AMD 5700 XT Hot spot/Instability

[Mitukczuk]

Hello,

I had to postpone my graphics upgrade so meanwhile I am looking into what may be causing occasional instability (reset when gaming) of my MSI AMD 5700XT.
See the attached reading from the CPUID HW Monitor - I am not much of a GPU guy in terms of knowing the voltages, OCing etc. I know the 5700XT run a little hot but is something wrong in the picture? Power draw? Other temps? The hot spot is a hot spot for a reason but 110-112 is up there.
The card runs great 9/10 times but sometimes the PC hard resets when gaming - RDR2 / Jedi Survivor. It is not often, but it happens from day 1 of owning the card (its been three years).

Any ideas?

 

[Kenrou]

I had the same issue with my old XFX 5700xt Thicc III, core temp was fine but hotspot was 20c-30c hotter, ended up returning it after repasting and not really getting anywhere with undervolting. I assumed bad cooler mount because there was a lot of kerfuffle around these at the time

In your case, you're hitting 1.2v under load, that would certainly explain those temps, undervolt, undervolt, undervolt...

 

[Mitukczuk]

I had the same issue with my old XFX 5700xt Thicc III, core temp was fine but hotspot was 20c-30c hotter, ended up returning it after repasting and not really getting anywhere with undervolting. I assumed bad cooler mount because there was a lot of kerfuffle around these at the time

In your case, you're hitting 1.2v under load, that would certainly explain those temps, undervolt, undervolt, undervolt...

Thanks for your reply.

I don't have much experience with undervolting and generally messing with the GPU stats.
Do you have any recommendations where to start? I saw some settings in the Adrenaline Software. Any safe numbers that might work?

 

[Kenrou]

Check the video below, different GPU but same principle. Put everything at stock. I would advise starting with memory, increase by 20, test, increase by 20, test, until you can't go any higher, then lower by 20 and lock, you will see image artifacts almost straight away if it's unstable. For power, lower voltage by 10mv, test for ~30m, lower again, test for ~30m, rinse and repeat until it crashes, then raise the voltage to the last stable setting. After you lock both power settings and memory, test again for ~2h, this SHOULD be more than enough for extended gaming sessions, if not, increase by 10mv and re-test. Congratz on your 1st successful undervolt

For overclocking, same principle as memory, increase frequency while testing for power stability, increase, test, increase, test, rinse and repeat until it won't go any higher without crashing while keeping temps/fan noise under control. Yes, it's very tedious, but it's usually well worth dropping power and temps while keeping the same stock performance or better.

You can download Unigine for free on their website for testing, use the setting 1080p Extreme for maximum heat ⇾ https://benchmark.unigine.com/superposition

 

[Mitukczuk]

Check the video below, different GPU but same principle. Put everything at stock. I would advise starting with memory, increase by 20, test, increase by 20, test, until you can't go any higher, then lower by 20 and lock, you will see image artifacts almost straight away if it's unstable. For power, lower voltage by 10mv, test for ~30m, lower again, test for ~30m, rinse and repeat until it crashes, then raise the voltage to the last stable setting. After you lock both power settings and memory, test again for ~2h, this SHOULD be more than enough for extended gaming sessions, if not, increase by 10mv and re-test. Congratz on your 1st successful undervolt

For overclocking, same principle as memory, increase frequency while testing for power stability, increase, test, increase, test, rinse and repeat until it won't go any higher without crashing while keeping temps/fan noise under control. Yes, it's very tedious, but it's usually well worth dropping power and temps while keeping the same stock performance or better.

You can download Unigine for free on their website for testing, use the setting 1080p Extreme for maximum heat ⇾ https://benchmark.unigine.com/superposition

Ok, so its pretty much like CPU overclocking but it seems a little bit safer given its not done in BIOS which always seemed a bit more serious to me. But that's just my personal feeling. I WILL be playing around with the GPU in the upcoming days. Thanks a lot!

 

[EarthDog]

Keep us posted on your progress!

 

[Mitukczuk]

I didn't have much time but I tried to get at least something though with not much of a result. The hotspot runs wild and the temps are basically the same.
I lowered the Voltage to 1150mV and increased the clock to 1790Mhz and the readings in HWMonitor are from about an hour long session of RDR2 so it seems to be stable.

Is any of the settings completely wrong or should I experiment even more? Is FurMark a good testing tool in this case? Or is it too short of a test?

 

[Zerileous]

It still seems to be hitting 112c and drawing 1.2v. Sometimes the drivers are strange with that. If you're working well now then good to go and you can disregard these suggestions. I do have some other suggestions though. In a different thread you mentioned some unusual stuttering behavior and I'm now wondering if that is related to thermal throttling.

Since this card has been in service for a bit, you can give the cooler a good cleaning and also check your case fans etc. IIRC the hotspot issue was common with the generation to have such a large delta, but if it were me I'd like to have it a little below the limit. Repasting might help but it can be a lot of work and probably won't do much, but a quick dusting of the fins is needed for sure.

