Radeon VII Owner's Club!

Did the washer hack on my Radeon VII. Didn't seem to do much. The drivers will still crash when I try to use the AMD Graphical settings-> auto under volt button. Leaving things stock seems to be the only way I can get my card to run properly.

Try doing it manually, also post your gpu-z screen with your bios version. You may want to update if its outdated.

I'd try setting it to 990mV @ 1800MHz first, see if that runs.
If stable try 980mV @ 1800MHz, and so on, until you find the spot it crashes.

You can adjust fan to your preferences, and possibly get better clocks and/or lower temps or vice versa.

I recommend running heaven in window and looking for artifacts. If you see artifacts, stop heaven, and change your settings.

You can use my stress/undervolt/overclock testing if you wish as template. (as its unmodified radeon vii stock air cooling)


or you can use this community table, tho cooling is not listed on all of them, and you can't be sure its not just 1min undervolt/overclock for one case.



//Note: On my XFX radeon vii i can reach 2GHz on core, on air. I am thermally limited, and i feel there's a lot of potential once i'm not.
 
Last edited:
Manually undervolted my Radeon VII down to 980 mV. Had a Division 2 crash, but it looks like a game issue and not a driver issue. I assume that my bios is the latest since the card is an anniversary edition. I'll continue to jump down 10 mV to see how low I can go.

With auto undervolt fallout 4 would crash the system during launch. Ran with no issues at 980 mV
 
Manually undervolted my Radeon VII down to 980 mV. Had a Division 2 crash, but it looks like a game issue and not a driver issue. I assume that my bios is the latest since the card is an anniversary edition. I'll continue to jump down 10 mV to see how low I can go.

With auto undervolt fallout 4 would crash the system during launch. Ran with no issues at 980 mV
you should not being getting crashes with the division 2 these days.....i think you hit a voltage thing with that game cause i got a hundred hours in that game with no crashes.
 
AIDA64 stress test tool may be able to show you errors (and stop the test, if it detects errors in compute - its quite sensitive to errors, errors would be something akin to artifacts)
 
IT LIVES!!!!

After all that headache, it looks like I have a good one. Might see how far I can push it on reasonable voltage / temps.

New mobo/cpu and fresh windows did the trick.

it lives.png
 
Great news! Alive at last.

How do you like the 3600X? I got one as well and it seems like a nice kit.
 
Great news! Alive at last.

How do you like the 3600X? I got one as well and it seems like a nice kit.

Faster than the outgoing 1700x, and memory overclocked via xmp profile in one click. Less power consumed. It's good enough for what I do. I bought a cheap x570, so if I want to swap out next gen.
 
i'd set it to manual, but if you don't mind then ;)

Maybe next steps, just making sure everything is stable at the moment. As for the Radeon VII, think I am going to stick to stock speeds @970-990mv. It is quiet under water and has enough performance to where I am happy with it given how much I ended up paying for it out of pocket in the end.

Decided to pull up F@H, pulling down ~1.1M ppd @ 990mv.

(y)
 
Last edited:
waterblock installed

<palatine laugh>
giphy.gif


upload_2019-7-16_0-22-30.png

1h overclocked stress | prior this was hitting 110-115'C jT, and around 90'C core. (only 2 low rpm fans, 480 + 240 rads)
(80-84'F ambient, AC is dying when its over 100'F outside)


2GHz easy peasy
 
Last edited:
Nice, those are really good results. I want to shoot for 2050Mhz when I put my block on. I want to see it maintain >2000Mhz clocks in games
 
Hi all just logged in to see if anyone was having issues with the last couple of driver updates. Mine has been doing a lot of driver resets. so dropped back to 10.5.2 seems to be doing mo better in the last few hours.
Saw some temperature post so I thought I would throw mine in at idle - EK waterblock . used some thermal pads and compound around the GPU
 

Attachments

  • Ray 7 temp 1.png
    Ray 7 temp 1.png
    34.7 KB · Views: 0
  • 10.JPG
    10.JPG
    474.3 KB · Views: 0
Hi all just logged in to see if anyone was having issues with the last couple of driver updates. Mine has been doing a lot of driver resets. so dropped back to 10.5.2 seems to be doing mo better in the last few hours.
Saw some temperature post so I thought I would throw mine in at idle - EK waterblock . used some thermal pads and compound around the GPU
why this?

upload_2019-7-21_19-11-12.png


These are just smd's;
 
Been working with mine a bit lately, doing a loop change and will be playing a bit more. Running an R VII with an EK block currently on an XSPC EX360 rad with 3x Nocuta NF-A12x25's with an EK 3.2 DDC pump and have been running 2000/1200 @ 1060mv with junction temps maxing out around 80-83c. Adding a second 360mm rad (XSPC TX360) to throw a block on the 3600X and play around some more with the VII.
 
I decided to sell mine...I really love the 5700XT and since I can do 2.2Ghz on air, I have no doubt it will be just as fast (hopefully faster) on water. The price they are fetching on ebay along with a 50% coupon made it worth it. VEGA was a misunderstood uarch, and I will miss it!
 
