Overclock.net banner

1521 - 1540 of 1621 Posts

·
Registered
Joined
·
15 Posts
Quick update after repasting: GPU core dropped 5C and Junction by 20C. Now it never goes past 95 (within 22C from GPU temp) whereas it throttled at 110C (within 40C from GPU temp) before worst case scenario. In normal scenarios it stays under 90C, including Time Spy Graphics test 1 and 2 where it peaked to 108C earlier.

Now, I realized I put 1mm thermal pads on the VRAMs and I think it's not touching the heatsink properly or very very slightly, so I will be changing them asap. 2mm pads are out of stock in Japan right now, so I have to run it as is for a while, probably no OC on VRAMs and undervolt the card. Regardless, the junction temps and temps overall significantly dropped significantly so the success here is that I know I can get this card running perfectly with a little bit of tuning.

Happy days!
 

·
Registered
Joined
·
1,278 Posts
Sorry to ask again but has anyone tried liquid metal ? Is it worth it vs gc-extreme for example?
 

·
Professional Lurker
Joined
·
184 Posts
Sorry to ask again but has anyone tried liquid metal ? Is it worth it vs gc-extreme for example?
For me, it's not worth the trouble, no. It would work better, sure, but for the trouble I would just use that time and research custom loop water cooling. It's not near as bad nor expensive as it once was. Start saving and go for it when the time is right :3
 

·
Professional Lurker
Joined
·
184 Posts
Thanks again for a lot of useful info. Just to confirm before I start my operation (which is in an hour or so): I should not change the pads in the front? I got Minus8 pads I was thinking of putting in the front and moving Red Devil's own used pads under the backplate, backside.

Interestingly, I could run Superposition at 2775mhz, fans at 100%, memory at 2115mhz fast timings, and MPT 350W/380TDC. Got 17914 points which is actually not bad. With further testing, it's fully stable at 2770mhz for rasterization without RT, and with RT 2710mhz. I guess it's not a bad card after all, it's just that Timespy doesn't like it, but everything else does. Temps stayed under 100C just about, but again, a delta of 33-38C, so I have to fix that.
I should have clarified when I mentioned not changing pads out: For the reference model, one should not change out the pads on the front of the card, nor should anyone sub-in aftermarket pads for ones that come with EK's water blocks. Use the included pads, especially for EK water blocks, else there will be mounting issues. As for aftermarket cards, I don't have any experience there, so I can't comment on that with any amount of certainty.

Regarding benchmark scores: Superposition is VERY forgiving with stability, and IMO Superposition should only be used as a performance metric, not a stability test.

I see so many posts of "I can only run xxxxMHz in Time Spy/Port Royal, but i'm stable at '+100MHz' in everything else".

In my experience, some/most of yall are not actually stable at +100MHz over what TS/PR are stable at, you just haven't crashed yet. I highly recommend tuning your global profile based on Time Spy and *Heaven benchmarks, and only increase clocks after a few tens of hours of confirmed stability in each game that you play.

This has been my *experience, anyway, with my reference RX 6900 XT under water, and my home-office-pc's reference RX 6800 on air in my Ncase M1.

* The only caveat to my opinion above is that non-RT games can run a bit faster core clock speed than RT games. But! I would still base my non-RT tune off of Time Spy/Firestrike stability, and likewise my RT tune off of Port Royal and/or Control stability (w/RT Reflections, Transparent Reflections, and Indirect Diffuse Lighting enabled).

* My best Heaven stability results are using the exact settings pictured (yes, it needs to be fullscreen too, else it doesn't push the GPU as hard as it can otherwise).
1619878546689.png
 
  • Rep+
Reactions: ptt1982

·
Professional Lurker
Joined
·
184 Posts
What clocks? And did you use MPT? STILL can't use MPT with my 6900XTXH card. Waiting for GPU-Z support so I can extract the BIOS.
Check out the Igor's Lab RBE/MPT thread. I think they found a flash tool that can talk to the XTXH cards via linux. I believe if it can see the card, it'll be able to extract the bios.

I don't know if MPT is updated quite yet, you'll have to read through the most recent few pages of the thread for that one.
 

·
Registered
Joined
·
1,278 Posts
For me, it's not worth the trouble, no. It would work better, sure, but for the trouble I would just use that time and research custom loop water cooling. It's not near as bad nor expensive as it once was. Start saving and go for it when the time is right :3
My card will be under water as soon as Alphacool releases the block ( middle of May ).

