PrimeGrid PPS LLR (CPU temp) challenge (for Ryzen 3000-series owners)! (Testing heat density of TSMC 7nm)

[VirtualLarry]

Ok, I've been tearing my hair out (well, thankfully not quite literally), trying to keep my R5 3600 CPU cool, while crunching 12 threads of PrimeGrid PPS LLR.

At this point, I would like to see other people replicate my setup, and see what their temps are.

I've used both a Gigabyte B450 AORUS PRO WIFI ATX, with both AGESA 1.0.0.2 (BIOS F40) and 1.0.0.3AB (BIOS F41b), and an Asus B450-F ROG STRIX Gaming ATX, with AGESA 1.0.0.2 (BIOS 2501).

I've tried both stock settings, as well as a mild 4.000Ghz manual OC with 1.365V, using the "AMD Overclocking" menu, "manual OC".

I've used both 4x8GB DDR4-3000 @ XMP, as well as 2x8GB GSkill RGB DDR4-3600 (with FCLK 1800 as well), in current setup (Asus B450-F).

I've used four different air coolers, as well as finally, on current rig, 240mm AIO WC (CoolerMaster). Still 97.5C running 12 PrimeGrid tasks. On water. Before the water even got a chance to "heat up", after being idle all night, 44C idle temp.

So if you would like to replicate, if you have a 3600 (or higher) 3rd-Gen Ryzen CPU, head on over to http://www.primegrid.com/ , sign up for an account, and download BOINC, from boinc.berkeley.edu , newest version for Windows, and then "Add Project", put in your PrimeGrid credentials. Note that you should also go to your PrimeGrid account page, on their site, and select "PPS LLR (CPU)" as your project type, and uncheck EVERYTHING ELSE.

Then in BOINC, under "Compute Preferences", make sure that "CPU count" and "CPU time" are both "100".

Edit: Oh yes, you should be running Win10 1903 64-bit, and download and installed the newest (7/7/2019 release or newer) AMD Chipset Drivers, and also, newest AMD Ryzen Master. Measure temps with Ryzen Master.

Edit: Btw, does anyone know, does the Ryzen R5 3600 use TIM/paste, rather than solder, on their lowest-end model? I suspect that they might.

 

[maddie]

Ok, I've been tearing my hair out (well, thankfully not quite literally), trying to keep my R5 3600 CPU cool, while crunching 12 threads of PrimeGrid PPS LLR.

At this point, I would like to see other people replicate my setup, and see what their temps are.

I've used both a Gigabyte B450 AORUS PRO WIFI ATX, with both AGESA 1.0.0.2 (BIOS F40) and 1.0.0.3AB (BIOS F41b), and an Asus B450-F ROG STRIX Gaming ATX, with AGESA 1.0.0.2 (BIOS 2501).

I've tried both stock settings, as well as a mild 4.000Ghz manual OC with 1.365V, using the "AMD Overclocking" menu, "manual OC".

I've used both 4x8GB DDR4-3000 @ XMP, as well as 2x8GB GSkill RGB DDR4-3600 (with FCLK 1800 as well), in current setup (Asus B450-F).

I've used four different air coolers, as well as finally, on current rig, 240mm AIO WC (CoolerMaster). Still 97.5C running 12 PrimeGrid tasks. On water. Before the water even got a chance to "heat up", after being idle all night, 44C idle temp.

So if you would like to replicate, if you have a 3600 (or higher) 3rd-Gen Ryzen CPU, head on over to http://www.primegrid.com/ , sign up for an account, and download BOINC, from boinc.berkeley.edu , newest version for Windows, and then "Add Project", put in your PrimeGrid credentials. Note that you should also go to your PrimeGrid account page, on their site, and select "PPS LLR (CPU)" as your project type, and uncheck EVERYTHING ELSE.

Then in BOINC, under "Compute Preferences", make sure that "CPU count" and "CPU time" are both "100".