I decided to sell mine...I really love the 5700XT and since I can do 2.2Ghz on air, I have no doubt it will be just as fast (hopefully faster) on water. The price they are fetching on ebay along with a 50% coupon made it worth it. VEGA was a misunderstood uarch, and I will miss it!

In my terms, i'll keep my radeon vii with 16GB vram until i can grab 24-32GB card.
:)

// I wonder what are the limits of 5700xt, i assume its around 2400MHz on good water.
 
why this?

View attachment 175648

These are just smd's;

On my Radeon VII going nuts with the paste around the edge of the die helped temps. “junction temp” is the highest temp of 80+ sensors. I wouldn’t have used the thermal pad though.... I’d be worried about that causing standoff or extra pressure on certain parts.

Others reported similar, that relatively unusually high amounts of thermal paste helped. Personally I was aiming to make sure there was heat transfer all over, you’re right the smds don’t need it lol.
 
In my terms, i'll keep my radeon vii with 16GB vram until i can grab 24-32GB card.
:)

// I wonder what are the limits of 5700xt, i assume its around 2400MHz on good water.


The 5700xt is faster, especially when clocked at 2+GHz, and this is without a ram OC (can't discuss that at this time) with way less power usage.

I can say that 2.2Ghz on air is stable though
 
The 5700xt is faster, especially when clocked at 2+GHz, and this is without a ram OC (can't discuss that at this time) with way less power usage.

I can say that 2.2Ghz on air is stable though

You can OC the VII too. VII is still faster OC to OC but obviously the 5700xt is a way better buy.
 
Yea, VII will still (barely) outpace a 5700 xt even at 2.2ghz. However, as you say, VII OCs as well (mine goes to 1900mhz/1110mhz stable at stock vcore). Both cards redlined, you're still looking at a 7-8% performance advantage on the VII.

VII does claw back more of a performance advantage in a few games at extreme resolutions and IQ with it's additional 8GB and higher mem bandwidth, but is that worth $250+? -- No sir!
 
Yea, VII will still (barely) outpace a 5700 xt even at 2.2ghz. However, as you say, VII OCs as well (mine goes to 1900mhz/1110mhz stable at stock vcore). Both cards redlined, you're still looking at a 7-8% performance advantage on the VII.

VII does claw back more of a performance advantage in a few games at extreme resolutions and IQ with it's additional 8GB and higher mem bandwidth, but is that worth $250+? -- No sir!

i've been stuck on fury-x for ages; Additional 8GB can make all the difference. :) // sure for gaming not that much, but for 3dsmax/vue it makes difference.
 
And that's where the VII is still a godsdamned steal!

VII is certainly a steal for content creation/modeling.

I did have a few weeks when I had my 2080 and VII to make some direct gaming comparisons. For 80% of games at 4k/highest IQ, the performance was equal for all intents. For those other 20% (Kingdom Come, Final Fantasy XV, and games I was using with high-res texture packs come to mind), there was a significant difference in FPS.
 
have we spoken about rocm (ML - Machine Learning)?
Its a monster once you get your project working. -- Sure matrix multiplication is better on NV but thats the only case, in rest of the workload vii kills. vram makes hell of a difference for neural net upscaling or colorizing old movies.
 
On my Radeon VII going nuts with the paste around the edge of the die helped temps. “junction temp” is the highest temp of 80+ sensors. I wouldn’t have used the thermal pad though.... I’d be worried about that causing standoff or extra pressure on certain parts.

Others reported similar, that relatively unusually high amounts of thermal paste helped. Personally I was aiming to make sure there was heat transfer all over, you’re right the smds don’t need it lol.

Just wish I had done all four sides - The pads were to level up to the waterblock so it would make contact. EK makes some nice pads over stock amd stuff did a simple heating pad test with my trusty laser temp gun.

This was the second time I set the block and it helped junction temps that's all I really know. Plus -- as of today yes I would buy the cheaper card >
 
Last edited:
I just got one open box from MC for pretty cheap. I'm glad I was able to score one before they get scarce.
 
It's a good card.

I ended up giving the Radeon VII rig to my brother. Trying to sell/donate some of my extra computers since I had too many. He's happy with it.
 
Finally got around to properly overclocking my card under water in the full new build. My GPU seems to take a bit more voltage than most, requiring 1209mV at 2101Mhz (which maintains ~2050Mhz in games) -- stock was 1117mV. I also can't seem to do 1200Mhz HBM -- I get artifacts in windows almost immediately -- even at idle temps of like 26C for memory!

Anyways I am getting 10,242 in Timespy, and 24,820 in Firestrike. Seem some people get quite a bit higher with 1200Mhz HBM. It is surprising that HBM clocks matter that much -- surely the card can't really need that much B/W and it must be memory latency improvements with higher clocks..? (+200Mhz on HBM clocks adds roughly as much b/w as a RX 580 has in total!)
 
Back
Top