[mike7877]

First: my 5800X3D is cooled by a 2019 Noctua D15 with the correct amount of Arctic Silver 5 in the right place. Two 140mm fans blow through its fins, and the case is well ventilated. And the room is ~20.5 deg C. PC is on the floor so it might be a degree or two colder.

OK, to the topic: Because my system is only a couple months old I haven't tuned everything perfectly yet, but I'm getting close; close enough that I began working on setting up final fan profiles for optimal cooling and noise. During this process I noticed something. I'm not sure if it's a problem.

So during some of the more memory intensive CPU loads I run, there can be a large delta between core temps and temps of the die "Tdie" as stated in HWiNFO64. In some cases it's 15-20 deg C!
Core temperature (the average of all 8) will be in the high 60s, while the first temperature in the CPU list ("Tctl/Tdie"), will be 85.

Control: Running AIDA64's CPU stability test (which is CPU checked off in the window opened by AIDA64 > Tools > Stability test), causes the cores (the average of all which none are far from), to rise to 64 degrees Celsius, and "Tdie" to reach 66.
FPU is the same but 19 degrees hotter: 83 and 85 degrees.

Smooth Video Project, a.k.a. SVP 4, is wanted (dead or alive) for the capital crime of "non-monolithic CPU die deg C delta being > 16.2411 Celsius before Biden takes his second afternoon laxative on the third Thursday in the Spring months of years ending in 2, 5, and 6".

This is unfortunate because other than an unintended failure to comply with the law, the program is fantastic. Almost divine, really. It's for video interpolation, which it's great at doing an exceptional job of all aspects of. It gives you gloriously exquisite results, that are always, always breathtaking, borderline sensational. Especially juxtaposed: the improvement is immense - it can only be described as immaculate. Every time.
It's what you deserve, so get SVP. You'll be pleased. Except maybe this temperature problem if you have a 5800X3D.

Operating condition that makes a problem:
19Gbps read and 14Gbps write are HWiNFO64's reported DRAM bandwidths (during its most visually appealing version of its real-time video revamp of a 4K24 video) resulting in the cores being at high 60s, with Tdie being ~85.


A little bit interesting is, pointlessly transferring a lot of data quickly to and from RAM doesn't caused raised temperatures and the aforementioned temperature delta. Basically the data being transferred has to be being used by the CPU for processing too. Or so it seems.


Has anyone else noticed this phenomenon with their 5800X3D? Or even possibly their non X3D version? I'm not sure how applicable the vanilla version necessarily would be, because it doesn't have that extra layer of silicon laid on its cores.

Is it bad to often have the CPU operating with this 20 degree delta? It'll be whenever I watch a 4K video either on YouTube or a movie ending in .avi, .m2ts, .etc etc.


Also, if you like smooth video, try SVP because it's amazing (like I said). To watch a YouTube video you can just copy the URL and it starts playing it. To watch a movie in .avi format you just double click it and interpolation happens and you view the result in MPC-HC

TL;DR Is it bad to often have the CPU operating with this 20 degree delta between the cores and "Tdie"? It'll be happening whenever I watch a 4K video either on YouTube or a movie ending in .avi, .m2ts, .etc etc. which, during this most recent stretch of existence, is probably 3% of the time. But that might grow in the future, depending.

 

[mike7877]

IMO that is normal
Tctl/Tdie is the hottest point (peak temp) of the whole CPU not only the cores and the motherboard uses that sensor to predict temperature spikes and ramp up the fan in time.
In the past that sensor has a offset of 10°c but now are in 1:1 relation.

What a coincidence, I noticed the other day that the GT 1030 I used for a long time in a PC for not games, that its hotspot is just a 10 deg C offset! Many many months I used that card and never noticed it was just a 10 degree offset lmao. After figuring it out I watched it more closely and also found it must be calculated in a weird way, beause very very very infrequently 10 changes to 9.9 or 10.1.

Anyway, about the 5800X3D - the Tctl/Tdie is definitely not an offset because I've been watching it a lot and it varies a lot. Unless a multiplier is used for the offset - a fraction derived from a temperature delta or something crazy. It especially annoys annoys me because the "CPU temperature" that the BIOS uses, is that... It's able to swing pretty wildly, up and down 10-15 degrees while the cores are only moving 1-4 while the power's not changing much (and I'm looking at the summed value of both regulators (in watts) that drive everything on the CPU package afaik).

I believe thats a hotspot.

It might be. If it is though, it's the weirdest hotspot I've observed -- most often I find it's pretty close to the temperature of the cores and the other CPU sensors, maybe 2-3 degrees warmer than the cores. Mostly its just a different temperature during high throughput data processing (as mentioned in OP). Though power doesn't increase much over normal use, temps rise high and make my fans noisy. At this point I basically have them only spin faster than idle once 72 degrees is reached. I have enough and they're big enough and moving enough air that even though they don't spin up until 72 degrees, at idle with 3-5% CPU utilization, (cores 15W, SOC 18W, package power 40W)

1.) average core temperature is 42.2 degrees
2.) CPU CCD1 (Tdie) is 48.9 degrees
3.) CPU die (average) is 48.4 degrees
4.) CPU (Tctl/Tdie) is 53.1
5.) L3 Temperatures average 41.6

I don't know what it is about Tdie... in #2 and #4 above it's included. #2 by itself, and #4 with Tctl. Tctl isn't listed anywhere by itself that I can see. If 4 is the average of Tctl and Tdie, then
hypothetical Tctl would be 57.3 degrees!
But that could be nonsense.. probably is.

