AMD CPU Thermal Throttling

[GreenSmoke]

In HWiNFO64 what stat(s) represent CPU performance?

To better explain, if I have bad airflow, the CPU thermal throttles at X% of its max ability. If I add better airflow, it still thermal throttles but now at Y% of it's max capability. I am thinking X < Y but I don't know what stats would represent X and Y in this scenario.

This all ties back to noise. If Y is only like 5% better than X but the noise is wonderful at Y and terrible at X. I would probably leave the settings at X.

What about the same question but on a GFX card (2060 Super).

 

[Zerileous]

You'll see marked improvement down to 70C. https://www.gamersnexus.net/news-pc/3492-ryzen-cpu-thermals-matter-coolers-and-cases

If you want to quantify the difference between two settings, then run Cinebench or something at both settings and see, or check frequency under some stable load like prime95. For gaming you could use a gaming benchmark instead. This would also more accurately reflect the thermal load in your case from the GPU while gaming. I suspect at 1440p it will not be a substantial difference.

Now to split some hairs, it sounds like you're actually asking about thermally constrained boosting. Your CPU will "throttle" as in sharply decrease frequency and voltage to save itself from a fiery death at 95C. You should keep your max load under this no matter what. If it's too loud then get better cooling (I know you're planning to do this anyway).

Now on to "boosting" which is going to be the gradual increase and decrease of clocks and voltage based on (roughly) three parameters (it gets more technical, but I'll spare the details): Power (measured in amps, or current), maximum boost clock, and temperature. It turns out that for almost every use case, regardless of stock or PBO settings, the constraining factor is temperature. This means that the CPU will "use" your cooling headroom to improve clocks. Only you can decide the balance of noise vs performance that is worthwhile to you.

 

[GreenSmoke]

Cinebench scored between 3412 and 3426.

The CPU stayed around 87c through the test.

This was with...
-Tuned case fans (temp goes up = fans go up, 100% at 90c)
-The Stealth Wraith
-XMP2 bios OC
-Default graphics card settings

My googling of gaming benchmarks was not definitive (Ill keep looking) I went with UserBenchmark so I'll share that.

UserBenchmarks: Game 109%, Desk 88%, Work 77%
CPU: AMD Ryzen 5 3600 - 86.1%
GPU: Nvidia RTX 2060-Super - 134.9%
SSD: Samsung 860 Evo 1TB - 101.6%
USB: SanDisk Cruzer Glide 32GB - 10.1%
RAM: G.SKILL Trident Z DDR4 3600 C16 2x8GB - 95.1%
MBD: MSI MPG X570 GAMING EDGE WIFI (MS-7C37)

I did order a D15S. If one fan won't cut it though, I will likely need to add a 120 to the intake since my RAM is too tall.

 

[Zerileous]

Well userbenchmark is appealing, but don't fall into the trap of their comparisons since it's not a controlled setting. The results for a specific component will include systems that barely function and only have one case fan and the cheapest components as well as the most robust water cooled builds. It also doesn't really stress the GPU much.

I think 3dmark is the standard by which everyone compares judges graphic performance, but it isn't free so I think your money would be better spent on things like cooling right now. Unigen heaven is free and pretty well regarded. On your system it will load the GPU but performance will actually be CPU constrained, so it is actually probably the perfect test for your scenario.

With 87C being your max in cinebench, I don't think you have a lot of room to turn down your fan curves. Once you have the D15S, you can play with temps and see if a fan curve favoring quiet is going to cost you much in terms of FPS but using the Heaven benchmark. While I really have no experience with the "big air" heatpipe towers (my last air cooler was an 800g chunk of copper), I have trouble imagining one fan being insufficient to cool a 6 core CPU.

In this review they tested different fan configurations https://www.gamersnexus.net/hwreviews/3061-fractal-meshify-c-review-vs-define-c. You'll want the rear exhaust fan to run in line with your cooler, essentially exhausting the hot air after it flows through the CPU cooler. Adding a second front intake fan helps the GPU by feeding it cool air, however there isn't great exhaust of air from the GPU, some hot air is recycled as it bounces off the side panel and PSU shroud. Really this is pretty common and not a big deal, especially for a 2060 Super, as it isn't a super "hot" card.

 

[GreenSmoke]

That is the review that led me to buy this case. I do plan to follow the advice, leaving the 2 front intakes.

I am interested to see the drop when I put in the D15S. After I get some stock benchmarks, I am also going to try shrouding the fan to see if it makes any difference. IMO, even a small temp drop from a shroud is a big bonus since it doesn't come at the cost of noise. It's probably wishful thinking or Noctua would have built something like that in but I won't know till I try it. I find a shroud for the D15 (double fans) and they say it has no effect on the cooling but I think the 1 fan is a whole different ball game and I don't find anything for it so far. The shrouds I find also seem to be cosmetic while I am thinking more functional.

