5800X3D temperature sanity check

[Stahlmann]

I just upgraded from a 5600X  to a 5800X3D today and saw significantly higher CPU temperatures, which is something I didn't expect to this degree, seeing how I'm running a full custom loop with 2x 360 mm radiators. Here are my operating temperatures during a few different load scenarios. Can any of you other X3D owners confirm if they're as expected?

 

Idle (~2% utilization): 42°C

Light gaming load (~30% utilization): ~60°C

CPU-heavy gaming load (~70% utilization): ~70°C

Cinebench R23: (100% utilization): ~90°C

 

Especially the Cinebench temperature is what sounded my alarm clocks. Even though it hovers around 90°C, according to HWiNFO64 it doesn't thermal throttle and stays around 4.2 Ghz. All cores hover around 85-90°C when running cinebench. So is this expected behavior?

 

(Yes, I double checked my waterblock mount and the thermal paste application. I also remounted once, showing that the first mount had good thermal paste spread and contact. At the point of testing I was using the default BIOS settings, not even DOCP enabled.)

[bal723]

maybe its rising it's voltage by too much?

[Stahlmann]

9 minutes ago, bal723 said:

maybe its rising it's voltage by too much?

I was using the "optimized default" BIOS settings at the point of testing, so it should behave according to spec.

[Hellowpplz]

1 hour ago, Stahlmann said:

I just upgraded from a 5600X  to a 5800X3D today and saw significantly higher CPU temperatures, which is something I didn't expect to this degree, seeing how I'm running a full custom loop with 2x 360 mm radiators. Here are my operating temperatures during a few different load scenarios. Can any of you other X3D owners confirm if they're as expected?

 

Idle (~2% utilization): 42°C

Light gaming load (~30% utilization): ~60°C

CPU-heavy gaming load (~70% utilization): ~70°C

Cinebench R23: (100% utilization): ~90°C

 

Especially the Cinebench temperature is what sounded my alarm clocks. Even though it hovers around 90°C, according to HWiNFO64 it doesn't thermal throttle and stays around 4.2 Ghz. All cores hover around 85-90°C when running cinebench. So is this expected behavior?

 

(Yes, I double checked my waterblock mount and the thermal paste application. I also remounted once, showing that the first mount had good thermal paste spread and contact. At the point of testing I was using the default BIOS settings, not even DOCP enabled.)

Your cpu is almost running like a intel laptop chip in a system with crappy cooling
Okay, but seriously, I believe it's max wattage is 140W, vs your old chips 65W, so a jump is expected. Plus it's just known to get hot.

 

https://community.amd.com/t5/processors/5800x3d-instant-90-c-under-stress-but-no-heat-dumped/td-p/629326#:~:text=Yes I know that the,as max for safe operations.

 

Plus the x3d chips all run hot

[podkall]

1 minute ago, Hellowpplz said:

Your cpu is running like a intel laptop chip in a system with crappy cooling

that is good, because intel laptop would be running at 0,5Ghz choking while Ryzen runs at 4+Ghz

[bal723]

5 minutes ago, Stahlmann said:

I was using the "optimized default" BIOS settings at the point of testing, so it should behave according to spec.

some motherboards increase voltage significally

[Hellowpplz]

1 hour ago, podkall said:

that is good, because intel laptop would be running at 0,5Ghz choking while Ryzen runs at 4+Ghz

Well now that's just offensive

It still runs like an oven, but it doesn't know when to give out. Runs at a minimum of 3.3ghz constantly, if not it's max of 4.2ghz.

@podkallall at the expense of battery life. It's a 30W chip and when pushed that far gives me a whole ~75 Minutes of battery

[Stahlmann]

1 minute ago, bal723 said:

some motherboards increase voltage significally

So what is considered normal?

 

 

This is the voltage readout during a cinebench run:

 

[bal723]

5 minutes ago, Stahlmann said:

So what is considered normal?

 

 

This is the voltage readout during a cinebench run:

 

idk i just know that if cpu is overheating (which it does) its often either voltage or bad cpu design

[Ripred]

I mean I get those kinds of temps but I only have a 212 evo air cooler...I would double check your loops and water block seating, also set negative in pbo minus 20 and test, if that looks good try minus 30

[WereCat]

Yes, that's normal. The limit is not the cooler but the cooler contact and mounting pressure. 

[YoungBlade]

The X3D chips will run hot. This is due to their nature: the 3D V-Cache is stacked on top of the CPU core complex. This means that, to transfer heat to the IHS, it first must transfer through the cache, which itself is hot.