That's why I'm trying to understand if I should use liquid metal or "normal" paste. I've heard the liquid metal will be better but at the same time... wanted to have some more feedback :)
 

·
Registered
Joined
·
26 Posts
My card will be under water as soon as Alphacool releases the block ( middle of May ).

That's why I'm trying to understand if I should use liquid metal or "normal" paste. I've heard the liquid metal will be better but at the same time... wanted to have some more feedback :)
I would personally use liquid metal. I have experience with it, and it's always better. Just got to make the necessary precautions and apply it correctly.
 

·
Registered
Joined
·
1,278 Posts
I've applied to cpu's ... never did it on GPU's though :) but my guess it's gonna be similar :)
 

·
Mentally Insane User
Joined
·
494 Posts
Problem with liquid metal is it can start to run once hot and short out nearby components as it's conductive.
 

·
Professional Lurker
Joined
·
184 Posts
My card will be under water as soon as Alphacool releases the block ( middle of May ).

That's why I'm trying to understand if I should use liquid metal or "normal" paste. I've heard the liquid metal will be better but at the same time... wanted to have some more feedback :)
Ah, I get you. Right now, I use Gelid GC-Extreme on my reference RX 6900 XT. Under full benchmark load (2750MHz on the Adrenalin Core-speed slider, core ~330-340W, Card ~390-410W), hotspot tops out at 73C, Core 55C-ish, VRM, Memory Junction, and all other temps all well under 60C. Knock 5-10C off of each number for normal gaming loads at native resolution on my 4k120 LG CX 48" OLED panel. Likewise, core/hotspot temps on the b/f's 3080FE under water are nice and cool with GC-Extreme on his LG CX 48" OLED panel.

In my opinion, a quality, thick paste like Hydronaut, Gelid GC-Extreme, etc., will get you there, won't pump out*, and relieves you of the trouble and worry that come with Liquid Metal.

*will pump out very, very slowly compared to a thinner paste like Kryonaut, NT-H1/H2, etc.
 

·
Registered
Joined
·
1 Posts
Couple of noob questions here:

1. Does Port Royal ever just stop the benchmark mid run but no errors show up in either 3dMark or The Radeon Software? I would take this as a sign of instability but I've literally only have increased the Power limit by 15% in the Radeon software and changed the fan curve, nothing else. EVGA Supernova 850 P2 Power Supply, so I don't think it's a power delivery issue? I'm just wondering if I should be taking this as a sign of instability or not...

2. I've only overclocked an R9 290 before and have heard the process has changed quite a bit on top of these cards having some nuances that need to be taken into account. Is there a good guide for these cards somewhere in this thread that I could follow. I have a reference card with the stock cooler so I doubt MPT is in the realm of feasibility at the moment, but I want to get as much performance as I can without it and see where my junction temps are.

3. When I've tried in the past to overclock this card, sometimes it seems to ignore the MHz range I set it to, is that a sign of instability or is it the Radeon software/drivers being weird?

Thanks!
 

·
Registered
Joined
·
26 Posts
Quick update after repasting: GPU core dropped 5C and Junction by 20C. Now it never goes past 95 (within 22C from GPU temp) whereas it throttled at 110C (within 40C from GPU temp) before worst case scenario. In normal scenarios it stays under 90C, including Time Spy Graphics test 1 and 2 where it peaked to 108C earlier.

Now, I realized I put 1mm thermal pads on the VRAMs and I think it's not touching the heatsink properly or very very slightly, so I will be changing them asap. 2mm pads are out of stock in Japan right now, so I have to run it as is for a while, probably no OC on VRAMs and undervolt the card. Regardless, the junction temps and temps overall significantly dropped significantly so the success here is that I know I can get this card running perfectly with a little bit of tuning.

Happy days!
Care to share how to tear down the gpu?
 

·
Registered
Joined
·
15 Posts
I should have clarified when I mentioned not changing pads out: For the reference model, one should not change out the pads on the front of the card, nor should anyone sub-in aftermarket pads for ones that come with EK's water blocks. Use the included pads, especially for EK water blocks, else there will be mounting issues. As for aftermarket cards, I don't have any experience there, so I can't comment on that with any amount of certainty.

Regarding benchmark scores: Superposition is VERY forgiving with stability, and IMO Superposition should only be used as a performance metric, not a stability test.

I see so many posts of "I can only run xxxxMHz in Time Spy/Port Royal, but i'm stable at '+100MHz' in everything else".