I think the fan would suck an awful lot of air from around the sides instead of through the front fins if it doesn't have a shroud.

As long as I maintain such a large positive pressure, I'll also leave off the expansion covers out to further eliminate the GPU cavity from recycling.

 

[Zerileous]

I usually see shrouds implemented in OEM cases where they are effectively used to combine CPU and case fans, either directing flow from a CPU HSF out of the case, or directing flow from a case fan over a CPU HS without a dedicated fan. I doubt it will do much when you already have two fans lined up promoting smooth laminar airflow. But it never hurts to try and see!

For the GPU, I wonder if you could put a fan there, really though see if its even a problem for your card first. An overclocked 1080Ti running Furmark is a lot different than a 2060S gaming.

 

[GreenSmoke]

Yeah, I'll see how it goes once my CPU isn't the bottleneck.

There would be 3 fans in total in teh CPU line. Front/upper case intake > CPU Fan between 2 rads > case exhaust. Just to be sure we are apples to apples.

 

[Zerileous]

You mean top of the front of the case? Or firing down from the top of the case?

 

[GreenSmoke]

From the front upper fan, sorry. All intake is from the front. Nothing on the top of the case.

 

[Zerileous]

Perfect, sounds like a good setup. I think you'll be really happy with the D15.

 

[GreenSmoke]

I loved my U12S and it's still in operation. Noctua is a great company both in product quality and support.

 

[GreenSmoke]

I got my new HS and installed it and it's pretty sweet. You will be shocked to hear it works much better then the stock wraith HS.

Running Prime95 Small FFTs, I have hit 79c max temp in 30 minutes using HWiNFO64.

The HS fan would be 70% at 70c and go to 100% at 90. It's audible at 980RPM (of 1500max) but not too terrible. I probably need to play with my fan curve a bit.

I wish I could see my PS fan and other PS info in HWiNFO. I went HAM on the PS to hopefully draw low enough power the fan wouldn't turn on. I think it comes on when Prime is running but I haven't heard it any other time.

 

[Zerileous]

not shocked in the least . As much acclaim as AMD has received for its good *stock* coolers, they're not that good, especially the stealth. They're just better than the stock coolers shipped with the competition's CPUs.

 

[Woomack]

AMD stock coolers are not enough to run AMD 3000 series processors at stock. However, CPUs are throttling to keep stability at specified 95°C so users won't see stability issues because of overheating but performance drop. Most won't even notice that.
Intel stock coolers are enough to run their processors without throttling but all higher Intel CPUs don't have a cooler included and Intel wouldn't be able to provide low-cost coolers for chips like 9700K or 9900K. All new higher series processors have ridiculous TDP in specification, much lower than real values. The same Intel and AMD.

 

[GreenSmoke]

I didn't have the wraith spire (from the X version) but based on my experience with the normal wraith, I bet the spire is right on the line of being able to keep the 3600x below the 95°C throttle point. As many Youtubers have pointed out, you would be better to save the $50 and spend it on a Hyper 212 for $10 less or something better.

 

[Woomack]

I was testing 3600 with 3700X cooler and it simply can't handle it. 95°C after 10 mins of AIDA64 stability test and clock drops to 3.2-3.6GHz.
Hyper 212 can barely handle it. I've tested similar cooler and the CPU was going up to 92-93°C. Actually, the same temps are on my 3600 and 3700X.
Noctua D9L is the smallest cooler that I was testing and could handle these chips without throttling. With 1 fan results were about 93°C, with 2 fans about 86°C.
My tests are pretty much confirmed by Noctua and as you can see on their website, because of ridiculous TDP in new CPU specs, Noctua removed cooler TDP list and work on the new one.

 

[Zerileous]

You make a good point. Actually I saw my 2600X hit 100C running Prime95 Small FFT on the stock air cooler, just with XMP enabled to 3200MHz (yes the CPU was seeing more voltage to the SOC, but it was a big increase for such a small change. I've honestly haven't seen an Intel stock cooler in 10+ years, so I can't say if they're totally terrible, I just got that impression I guess.

Since you mention TDP values, I thought I'd share this (also note the use of the phrase "at least," I never noticed that before):

Package Power Tracking (“PPT”): The PPT threshold is the allowed socket power consumption permitted across the voltage rails supplying the socket. Applications with high thread counts, and/or “heavy” threads, can encounter PPT limits that can be alleviated with a raised PPT limit.

Default for Socket AM4 is at least 142W on motherboards rated for 105W TDP processors.
Default for Socket AM4 is at least 88W on motherboards rated for 65W TDP processors.

Of course PBO could increase this further, in theory, however GamersNexus didn't find the CPU to encounter these limits in stock configuration in their testing.