Edit: Oh yes, you should be running Win10 1903 64-bit, and download and installed the newest (7/7/2019 release or newer) AMD Chipset Drivers, and also, newest AMD Ryzen Master. Measure temps with Ryzen Master.

Edit: Btw, does anyone know, does the Ryzen R5 3600 use TIM/paste, rather than solder, on their lowest-end model? I suspect that they might.

I think you overthinking it all. Basic logic goes a long way.

If no cooler tested works, including a water setup, AND the CPU is running hot, then the full heat load is NOT getting to the IHS.

Try and see if the motherboard around the socket is at elevated temps. A lot of the heat can be transmuted thought that path if the temp delta is high.

As to the TIM:
They have the expense of a separate assembly line for R5 3600 just for the (cents) in paste savings.

 

[IEC]

The entire Ryzen 5/7/9 lineup is soldered. They don't use TIM on the R5 3600. This was confirmed by der8auer who delidded one. Someone already linked you the video in a different thread.

Maybe you should take a picture of your setup so people can see if you're missing something.

 

[VirtualLarry]

If no cooler tested works, including a water setup, AND the CPU is running hot, then the full heat load is NOT getting to the IHS.

This is why I need some volunteers to run THE SAME SOFTWARE, and see if they get THE SAME or DIFFERENT results. If it turns out, that they get THE
SAME results, then this is a systemic "defect" in the CPU, or at least, "a furmark equivalent" / power-virus type app, that hits 7nm CPUs especially hard because of the heat density. But if no-one else sees the heat issues that I'm seeing, then the problem would seem to be limited to my sample. ("I got a dud.")

 

[maddie]

This is why I need some volunteers to run THE SAME SOFTWARE, and see if they get THE SAME or DIFFERENT results. If it turns out, that they get THE
SAME results, then this is a systemic "defect" in the CPU, or at least, "a furmark equivalent" / power-virus type app, that hits 7nm CPUs especially hard because of the heat density. But if no-one else sees the heat issues that I'm seeing, then the problem would seem to be limited to my sample. ("I got a dud.")

Just check and see if your motherboard next to the socket is getting very hot. Why is that so hard to do before going to all these convoluted tests?

 

[GhostPirateLeChuck]

Assuming I picked the right project, I predominately saw temps hovering between 65/66C. I briefly saw the temp jump to 80C when I shut down BOINC.

I've attached the before and after (I just ran that project for 30min, since I need to get working anyway).

FWIW, here's the hardware:
3900X
X570 Aorus Master (F5G BIOS)
64GB Vengeance LPX 3200CL16
NH-D15 CPU cooler (used the include Noctua thermal paste)
Stock Meshify S2 case with multiple NF-F12 fans


I disabled PBO in the BIOS, but everything else is as close to stock as possible. I could tinker with case fan placement and remove the front and top air filters on the case to possibly bring temps down further, but 66C is OK.

When all cores were in use, the 3900X never managed to hit 3.8GHz (it got close), but that initial 3xx Sieve project BOINC automatically loaded before I could finish setting things up spiked the temps to 80C, and the cores in use were hitting 4.3GHz.


EDIT: In all honesty, I've seen higher temps just loading up AT in Chrome/FF/Edge.

EDIT 2: Ran that Creator Mode stress test in RM a second ago. Temp stayed around 68C, and all cores were running at 4.1GHz for the duration.

 

[VirtualLarry]

And what about comments like this? This may be a more wide-spread problem than we realize.
https://forums.anandtech.com/threads/amd-ryzen-3000-builders-thread.2565687/page-48#post-39877196

I see too many times, when my CPU is high 80, low 90s C temperatures without stress testing
I wonder how long will it work

Remember, running 12 WCG tasks, only got to 78-81C on water. But 12 PrimeGrid PPS LLR tasks? 100C+!

