Its louder than I expected. I can't believe this 7nm chip with open air 3 fan cooler is louder than 1080ti FE but it is. Did a bit of undervolting and it was happy with 75mv undervolt but no more. Helped the noise and temps a little but once this thing gets to 2100rpm its very noticeable
Sounds like your card is in the same ballpark as my Vega FE. My Radeon VII undervolts like a champ. Default voltage is specced at 1.128V but I can run mine as low as .991V which is a nice drop. My VegaFE actually undervolted by more (.15V) but was set so high by default (1.25V) that it still needed a lot of voltage to operate.
How are your Junction temps? I hear the card throttles hard when past 100C or so? I'm not referring to core temps.
Luckily, this is fixable.
Depends on how I operate the card. If all I do is undervolt, I can get TJunct down to 80C or less running the fans full-bore. It gets a little warmer than that with the default fan profile, but nothing much above 90-100C (in other words, lower than the reviews). Performance is still normal with this configuration.
I have seen absolutely no thermal throttling yet. Of course, I have not pushed TJunct above about 105-106C in any of my testing.
Now, on to some other data:
It looks like testing the limits of this card at lower resolutions is a non-starter for me. I can't really push it at 1080p. It's hard (but possible, sometimes) to get it to clock high in light workloads or where my CPU may be a bottleneck. But I have found a partial solution to maintaining higher clockspeeds: overclocking.
Because my particular card undervolts so well, it's possible to set higher clockspeed targets with voltage set below defaults. You don't have to follow the voltage curve specified by Wattman when overclocking. In particular, I use this profile right now until I can find something better:
Code:
<?xml version="1.0" encoding="UTF-8"?>
<GPU DevID="66AF" RevID="C1">
<PPW Value="1"/>
<FEATURE ID="4" Enabled="True">
<STATES>
<STATE ID="0" Enabled="False" Value="1701"/>
<STATE ID="1" Enabled="False" Value="1825"/>
<STATE ID="2" Enabled="False" Value="1950"/>
<STATE ID="3" Enabled="False" Value="1701"/>
<STATE ID="4" Enabled="False" Value="1950"/>
</STATES>
</FEATURE>
<FEATURE ID="12" Enabled="False">
<STATES>
<STATE ID="0" Enabled="False" Value="950"/>
<STATE ID="1" Enabled="False" Value="1001"/>
<STATE ID="2" Enabled="False" Value="1070"/>
</STATES>
</FEATURE>
<FEATURE ID="5" Enabled="True">
<STATES>
<STATE ID="0" Enabled="False" Value="1050"/>
</STATES>
</FEATURE>
<FEATURE ID="9" Enabled="False">
<STATES>
<STATE ID="0" Enabled="False" Value="0"/>
</STATES>
</FEATURE>
<FEATURE ID="8" Enabled="False">
<STATES>
<STATE ID="0" Enabled="True" Value="0"/>
<STATE ID="1" Enabled="True" Value="0"/>
<STATE ID="2" Enabled="True" Value="0"/>
</STATES>
</FEATURE>
<FEATURE ID="17" Enabled="False">
<STATES>
<STATE ID="0" Enabled="True" Value="0"/>
</STATES>
</FEATURE>
<FEATURE ID="19" Enabled="False">
<STATES>
<STATE ID="0" Enabled="True" Value="0"/>
</STATES>
</FEATURE>
<FEATURE ID="20" Enabled="False">
<STATES>
<STATE ID="0" Enabled="True" Value="0"/>
</STATES>
</FEATURE>
<FEATURE ID="21" Enabled="False">
<STATES>
<STATE ID="0" Enabled="True" Value="0"/>
</STATES>
</FEATURE>
<FEATURE ID="22" Enabled="True">
<STATES>
<STATE ID="0" Enabled="False" Value="29"/>
<STATE ID="1" Enabled="False" Value="100"/>
<STATE ID="2" Enabled="False" Value="50"/>
<STATE ID="3" Enabled="False" Value="100"/>
<STATE ID="4" Enabled="False" Value="72"/>
<STATE ID="5" Enabled="False" Value="100"/>
<STATE ID="6" Enabled="False" Value="94"/>
<STATE ID="7" Enabled="False" Value="100"/>
<STATE ID="8" Enabled="False" Value="105"/>
<STATE ID="9" Enabled="False" Value="100"/>
</STATES>
</FEATURE>
</GPU>
If you are observant, you'll notice that I moved my lower clockspeed range up, dramatically. I have the minimum specific clockspeed target set to 1801 MHz @ .991v and the maximum set to 1950 MHz @ 1.080V . It's important to note that you do
not want to set the minimum and maximum clockspeed target to the same thing. It does not hold clocks at a higher level, and it can cause bizarre dips in memory clockspeed some of the time. Not sure why.
I've also noticed that it is very easy to overclock memory. I have no third-party tools to confirm that the memory overclocks are working, though some of my benchmark performance numbers seem to suggest that it helps. Right now I'm running memory at 1050 MHz. I may try more later. Sadly, I have no way of monitoring HBM hotspot temp at this time, so I'm a bit reluctant to push too hard, especially when I have no memory voltage controls.
In any case, manipulating clockspeed targets set my average GPU clockspeed closer to a constant speed of 1801 MHz than default settings. I still get dips into the high 1600s for reasons I can not explain. Temperature does not seem to correlate to the dips. I have the fewest dips in 1440p and 4k benches. I also get higher power draw in benches where I see fewer clockspeed dips (imagine that). Overall, performance is now improved. I picked up around 1100 points (~17000->~18100) in Unigine Superposition 1080p medium versus default settings.