Question regarding my ASUS TUF RX 6800 Memory Junction Temperature

[CyCloud]

Greetings everyone!
Maybe one of you can answer me regarding my concerns with my ASUS TUF RX 6800.
It seems that according to HWinfo that the GPU Memory Junction Temperature is always a lot higher than my GPU Hot Spot and when compared to other RX 6000 series GPUs that I found data on online.
Even when completely idle and nothing else running in the background except HWinfo, VRAM junction temps stay at roughly 54°C while my GPU hot spot is down to roughly 38°C.

Under a gaming load I've seen the VRAM junction temperature go up to roughly 84°C while the GPU Hot Spot is at roughly 77°C, so the gap is closing. However, when I drop the load, GPU Hot Spot temps go down fairly quickly while VRAM junction temperature drops only fairly slowly, leading me to believe there is something wrong somewhere. Just playing YT videos in Firefox will slowly heat up the VRAM junction temps to roughly 70°C where it stabilizes. The fan stays at 0 RPM in the YT scenario because the GPU Temperature is staying just below 55°C.

I haven't disassembled my card because I might have to RMA it but from the outside looking through the gaps I can at least confirm that all memory modules seems to have thermal pads which seem to be making contact with the heat sink. The backside of the PCB also has thermal pads connecting them to the backplate where the memory modules are.

While looking into different models I haven't come across someone posting anything about this specific model, only the one video on from Hardware Unboxed regarding the XT version (which is using the exact same cooler) praising the overall temperatures including memory temps, though I don't know what they used to measure the temperature.
According to GPU-Z, my GPU is using GDDR6 from SK Hynix, maybe those just run hotter than Micron or Samsung VRAM?
From what I understand, only the highest VRAM junction temps get reported, is that right? It would be really nice to see the readouts of multiple memory modules to confirm if one is behaving abnormally.

Anyway, thank you for taking the time to read this.

Edit: By the way, I've also written a mail to ASUS regarding this. Haven't received any reply yet but I will update this if they do reply.

 

[Zach]

Don’t count that ASUS will answer you and if they will, that their answer will have any kind of value.

When VRAM Junction temp is that high the backplate of card should be hot since thermal pads are in place. Try to touch the backplate to feel it.
It’s typical for GDDR6 to go in high values, up to 80-85C. Tjmax of those chips are usually around 100-110C.
Not that I would like to see them anywhere near that.

If you start the fan manually (from drivers) is VRAM temp dropping?
What is the CPU of the system? If it’s a Ryzen 3000/5000 are you using SAM(smart access memory) you can see this in Adrenalin drivers.

 

[CyCloud]

Thanks for your reply.
Yes the backplate is getting hot during load. If I disable the 0-fan and let it run normally all the time (roughly 800-900 rpm) in idle VRAM temps drop to the low 50s but I've never seen it in the 40s, even after a fresh cold boot.
I'm using a Ryzen 3700X and SAM is active.
Basically doing anything worthwhile on this system basically forces the temps into the 60-70s and I don't feel comfortable having the VRAM basically at that temperature all the time for no good reason.
So maybe the thermal pads aren't doing their job right, the readings are off or the calibration is off.
I haven't seen the temps go up higher than 84°C but I also don't know what application to use to really stress those temps to check if they really stay in the mid 80s at worst. Do you have a recommendation what I can use to really hammer those temps?

 

[Zach]

Is VRAM loading in HWiNFO high (MBs)?
Try to disable SAM and see if anything changes.

If backplate is hot then thermal pads are doing their job.

What is the clock of VRAM usually?

For some reason VRAM usage must be higher and 0-fan function for GPU doesn’t help.
I’m not saying to disable 0-fan. I wouldn’t like my GPU card to dust it self when there’s not really a reason…

 

[Zach]

Better post a screenshot of the the entire GPU section of HWiNFO sensors window on idle.

 

[CyCloud]

VRAM usage ranges from a couple 100 MB during browsing/watching YT to several GB when gaming, nothing odd there.
VRAM clock hovers from 2MHz to 900MHz during normal browsing and is ranging from roughly 1300MHz to 2000MHz during video playback, this is where it approaches 70°C.

Gonna disable SAM and see how that affects the temperatures.