I'm not sure why I'm getting such push-back on this, this "ain't my first rodeo", I've used AIO WC kits before (specifically the CM), I've built dozens of Ryzen systems, I know stuffing the PC in a cubby isn't the greatest, but when I get similar / same temps, between four different air heatsinks, and WC 240mm, something's up.

I know that when I disassembled the previous rig, with the Gigabyte B450 AORUS PRO WIFI, and the Gammaxx 400 heatsink, the mobo was kind of warm, the plastic side mounts were too.

 

[Kenmitch]

Seems like an unreasonable request when a person can get a quicker idea of their temps compared to others. I guess it's possible your onto something, but it's also possible it's just your cpu is a dud?

Using the latest version of Aida64 system stress tester selecting all but the stress system disc and gpu{s} options what do your temps do? I tested this with the stock cooler on my 3600 and they would climb up to 90c within a minute or two.

Humor me and try it out.

My 3600 is running stock. According to cpuz it's currently fluctuating between 3974MHz - 4024MHz @1.360v's and the temp is holding at 47c and I'm about 20 minutes into it. I'm only using air cooling so I'd imagine your temps would be lower or very similar to mine.

Not the same load your requesting, but it's good enough to get a temp comparison.

Edit: If you try Aida64 make sure it's reading the cpu diode temps under the options. I had to go back into preferences and enable mine as it was just reading the cpu temp. Not sure what cpu temp is? Maybe the socket? CPU diode temp looked to be what I'd expect as it was 74c @ 45 minute mark.

 

[VirtualLarry]

I disabled PBO in the BIOS, but everything else is as close to stock as possible. I could tinker with case fan placement and remove the front and top air filters on the case to possibly bring temps down further, but 66C is OK.

When all cores were in use, the 3900X never managed to hit 3.8GHz (it got close), but that initial 3xx Sieve project BOINC automatically loaded before I could finish setting things up spiked the temps to 80C, and the cores in use were hitting 4.3GHz.

Thanks for taking the time.

I guess I forgot one step, after installing BOINC, choose "View", "Advanced", and that will get you the full complement of menus and options, not that simplified view. Once you do that, click on "Options", "Compute Preferences...", and then under the "Computing" tab, make sure that under "Usage Limits", both boxes show "100" (%). Click "Save", then click on the "Tasks" tab, (under the Advanced view), and then on the left side, click "Show Active Tasks", and verify (count) how many active PrimeGrid tasks that you have. You should have one per thread of your CPU.

If that all checks out (24 threads?), and your temps are indeed only 66C, then I clearly do have a problem with my chip. Would this be grounds for an RMA, even though I've mildly overclocked it? (4000Mhz fixed OC, 1.365V), I tried 4100Mhz and maybe up to 1.400V in the Gigabyte board, but it wouldn't boot.

Another way to verify, is bring up Task Manager in Win10, click on "Performance" tab, right-click on the graph, "Show Kernel Times", and then right-click on it again, and select "Change Graph to" -> "Logical Processors". Then see if all of the graphs are "full up".

Edit: I noticed that you're using an X570 board, I wonder how much that factors in, as far as "mature" BIOS, versus all of the beta-esque BIOS releases that we've been getting for the older B450 and X370 boards.

 

[Abwx]

And what about comments like this? This may be a more wide-spread problem than we realize.
https://forums.anandtech.com/threads/amd-ryzen-3000-builders-thread.2565687/page-48#post-39877196



.

People who experience problems should post screen shots if they want accurate answers, with HWinfo, CPUZ and other Ryzenmaster, i have a hard time understanding that the only info is often textual, in an era of computing and internet.....

 

[GhostPirateLeChuck]

Using the latest version of Aida64 system stress tester selecting all but the stress system disc and gpu{s} options what do your temps do? I tested this with the stock cooler on my 3600 and they would climb up to 90c within a minute or two.

Wow! I just tried it, and averaged 81C, with spikes up to 90C, and I only let it run for 3 min.

