Inexplicably high temps on 5800x and 5900x

[harmattan]

I've got a dilemma where I just can't seem to get my CPU temps under control across multiple CPUs. First, I was running a 5800x (stock) where idle was around 45c and load would hit 90c (limit) after a very short time. Temps spike in the 60cs simply by running a video in a browser.

I thought it may be the CPU as 5800xs are known the run hot, so I got a 5900x -- same high temps, albeit 2-3c cooler idle. I've tried using curve optimizer (-30), and all other settings I've seen that can help with temps. Curve optimizer helps somewhat, but I still hit 90c after about 6-7 minutes of gaming. The only thing that has brought the temps under control is an exceedingly aggressive vcore undervolt (1.23v) and reducting CPU multiplier across all cores to 4.3ghz, or ECO mode which effectively does the same thing -- not optimal.

I've tried two cooling solutions: 1) Adata XPG Levante 240mm AIO, 2) Alpenfoehn Brocken 3 (both using AC5 Silver). Same general results i.e. high temps on both: either of these solutions should handle the 5800x and 5900x well. I running this setup in a Corsair 650D with very good cooling in a well ventilated area.

The only new component I installed since I've seen the high temps is a Gigabyte x570 GamingX motherboard. But I don't think this would be the culprit since I was running a 3800x in it previously at super cool temps.

 

[Zepher]

I use Corsair AIO's with my 5800X and 5900X. I did the vcore voltage offset, I think -0.125v.
encoding with the CPU hits around 75C with the fans set to quiet.

 

[pendragon1]

check you mounting and that youre using enough paste, maybe try another paste too to save a couple more C.

 

[Rvenger]

Get some MX4 and quit using Arctic silver. Also, use voltage offset and not curve optimizer. Make sure pump is running 100% speed in bios. Looks to be an older Asetek design.

 

[harmattan]

Get some MX4 and quit using Arctic silver. Also, use voltage offset and not curve optimizer. Make sure pump is running 100% speed in bios. Looks to be an older Asetek design.

I've tried setting an offset as well, but no improvement. The mounting on both my AIO and Alpenfoehn are solid and squarely affixed. Switching paste isn't going to do much, but I have tried Grizzly as well - no difference.

How do I check pump speed in bios? Didnt know you could do that...

 

[Mackintire]

Try these settings:
PPT 150
TDC 115
EDC 130

You should see a 7c drop on Air before you touch the processor voltage.

 

[Phazer Tech]

It's probably PBO being too aggressive. I disabled it and did an all core overclock, spend the time to tweak PBO till it doesnt randomly crash.

 

[chameleoneel]

I seen the high temps is a Gigabyte x570 GamingX motherboard.

The problem is this.

Gigabyte probably has the bios set to default to aggressive power/voltage/LLC, to gain higher scores on reviews.

Make sure all performance enhancers are turned off. Try to mimick stock settings.

If PBO is turned on, make sure it is set to the default values. Not the "motherboard" setting.
Ryzen mobos usually have two very different places to activate PBO. The setting I am talking about are usually in an advanced menu, where you acknowledge an overclocking warranty void usage agreement----and then you get access to more granular PBO and overclocking settings.

 

[harmattan]

The problem is this.

Gigabyte probably has the bios set to default to aggressive power/voltage/LLC, to gain higher scores on reviews.

Make sure all performance enhancers are turned off. Try to mimick stock settings.

If PBO is turned on, make sure it is set to the default values. Not the "motherboard" setting.
Ryzen mobos usually have two very different places to activate PBO. The setting I am talking about are usually in an advanced menu, where you acknowledge an overclocking warranty void usage agreement----and then you get access to more granular PBO and overclocking settings.

This was it. I traced back the start of the high temps to installing this motherboard. I set PBO to default and used the PPT, TDC, EDC settings above. Damn you Gigabyte.

 

[chameleoneel]

Excellent

 

[freeagentt]

Try these settings:
PPT 150
TDC 115
EDC 130

You should see a 7c drop on Air before you touch the processor voltage.

Not only will it run cool but you will kneecap it’s performance with limits like that.

 

[Phazer Tech]

Not only will it run cool but you will kneecap it’s performance with limits like that.

In what scenarios besides benchmarks? Even things such as video rendering will see minimal impact in performance. Efficiency drops fast once you get past a certain point, and these chips are more or less already at their max efficiency by default.

 

[freeagentt]

In what scenarios besides benchmarks? Even things such as video rendering will see minimal impact in performance. Efficiency drops fast once you get past a certain point, and these chips are more or less already at their max efficiency by default.

You can easily see GFlop output in Linpack.

Define efficiency.

 

[xx0xx]

Motherboard manufacturers maxing out stuff these days is such an annoying development. I hate that the industry has turned to that. My i7-10700 was running quite hot compared to typical temps. Found out that by default the motherboard just maxes/disables all the limits and lets the thing run at PL2 / high temps / high power 100% of the time. I've been able to cut my temps, power usage, etc down a lot while maintaining turbo clock speeds with extremely small performance losses.

Dropping power limits/etc vs. losing performance are not 1:1. Sometimes you can save a ton of power/heat for a small performance loss- at least on the Intel side, but it wouldn't surprise me if it was also true for AMD

 

[Phazer Tech]

You can easily see GFlop output in Linpack.

