• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

The Radeon RX9070XT / RX9070 Owners Thread

ICDP said:
No worries, as long as you understand I’m only joking.

My own 9070 XT is at best adequate for overclocking it can run Steel Nomad all day at -80mV and 2730 VRAM. Yet some games crash and show VRAM performance drops at anything above 2700 and -45mV.

Though for my failed overclocks it can take about a minute to recover, is yours doing this at every driver lockup?
Click to expand...

-45Mv global undervolt is honestly above average. Most people saying theirs is stable at ridiculous figures of -100mv or so on every game are probably either lying, or haven’t tested enough.

My Taichi got top 40 in the world on Timespy and Steel Nomad so it’s very good silicon. Could push it as far as -165mv in Timespy without crashing. However, most games are stable around -85 or -80 but there’s a fair few that need -50 or even -45. It’s a very broad spectrum and depends on workload and the individual game. For instance, -75Mv in TLOU Part 2 is stable.

Honestly, a per game undervolt is the way to go, rather than global and forget. You are probably leaving quite a bit of performance on the table in some games that could go much lower. It’s more time consuming because you have to adjust it per game, and you will get crashes until you figure the correct value for that game, but, it’s the best way to ensure maximum performance on a game by game basis.

My VRAM ended up best at 2810. I simply used Memtestvulkan and adjusted the VRAM overclock on the fly until it showed the highest GFLOP throughput. If I went higher it started dropping throughput due to error correction. If I went lower, performance went down.
 
Last edited:
MCFC_ANDY said:
PXL-20250404-154836996.jpg

Nice view of you gou, I don't get luxury of seeing mine
Click to expand...
Nice racing setup! I have to fit mine to my desk and have had to remove the wheels on my chair and replace with feet. Pressing a load cell brake with wheels doesn't really work haha. No luxury of having a full time race setup for me unfortunately.
 
matty007 said:
-45Mv global undervolt is honestly above average. Most people saying theirs is stable at ridiculous figures of -100mv or so on every game are probably either lying, or haven’t tested enough.

My Taichi got top 40 in the world on Timespy and Steel Nomad so it’s very good silicon. Could push it as far as -165mv in Timespy without crashing. However, most games are stable around -85 or -80 but there’s a fair few that need -50 or even -45. It’s a very broad spectrum and depends on workload and the individual game. For instance, -75Mv in TLOU Part 2 is stable.

Honestly, a per game undervolt is the way to go, rather than global and forget. You are probably leaving quite a bit of performance on the table in some games that could go much lower. It’s more time consuming because you have to adjust it per game, and you will get crashes until you figure the correct value for that game, but, it’s the best way to ensure maximum performance on a game by game basis.

My VRAM ended up best at 2810. I simply used Memtestvulkan and adjusted the VRAM overclock on the fly until it showed the highest GFLOP throughput. If I went higher it started dropping throughput due to error correction. If I went lower, performance went down.
Click to expand...
That's a good tip that last one. @Alexrose1uk recommended that to me the other day.
 
FSR 4 quality in TLOU Part 2 is incredible btw. Completely maxed out and with an upscaling sharpness of 2, it looks virtually indistinguishable from native. Hovers around 100fps in most scenes, and around 85fps in intense ones.

This card can run it completely native at 4K maxed for sure, but in extremely intensive scenes on a OLED, I can feel a tiny bit of frame drop into noticeable territory. Most wouldn’t though.

FSR4 Quality looks so good you might as well use it.
 
Last edited:
Kirkatron said:
Nice racing setup! I have to fit mine to my desk and have had to remove the wheels on my chair and replace with feet. Pressing a load cell brake with wheels doesn't really work haha. No luxury of having a full time race setup for me unfortunately.
Click to expand...
The joys of being old, only issue is when I play other games I have to move PC to bedtro out the way :cry:
 
mark 11:11 said:
is there a link to a tutorial on how to do this? that sounds like a really neat tool to use for this lol
Click to expand...

Just download Memtestvulkan from Github and run it. It will automatically start testing.

Look to the right side of the window for the ‘Written’ section, and next to it, it will show the speed in GB/S. It also has a ‘checked’ section with the speed also.

It will complete each iteration and make a new pass every 20- 30 seconds or so, but it depends on the GPU speed. During testing, you can adjust the VRAM overclock on the fly and look at the speed figures. Keep increasing it by 10mhz until you come to a brick wall where the speed will either stay the same, or drop, due to error correction. If you already have a VRAM overclock, make sure to test it slightly under the value you have now, incase you are already in memory correction territory.