 

[CyCloud]

Just disabled SAM in the BIOS, doesn't seem the affect the temperatures at all.
Other than those temps I really like this card. It undervolts fairly well and can even be pushed to the max OC the official AMD Software allows without becoming unstable. Coil whine is also at a minimum. With my luck, chances are when I RMA this one, I'm getting one with just slightly better VRAM temps but much worse coil whine and UV capabilities.

 

[Zach]

I’m not near my PC now but I will see later how VRAM behaves under various situations. I have an R9 5900X, X570 and RX 5700XT with SAM enabled.

Sorry for not mentioned it before, but it’s better to leave HWiNFO running for long time (1-2 hours) so readings are more accurate through your typical system usage.
By first look on that screenshot everything looks normal though.

 

[CyCloud]

I have been running HWiNFO basically all day the last couple days, that's the data I was referring to during my posts. Didn't have a screenshot of those days so this one relatively fresh after boot is all I could offer at the moment. But all the metrics fall in line with what I experienced the last couple days.

 

[Zach]

I think we can rule out the usage as on mine, memory usage, voltage, clock, current is higher and yet temp is lower.
So only thing left is cooling.
Maybe this specific model (the nonXT TUF) has a bad design on memory cooling or the specific item you have have some cooling defection.

Can you inspect the card without tearing it down for any gaps? As in for loosen cooler, screws etc...

 

[CyCloud]

I can tell that the cooler assembly is pretty much identical to the 6800XT and 6900XT variant of the ASUS TUF without taking it apart. There are enough little gaps to look through most of the stuff and determine that. And from the research I've done the whole heat pipe and fin stack assembly is exactly the same they use on the Strix variant as well.

This is a picture from someone else who took the XT variant apart:

One other thing I'll be trying soon is relocating my M.2 SATA SSD, which is placed directly underneath the GPU. It generates some heat as well and usually sits at 50°C+ which is probably not too healthy for it either. Ordered a M.2 to SATA adapter and will report back if it made any difference.

ASUS also replied today, telling me that the junction temperature can rise up to 95°C and thus can be higher than the GPU Hotspot and that this property is taken into account from the design point of view of the graphics card.

Would be great to know if this really is typical behaviour for this cooler assembly but I couldn't find any HWinfo posts about this one from other people, so I have no real comparison. Probably doesn't help that even if we find find one, they may have a model with Micron or Samsung VRAM, not SK Hynix like mine. I feel there is still a slim chance that the readouts may not be entirely accurate for my VRAM but that's not something I can validate myself.

Still need to find something to test how hot these modules can potentially get in my configuration. If mid 80s under load really is the ceiling, it wouldn't be great but at least not nearly as terrible as the PS5 VRAM. lol

Anyway, I still have over 2 years of warranty on this card, so should the VRAM kick the bucket I'm going to make it ASUSs problem. In case it outlasts its warranty I'll take it apart, do some more direct measurements and possibly replace the thermal pads.

Made some more light test, one roughly 3h session of browsing, YT and office work on 2 monitors and the other playing Quake II RTX on max settings while watching a video on the second monitor.
The screenshot of the gaming session was made about 15 minutes after I closed the game and went back to browsing. During the game VRAM temps stayed in the 80-83°C range. These tests were done with default settings in the AMD Software, so no UV, OC or fan curve adjustments.

Once again, thank you for your time.

Edit: Taking another look at the screenshots, the memory clock peaked at an odd frequency. I'm certain that I didn't have any memory OC on and even if I did, the AMD Software only allows up to 2150MHz. Strange.

 

[Zach]

About a year and a half ago I run some compute software on the 5700XT for like 4-5months straight every day except the time I played games. It was running on GPU memory mostly. With 50~60% fan speed (depending ambient) memory junction temp was constantly around 80C. So imagine that temp for 4-5months on most hours of the day(out of 24h).
What I'm saying is that GDDR6 memory can take this (80~85C) without issue as its operating Tjmax is usually 100~110C (depending brand and specs).

The thing is that I never saw mine above 70C on typical tasks, even on long gaming. Usually is around the same temp as GPU temp (not GPU hotspot). But I tend to set fan speed high (60~80+%). I'm wearing over-ear headset with some noise cancelling so I dont mind noisy fans much. Most weird thing to me is that your VRAM is increasing temp on nonGaming tasks. Never saw this before and I havent heard it from others.

You can always ask a question on TPU forums (I'm there too). Its a much bigger (more users) forum.

