Download: Radeon Software Adrenalin 2019 Edition 19.9.3

#5717326 Posted on: 10/05/2019 02:33 PM

---

Speaking of Wattman, fan curve still broken? Last time I used it when the fan speed changed in steps, not smoothly. Like this:

Source: https://forums.guru3d.com/threads/fan-temperature-hysteresis-questions-steps-linear-curves-and-speed-changes.428368/

You don't have manually set FAN profile for Wattman I hope? That would interfere with MSI Afterburner manual FAN settings.

#5717330 Posted on: 10/05/2019 03:06 PM

---

Perhaps you misunderstood me. I found this screenshot on the Internet and added it just to visualize how Wattman controls fans. Well, at least earlier.
I never use similar software together. Previously (about 4-5 months ago), when I tried to configure something in Wattman, Afterburner was turned off and removed from startup. Problem, if I may call it that, was in Wattman, no other software could affect it.

#5717427 Posted on: 10/05/2019 11:15 PM

---

I'll try those when I get home.

#5717577 Posted on: 10/06/2019 05:47 PM

---

Speaking of Wattman, fan curve still broken? Last time I used it when the fan speed changed in steps, not smoothly. Like this:

Source: https://forums.guru3d.com/threads/fan-temperature-hysteresis-questions-steps-linear-curves-and-speed-changes.428368/
It still does that on Vega.
Navi appears to have a curved fan curve instead of them straight lines connecting da dots, maybe it works properly on that?

#5717586 Posted on: 10/06/2019 06:15 PM

---

For Navi it uses I think it's four stages from lowest to highest between 30 Celsius for minimum and 100 Celsius for maximum but this works for the reported junction temperature which is different from the GPU core temperature sensor reading though from testing both seem to be pretty close to one another hovering between 30 Celsius at idle upwards of 60 to 70 Celsius under load though I do run a cooler ambient air temperature in this room and the fan is a bit higher than stock although I believe AMD kept a 40% curve for the stock Navi GPU models.

Gradually increasing as the temp goes up but once it hits peak I believe the GPU starts trying to throttle to maintain around this level if at all possible instead of as earlier ramping up to higher and then if that fails just pushing near 100% fan speed though if the GPU junction temps should get close to the threshold I would expect Navi to also ramp up fan speed as needed.

Can be important if setting a fan curve that is silent but the GPU is edging close to the limit and thus starts trying to reduce overall clock speeds for how the GPU is constantly re-adjusting its speeds and you're not getting the full boost performance as a result.


Oh and VRAM tends to be a bit hotter sitting at around 5 - 10 degrees above the reported GPU temperature ratings that I am seeing at least through GPU-Z's sensor information on this so with a lower fan curve the GPU might be nice and all but memory might be pushing near 80 Celsius instead.
(Stock voltage on the Samsung models and I guess the Micron ones too appears to be around 1.35v from what I am finding. Assuming that it's over the default same as the base clock speeds are above actual specs thus a bit more heat generated.)

No Wonder the Nitro+ cooler utilized a third fan and a even bigger secondary block of heatsinks for the VRM's and memory modules and the result is a good drop in overall temps compared to Pulse which I'm using.
(Undervolted and slightly underclocked but barely any noticeable performance differences, stock though junction temperature can hit near 90 Celsius and memory is probably not far behind and might be sensitive to high temps affecting their performance or at worst stability.)


Vega should be doing a similar thing but instead it just switches between states as temps get closer to these little "steppings" so that makes it a bit more problematic to set something akin to a fan curve instead of this little design and AMD not revising it since the Adrenaline 2019 drivers came out.



EDIT: As for performance comparing some reviews and then differences in clock speeds between 5700 and the 5700XT but also the higher quality (Better binned?) 5700XT Anniversary Edition it seems the default can see boost clocks drop around 1700 Mhz on more demanding titles and less demanding ones having the GPU run cooler hit 1900 to 2000 Mhz.

So there's definitively something of a effect although it's not just heat affecting how the boost function works but it's part of maintaining a higher game clock and full boost clock performance value instead of having it decrease.

Although it can fluctuate all over the place due to taking into account several other factors but on average it's possible to maintain around 1800 to 1900 Mhz boost with some tweaking even for more demanding titles possibly higher.

I assume water can keep even the most demanding workloads under control far as temps are concerned at least, might be important for Big Navi if it has more cores and other bits in the GPU chip running hotter than the 5700 variants do now but capable of another 30% some performance gain or whatever it might end up at.
(Might be less, could be higher depending on core count total say back to 3072 instead of 2560 or going above 64 = 4096 to some higher value but also how well it then scales and what it would need and then this is still RDNA 1 and the 5800 perhaps and not RDNA 2 and what that might bring.)



EDIT: But then again Vega has a similar thing and Vega20 or Radeon VII utilizes some of the features that Navi would then bring out plus OverDrive 8 but the same effect holds true, GPU can be limited from a number of conditions so P6 clock speeds and P7 boost can be tough to maintain and this little issue with fan curvature or stair steps isn't the most optimal for a optimized curve and fan rampup and cooling though it can be worked with getting some effect.
(Although it then switches up and down quickly which affects noise levels at least if it ramps up suddenly like that to the next "step" in the fan setup for temps and speed.)


Heh HBM is also quite sensitive so that shares similarities too for needing some extra care as to not lose performance particularly if running mods or overclocks on it's default settings.

(To the surprise of absolutely nobody I guess but yeah sensitive cards far as getting ideal performance out of them needing a bit of tuning and tweaking to not hit this or that bottleneck or threshold for anything from temps to power to overall utilization and other factors.)




EDIT: Think I tracked down the display corruption issues on my end, it's another CEF or Chromium Engine Framework issue which AMD hasn't resolved 100% yet, pretty sure the Discord one would use a compatibility flag or profile but then other programs are susceptible.

Steam has hardware acceleration off and I also disabled the browser component entirely via a property flag.
Uplay has neither of these and the issue does tend to manifest from when the UPlay client pops up after exiting any game that isn't bound via Steam but their client directly so after a few more days of simply not using UPlay that should be on the way to being confirmed though could be coincidence too but I suppose in a few days of testing that will be confirmed.

EDIT: Though that said the Discord issue is report as improved not fixed with users still hitting display corruption, driver crashes, black screens and BSOD's though since it's a mix of a number of different issues some of these would likely be from other problems.

EDIT: GPU acceleration is also used by many other programs in various ways, Direct2D and so on, might be something overall here since 19.7.5 -> 19.8.1 but I don't know for sure though 19.7.5 would in turn have media playback hardware acceleration issues as unresolved so it's one problem or the other until later drivers can fix whatever this is coming from.



EDIT: Hmm but then that should be occurring more during the start of Uplay/Orbit iself and less in this method unless it's tied to how the GPU is working or when it's clocking down from exiting a game. Needs testing.