Ryzen 7 3800xt Prime95 Temps

jandcando · Jul 25, 2022

Hello, I recently upgraded from the stock cooler to the Noctua NH-U14S which claims on their site to be in the "best turbo/overclocking headroom" category for this processor and I was wondering if my stress test temps are normal or something to warrant re-applying thermal paste. Installing this new cooler helped a lot for my idle temps (from 50c to 40c, haven't touched fan curve yet). It also helped a lot for my Cinebench stress test (held at 85c before, now maxes out at 74c). These are all the non-overclocked temps.

Then I discovered the world of overclocking and wanted to see if I could do some more stress tests to validate the stability. I tried out Prime95 and WOW did my temps and power consumption skyrocket. I was using 120 watts and the temps jumped to 94c and crept up the longer I let the test run, all the way to about 96c where I got nervous and shut it off.

Is this alright? For further reference I'm now at 4300 MHz where the normal for this CPU is 3900 MHz. My VDDCR is at 1.31875 volts. I was hoping to push a little further than this since my UserBenchmark scores haven't really budged after all this, but I am starting to gather that's not a great benchmark tool from reading other forums.

Any help, input, advice, or just anecdotal experience is very appreciated!

MishelLngelo · Jul 25, 2022

jandcando said:
Hello, I recently upgraded from the stock cooler to the Noctua NH-U14S which claims on their site to be in the "best turbo/overclocking headroom" category for this processor and I was wondering if my stress test temps are normal or something to warrant re-applying thermal paste. Installing this new cooler helped a lot for my idle temps (from 50c to 40c, haven't touched fan curve yet). It also helped a lot for my Cinebench stress test (held at 85c before, now maxes out at 74c). These are all the non-overclocked temps.

Then I discovered the world of overclocking and wanted to see if I could do some more stress tests to validate the stability. I tried out Prime95 and WOW did my temps and power consumption skyrocket. I was using 120 watts and the temps jumped to 94c and crept up the longer I let the test run, all the way to about 96c where I got nervous and shut it off.

Is this alright? For further reference I'm now at 4300 MHz where the normal for this CPU is 3900 MHz. My VDDCR is at 1.31875 volts. I was hoping to push a little further than this since my UserBenchmark scores haven't really budged after all this, but I am starting to gather that's not a great benchmark tool from reading other forums.

Any help, input, advice, or just anecdotal experience is very appreciated!

There are only few reasons to manually OC Ryzen 3000 and up. One of them is if you need reasonable all core GHz (about 4.4GHz) with SW that can't push it to.
Considering 4.7GHz max boost (on one core) while others running up to 4.2-4.4GHz, at 3.9 or4GHz it's a massive underclock.
Best way to "OC" Ryzen is to cool it and with 3000 series make it stay under 70c, over that boost gets progressively lower (all core and single).

jandcando · Jul 25, 2022

MishelLngelo said:
There are only few reasons to manually OC Ryzen 3000 and up. One of them is if you need reasonable all core GHz (about 4.4GHz) with SW that can't push it to.
Considering 4.7GHz max boost (on one core) while others running up to 4.2-4.4GHz, at 3.9 or4GHz it's a massive underclock.
Best way to "OC" Ryzen is to cool it and with 3000 series make it stay under 70c, over that boost gets progressively lower (all core and single).

I did disable PBO for my manual overclocking efforts, should I turn that back on to get some single core boosts? I'm not sure that works with manually changing core ratios so correct me if I'm wrong but it'll have to be one or the other (PBO OR core ratios)?

From reading around a good practice is to just enable PBO and set the max boost offset as high as it will go and it'll just go to what it is thermally comfortable with. I might just go ahead and try that when I get home.

Slaughtahouse · Jul 25, 2022

What is your objective? Just for fun? Gaming? Benchmarks? Day to day usage?

Everyone talks PBO but I find all-core OC better for my day to day usage. When I say better, i'm talking* rounded results. Best performance to temperature to power consumption.

I'm at 4.3Ghz at 1.2v (3800X). Of course every chip varies so results will vary but I've done a lot testing to get there. 1.3v is a lot, even for good quality air cooler so I'm not surprised you're seeing those temperatures.

You will have to play with the fan curve when stressing at max load if you want to knock those temps down a tad. However, i'd start with the voltage if you want to keep all-core.

For your visibility, see my results at various voltages with high LLC (gigabyte). Again, you'll need to do your own testing. As each chip / board varies.

Clock Speed [GHz]	4.25 GHz	4.275 GHz	4.30 GHz	4.325 GHz	4.35 GHz	4.375 GHz	4.4 GHz
Bios VCore [V]	1.166	1.178	1.196	1.22	1.244	1.292	1.34

jandcando · Jul 26, 2022

Slaughtahouse said:
What is your objective? Just for fun? Gaming? Benchmarks? Day to day usage?

Everyone talks PBO but I find all-core OC better for my day to day usage. When I say better, i'm talking* rounded results. Best performance to temperature to power consumption.

I'm at 4.3Ghz at 1.2v (3800X). Of course every chip varies so results will vary but I've done a lot testing to get there. 1.3v is a lot, even for good quality air cooler so I'm not surprised you're seeing those temperatures.