In my experience, some/most of yall are not actually stable at +100MHz over what TS/PR are stable at, you just haven't crashed yet. I highly recommend tuning your global profile based on Time Spy and *Heaven benchmarks, and only increase clocks after a few tens of hours of confirmed stability in each game that you play.

This has been my *experience, anyway, with my reference RX 6900 XT under water, and my home-office-pc's reference RX 6800 on air in my Ncase M1.

* The only caveat to my opinion above is that non-RT games can run a bit faster core clock speed than RT games. But! I would still base my non-RT tune off of Time Spy/Firestrike stability, and likewise my RT tune off of Port Royal and/or Control stability (w/RT Reflections, Transparent Reflections, and Indirect Diffuse Lighting enabled).

* My best Heaven stability results are using the exact settings pictured (yes, it needs to be fullscreen too, else it doesn't push the GPU as hard as it can otherwise).
View attachment 2488753
Thanks for the reply. I changed the pads to 1mm pads and they didn't touch heatsink and VRAMs went up to 90C as they were. Now, I bought Thermalright's 2mm pads and changed them, and now the hotspot on GPU goes way too high, so I think the original pads were simply 1.5mm, and there is no good connection to the heatsink. I do have the original pads still, but they are all squished. They still work, though, and I can put them on.

On Red Devil, I advice you guys not to change the thermal pads or tune it too much, but repasting it makes wonders to the GPU temps. Just leave the pads as they were. Now, for the fourth time, I need to open it up and tune it back to normal to get it working. Jeesus, took 4 hours already today of my time to get back to normal temps, but I'll do it.
 

·
Registered
Joined
·
15 Posts
Care to share how to tear down the gpu?
I did a teardown on Red Devil 6900xt (not ultimate).

1) Unscrew all the screws you can see on the backplate
2) Remove two cables, fan (white with colored cables) and another white one on the side of the PCB
3) Carefully lift the heatsink, there’s another black cable under the top of the card (the red devil rgbs), disconnect the cable (it’s linked to another black cable) before fully lifting the heatsink (or lift the heatsink to a standing position next to the pcb and then remove it)
4) (optional) if you want to access what’s under the backplate remove five or six screws (forgot how many) that hold the backplate on the PCB, they are near the gpu chip and memory modules (make sure you don’t remove the hdmi/dp holder screws)
5) (optional) remove one (white) cable from backplate, it gives led light to the red devil logo before fully removing the backplate from the pcb

If you add pads: ONLY use 1.5mm pads, 2mm are too high and 1mm too low for the heatsink to be mounted well (believe me, tried all of them).

My temps for changing to high quality pads for Vram dropped around 5-7C, and changing thermal paste dropped gpu hot spot by 15-20C. In my case the original paste job was sloppy which is why I did this operation. My card is also used, so I had nothing to lose (too hot card, or a broken card, it’s the same to me). Now the temps are great. I used Thermal Grizzly Kryonaut, and I’m aware I have to repaste once a year or so.

Update on clocks and further testing:
My final stable clocks 2620mhz core, 2100mhz memory. Daily use 2600mhz / 2080mhz. Boost between 2480mhz-2580mhz (higher lows and higher highs after repasting). It does work in games at 2750mhz core (no crashes without vsync, 4K couple of hours of testing, crashes may come later) and ray tracing with 2670mhz (4K no vsync).

Previous driver gave me 2640mhz at Timespy, now it’s lower at 2620mhz, but scores are the same. Instant crashes start at 2780mhz, with ray tracing at 2750mhz. Best TS score 20738 using MPT. My card likes power up to 350W/380TDC, after that scaling ends due to the silicon not being able to push clocks higher.

It’a sad that I can’t push the card past 2750mhz at TS and ray tracing, because the temps even on air would allow it with a good 50% fan speed. I’m hoping a new Vbios from powercolor and better drivers from AMD would fix this (for example 1.2 vcore could help.) My clocks seem quite typical for red devil basic though, and most of the extra performance came from the MPT. Daily use TS is around 20350 which is 11% higher than my previous fine-tuned TUF Rtx3080 which I sold due to realizing I don’t need DLSS or Ray Tracing, and like higher raster performance and VRAM capacity.

Done with tuning and testing (used 4 days of my 11 day annual leave on it) until new drivers hit. I think I’ll play some games now!
 
1521 - 1540 of 1621 Posts
Top