 

Basically, you've got an extra layer between the cooling and the cores. This is the main reason AMD does not allow overclocking for the 5800X3D - people would be cooking their chips to death left and right. You just can't get rid of the heat fast enough.

 

As long as the CPU isn't thermal throttling, it's fine. It will be hot, but it's not in any danger.

 

If you want to reduce thermals, and potentially improve performance, you can try undervolting the chip.

[Agall]

1 hour ago, Stahlmann said:

I just upgraded from a 5600X  to a 5800X3D today and saw significantly higher CPU temperatures, which is something I didn't expect to this degree, seeing how I'm running a full custom loop with 2x 360 mm radiators. Here are my operating temperatures during a few different load scenarios. Can any of you other X3D owners confirm if they're as expected?

 

Idle (~2% utilization): 42°C

Light gaming load (~30% utilization): ~60°C

CPU-heavy gaming load (~70% utilization): ~70°C

Cinebench R23: (100% utilization): ~90°C

 

Especially the Cinebench temperature is what sounded my alarm clocks. Even though it hovers around 90°C, according to HWiNFO64 it doesn't thermal throttle and stays around 4.2 Ghz. All cores hover around 85-90°C when running cinebench. So is this expected behavior?

 

(Yes, I double checked my waterblock mount and the thermal paste application. I also remounted once, showing that the first mount had good thermal paste spread and contact. At the point of testing I was using the default BIOS settings, not even DOCP enabled.)

The 3D v-cache mod effectively adds a +25C insulator on top of the die. Something I've tested with the 5800x3D and my 7950x3D. Reaching the thermal limit at full power without using a phase change or direct die is expected.

[Stahlmann]

Now that I've applied a PBO2 curve optimizer offset of 30 (max value) in my BIOS, the temperatures are a bit more reasonable and it's still stable. It doesn't get as hot under light and medium loads and boosts higher in all situations. A 100% all-core load like Cinebench will still shoot it to 90°C in an instant, though. But if that's expected, I guess I don't mind.

 

(I had to change my motherboard due to another problem, so I've remounted the cooler 3 times now. I'm absolutely certain that the cooler mounting is not the problem at this point.)

 

Maybe I just have a "bad" chip when it comes to the silicon lottery, because even with the voltage offset and the resulting indirect overclocking, I'm still missing almost 1000 points in Cinebench compared to what reviewers get out of the box.

 

My all-core score out of the box was 12500. With the curve optimizer it comes up to about 13700. All reviewers I looked up posted scores of >14000 out of the box.

[freeagent]

Lol..

 

Well, you guys might be used to seeing temps like that. But my 58X3D idles in the 20s and R23 barely brings it over 60.

On 5/22/2024 at 9:12 AM, Agall said:

The 3D v-cache mod effectively adds a +25C insulator on top of the die

Nope. Not even remotely close.

 

Wait a minute.. are your running it at stock? If you access the curve and bring all cores down to say -25 that should help you quite a bit. If not, then your cooling situation sucks.

[Stahlmann]

56 minutes ago, freeagent said:

Lol..

 

Well, you guys might be used to seeing temps like that. But my 58X3D idles in the 20s and R23 barely brings it over 60.

Nope. Not even remotely close.

 

Wait a minute.. are your running it at stock? If you access the curve and bring all cores down to say -25 that should help you quite a bit. If not, then your cooling situation sucks.

What cooler are you using?

 

Like I said, i double and triple checked the mount/contact. My 5600X was cooled just fine before and it has just slightly lower heat output.

Afaik the 5600X and 5800X3D should share the same heatspreader design and cooler support.

 

The lower R23 scores would suggest I'm thermal throttling under all-core load...

 

Doing a little more research it seems that the X3D chips run hot, but hitting 90°C still isn't the norm, not even with air coolers. So I shouldn't hit these kinds of temps, especially not with a custom water loop and 720 mm worth of radiators.

[Agall]

1 hour ago, freeagent said:

Lol..

 

Well, you guys might be used to seeing temps like that. But my 58X3D idles in the 20s and R23 barely brings it over 60.

Nope. Not even remotely close.

 

Wait a minute.. are your running it at stock? If you access the curve and bring all cores down to say -25 that should help you quite a bit. If not, then your cooling situation sucks.

Its as simple as comparing multiple different 7950x3D's CCD0 vs CCD1 temperature, including ones that are direct-die cooled. Also considering something like a 7800x3D vs 7700x at the same wattage and clock speed, and seeing the differences in temperature, all comes out to about +25C. I'll make it even clearer and say that its +25C under full wattage draw and boost clock.