Define efficiency.

Again, that's a benchmark. There aren't very many real world applications that run your CPU 100% full tilt.

Efficiency as in power efficiency. Performance per watt. With these CPUs if you add 50% more power over the stock settings you'll be lucky to see 20% improvement in benchmarks, and temperatures will be a lot higher too.
It generally isn't worth it unless benchmarks are that important to you. Real world performance will barely be noticeable.

 

[freeagentt]

Again, that's a benchmark. There aren't very many real world applications that run your CPU 100% full tilt.

The point was you can see the effects of cutting down power limits. It does have a benchmark section, does anyone really use it though?

With these CPUs if you add 50% more power over the stock settings you'll be lucky to see 20% improvement in benchmarks, and temperatures will be a lot higher too.

If you can cool it, there should be no problem then, right?

Real world performance will barely be noticeable.

Its noticeable.

 

[Phazer Tech]

Its noticeable.

Which real world applications do you use where it's noticeable?

 

[freeagentt]

Which real world applications do you use where it's noticeable?

Any crunching program such as F@H or WCG, Rosetta.. etc.

 

[TheSlySyl]

I use 100% of all 32 threads when transcoding or rendering on my 5950X pretty often. With my Arctic Liquid Freezer 420 and after some PBO tweaking it stays at around 80'c max.

 

[Mackintire]

PPT 150
TDC 115
EDC 130
These settings will limit your processor to approx 150 watts. That'll keep you at or near 80c with most of the decent air-cooled heatsinks. You can push PPT higher, but your temps will rise as well. With my older Noctura U12 raising the PPT to 165 raised the temperature 8c and resulted in a minuscule 3% performance gain due to thermal throttling. These settings are skewed for higher PBO under lower thread counts as my heatsink is limiting performance at the all core value. Upgrading my heatsink to the latest Noctura would probably only net me another +5% under multi-threaded benchmarks. I'm already hitting 5.0Ghz on a single thread and my score is only lower because I have 64GB installed requiring slightly lax latencies PC3600 CAS 18. If I went with water cooling I could push the system 8% faster but that's with a 35-50% increase in power draw. Personally winning the highest benchmark for a +30-50% in power draw isn't worth it. If I wanted to spend money and actually get a useful performance jump I would have either spent the money on a 5800x3d or a Zen 4 chip and gone straight to using a 250 watt+ capable closed loop water cooled setup.

TheSlySyl is running 50% more cores than I and so yeah he's going to pull at least 50% more power, but he's also going to be able to perform 50% more work. It doesn't make sense trying to cool at 5950x on air.​

 

[freeagentt]

It doesn't make sense trying to cool at 5950x on air.

Why?

 

[Mackintire]

Why?

All your turbo speeds will be thermally limited. As a [H]ardOCP member and gamer type It doesn't make sense to spend $550 on a processor and leave 10%+ of the processor's potential performance on the floor. Unless you don't care about clock speed and are one of those users who are exclusively aiming to run widely scheduled and threaded loads, aka DB or video editing/rendering.

.....Then there's the fact that the 7900X starts to look really good at that price point and should make up the difference fairly quickly.

 

[freeagentt]

All your turbo speeds will be thermally limited. As a [H]ardOCP member and gamer type It doesn't make sense to spend $550 on a processor and leave 10%+ of the processor's potential performance on the floor. Unless you don't care about clock speed and are one of those users who are exclusively aiming to run widely scheduled and threaded loads, aka DB or video editing/rendering.

.....Then there's the fact that the 7900X starts to look really good at that price point and should make up the difference fairly quickly.

Ahh ok. On my 5900X and FC140 I run 240/160/190 and get the full 5150 boost, under Linpack load it will run at 4500MHz. R23 is usually around 4600-4800MHz depending on my ambient.

 

[Mackintire]

On my 5900x I'm running 150/115/130 I see 5Ghz boost with an occasional 5.05Ghz, R23 runs 4.65-4.8 depending on the core. I'll have to run a Linpack to compare. My goal was to find the sweet spot between performance and efficiency.

 

[freeagentt]

On my 5900x I'm running 150/115/130 I see 5Ghz boost with an occasional 5.05Ghz, R23 runs 4.65-4.8 depending on the core. I'll have to run a Linpack to compare. My goal was to find the sweet spot between performance and efficiency.

I snapped this last may, it’s one of the few that I have on my phone unfortunately. Core temp says a min of 0 because I forgot to turn off sleep.

 

[harmattan]

Those PBO settings (PPT 150, TDC 115, EDC 130) are holding down my temps. However, this board now has a frustrating habit of resetting my vcore setting, which I keep at 1.3mv back to stock, thus increasing my temps again to uncomfortable levels.

This Gigabyte x570 is the worst mobo I've used since, dare I say it, the EVGA 680i. At least I can't smell ozone coming off the Northbridge, but it's quirky as heck.

 

[freeagentt]

dare I say it, the EVGA 680i

You may..

I hated that POS board lol. Was great with a dual core, couldn't run my Q6600

Traded it and a hundred bucks for an 8800GTX