Processors

[govvern]

Hey everyone,

I had a few concerns regarding my new 5950x. Recently I just finished a new PC build with the following specs:

• MOBO: ASUS TUF Gaming x570-Pro
• CPU: AMD Ryzen 9 5950X
• GPU: EVGA RTX 3080Ti FTW3 Ultra Gaming
• RAM: Crucial Ballistix 2x 16GB CL16 @ 3600Mhz
• Storage: 2x Samsung 980Pro NVME M.2
• CPU Cooler: Noctua NH-D15 (Noctua Thermal Paste)
• Case: Corsair 5000D Airflow
• Fan Setup: 3 Front Intake, 3 Top exhaust, 1 Back Exhaust

PBO is off, Windows power settings are set to balanced, and Nvidia Control Panel settings are all default.

I’ve been running benchmarks from Cinebench R23, Heaven Benchmark, and Superposition using HWInfo to monitor.

Using Cinebench, I’m seeing CPU temperatures peak at 65C and hover around 61C under 100% load across all cores. Also all core voltage are right around 1.025V @ 3.5-3.8GHz. This is what I would expect to be normal behavior.

Now, when running Heaven Benchmark, everything is completely different. Granted, my GPU temps are around 78-82C, but my CPU temps are jumping up to 75C, sometimes spiking close to 85C! Also at 5-10% usage I’m seeing and constant 1.4V across all cores with clocks jumping as high as 4.5GHz. Some individual cores even reaching 5GHz. As far as I’m concerned this is not a CPU benchmark.

Another thing in Heaven benchmark is I’m seeing my Power Reporting Deviation dip as low as 55% which is really concerning since it’s meant to stay between 90-110%.

Again, everything is stock and I don’t have PBO or OC settings enabled.

This happens in games as well not just benchmarks. My most recent was Outerworlds where I saw my CPU temps spike to 86C just in a loading screen.

I’m really not sure what to do here. This is the closest I’ve gotten to figuring out why my PC is running so hot.

I would really appreciate any feedback.

[ryzen_type_r]

How many FPS are you seeing in the Heaven benchmark?  If it's in the hundreds, then I think what you're seeing is normal.

Even though it's primarily a GPU benchmark, a very fast video card needs a fast CPU to send instructions to it.  From what I've observed with Heaven, it's very lightly threaded, just 1 or 2 CPU cores do all the work.  In a lightly threaded workload, Ryzen CPU's will boost those cores way up, as you've seen.

Temperature can spike up quickly even with just 1 core doing a lot of work - you can see this if you run Prime95 stress test with just a single thread.  Think of the physics of this - one tiny little spot is generating a lot of heat.  If you had 16 cores doing the same total amount of work, the same heat would be spread over a much larger surface area, making it easier to dissipate.

As for Power Reporting Deviation, it is not accurate unless all cores are running a constant full load, which Heaven is most definitely not.  So ignore that metric.  It's a motherboard issue anyway, there may even be some kind of correction for power reporting in the BIOS settings somewhere.

As for Outer Worlds temp spiking during loading screen, maybe the same thing as with Heaven - if you're running with Vsync off the framerate could be in the hundreds.  But that's just a guess, I don't have that game so can't tell you what it's doing during the loading screen.

Ryzens boost very aggressively to run as fast as possible for any given workload, and you bought the biggest baddest Ryzen that is available and combined it with one of the fastest video cards on Earth, so yeah, everything's gonna run fast and hot.

If it really bothers you, you could try capping frame rates to see if that helps (you can do it using the Nvidia Control Panel), I mean there's really no reason for a game to be running at 400fps sitting at a menu screen.  And in the motherboard BIOS, there's a setting called Platform Thermal Limit where you can set a max temp and the CPU will not exceed it (by default, your 5950X has a limit of 90C), but you do lose some performance depending on how low you set the temp limit.

You might also want to look into a good AiO liquid cooling solution, but I can understand not trusting them entirely, I've seen what failures can do LOL

[govvern]

Holy **bleep** this is easily one of the best responses I’ve ever gotten. Thank you for being so in-depth and hitting every question/concern I had. I’ll definitely look at trying to limit the FPS in some games and check out that “Platform Thermal Limit” option as well.

On a side note; Do you think that it’s worth undervolting? I would only consider it if my temps are going to shorten the lifespan of my CPU. If not then I don’t see a reason to. 

[goodplay]

I would suggest your best bet is asking for optimal (non OC) bios settings either from asus forum or reputable overclockers forums.

[ryzen_type_r]

You can try undervolting if you want, but there are a few different ways to do it, and you'll need to spend time tweaking and testing for stability.  If that's your kind of thing, go for it.

It's probably easier just to set a Platform Thermal Limit if you don't like your CPU exceeding a certain temperature, easy peasy.  It just works.  And for example if you were to upgrade your cooling, that gives the CPU a bit more thermal headroom, and it will boost a bit higher to use that headroom without exceeding that temp limit.

[freedomfries]

Undervolting is one approach I hadn't considered. That and setting thermal limits seem a bit iffy. For one, the CPU temps spike so fast, that you basically will  be constantly revving your CPU up and down...

I have a brand new 5950x paired with DDR4000 and a gigabyte AORUS x570s.

