My New Vega 56 Cooling Mod

[Woomack]

I was correcting thermal pads and grease a couple of times. My block had too thick thermal pads but when I removed part of them and just used thermal grease then I thought that all is fine... well, almost. When I set a higher power limit and the card went up to 250W then VRM had 100-102°C (~68 at default settings). All other temps were fine. I have to buy some 0.5mm thermal pads today. Looks like part of the phases have good contact, part not.

Edit:
I have no idea why but after replacing thermal pads, VRM temp went down by ~30°C ... but hot spot temp went up by ~30°C. I only replaced thermal grease to thermal pads on VRM. I wonder if it's just bad reading or something else.
After 20x 3DMark FS, max temps were about 45°C GPU, 48°C HBM, 65°C VRM but hot spot 105°C

 

[Zerileous]

If hot spot is 105 then you're throttling.

My guess is the thermal pads are getting good contact to cool the VRM but now decreasing GPU contact pressure increasing hotspot temps. Some people have reported luck by tightening screws or using the stock x-shaped bracket to increase mounting pressure. Otherwise could try re-applying thermal paste, or even going RVII with a graphite pad.

 

[Woomack]

GPU and HBM temps dropped by about 5C (are no more than 50C now), VRM too ... pads are just right and I was checking that 3 times already. All seems perfect now but somehow hot spot temp is 100C+ at +50% power. When VRM were at 100C+ then hot spot was 65C max. The only thing I changed were pads on VRM because was not good enough contact. VRM temps went down to ~65C and hot spot went up to ~105C.

Edit:
I could've mixed some of the results and that 63°C hot spot could be at stock BIOS/settings. After next try on Vega64 BIOS/1.2V and +50% power, GPU temp went down to 43°C, HBM to 45°C while hot spot is 110°C. That's after 5-10x Fire Strike. Now I used that x-shaped bracket. Maybe it's normal as it's Nano PCB and all is more "packed", also there are less power phases. I don't really remember what hot spot temps were on stock cooler but typical for Vega 56/64 is 90°C+ under full load at standard settings (at least what I've seen around the web).

Edit2:
I guess I found out the reason of weird hot spot temp. When I replaced thermal grease on VRM with thermal pads, I also changed drivers to the latest one. Looks like Adrenalin 19.3.1+ is causing that issue. The same was on 19.3.1 and 19.4.1 ... now I back to 19.2.3 and hot spot temp is about 85-87°C after some FS in loop.
On Vega 64 BIOS or at +50% power it still goes to 100-105°C but at least it takes some time so can be improved with a larger radiator or something, while before it was 30->110 in 3 seconds. Will check other drivers.

Edit3:
One more try without pads on VRM but better pressure from x-shaped bracket and I think that did the trick. Still checking drivers but I'm at 53°C hot spot, 61°C VRM, GPU 39°C, HBM 45°C / stock BIOS and default settings.
I got 10g tube of NT-H2 from Noctua and I think I just used half of it for remounting this block ... btw. really nice TIM, spreads really well

 

[Zerileous]

Nice. Glad you got it figured out. I do have hotspots run 30c over my GPU, but as long as I can't find a reason that it's harming performance I don't want to mess with my block.

Do you spread the TIM manually or use an X or other method? I've recently used just hydronaut which is very thick and cannot be spread manually, and EKTIM ectotherm which is very easy to manually spread. I used manual spread for a laptop and PS4 recently. For the Vega I used the ectotherm and the X on GPU and each of the HBMs getting their own X. This is the paste that shipped with the EK block and also the method recommended with the instructions.

 

[Woomack]

Noctua says that for large cores/IHS it's enough to make an X with 5 rice grain size spots and it should be enough but I've noticed that not whole IHS was covered when I did that on the CPU ... or my dots were too small. I'm used to spread the TIM and with that Noctua stuff it can be really thin. On VRM I just left half a pea without spreading and it seems enough.
I don't like Grizzly stuff. It's not spreading really well and I had a bad luck to get a bit dry tube of kryonaut. I had 2, one was fine but the other one was a bit dry, no idea why. I like Gelid Extreme but it's hard to get it lately. Arctic MX-4 is still good if there is nothing better around and is not so expensive.
I'm not a fan of EK products (various reasons). I buy it from time to time only when is nothing else and I need it.

