Ok @mackbolan777 , good advice.
I did an undervolt and decrease the gpu clock from 2100 to 2000, following the youtube ancient channel, and increase the vram from 1750 to 1800.
I will try during the next week.
85c? Sounds like you might need some case airflow, like a side intake fan.
You shouldn't be hitting more than 70c under full load with an undervolt as well.
I have the ASRock RX 5700 XT Challenger Pro with triple fan design.
• the card is delivered with factory overclock
• the default fan curve which was showing up in Radeon Software was limited to 54% RPM
The factory settings of this card were causing thermal throttling.
I have made my own custom fan curve now and undervolted the card:
GPU temps under load:
• core / mem / VRM: between 60°C and 65°C
• junction temp (hot spot): 85-87°C
• Radeon Software is set to ramp up the GPU cooler fans to 100% RPM at 85°C
• max core clock is set to 2000 MHz and the card is reaching stable 1921-1951 MHz while gaming.
• voltage is currently set to 1025 mV
In fact the 5000 series are considered EOL,
so no driver moving forward will have anything new to add for a 5000 series card.
5000 series is the old model now. So that is no surprise.
The problem with driver version numbers, starts when game developers force you to use the latest driver.
Sure, it is recommended to use the latest one - but sometimes this causes more problems.
My GPU is currently working fine with 20.11.2 driver.
FPS are good and GPU load is working fine and stable.
Too bad that there is no "one size fits all" solution to make that possible for everyone.
Black screen and freezes retorns or wattman crash again... using 20.9.1 or 20.8.3. I update to the last 20.11.2 and ramdom freeze again (playing valorant with competition settings: all set at low).
I'm not sure but, is it possible that using two fixes at the same time I'm doing something wrong?
When I try to undervolt and go with too low with voltage, wattman crash and restore the default settings (after a few seconds I can continue without force a restart). But playing games like Valorant (no gpu demanding) , sometimes it's a wattman crash (specially with early drivers), sometimes black screen and freeze (forcing a restart).
@iTruji are you still overclocking your VRAM?
- this might be the reason for your current crashes and wattman resets.
My own RX 5700 XT has VRAM set to 1750 MHz by default and it is working fine.
But as soon as I try to increase that value, I get crashes in less demanding games.
VRAM overclock of 1800 MHz is able to pass TimeSpy and Firestrike Benchmarks,
but playing less demanding games, causes random crashes for me.
So I leave my VRAM at 1750 MHz and it is working fine.
I am also now using driver 21.1.1 without any issues.
Updated my BIOS to AGESA 188.8.131.52 to use Smart Access Memory (resizeable BAR), too.
I am still able to force blackscreens / system freezes with the wrong settings, tho.
So I keep Radeon Software on "Standard" profile and have my card undervolted.
The settings inn that fix apply to a RX 5600 XT, so your voltage would need to be higher. 1.050mv sounds about right for a RX 5700 XT. When you go removing drivers, be sure to use DDU each time. You should only need the 20.8.3 or 20.9.2 as the newer one's are geared towards the 6000 series and offer no gains to your card, if not more headaches. Yes, you will lose a few FPS and that's more or less by design to lower the temps and increase stability but 950mv might be too low.
The video on Ancient Gameplays via YouTube explains the process in the best detail. 2100Mhz is a bit crazy for a factory OC and probably a "wish list" item from MSI. It's unrealistic with the cooling solution they or any manufacturer uses. In the video I describe above, the OP says to lower the clock to ~1900Mhz then do the undervolt of ~1.050mv or so. Fan curve can help with the heat, try to go more aggressive with less jumps per degree, this lowers the polling by the controller and increases stability.
@RPX100 makes a point with VRAM, don't mess with it or try his setting for that, the 5000 series aren't good with VRAM OC at all. They have plenty of VRAM speed at ~1750-1800, which translates to 14.5Gbps-16Gbps.
Make each change slow and don't flip driver to driver like a madman. If you had a max stability at 1905/950mv that's what I'd use, the few FPS aren't worth the aggravation. But if you still had stability at 1999/1.050v, that's ok too. Mess with the fan curve and you should be good to go with that. The factory 2168/1168mv is insane, however even at the high 95c, still within operating range. A fan curve adjustment could lower that temp to 80c and be stable. Remember to always use a game as your "real world" stability test, because these 3D Mark TimeSpy type tests are the worst case scenario and your card isn't going to be running at those extremes as long or ever.
Games like Valorant are very CPU demanding, so if you're crashing out there, it could be a power supply issue or CPU cooling/setting issue you haven't looked at. You should always use 2 X PCIe power plugs vs. the split dongle for the GPU. One cable can only supply 300w and 9A. The card can pull the wattage relatively safely but you're looking at a ~14A draw which exceeds a single cable's capacity. Cards of these class pull roughly 2.5W per frame, so let that be a guide as why 2 are better than one cable wise.
@RPX100 No VRAM overclocking the last two weeks (stock 1750).
Interesting to read you that stress programs can pass the test but a real game causes crashes (I tested undervolt with furmark).
Have your card undervolted and underclocked?
Reason for underclocking your particular card is because MSI decided to offer more than they can realistically deliver. Also, you'll find it true with XFX THICIII Ultra, they have about the same result of a heat issue but they did reduce their clock to 2025Mhz but run best at 1950Mhz.
If you really want the full ludicrous OC they advertise, complain to MSI and they'll RMA it. The next one might handle the OC out of the box, but I doubt it. How many FPS are we talking about? Less than 20. Not worth the effort to beat the card for that. Driver wise, you'll probably find more stability using the 20.8.3 or 20.9.2. Some have had no issue with the latest one but it offers nothing new for the 5000 series as they are EOL (End Of Life) products, AMD no longer makes but still supports for now.
Check your PSU. Check the CPU/RAM for timing issues like the set the "IF" in BIOS to match half your RAM speed not to exceed 1800Mhz with DDR4 3600. If you're running any 3000 series CPU, try setting the SOC to 1.10v manually. Swap the card into another PC and see if it happens. 1750Mhz VRAM is perfect. Ironically, try 1762 VRAM, the card needs a 12Mhz offset up for some reason I found out the other day but it never presented as an issue for me. That applies to any VRAM OC, so 1800=1812, etc.
Maybe my problems is to jump driver version to another version too fast...
I will start with 20.9.2 and test the undervolt and maybe the underclock later.
With only the GPU undervolted (2100 mhz and 1116 mV) the 95ºC is with the fan at 100% playing COD Warzone and higher than 100ºC using furmark. In Valorant, temps are down 80ºC.
I try again with 1905/950 and going slowly with the test. But if I have a crash, what would I do? Up +10mV? I tested in furmark and it's really stable for 15 minuts but as @RPX100 says, testing in real game can crash even if in furmark is stable.
I will not touch the VRAM clock.
My PSU is a corsair 750 gold and the gpu has two different wires and not the split. Despite this, I'm using this cable a month ago: https://www.amazon.es/LINKUP-Juego-extensión-Cable-Personalizado/dp/B07J9Q9RMG/ref=sr_1_1_sspa?__mk...
And CPU (i5 10600k without OC), has a Noctua D15S and temps are low.