 

[Kenmitch]

Wow! I just tried it, and averaged 81C, with spikes up to 90C, and I only let it run for 3 min.

Quick and easy, no hoops, and we can compare temps.

On another note I'm into it for 35 minutes @48c

Edit: Looks like AIda64 is viewing some other cpu temp maybe socket? I had to enable the cpu diode in the temperature preferences to get what looks like the correct temps.

 

[VirtualLarry]

EDIT 2: Ran that Creator Mode stress test in RM a second ago. Temp stayed around 68C, and all cores were running at 4.1GHz for the duration.

I don't see any Stress Test in my Ryzen Master. Where is it located? I clicked "Creator Mode" on the left, but I don't see any "Stress Test" options or buttons.

Edit: I see the options to configure "Stress Test" under Settings on the left left, but where is the button to actually invoke it?

 

[GhostPirateLeChuck]

Thanks for taking the time.

I guess I forgot one step, after installing BOINC, choose "View", "Advanced", and that will get you the full complement of menus and options, not that simplified view. Once you do that, click on "Options", "Compute Preferences...", and then under the "Computing" tab, make sure that under "Usage Limits", both boxes show "100" (%). Click "Save", then click on the "Tasks" tab, (under the Advanced view), and then on the left side, click "Show Active Tasks", and verify (count) how many active PrimeGrid tasks that you have. You should have one per thread of your CPU.

If that all checks out (24 threads?), and your temps are indeed only 66C, then I clearly do have a problem with my chip. Would this be grounds for an RMA, even though I've mildly overclocked it? (4000Mhz fixed OC, 1.365V), I tried 4100Mhz and maybe up to 1.400V in the Gigabyte board, but it wouldn't boot.

Another way to verify, is bring up Task Manager in Win10, click on "Performance" tab, right-click on the graph, "Show Kernel Times", and then right-click on it again, and select "Change Graph to" -> "Logical Processors". Then see if all of the graphs are "full up".

Edit: I noticed that you're using an X570 board, I wonder how much that factors in, as far as "mature" BIOS, versus all of the beta-esque BIOS releases that we've been getting for the older B450 and X370 boards.

I found the tip on using Advanced View to change the compute settings before I ran the test. ; )

I also double-checked Resource Monitor, and made sure all 24 "CPUs" were pegged at 100%.

 

[GhostPirateLeChuck]

It's the "Apply & Test" button at the bottom. After you click that, it'll show "Running Stress Test", with the normal meters still semi-visible behind that.

 

[VirtualLarry]

Using the latest version of Aida64 system stress tester selecting all but the stress system disc and gpu{s} options what do your temps do? I tested this with the stock cooler on my 3600 and they would climb up to 90c within a minute or two.

I went to aida64.com, downloaded "Aida64 Extreme" (Trial edition), ZIP version, unzipped it, and ran it, and I keep getting BSODs, with "WDF_VIOLATION". (Windows Driver Framework? Foundation? Violation?)

I'm running Win10 1903, any ideas how to get Aida64 to actually run? (It displays the splash screen, but then BSODs.)

Note that if I run CrystalDiskInfo on this rig, it will hard-hang the system as well. Something to do with the AMD NVMe RAID drivers.

Edit: Thanks, I ran the Creator Mode Stress-Test, if I left the selector on top on "Default", then temps were 65-66C, clock speed was a little over 4000, 4040Mhz. If I clicked over to the right, to "Manual", and set each core to 4000Mhz, and ran the stress-test, temps were basically.

 

[VirtualLarry]

People who experience problems should post screen shots if they want accurate answers, with HWinfo, CPUZ and other Ryzenmaster, i have a hard time understanding that the only info is often textual, in an era of computing and internet.....

I did post a screen shot. :|

 

[VirtualLarry]

Quick and easy, no hoops, and we can compare temps.