 

If you don't understand why that is, I would recommend looking into thermodynamics. A 5800x3D in the 60C range at full load sounds like its not under at max wattage, potentially a severe undervolt to the point of less performance. I would recommend verifying your undervolt with Cinebench R23.

 

In my case, I can undervolt my 7950x3D to the extreme and get more performance with the same thermal result, since I've been able to get 37531 in R23 while air cooling, which is almost into 250W draw 7950x territory. That's with direct-die though, which removes several millimeters of heat transfer layers. They're still TDP limited either way.

[Agall]

28 minutes ago, Stahlmann said:

What cooler are you using?

 

Like I said, i double and triple checked the mount/contact. My 5600X was cooled just fine before and it has just slightly lower heat output.

Afaik the 5600X and 5800X3D should share the same heatspreader design and cooler support.

 

The lower R23 scores would suggest I'm thermal throttling under all-core load...

 

Doing a little more research it seems that the X3D chips run hot, but hitting 90°C still isn't the norm, not even with air coolers. So I shouldn't hit these kinds of temps, especially not with a custom water loop.

That is the case, yes. When I had a 5800x3D, I tested a Hyper 212, an EKWB 280mm AIO, and an NH-D15. The 3D v-cache mod is effectively a +25C insulator on the CCD.

[Stahlmann]

1 hour ago, Agall said:

That is the case, yes. When I had a 5800x3D, I tested a Hyper 212, an EKWB 280mm AIO, and an NH-D15. The 3D v-cache mod is effectively a +25C insulator on the CCD.

So your 5800X3D also ran into the thermal limit, no matter the cooler?

 

It's still weird that there are so many reports of people having it max out under 80°C even when running stock, so no undervolt.

 

It's also weird that I need an undervolt to even come close to what most reviewers score with stock settings. And that I lack around 2000 points when running stock.

 

These results were not at all what I envisioned when investing in a full-on custom water loop a few years ago.

[freeagent]

1 hour ago, Agall said:

In my case, 

But this is about 58X3D and AM4..

[Agall]

38 minutes ago, freeagent said:

But this is about 58X3D and AM4..

3D v-cache between 5000x3D and 7000x3D have no difference I'm aware of, which is what I'm speaking of specifically. We have no way of comparing a non-3D and a 3D CPU on the same chip since there's no R9 5000x3D.

 

That could be done by simply recreating the same thermal conditions and testing a 5800x vs 5800x3D at a fixed wattage, voltage, and clock speed, but its still possible to have differences between the chip. 

 

In comparison, a 7950x3D direct-die cooled would eliminate the most amount of variables when determining the maximum draw difference in temperature between a 3D v-cache CCD and normal CCD since it has one of each.

 

Example, my direct-die 7950x3D with an NH-D15 versus @PDifolco with a PA120SE:

 

 

I'd say given the circumstances, having similar approximate deltas between CCD0 and CCD1 in these scenarios is quite definitive. I grossly approximated with my +25C 'rule', so if you want to extrapolate this dataset to find a more accurate value based on different core temps average/minimum/maximums, then please, I'm terrible at statistics  

 

Noting that my 7950x3D is on the left, which might be a better dataset since the latter is a bit buggy with those 0C readings. This is just a snip from this screenshot, where I reset the dataset immediately after hitting 'start' and screenshot immediately after it finishes.

 

 

 

[freeagent]

At 143w PPT on my X3D I see under 80.. I am not running it right now but I can plug it in when I get home.

 

R23 is a pretty light load.

[Stahlmann]

51 minutes ago, freeagent said:

At 143w PPT on my X3D I see under 80.. I am not running it right now but I can plug it in when I get home.

 

R23 is a pretty light load.

The fact that you said it idles in the 20's makes it pretty hard to believe either way. You can never cool a processor under ambient. I my experience not even close to ambient, really.

 

How you get your CPU to go to 143W PPT is also weird. This CPU is completely locked down to 105W afaik. The top PPT recorded during my 10 minute Cinebench loop was a bit over 106W.

 

 

 

And just to verify again, I just did another final remount. This time I applied the thermal paste with the pea method instead of manually spreading it out and it helped a bit. Now the CPU doesn't instantly go to 90°C and starts more around 80°C when running CinebenchR23, only creeping to 90°C after looping for approx. 10 minutes. And the score increased to 14700, so now I'm at a point where it at least performs as expected, albeit with the -30 voltage offset already applied.

[freeagent]

Alright I'm going to plug it in

[freeagent]

All done.

 

Cinebench is weak.

 

If you want to see the limits, you need to run something like Linpack. Prime95 wont do it.