Second, just for the sake of eliminating the problem, there might be a default "quiet" mode in the Radeon control panel. (I got mad and don't currently have it installed or I would be able to give better directions). I would avoid the one that says "undervolt my GPU" though, as I believe that is an automatic overclock thing and those rarely work well.

The next quickest/easiest way to drop the temperature is to move the power slider into the negative.

The theory with undervolting OCs is to trick the card into overclocking itself, using the built in boosting behavior, drawing less voltage at a given power table allowing it to pull more power/wattage (with an increased power slider). Sometimes this can help temps, but other times it can increase them. It all depends on what is holding things back in a given setting, and I'm not familiar enough with the card to say for sure. However, I can say that the best performance will be at a setting that doesn't throttle or cause the GPU to shut down, even if the overall clocks are lower.

 

[Mitukczuk]

It still seems to be hitting 112c and drawing 1.2v. Sometimes the drivers are strange with that. If you're working well now then good to go and you can disregard these suggestions. I do have some other suggestions though. In a different thread you mentioned some unusual stuttering behavior and I'm now wondering if that is related to thermal throttling.

Since this card has been in service for a bit, you can give the cooler a good cleaning and also check your case fans etc. IIRC the hotspot issue was common with the generation to have such a large delta, but if it were me I'd like to have it a little below the limit. Repasting might help but it can be a lot of work and probably won't do much, but a quick dusting of the fins is needed for sure.

Second, just for the sake of eliminating the problem, there might be a default "quiet" mode in the Radeon control panel. (I got mad and don't currently have it installed or I would be able to give better directions). I would avoid the one that says "undervolt my GPU" though, as I believe that is an automatic overclock thing and those rarely work well.

The next quickest/easiest way to drop the temperature is to move the power slider into the negative.

The theory with undervolting OCs is to trick the card into overclocking itself, using the built in boosting behavior, drawing less voltage at a given power table allowing it to pull more power/wattage (with an increased power slider). Sometimes this can help temps, but other times it can increase them. It all depends on what is holding things back in a given setting, and I'm not familiar enough with the card to say for sure. However, I can say that the best performance will be at a setting that doesn't throttle or cause the GPU to shut down, even if the overall clocks are lower.

Thanks for your thorough reply man!

The card has been troublesome from day one with the random shutdowns so I dont think its an issue of the fans being dirty (I clean them at least once every three months. I have filters and the case is quite clean as well Id say.)
But they sometimes happen, sometimes not. Sometimes after 20 mins, sometimes after 3 hours of 100% use. Its been very inconsistent.

Also, when I did some tweaking in the MSI afterburner, the temps did respond a bit better with hotspot running at 105c. But I tested it only for about 15-20 mins. Could the card respond better with a different software?

 

[Zerileous]

Yeah it is a possibility. I would just not advise changing things in both afterburner and radeon, as they could conflict. I have AB installed right now. I have had both running before without issue, but I would just pick one and stick with it. I think 105c is just fine for the hotspot and should keep you from throttling.

Yes it's high, but the most important thing to pay attention to is the specified temperature parameters. Due to differences in how each product (hardware) measures, calculates, and reports temperatures, it isn't apples to apples across platforms or devices. If you can play for a while and not experience shutdowns, then you know it was probably thermal. If not, then possibly the thermals were functioning as intended and something else is the problem.

 

[Kenrou]

As far as I remember, the hotspot on the 5*** series was sorta designed to run full tilt and is usually hovering on the 90c-110c margin even with good cooling, it was stated it was a "perk" of the new gen (RDNA 1 I think?) that caused a lot of arguments when it came out because, ofc, people thought their GPUs were burning up. Supposedly, as long as core temps are 90c or below (lower the better), hotspot 110c or below (starts throttling at 110c), AMD says you are technically "within spec". Afterburner caused issues with the old AMD control panel, honestly don't know about this one because I sold mine right around the time it came out

One thing that really ticked me off about mine (besides the really crappy drivers at launch) was that the fan speed was going off the hotspot instead of the core, so (at stock) the fans were going full bore pretty much most of the time, with a little luck it's different with other brands. Good video on your GPU:

 

[Mitukczuk]

At first I thought I can't make it but at the end, I was able to buy a replacement and got Gigabyte 7800 XT.
It is very interesting to see the card perform when compared to the 5700 XT. The Hotspot sat at 83 when playing Jedi Survivor. The other temps looked about 20 degrees lower as well. It runs at 1.1V but draws circa 350W.
The card is MUCH quieter too.
Interesting stuff. Hopefully my PSU holds up

 

[Zerileous]

PSU should be plenty, your CPU only uses about 60W iirc. I bet you barely pull 500W from the wall. Is it very old?

 

[EarthDog]

That 342W is a blip id imagine. It's a ~265W card. Still, even if it was 350W, your PSU is fine so long as it's working properly.

 

[Mitukczuk]

PSU should be plenty, your CPU only uses about 60W iirc. I bet you barely pull 500W from the wall. Is it very old?

It is actually 3 years old TODAY.

 

[JLK03F150]

It is actually 3 years old TODAY.

Your Platinum PSU has a 10 year warranty, so I would say you are fine with that unit for several more years.