On another note I'm into it for 35 minutes @48c

You're running a DVID undervolt, right? According to the guy from AMD that posted on Reddit (was in the builder or speculation thread), you will get lower temps with an undervolt, and it won't show in the clock speed, but you'll also get lower performance. Check out CB scores, with and without the under-volt.

 

[Kenmitch]

You're running a DVID undervolt, right? According to the guy from AMD that posted on Reddit (was in the builder or speculation thread), you will get lower temps with an undervolt, and it won't show in the clock speed, but you'll also get lower performance. Check out CB scores, with and without the under-volt.

Nope. I went back thru my uefi and made sure everything except my memory was on the stock settings. Vcore is on auto.

Decided to just run stock util MSI updates the uefi. Everything looks to work on the launch update, but waiting for and update before I play around with the chips PBO, etc.

Your Aida64 bsod is puzzling. Google search shows that BSOD is most likely caused by a bad driver.

 

[GhostPirateLeChuck]

I used Aida64 Extreme in Win10 1903, and didn't get a BSOD or a CTD, either. Not sure what the deal is there. The reported CPU and CPU diode temps in it were off by just 1 degree from each other in my tests.

 

[IEC]

Your screenshot is showing you are giving your CPU vcore of 1.395V-1.417V rather than 1.365V (already too high for 7nm). You also have PPT, TDC, and EDC way above stock limits.

What happens if you go to BIOS, load optimized defaults, and reset your CPU to stock settings?

Relevant parts of screenshot snipped out:

 

[VirtualLarry]

Your Aida64 bsod is puzzling. Google search shows that BSOD is most likely caused by a bad driver.

I was just generally assuming that the AMD NVMe driver, which didn't play nicely with CDI, was the culprit, so I pulled out a SATA SSD, made a fresh USB stick with Win10 1903 (updated 7/9/2019, btw, so this is a newer build than initial 1903 MCT), and installed AIDA64 Extreme, Ryzen Master, AMD newest chipset drivers, and I left the 4000Mhz 1.365V OC.

What happens if you go to BIOS, load optimized defaults, and reset your CPU to stock settings?

See third pic.

Edit: Now Ryzen Master is reporting ZERO clockspeed, for all cores. Are we dead? WTF? (See fourth pic.)

Edit: Idle temps after a reboot of 55C, on water.

 

[GhostPirateLeChuck]

Your screenshot is showing you are giving your CPU vcore of 1.395V-1.417V rather than 1.365V (already too high for 7nm).

Huh. I just checked mine (it's definitely not at idle right now, but I'm not pushing it hard), and with basically everything at stock except enabling SVM and an XMP profile, and disabling PBO, the vcore on this 3900X is pulling more than that:

 

[IEC]

Huh. I just checked mine (it's definitely not at idle right now, but I'm not pushing it hard), and with basically everything at stock except enabling SVM and an XMP profile, and disabling PBO, the vcore on this 3900X is pulling more than that:

View attachment 8717

I should clarify I am talking about all-core OC voltages. The "safe" 24/7 OC voltage for all-core or per-CCX overclocks is thought to be ~1.325V based on a comment Robert Hallock made on the AMD subreddit. Under stock behavior it's not uncommon to see voltages up to 1.500V for transient ST/lightly threaded workflows. Which as stock behavior is SAFE per AMD.

Of course no one knows what's truly safe for a 24/7 OC yet so we'll have to wait to see what the more adventurous pull with 1.35V+ all-core or per-CCX OCs and if that has any degradation down the road.

 

[GhostPirateLeChuck]

Ah! Cool. I was starting to wonder if I should be very concerned.

 

[IEC]

AVX2 loads will heat up your CPU quickly. Throwing > 1.36V at your chip without adequate cooling is an exercise in futility. Especially because your maximum achievable frequency almost stops scaling beyond 1.3V (without exotic cooling).

3900X at 1.25V fixed with full AVX2 load. The difference in chiplet binning becomes readily apparent when your best chiplet does +150MHz AVX2 versus the other... this is with a 27°C (81°F) ambient.