It’s a pretty exact science and will enable you to get the highest throughput possible before error correction. Just look at both the written and checked speeds and stop increasing at the highest average throughput speed of both. You can also look at the time it takes to finish each iteration. The highest throughput should yield the fastest iteration complete times.
 
ok cool that is pretty much what I thought the instructions would be.
weirdly where I have autism stuff like this especially when its something you can do in a systematic way like you just explained as long as i stick to the instructions and dont decide to wander off on some tangent i find that kind of thing not as mind numbing or annoying as a lot of other people
thank you muchly
il report back with what appears to be the sweet spot this will keep me entertained for ages lol
 
I also grabbed one of the last Kolink Rocket Heavy cases they were obviously just clearing out when it was £20. It only comes with a gen 3 riser but a couple of extra holes in my loque riser pcb and now it fits the bracket just fine :D
 
What's the warranty length on the 9070 XT Sapphire Pulse?

I've heard it's 2 years and this is putting me off buying it. Every other card I've bought has had a 3 year warranty.
 
Im debating getting a 9070 non-xt before the prices go up because of the tariffs but I see that there's a lot of talk about the 9070 series having high VRAM temp's. Do you think any of the cards will be viable in a low airflow enviroment ( Fractal Design R6 ) ?
 
matty007 said:
-45Mv global undervolt is honestly above average. Most people saying theirs is stable at ridiculous figures of -100mv or so on every game are probably either lying, or haven’t tested enough.

My Taichi got top 40 in the world on Timespy and Steel Nomad so it’s very good silicon. Could push it as far as -165mv in Timespy without crashing. However, most games are stable around -85 or -80 but there’s a fair few that need -50 or even -45. It’s a very broad spectrum and depends on workload and the individual game. For instance, -75Mv in TLOU Part 2 is stable.

Honestly, a per game undervolt is the way to go, rather than global and forget. You are probably leaving quite a bit of performance on the table in some games that could go much lower. It’s more time consuming because you have to adjust it per game, and you will get crashes until you figure the correct value for that game, but, it’s the best way to ensure maximum performance on a game by game basis.

My VRAM ended up best at 2810. I simply used Memtestvulkan and adjusted the VRAM overclock on the fly until it showed the highest GFLOP throughput. If I went higher it started dropping throughput due to error correction. If I went lower, performance went down.
Click to expand...

These are similar to my findings. I could happily play HZD Remastered and AC shadows at -80mV, yet Elden Ring was crashing after a few minutes and is only stable at -45mV. Other games were somewhere in between. What people should also be aware of is that some of the OC higher TDP models have higher stock voltage to begin with.

My own take on it is testing each game for max stability is a pain in the hole. So I settled on -40mV to be sure. Honestly the difference between -80mV and -40mV is only a few percent at best. When you are getting 60 vs 62 FPS in AC Shadows, or 140 vs 145 in HZD, the juice ain’t worth the squeeze in my opinion.

Obviously to some that free performance is worth pursuing.
 
Last edited:
matty007 said:
-45Mv global undervolt is honestly above average. Most people saying theirs is stable at ridiculous figures of -100mv or so on every game are probably either lying, or haven’t tested enough.

My Taichi got top 40 in the world on Timespy and Steel Nomad so it’s very good silicon. Could push it as far as -165mv in Timespy without crashing. However, most games are stable around -85 or -80 but there’s a fair few that need -50 or even -45. It’s a very broad spectrum and depends on workload and the individual game. For instance, -75Mv in TLOU Part 2 is stable.

Honestly, a per game undervolt is the way to go, rather than global and forget. You are probably leaving quite a bit of performance on the table in some games that could go much lower. It’s more time consuming because you have to adjust it per game, and you will get crashes until you figure the correct value for that game, but, it’s the best way to ensure maximum performance on a game by game basis.

My VRAM ended up best at 2810. I simply used Memtestvulkan and adjusted the VRAM overclock on the fly until it showed the highest GFLOP throughput. If I went higher it started dropping throughput due to error correction. If I went lower, performance went down.
Click to expand...
Agreed.
Timespy is so easy to run with a ridiculous undervolt. I managed -180mV which got me in the top 30. My card however is only stable at -95mV in all the games I have so far played. Some allow a lower voltage but I can't be bothered setting individual settings so I've settled for -95mV until I find a game that crashes on.
It works on Indiana Jones, Horizon Zero Dawn Remastered, Forbidden West and Space Marine 2, ACC and Forza Motorsport. All bar Space Marine 2 run with a lower voltage offset though.
Timespy is good fun for seeing how fast ypu can get it though. 34k GPU score was my best but that is in no way game stable haha. Same as my 7900 XTX, that could go below 1090mV easily in Timespy but in games, 1090mV turned out to be the stable max voltage. I had 35k GPU score with that card, again, not game stable at all.

It does however give you a decent idea of how good your silicon is though.

I need to give memtest vulkan a try as I've just picked a figure on my VRAM that didn't reduce FPS performance in game and didn't give instability and then knocked a little off.
Although with the voltage offset i found that without a memory overclock it was stable at lower voltages. I could also get lower voktages with a negative core frequency offset.
This is why I want a voltage frequency curve adjustment tool to manually adjust the frequency at ech voltage point to get the very best from the card.
 
Last edited:
You must log in or register to reply here.
Back
Top Bottom