[SOLVED] Can´t undervolt my Rx 480 (crashes)

tinoferrol

Honorable
Sep 16, 2017
17
0
10,510
-Asus Dual RX 480 4GB OC
-Driver 19.9.1
-Windows 10 64

I can´t even lower voltage of te P-States not even a little bit, because suddenly temps begin to raise (when it should be the opposite) and finally it crashes and restarts Wattman settings to default. last 3 P-states are this ones: 1100, 1156, 1150 mV

Other times my computer just freezes and restore default. It seems I can´t touch anything in Wattman because it crashes even with minimal variations... What's the problem? I had another 580 and had no trouble at all downvolting...
I would like to do it because the card can't reach the 1300 clock speed almost never...
 
You have bad luck, you have the worst quality silicone in your specimen. That is why AMD put in too much voltage for most cards because the few ones of low quality should also run with standard settings. bad luck man! :(
Rather than worse what he got probably the standard for most of AMD chip. TBH i dont think just because let's say 15 to 30 percent of AMD chip need higher voltage for stable operation AMD end up pushing the voltage higher for the rest of other 70 to 80 percent chip that can operate on lower voltage. Worse chip actually end up being use in cards like 470D and such.
 
Rather than worse what he got probably the standard for most of AMD chip. TBH i dont think just because let's say 15 to 30 percent of AMD chip need higher voltage for stable operation AMD end up pushing the voltage higher for the rest of other 70 to 80 percent chip that can operate on lower voltage. Worse chip actually end up being use in cards like 470D and such.
They actually did. They set voltage for the worst chips, and everything else just gets pumped hotter than necessary.
 
You sound fairly knowledgeable, so forgive me for the following questions:
  • Have you tried using a different driver version?
  • I don't see you mentioning anything about temps. What are temps?
  • I'm familiar with P states and voltages, but your description was difficult to understand. You listed 3 voltages but only one of the 7 P state frequencies.
  • Are you leaving memory timing on auto?
  • What did you set for VRAM voltage?
 
They actually did. They set voltage for the worst chips, and everything else just gets pumped hotter than necessary.
Except the way indutry work where they will not going to do to that. As i said the worst silicon are regelated to product likes 470D. The performance is similar to 470 but power consumption is similar to 480. AMD is aware about their disadvantage in power efficiency that's why AMD is pushing hard to make RX480 into a 150w GPU which later create a little bit of fiasco when the card actually consumes a bit more than 150w. Increasing the voltage for the majority of their chip when only small of their chip can't operate at such voltage defeat the purpose when AMD try to make their GPU as efficient as they can. Hence most often the "bad" chip was turn into another product so they did not just throw them away because of certain defect that will not going to affect how the chip work correctly. He said that his RX580 able to downvolt successfully. That's most likely due to the improvement AMD has done with polaris 20 chip over polaris 10. In a way that might be similar to what nvidia has done with GF100 to GF110 before. Just in RX580 case there is no reference design for it. Board partner have full control since the very beginning with their custom board. Most often board partner push thing higher so RX580 can operate at much higher frequency that not possible with RX480.
 
So you're saying they set voltage for the average chip, and the worse chips get....?
A) discarded - now the average chip is the worst again, but your average has now shifted and the voltage can be lowered, so you end up successively discarding every chip you make.
B) the worse chips get shipped malfunctioning because they're not getting enough voltage to be stable?

Conversely, if Polaris' AVFS was as good as they claimed, not a single Polaris GPU should be able to be undervolted (zero safety factor) OR they all would (fixed safety factor, like 50mV)

I'm in no way disagreeing with the possibility that the OP got one of the worst RX480s in existence.
 
Last edited:
  • Like
Reactions: TJ Hooker
worst RX480? probably not. take this for example: how many RX480 able to hit 1500Mhz and use that clock on 24/7 basis? then compare it to RX580. i still remember when people raging about XFX RX480 (i don't remember which one) able to hit 1475mhz on jayz videos (hand picked by XFX of course) and they said this specific model probably guaranteed to hit at least 1450mhz, but there are people with the same exact model can't even get passed 1400mhz. undervolting is also like that. i see this wrong assumption by majority of people that almost all AMD chip can be undervolt successfully. only very very small percentage of chip unable to do it. but because of this tiny percentage cannot operate at much lower voltage AMD decided to increase the voltage for the majority of their chip which else are doing just fine with much lower voltage. it does not make sense at all. it's like putting their majority of chip at disadvantage just because of this small percentage chip that cannot operate at much lower voltage. AMD has such voltage for majority of their chip because majority of their chip need that much voltage for stable operation.
 

tinoferrol

Honorable
Sep 16, 2017
17
0
10,510
You sound fairly knowledgeable, so forgive me for the following questions:
  • Have you tried using a different driver version?
  • I don't see you mentioning anything about temps. What are temps?
  • I'm familiar with P states and voltages, but your description was difficult to understand. You listed 3 voltages but only one of the 7 P state frequencies.
  • Are you leaving memory timing on auto?
  • What did you set for VRAM voltage?
-I tried whit the previous one, 19.5.2; no changes
-In FurMark, preset 1080 -> 73ºC fans 100% (I replaced thermal paste)
- p7 (last): 1300 MHz 1150 mV
p6: 1270 MHz 1156 mV
p5: 1225 MHz 1100 mV
- Yes in auto (as I allways did in undervolting)
-VRAM voltaje is at stock, 1000 mV 1750 MHz

Thank you for the answer...
 
Sorry, I've got nothing.
Your voltages seem plenty high from my experience and temps are plenty good, but as Renz and I said, AMD had to set their "Auto" voltage profile based on the worse/worst reference chips. It would seem you're in possession of a trend-setter!

If you're looking to save power/heat/noise, you're probably just going to have to rely on Frame Rate Target Control (for fixed refresh monitors) or Chill (for FreeSync monitors).

There's also a "Power Efficiency" toggle that can be used when on Auto voltage, but that's a black box. I've tested it before (data logging on GPUz and FRAPS), but I have no idea how it decides what to do how much and when. There's a possibility that enabling this feature could result in instability also, but it's worth a shot.
 
I forgot about that article.
I did skim back over the article, but don't have enough time to fully analyze the issues they had (it would seem they're still allowing something to behave "automatically").
[Edit]: Ok, it looks like in that review, the Balanced/Auto setting was already applying ~1.025V anyway, so if you call 25mV an "undervolt".... Still odd that a 1V clamp wasn't being treated properly without the power limit increase. Then you look at the frequency charts and undervolting by 25mV with a max power limit gained them 125MHz core. This article was definitely using/relying on auto frequency, and there's definitely some weird stuff going on with the AVFS thinking it can go....higher frequency than default with...less voltage just because it got tricked into thinking it could apply more power....but it wasn't....

I haven't experienced those sorts of problems on Polaris cards using dynamic frequency and manual voltage though. They expose all 7 frequency states in WattMan and you can set/clamp the voltage for each. (Again in my experience) the cards never deviate from the Frequency/Voltage that you've set. In this method, you're completely taking the reigns from the AVFS and telling the chip what to do yourself. But heck, it's not difficult at all to try sliding the Power Limit slider up and see what happens!!
 
Last edited:
  • Like
Reactions: TJ Hooker