You will have to play with the fan curve when stressing at max load if you want to knock those temps down a tad. However, i'd start with the voltage if you want to keep all-core.

For your visibility, see my results at various voltages with high LLC (gigabyte). Again, you'll need to do your own testing. As each chip / board varies.

Clock Speed [GHz] 4.25 GHz 4.275 GHz 4.30 GHz 4.325 GHz 4.35 GHz 4.375 GHz 4.4 GHz
Bios VCore [V] 1.166 1.178 1.196 1.22 1.244 1.292 1.34

The fan curve helped my Cinebench results a lot! Such an obvious tip but somehow that's like the one place in BIOS I hadn't visited yet lol. I also dropped my voltage to be at 1.3v instead of a couple hairs higher and that combination helped the temps stay around 71 during the Cinebench test. I remember seeing some rounding errors being reported in Prime95 and I was able to make them occur less often by upping the voltage which is why my voltages got so high. I'll likely keep lowering and see if Cinebench can at least be stable even in Prime95 can't.

As for goals, I'm at the moment just having fun with it. I was looking at getting a new processor to upgrade to but that would require a new motherboard, so the CPU cooler was a pretty cheap alternative for an upgrade. I had a little obsession with seeing the benchmark numbers get higher and higher, but I think I'll have to drop that for now haha. I just want a modest performance boost that still stays cool enough in day to day use. I do gaming and also run simulation software at times for school so upping the clock speed does have some utility there.

Thanks for posting your voltage tests too! Those are very nice to see for reference but I'll still do my own testing. They definitely tell me I went a little high on the voltage. I might report back on my Cinebench results and temps with PBO enabled too just as a reference for others in my boat.

jandcando · Jul 26, 2022

Ok, I'm back with my results just mapping the landscape of PBO vs. no PBO. A little more background on my system first:

MOBO: ASUS B550M-A AC
RAM: 4x8 GB DDR4 3600 MHz 16-19-19-39 but clocked at 3733 MHz. no changes to timings beyond DOCP profile
Processor: Ryzen 7 3800XT

I include the RAM because I've heard it can put additional strain on the CPU so take these results with that in mind. These are not extensive tests but hopefully paint a picture. I did 3 tests: one with my manual OC settings, one with those and PBO enabled with a +200MHz offset limit, and one with just PBO enabled. That second test was exactly the same as the first within margin of error of Cinebench's scores, so I'm confirming you can't do manual OC and use PBO to great effect.

Cinebench Test Results

Clock Speed (GHz)	Bios VDDCR	All Core Score	All Core Temp	Single Core Score	Single Core Temp	Notes
4.3	1.3 V	12695	71 C	1271	50 C	core clock = effective core clock here according to HWiNFO
PBO+200MHz offset	AUTO. It averaged 1.35 V in multi core test and 1.47 V in single core test (!)	12643	73.5 C	1339	58.6 C	4.6 GHz max core clock and 4.34 GHz max "effective" clock according to HWiNFO

What I'm gathering from this is that PBO is pretty great for single core performance but it cranks the voltages pretty high. The cpu was very power hungry in that last test and got much warmer with a slightly-worse all core score.

I don't want to touch Prime95 since it seems to do MUCH different things to my CPU but maybe I'll look at it again once I can inch the voltages down a bit more.

Slaughtahouse · Jul 26, 2022

That is what you're going to find and not easily understood by many users. Effective clocks. This is the challenge I was facing with PBO. The results are a bit artificial outside of benchmarks.

A) They're typically "boosty" (not sustained) and;
B) When you average out the clock rate over X period of time, you'll find you wont achieve the max "core clock" for the entire duration of the test.

For example with PBO, my CPU would struggle to hit 4.3 GHZ on both CCDs (clusters of 4 cores) for all cores. Once CCD would hit 4.3 GHz (boosty / unsustained) and another would hit 4.25 GHz. All at much higher voltages (and heat).

Sure, single core results were better. No denying that 4.6ghz > 4.3ghz but if you average it out over a duration of time, you'll find in the real world the results are negligible. Hence why I stick to my all core OC, which effectively limits voltage to what I set it too and can sustain the clocks. I use 3D productivity applications a lot so this is my preference.

If you were only gaming or benching, you may want to stick to PBO because in those instances where X frames are being called for... 2 seconds, the boosty performance will help reduce the performance loss. Example: an explosion on screen with partials and physics.

Everyone's mileage will vary but I'm glad you found out sooner than later how to utilize HWiNFO and didn't just share "core clocks". Which is simply a value of BUS Clock X Multiplier.

There is more you can do in PBO, such as modifying the EDC, PPT, TDC (amps, power, and thermals) limits to get better performance btw. But its quite a rabbit hole and IMO, not worth the time for the average OCer.

Edit: I forgot to mention, ~1.5v is 100% normal for single core loads. Just think about it this way, as workers / cores scale up, voltage scales down. When additional amperage is required for the workers, voltage will have to drop to better manage the power budget and heat.

This is a slide for Zen 3 but its comparable to Zen 2 behavior.

Ryzen 7 3800xt Prime95 Temps

jandcando

MishelLngelo

jandcando

Slaughtahouse

jandcando

jandcando

Slaughtahouse

Top Contributors this Month

Recommended Communities