I always run prime 95 to test for system stability and was horrified at the temperatures I was getting. 4k FFT tests would push temps to 90c, my cores would go waaay past 4ghz and the CPU was using >175w of power! Of course at that point the CPU throttles down. And they advertise a 105w tdp.... Yeah right.

Even idle, I was looking at 50c and temps would spike whenever something was being done, however briefly. After much tinkering and reading on forums I came to the conclusion that these CPUs come out of the box cranked all the way to 11 and you probably want to spend some time tweaking this to your needs.

I plan on doing some rendering with this, so long sustained workloads, and I can't be running this cpu to 85c for hours on end. I opted for another approach, which was to turn off the core boost. If you leave everything else to default, your CPU is basically pegged at 3.4ghz, while core boost can bring this up to 4.8ghz, so while your temps suddenly become very mild, you give up a LOT of performance. I started manually raising the clock multiplier until I found a stable baseline, which so far is 3.8-4ghz.

At this speed on all cores, prime 95 large fft tests run in the mid 60c, and small fft tests can push that up to 85-88. Bear in mind there's no natural workload that will do this to your cpu. When I run a Blender render test, the highest it went was 71c.

I ordered a bigger cooler in the hopes of shaving off a few degrees and having a 4ghz pc that runs any workload at decent temps. I'm currently idling at 33c, with a browser and email up. When I play fallout 4, i'm in the 45c range. I will try to turn core boost back on when my Fuma 2 cooler arrives, to see what I can get away with.

I don't like replacing my computer every couple years, so I want this CPU to last me a while, and frankly, 16 cores at this speed is plenty for my needs. I do wish AMD was more honest about the power consumption and TDP of this chip at the stock settings. I think this would prepare people better to deal with it.

[ryzen_type_r]

---

@freedomfries wrote:

Undervolting is one approach I hadn't considered. That and setting thermal limits seem a bit iffy. For one, the CPU temps spike so fast, that you basically will  be constantly revving your CPU up and down...

Setting a platform thermal limit absolutely works.  Try it and you'll see.  'Revving your CPU up and down'?  CPU's are designed to do this - they adjust their frequencies and voltages to adapt to the workload.  AMD says current Ryzens do this 1000 times a second.

I always run prime 95 to test for system stability and was horrified at the temperatures I was getting. 4k FFT tests would push temps to 90c, my cores would go waaay past 4ghz and the CPU was using >175w of power! Of course at that point the CPU throttles down. And they advertise a 105w tdp.... Yeah right.

Those temps are normal for Prime 95.  It's a stress test, so you're intentionally stressing the CPU in a way that workloads in typical applications for most people do not.  FWIW, my old Intell Haswell would hit 100 instantly when Prime95 was set to use AVX2.

AMD's TDP numbers are not accurate.  By default, '105W' CPU's have a fused PPT of 142W, so under certain workloads, you can see up to 142W at the CPU socket.  If you're seeing values of 175W just for the CPU, it means the PPT limit was changed to 175, probably in the BIOS (lots of mobo's do this out of the box).  Ryzen Master can tell you what it's set to.

The PPT can be lowered, so you can in fact specify a lower maximum power at the CPU socket using BIOS or Ryzen Master, and in that way directly control power consumption.

---

I don't like replacing my computer every couple years, so I want this CPU to last me a while, and frankly, 16 cores at this speed is plenty for my needs. I do wish AMD was more honest about the power consumption and TDP of this chip at the stock settings. I think this would prepare people better to deal with it.

---

I agree with that.  Both Intel and AMD have been playing it a bit loose with their TDP definitions over the past decade or so.

Take a look at some of the temperatures and power consumption numbers for high end Intel Alder Lake CPU's.  They make AMD's look reasonable.

That said, AMD has given us a number of good ways to keep temps and power consumption in check.  Setting a Platform Thermal Limit allows the CPU to maximize performance but not exceed a set temperature.  Setting a different PPT allows us to directly limit the power consumption of the CPU at the socket.  And you can set the maximum operating frequency to whatever you want using Windows power plans (use Quick CPU if you want control over every single one of the many power plan settings).

You can use any combination of these to tweak the CPU's behavior to suit your use case.

[freedomfries]

Yeah I'm trying to get familiar with all the settings in my Gigabyte motherboard and what they mean. Most of them are acronyms that I have to look up. I'm not a seasoned overclocker, I usually content myself to finding the balance between performance and stability, the latter being more important for my needs.

Ryzen, as a platform, is offering waaay more tweaking options than I'm used to. My last box was a skylake 4 core, and it never went above 55c. I guess I must bet used to seeing higher temps.

My current goal, if possible, is to run the RAM and CPU at 4ghz, while keeping the temperatures under 80c at load. I have some pretty good Gskill Trident DDR4000 sticks, and I ordered a beefier cooler than my Noctua NHu12, which should knock down a few degrees.

You seem to know the platform well. What kind of settings in RAM and CPU would you recommend for me to try and achieve this? I feel like it's more a question of the RAM being able to run at the binned speed.

When I had my CPU clock multiplier at 40, and the DRAM XMP profile on (4000mhz), Prime 95 threw an error after 25 minutes. What do you think I could do to stabilize this?

Any suggestions appreciated.