for DC work with this many cores.

All in all, I went with the 5900X, because it had higher cache-per-core ratios than all other AM4 CPUs, save perhaps the 3900X, but it should have 15-20% higher IPC than the 3000-series CPUs.

My trusty 3600 has served me well, but time to pass it down to another rig.

Edit: Three 8-thread tasks of whatever PrimeGrid project we just had a race on, temps are hitting 80C on my 240mm AIO cooler. Maybe I should change the thermal paste.

 

[VirtualLarry]

This one is weird. Can BOINC "uversubscribe" cores? I had PrimeGrid prefs set to 1 task of 10 cores/threads, had computation suspended, and then allow new task, and I had one 10-core task queued up. Then I went to my Prefs page, changed it to 3 tasks of 8 cores/threads ea., then Updated PrimeGrid in the client, and it downloaded two more tasks, 8-cores ea. I then proceeded to un-suspend CPU computations, and all three tasks resumed. But 8+8+10 = 28 threads, and I'm on a 24-thread CPU. Shouldn't one of those tasks suspend itself?

I'm kind of curious how this is working, now, with modern (newest) BOINC for Windows. Does it allow over-subscribing thread counts? Or is this some sort of artifact on how PrimeGrid applies their limited tasks/treads options in their Prefs.

 

[Ken g6]

Does it allow over-subscribing thread counts?

I dunno about the latest BOINC, but I always use "<ncpus>100</ncpus>" in cc_config.xml. That lets me set the number of CPUs to use instead of setting a percentage. Or over-subscribe temporarily for bunkering or something.

 

[StefanR5R]

I haven't seen the behavior which you describe. But:

'Suspend work -> change settings -> resume work' works nicely if the settings are controlled by a local app_config.xml. (Must be 'suspend to disk', not 'suspend to RAM'.) IME it works only partially if the settings come from the web preferences. In the former case, resumed tasks take over the new thread count per task. In the latter, the application does not, at least not with the client version which I tried at the time.

(1) What the application does, (2) what the client assumes it does, and (3) what boincmgr is displaying are three different things and may be largely unrelated.

You can check on the application by inspecting the stderr.txt files in the slot directories.

On another note:
The LLR based PrimeGrid subprojects which work on larger prime candidates work best on the two-CCX computer which 5900X is if there are two tasks at once, and each task is tied to logical CPUs which belong to the same CCX per task. This latter step is unfortunately not available as a feature in the application or the client, and not very well (or not at all) pursued in the kernels of popular operating systems. Hence, if desired, the user needs to implement it with a third tool. Restricting one task to a single CCX is effective because it avoids cross-CCX synchronizations which are costly on Zen/ Zen2/ Zen3. And in Zen3, each CCX has got a large enough level-3 cache to properly host even the biggest PrimeGrid LLR tasks.

 

[VirtualLarry]

Well, seems things aren't 100% with this new CPU and my rig. Something else may need replacing/upgrading, including the mobo or PSU.

I was getting BSODs, so I swapped out the 4x8GB Trident RGB DDR4-3600 @ XMP 3600 RAM, for 4x16GB Team Vulcan Z DDR4-3200 @ XMP 3200.

Still getting BSODs.

I AM mining, on both GTX 1660 Supers (power-tuned with AB), and all 24T of my CPU. I have Firefox (newest) open constantly, and use ProcessLasso to keep Firefox at "AboveNormal" priority, so it's always responsive, even when mining.

Not sure what's going on, but if it keeps crashing, I may swap out the PSU next.

Edit: Bumped Vdimm to 1.3600V, and set Vsoc to 1.10V.

 

[Icecold]

I would maybe set a PPT limit if you haven't already and see what happens. It'll limit the Monero mining performance somewhat, but will keep PBO from going crazy and using a ton of power, generating excess heat, etc. What motherboard are you using?

 

[Fardringle]

I set the PPT limit on my 3900X to 105 watts (the 'standard' max) instead of letting it boost all the way to 142 watts.

I only lost maybe 5% in benchmark scores, but CPU temps dropped by around 10C and power usage (and therefore electric bills) dropped by a lot. It does mean not being at the absolute top of possible benchmark scores, but the tiny drop in benchmarks is more than worth it to me for all of the gains.

The 5900X isn't exactly the same as my CPU, but it's close enough that I'd expect to see similar results.


What are the error codes on the blue screens? A lot of the time they are not helpful but once in a while they give an error code that is actually meaningful.

 

[Assimilator1]

Congrats on the nice CPU upgrade!
I hope you can sort out the gremlins quickly.

Btw, before I let new CPUs or over/underclocks loose on DC I run a 48hr stress test on OCCT, that usually highlights any problems!

 

[BadThad]

It's pretty amazing, seeing 24 CPU thread-charts in Task Manager. Tbat's the most that I've ever personally seen on rigs that I have owned or currently own.

I'm CPU mining (XMR through NH) right now, at 100% CPU load, and my temps have only reached 70C, 75C peak, with a 3 or 4 year old CoolerMaster MasterLiquid Lite 240mm AIO.

(I didn't even change the thermal paste when I installed the new CPU! Shh, Don't tell anyone....)

I have yet to try some PrimeGrid, maybe I will soon, and see if the PC reboots on it's own again.

Memory is 4x8GB Trident RGB DDR4-3800 CAS18-22-22-38 (39?). But it beeps when I cold-boot it, not sure if the RAM is syncing properly, I have it set to XMP.

I have a couple of kits of Team Vulcan DDR4-3200 2x16GB (so 4x16GB total, for 64GB) that I can throw in, and I just might. Might need the extra RAM for DC work with this many cores.

All in all, I went with the 5900X, because it had higher cache-per-core ratios than all other AM4 CPUs, save perhaps the 3900X, but it should have 15-20% higher IPC than the 3000-series CPUs.

My trusty 3600 has served me well, but time to pass it down to another rig.

Edit: Three 8-thread tasks of whatever PrimeGrid project we just had a race on, temps are hitting 80C on my 240mm AIO cooler. Maybe I should change the thermal paste.

Like you I LOVE seeing 20 threads running on my 10850 I built in April! This one machine is cranking out Rosetta work like crazy.

 

[igor_kavinski]

Not sure what's going on, but if it keeps crashing, I may swap out the PSU next.

So what happened? Problem solved by swapping out PSU?

 

[igor_kavinski]

So what happened? Problem solved by swapping out PSU?

Anybody home???

 

[ericlp]

I'm CPU mining (XMR through NH) right now, at 100% CPU load,

So whats the daily payout for just running the CPU? When the 3090 dropped to around 4 ~ 5 bucks a day at 350 watts, I figured it wasn't worth it. Esp, when you add in the electric bill.