Sapphire PULSE RX 580 Throttling & Fan Speed Problem

[AhsokaTano]

I have a very weird problem with my Sapphire PULSE Radeon RX 580 8GB card.

I bought the card couple of months ago from a feller who used it for mining Ethereum. At first everything seemed fine, except the card's fans would quickly reach 3,000 RPM and were very loud while gaming. After I close the game (i.e. Red Dead Redemption 2), fans still keep running at full speed, and I always have to shutdown my PC to reset their RPM to normal levels. I tried to apply custom fan curve via AMD Radeon Software, however, the fans still kept behaving the same, as if they were able to completely ignore the custom fan curve settings. Then I tried using MSI Afterburner to set the custom fan curve, and then I got my problem solved.

Last week, however, I decided to overclock my GPU. I ran some benchmarks and closely monitored voltages, frequencies, and power consumption. Then I found out something strange. Apparently the card starts to throttle once it reaches temperature of 70 C (regardless of custom fan curve being enabled or disabled). Until then the GPU runs a 1,366 MHz as expected, but once it reaches 70 C, it starts to gradually decrease clock to 1,250-1,300 MHz while gaming, and it even goes all the way down to 800-900 MHz while running OCCT 3D stress test. The card NEVER reaches 75 C. Another weird thing is that the max. GPU voltage it can reach is 1.075 V (maybe it peaks at 1.1 V according to HWINFO64, but it only lasts for like one nanosecond). AFAIK, the Sapphire Radeon PULSE RX 580 is designed to run at 1.15 V at max. speed. Right? Also, it shouldn't start to throttle at 70 C, as its max. temp is rated to 84 C, while its target temp is rated to 75 C.

I decided to check whether I have modified GPU BIOS installed or not. GPU-Z reported that my Device ID is "1002 67DF - 1DA2 E387" (see attached images). It also reported that my Board ID is "113-1E3870U-O4B", and that my BIOS Build Date is "2017-11-15 01:27". I immediately checked TechPowerUp BIOS DB, and found that same BIOS here. Then I compared checksums of both BIOS files, and they turned out to be completely identical. So I guess I have stock BIOS, right?

I also took out the GPU from my PC to check its SKU, part number, and serial number. The SKU of the card is "11265-05-20G" (the full SKU is printed on the box it came with), and its part number is "299-1E387-000SA" (see image). I'm still not sure if my card features Samsung's or Micron's memory, but according to GPU-Z it is Samsung's. All in all, it seems that my BIOS is legit, and is not the cause of thermal throttling and weird fan behavior.

I'm not sure what to do next. Should I flash a new BIOS? But which one? I also reinstalled Windows 10 two times, and the same thing always happens. I also installed multiple versions of AMD Radeon Software, but it didn't help. I mean the card runs fairly fine. Gaming performance is fairly good. However, I was hoping to overclock the card to 1,400 MHz, and undervolt it to 1.075 V. But if the card already throttles at 1.075 V, and at 70 C, then doing this would be useless.

Should I re-apply the thermal paste and change the thermal pads on memory modules? I thought maybe GPU VRM is overheating, however, HWINFO64 says that it only goes up to 59 C under full load.

Why would the card start throttling at 70 C? I'm completely puzzled. The more I know, the less I'm sure what to do next.

 

[Caring1]

Sell it and buy one that isn't a mining card.
Google the part number and you will find multiple people with issues with that card.

 

[DeathtoGnomes]

when OC'ing try to play with a reduced the power limit, 5% after first then 1% increments until max temp is around 68 degrees.

have you repasted?

 

[AhsokaTano]

I have fixed the problem! I tore down my card and found a huge hotspot in the middle of a GPU. The thermal compound completely dissolved at the center (see images), and there was nothing to draw away the heat from that area. I reapplied some Cryorig CP7 thermal paste, and put the card back together. The temperatures are now around 68 C under full load, and fans are MUCH quieter than they were before. I no longer need MSI Afterburner to apply the custom fan curve. The card now runs rock solid at 1366 MHz while gaming. All I had to do was change some thermal paste. How simple!

There are few interesting things worth mentioning. The card was clearly used for mining, however, it has never been torn apart until now, since warranty stickers were present on the screw heads. The heatsink was very clean. There was very little dust inside it. However, the thermal compound on the GPU was extremely dry. My guess is that nobody reapplied the paste ever since the card left the SAPPHIRE factory in 2017. I also noticed that the thermal pads under the VRM portion have considerably dried up, and will probably need to be changed in the near future (see images).

I was worried if the card will throttle again once it hits 70 C, however, that didn't happen. The card now operates according to Sapphire's specifications, which means it doesn't throttle until it hits 84 C. The fans now also work very quietly, and there is no crazy spinning at 3,000 RPM once it hits 71-72 C. In fact, fans are now spinning at around 1,600 RPM at 75 C.

It is weird how sensors weren't reporting excessive operating temperature, even though the GPU was overheating. However, diagnosing thermal hotspots is very difficult without using a thermal camera. I found a very interesting article on this subject at IgorsLab mentioning RX Vega 56 cards. If I understood correctly, the sensors aren't always able to report temperature values across the whole surface of a GPU. For example, if there is a small area of the GPU that is not covered with a thermal paste, it will overheat, but sensors will still report the temperature of the whole package, which may be within safe values. This way, you can experience thermal throttling, even if the temperature readings seem normal. Diagnosing thermal hotspots is much easier with CPUs, since each core has its own temperature sensor, and you can easily diagnose which core/area of the CPU is overheating.

Anyway, I'm happy I have fixed my RX 580. Especially since buying a new card these days would be extremely expensive and difficult due to cryptocurrency craze. I have already overclocked the card, and it runs fine. The GPU core now runs at 1,400 MHz (1.075 V), while VRAM runs at 2,100 MHz (950 mV). Power limit is set to max (+30%). Overall, it's a mild overclock. Fans are working in accordance with settings from BIOS. I ran OCCT stress test for an hour, and the card's temperature was locked to 75 C, and there was no throttling. I'm slightly disappointed that overclocking the card didn't improve performance in games at all. I gained more performance improvement by just reapplying new thermal paste and eliminating the thermal throttling. Only The Division 2 achieved +2 FPS on average. Nevertheless, the card now consumes about 5-15 W less, and it runs much quieter and cooler.

I hope this post helps someone out there. Lastly, I would just like to say that SAPPHIRE makes absolutely amazing cards, and that they are well worth the money. My PULSE Radeon RX 580 was probably throttling for months, maybe even years, and it is a miracle that the card didn't sustain any damage. The card's BIOS did a great job protecting the GPU, and the card now operates perfectly, despite struggling with overheating for a very long time.