Graphics Cards

[offler]

Hello.

I installed my Radeon VII on saturday and did few tests during sunday.

Witcher 3 @ 4k
Does run well at 59-60 FPS, after some time the card start to throttle and FPS drops to 50 or below.

Kingdom Come: Deliverance @ 4k

It runs on 47 FPS as its unoptimized and i use high-resolution texture pack. But even here its obvious that thermal thtrottling starts.

In both cases the throttling starts due Tjunction temperature above 110°C. From the relesed data is known that the VII chip has 64 thermal sensors and throttling starts regadless which of these sensors report high temperature.

The same behavior was visible during synthetic stress test using MSI Kombustor at 1080p. Over 45 minute test when i decreased power to -20, overall GPU temperature was not higher than 83°C, but Tjunction was over 110°C...

From the screens I have seen at YT channel by DerBauer is clear that vapor chamber has been cut and surface is quite flat. Replacing thermal compound even with liquid metal may help a bit, but the primary cause here seems to be not just the type of thermal paste/TIM but the pressure distribution.

amdmatt‌

a) Is it possible to add to Wattman a view which allows me to check each of the sensors individually?
b) I will not replace thermal compound (i guess the graphite should be OK), but there are mods which increase the pressure a bit. Apparently bad contact seems to be an issue, even when its obvious that Thermalpad is glued to both cooler and chip.

Apparently this might help:

[offler]

I did the mod with plastic washers with 0.2mm thickness.

When power is set to -20 and GPU is stressed with MSI kombustor, no throttling occurs as top temperature on the hotspot is 107-108°C. Overall GPU temp decreased from 83°C to 78°C.

Techpowerup used in their review metal washes 0,5mm thick and they claimed roughly 10°C improvement.

DerBauer replaced the thermal compound with liquid metal and had 5°C improvement.

Basically the cooler itself is OK, even a bit noisy, but everyone should be looking for a way how to get much better thermal diffusion on every sensor spot.

[offler]

amdmatt‌

Ok, after few more days, the heat is still high, so i assume the hotspot on the chip is either:
a) due too high pressure on the vapor chamber
b) A bubble under the TIM...

So i guess i have to remove the cooler and check it...

[averagebuild]

I am having issues with high heat myself, and it is concerning! I really wanna know what they were thinking when they introduced this card! I love AMD and have gone as far as investing in them, but now I have to reconsider getting a NVIDIA card. I am dissapointed. I am giving them some time due to drivers possibly need to be ironed out, but I am losing my paitence.

Seperate note: Do you think that if there was some sort of design flaw that they would do a recall and repair it? 

[hardcoregames_]

Thermal pads are not all that good, which is why I have had to refurbish a lot of video cards

[hardcoregames_]

I use MX-4 and I have done a few video cards that were overheating, works well

post pics if you dissect your card

[offler]

I removed the cooler to see whats wrong.

a) The contact surface was (and sligtly still is) concave.
Carbon thermal pad lost contact with the cooler in the center.

b) I lapped it a bit, used Arctic Silver 5, and after one week when the thermal paste was slightly cured, i got to like 29-36 delta between GPU temp.
On Tjunction it was slightly worse, but GPU temp went quite much lower to 68°C

c) I lapped it for the second time, this time I used Noctua NH1 which does not need any curing time.
Tjunction remained absolutely same, GPU temp went to 78°C, delta decreased to 28-34°C.

This time the lapping is much better...

So i will reapply again Arctic Silver 5 again, and I will look forward for Thermal Grizzly Carbonaut or Hydronaut.

[hardcoregames_]

I use MX-4 which is not conductive and it seems to last 

[offler]

So...

These are results after lapping. Power minus 20 percent, fans on full speed, rest on default.

https://www.overclock.net/forum/attachment.php?attachmentid=258336&stc=1&thumb=1&d=1552226157 

Used thermal paste is Arctic Silver 5, readings were taken one hour after it was applied. Today it was 63°C GPU and 93°C Tjunction. It will be probably slightly better in few days.

[harbi]

I'm very tempted to try out the "washer mod" as I want some room to overclock. I only undervolted my Radeon Vii, playing at 4k, uncapped fps, no bottleneck. 

Mine is stock 1075 mV, manual undervolt to 989 mV

Edge Temp went down from 75c to 64c
Junction temp from 105c to 88c

.

^ No overclock + manual fan curve ( max 59% / 2750 rpm ) when hitting 85c Junction temp.

[offler]

Oops. Wrong link.

https://www.overclock.net/forum/attachment.php?attachmentid=258336&d=1552226157 

Also now are temps a bit better. its 62°C and 92°C as the paste cures.

[hardcoregames_]

have you tried the fuzzy donut yet?

how hot does that get?

[offler]

Test details:
1920x1080

MSI Kombustor logo
Not sure about API, the default one.

Driver is set to defaults, then minus 20 percent to power, maxed RPM on fans and the numbers after 10 or so minutes on MSI kombustor are 62°C GPU temp and 92-94°C.

In this scenario the GPU isnt throttling at all, but is power capped. Each more 5% on Power equals like 3-4 more °C on Tjunction on the same settings above.

A tthis point its like 10-20°C better like with original carbon thermal pad.

[hardcoregames_]

Furmark (fuzzy donut) is what I use to test cards for thermal stability. If they can handle 15 minutes or more than they are deemed game ready.

My EVGA GTX 1060 card gets warm but it handles the load so the card should be good for a while. I check my cards all the time to be sure no faults have emerged.