TechPowerUp Forums

Computer hardware discussion forum, focused on tweakers, overclockers, gamers and enthusiasts.

www.techpowerup.com

As for the frequency of VRAM are you sure that 2150MHz is the max speed? BTW I tried to search web for a review of your card and I only found an ASUS TUF RX6800 OC Edition (nonXT).
Is your GPU OC Edition? Is it maybe a dual vBIOS (with a switch?).

How drivers look like on the below page?

 

[CyCloud]

Yeah, it's the OC Edition. I don't think there is a non OC TUF, but I'm not sure.
It does have a dual bios. Was set to P (for performance, I guess) by default, but it also has a Q (quiet) mode that doesn't look appealing, especially with the temps as they are. The fans are really quiet in P mode anyway and when it gets somewhat louder the pitch is not nearly as annoying as many other cards I had before.

The high memory frequency is really weird, it doesn't seem to happen most of the time playing games. I'm gonna keep my eyes open for a while to figure out what triggers it.

Besides the weird memory temps this card is a champ. Even does fast timings and "full" memory OC without issues so far. (need some further testing, though)
If there is anything you want me to test which doesn't involve flashing the bios or disassembly I'd be happy to.
Gonna look into dumping the BIOS this weekend when I got the time for that.

The default settings look like the first attached picture(expanded for visibility), and the second demonstrated the limits the software gives me.
The main reason why I posted here is because HWinfo is the only tool that I know that even displays the memory junction temperature for my card.

 

[Zach]

The reason I suggested TPU forums is that there the knowledge (and ideas) pool is a lot larger because of so much more users than here.
HWiNFO is an awesome monitoring tool and it is very well known there too.

I think that "Quiet" BIOS is lowering clocks too beside fans. But this can be achieved by settings in drivers...

Honestly I'm out of ideas. Nothing really seems abnormal in your HWiNFO readings, except the 2150+ speed on VRAM but that only occurs on gaming(?).
It does not justify the temp increase creep during other simpler tasks, like watching a vid.

For sure I wouldn't ever recommend a BIOS flash for such a matter. And I understand about not wanting to loose warranty by flashing or removing heatsink. I would hesitate to tear down a new card unless I was 100% sure that by doing it it will solve the issue. With the cost of GPUs today, warranty is precious.

Did you tried to tight (all) screws on the backplate? I know one of them has that paint(or sticker) on it, but all others maybe.
Not too much tightening though...

For the record, when ASUS 5700XT ROG STRIX came out almost 3years ago it had a flaw on the cooler/GPU assemble process. I cant imagine they done it again...

Oct 12 2019

Jan 9 2020

 

[Zach]

 

[CyCloud]

Thanks for that suggestion, checked all screws and tried pressing/pulling on different point did anything, but that thing is built like a tank, nothing moves and all screws are tight. It was worth a shot, though.

I relocated the SSD from underneath the GPU, didn't seem to help too much with the GPU, but VRAM temps seem to rise a tiny bit slower and the SSD dropped temperature by 10-15°C, so not a total waste of money. I originally was planning to do that anyway to free up that slot for an NVME drive, but I guess I might drop that idea since it's getting toasty in that location.

Regarding the higher VRAM frequency I didn't find out anything new, seems to be a rare occurrence that only happens for the shortest of burst and never actually stays at that speed.

Another thing I've observed is that the VRAM temps refresh in a odd pattern. They seem to be stuck at one specific value for a couple minutes, seemingly not refreshing at all and then they start updating like the rest of the sensor values for a couple seconds/minutes. But even after are no longer seemingly stuck they don't suddenly skip ahead a few degrees, making me wonder if the values are actually not refreshing or it just happens to be super stable for a little while.

I might put up a post on the TPU forums in a couple of days, but I have to consolidate all the findings into one big post first to not needlessly go over the same points with the people there.
This is really an odd case, if the peak VRAM temperatures were worrisome it'd be easy to just RMA it, but that's not the case.

 

[Zach]

Yes you have to be detailed about it if you go on TPU or you will answer all my questions all over again. PC specs and all your (up to that point) findings/observations.
And also yes about VRAM temps. Its not that bad considering that typically GDDR6 operates around 75~85C.

But I bet now you're like... why it can't be lower at least when not gaming. I would...

 

[CyCloud]

Indeed.
On the plus side, the temps only staying in the roughly 60-80°C means less physical stress due to thermal expansion. That aspect would be much worse if the VRAM would go down to the mid 45s, only to then go up to 80°C and back down to the 40s. lol