 

[Zerileous]

Specifically EK TIM products or all products?

I'm half tempted to see how my hyrdronaut spread since it's so thick, but there is pretty good mounting pressure with my blocks and I'm hitting 70c in small FFT, so it seems to be performing appropriately. I'll probably leave it alone.

 

[Woomack]

All EK products. They raise prices each year and I find blocks of for example Watercool.de to have better quality/more interesting design (at least products which I was comparing or using). I like their ultra clear series tubes. More expensive than competitive but look really nice. Another thing was a clear answer from their marketing. Competition works with me without problems, EK doesn't care. They don't want to support me so I won't support them. Not really a personal thing as I understand their reasons. I'm not a fan of their design and high prices and this is the main reason why I would pick something else. EK quality is high so it's hard not to recommend them if you are looking for a good block for your CPU/GPU.

I wouldn't care if it was 5°C but 30+ caused me to waste 3 evenings.
I don't remember what temps I had on the 2600X. On the 1920X with every Noctua cooler they have, I had 68°C or less. On Enermax AIO was a bit lower temp. Temps without offset as that CPU has +25°C and then it looks bad As I remember, 1700X was not much different. Simply everything below "safe" temp is fine.

Btw. 18.12.xx series drivers don't show VRM temps. On Vega 64 I can't see VRM temps at all but that's other thing.

 

[Zerileous]

Yeah I don't blame you on the pricing aspect. I bought my EK products on black Friday/cyber Monday with very nice sale prices. Even when one of my products was discontinued, I explained that I had ordered for sale pricing on black Friday and the website indicated that stock would return soon. I requested sale pricing on the new model, and instead they just gave it to me for the same price I had paid for the old model. I have heard other people dissatisfied with their customer service, but I can honestly say that I had a better than expected experience with them.

Sucks that the drivers are hiding that temp now.

 

[crull]

From my previous shim experiment only on the hbm. The hbm temperatures were over 10c lower than the gpu, because with the shim I was making better contact on those but not as much anymore on the gpu.

I also noticed when I tried a load test the hbm temperature remained somewhat low, but BOTH the GPU and HOTSPOT temperatures went a lot higher than usual. To me from that alone makes me believe the hot spot temperature is in the GPU or associated in some way with it's temperature.

I tried making the shim thinner but it was so thin that it broke on me.

So I'm now going at this in a different way by covering both hbm and gpu with a shim.

Here is my new custom shim, it hasn't been tested yet though. The hbm side is slightly thicker then the gpu side, around the thickness of a piece of standard copy paper.



After everything I have done so far here is what I think about the hotspot temperature. The whole time I was thinking it had something to do directly with the hbm so I would apply more ic diamond on them instead of the gpu and that would lower the hotspot temperature but not for the reason I thought. What I think now is that putting more paste on the hbm helps to level the heatsink so it makes better contact with the gpu lowering the hotspot temperature. When I did the paste test without any shims and took off the heatsink I could clearly see that the spread on the gpu was making better contact more towards the gap than the other edge of the gpu. Like the heatsink was tilted more towards the hbm because of the lower height of the chips.

 

[Zerileous]

Wow. Interested to see how this turns out.

 

[crull]

Wow. Interested to see how this turns out.

I had to let the thermal adhesive dry really well for the divider strip that will go in the gap. I'm going to lap the whole thing really and then try it out today. This time because I think I will have better contact on everything I'm going to try a thinner paste like kryonaut and only use ic diamond in the gap.

After I do some testing using Firestrike ultra loop as our common test I'll let you know if the delta of 30c on the hotspot has gone down, stayed the same or hopefully gotten better.

 

[Zerileous]

good luck!