 

[mike7877]

you can quick google it, it might be they have now multiple hot spot sensors, after all amd has chiplets. as far i know, nvidia cards have hotspot temp too, and every single card has hotspot 15c warmer ( give a take 3c ) then avrange core temps. hotspot is hotspot, its temp in split of a second, and its not always on same spot on the die, its just highest reported temp across die at any given time, understand?

I think so. To confirm, are you saying that the temperature I'm seeing when the mentioned "problem" type of CPU+memory intensive workloads I run is from a different sensor/location in the chip than the value is usually reported from (say when the system is idle)? That'd be extremely frustrating if the sensor it was currently focusing on wasn't available in the list of the rest of the chips's temperature sensors.

I think it would be nice if they named the sensors properly, according to their locations. Actually, they should probably standardize the practice of sensing temperatures in chips, or at least come up with "good practice" guidelines.
Ex. "One T. probe on on every ___". And then there should be a minimum number of them per watt per square milimeter. Detailed descriptions of all the sensors and their functions and their locations and what each is relevant to or useful for so that they can lay out thier sensor pages in logical way for users, who should have the ability to access all sensor data, even if not available immediately in efforts to minimize clutter and maximize efficiency and aesthetics.
That's a dream lol

I still think my GT 1030's "Hot Spot" is nonexistent. In fact, when it was new, I don't remember there being a "hot spot" reading lol.

 

[mike7877]

The CPU runs hot. Mine will hit 90C very quickly no matter what I do, so... uh... surprise?

Surprise! lol

 

[mike7877]

Tctl/Tdie like a said is the peak temperature of all the sensor integrated in the CPU and modern CPUs has a LOT of sensors. Meanwhile CPU CCDx and CPU die the average temp registred in X period of time. We don't know how those values are calculated because documentation is very scarce. Only some random articles from AMD and comments from AMD Reps on reddit .
I think the main funtion of that sensor is to control fan speed efficiently. In our gaming pc we have massive cooling capacity so changes in temperatures are usually "very slow" . But in laptops with those tiny coolers need some "prediction" to ramp up the speed before the CPU/SoC gets too hot.
Old Ryzen CPU (1000 - 2000 Series) has a 10°C offset and Threadripper 1000 series 27°C offset for that reason

A solution to make the fans don't go like a 16yr with a Honda Civic is chaging the temperature hysteresis in relation to the fan speed. You can change the value on the BIOS to like 10°C

I've got the fans behaving. Wow, 27 degrees. What an offset!!

I'm getting tired of this lack of documentation. Again!
Not that it was specifically this processor (or even AMD) that irritated me in the first place.

I have this feeling that X3D isn't coming back - it was just done to beat Intel. Intel had its own "X3D" with Broadwell. They didn't bring it back - just relied on higher core frequencies and the fact that their 5000 series didn't overclock past 4.2GHz
Zen 4 architecture is so similar to Zen 3, I think that having fast RAM (DDR5 6000 for example) is going to negate a lot of what the extra cache would do for Zen 4 if it was on DDR4 3600

 

[mike7877]

X3D doesn't run very hot, I don't have my new FC140 yet, but this is PA120. You should check your cooling

I think the cooling's fine - under most "full" loads (benchmarks etc.) peak temps are like 78-83, and I use a very quiet custom fan profile I designed. Two 140 mm fans spin always at 400-500RPM, keeping idle temps to around 40. Because quick heavy CPU usage (opening a program, using a CPU-heavy feature in a program) causes temps to rise 20-25C , the fans were spinning up effectively doing nothing because the heat generated by the chip was still in the heatspreader until after the fans spun back down. I made the fan profile keep speeds down at 400-500 until 68 C, which is only reached (with the fans idling there) under sustained heavy load. 97% of the time temp is 40-45C, the rest of the time is those 3-5 second 20-25 deg C swings (which I'm thinking would happen under even water cooling - only lapping the actual CPU and using a custom block that makes direct contact would change anything. Over 68 and before 87 the CPU and case fans never go above ~70%

The 90C seems to happen in only very specific circumstances

 

[mike7877]

Get rid of that awful Arctic Silver 5 paste. It's actually quite bad due to their dodgy usage of chemicals that do not pass on the heat through its application. In fact, there was a warning released by that Excremental Company, in relation to NOT using their Arctic Silver 5 concoction.

Your best paste would be the Noctua H2, without a doubt the easiest available paste, with true performance. No cost-cutting junk was ever released by Noctua - Proven Fact.

Have a look at how many pieces of Garbage that were released out there by Arctic, and they start looking like my 'Uncle Dodgy' selling secondhand plaster at a 4th Rate Hospital .

I haven't heard of any bad revelations (til now obviously), but I guess I could switch paste next surgery. A pretty big tube came with my D15 so I could use that. Worst (or best) case though, the difference in temperature from switching thermal pastes has usually been found to be 1 degree, maybe 2 C