 

[crull]

I tested the full shim in 2 ways running Firestrike Ultra Stress Test with my Corsair H55 Cooler. (26c Ambient)

P7 at 1662-1125mv, SPPT 220W (+142% PL), , 280W-300W, Rad Fan 1700rpm

IC Diamond on dies & Kingpin cooling Kpx on top of shim

GPU: 66c
HS: 92c
-------------------------
TG Kryonaut on dies and top of shim

GPU: 70c
HS: 95c
---------------------------------------------------------------

I think the shim idea is a failure. With both tests I have worse performance then without it. I think it's because instead of the heat dealing with one contact point with the heatsink it's now dealing with two because of the shim.

----------------------------------------------------------------

I thought of another idea to try. Instead of using a shim put thin washers on the 2 HBM side mounting bolts to raise that side slightly.

 

[Zerileous]

I just skimmed your thread and couldn't find the load temps with just the original mounting, although I believe you mentioned the Hotspot was around 100C. Maybe it needs a lower delta T with more fan speed or a larger cooler. Is the air hot leaving the rad?

 

[crull]

I just skimmed your thread and couldn't find the load temps with just the original mounting, although I believe you mentioned the Hotspot was around 100C. Maybe it needs a lower delta T with more fan speed or a larger cooler. Is the air hot leaving the rad?

The original heatsink was horrible not only with noise but temps. The air is leaving the rad, using a corsair static pressure fan on it. I also think the H55 can handle it because I had another one with an overclocked R9 290x and that card was around 300 watts tdp I think.

The best paste I have used so far is the IC Diamond it seems to be always consistent with temps and always the lowest temps. I have tried it a few different ways also which didn't seem to matter. It's thickness also helps with the die heights not being even. I ran out and have another batch coming which I will try using with raising the HBM side of the heatsink slightly higher.

 

[crull]

My card died. I was running firestrike and it started acting strange. So I shut off the system and it wouldn't post anymore. I tried everything I could think of such as;
1.) Cleared cmos with jumper & Took Battery Out
2.) Tried Secondary Bios
4.) Cleaned the PCI Connector
5.) Checked the PCI Express Power Connectors & Cables
6.) Tried the card in another system
7.) Cleaned the card with canned Air
8.) Re-seated the heatsink multiple times

What is strange though is it posted once after trying different things, but after that one time it wouldn't post again no matter what I tried.

 

[Zerileous]

That sucks, sorry to hear. I don't remember where I read/heard about the HBM cracking, but that would be my guess.

 

[crull]

That sucks, sorry to hear. I don't remember where I read/heard about the HBM cracking, but that would be my guess.

Thanks...appreciate it. I'm using an R9-270X for now, just got done taking everything off I could salvage off the card. I can't go backwards now so I bit the bullet and bought another 56 same as I had. It should here next tuesday. I intend on modding it again but this time I'm not going to take it to extremes with all the gluing. I'm going to try using the front plate with just better thermal pad and no back plate at all.

I've read that the back plate is aluminum but the front plate doesn't seem like it is. It seemed to be cast steel to me, what do you think?

https://www.tomshardware.com/reviews/radeon-rx-vega-56,5202-2.html

 

[Zerileous]

Not sure, I can tell you the front plate is non-magnetic on my reference Sapphire stock cooler, but it doesn't feel the same as the aluminum back plate. Maybe made by different production methods (cast vs stamping). I never removed the shroud, HS and blower from my stock cooler, so I can't tell you how the stock front plate would do with the liquid cooler.

Also, this article talks more about the hotspot, and the author's belief that the hotspot is in fact the real GPU core temp. Maybe AMD hides this because they don't want to further the idea that it is a hot and power hungry card lol. There was a hint at a possible performance difference in hotspot temp on my loop, based on pump speed, but I'll need to do additional testing to see if its a real result or a variance.

 

[crull]

They use thermal pads on the front plate, not sure they would do that if it wasn't some form of aluminum...right or would they? My plan is cutting off the section that has the blower and using whats left to cool the vrm's & coils with better thermal pad.

Cut this section off.