The Radeon RX9070XT / RX9070 Owners Thread

[Fizz]

Can I be pedantic and say that I got my GPU on Sunday the 9th not Monday the 10th when I posted in this thread, I'd had it a whole day but I was too busy playing with it

 

[Fizz]

So my 9070 (Sapphire Pulse) frequently boosts beyond 3GHz, not just in the stress tests but in games too. I've cross checked the numbers with different monitors and they all agree.

I thought the boost clock was ~2.5GHz.

I'm running the stock clocks with a -75mV voltage offset and a 10% power limit increase.

 

[mite_jan]

how do you check for vram errors , i like this /memtest_vulkan v0.5.0 by GpuZelenograd but i dont think it always detects errors , what i try to do is i pause on a scene and follow when the fps start dropping

 

[Bidley]

So my 9070 (Sapphire Pulse) frequently boosts beyond 3GHz, not just in the stress tests but in games too. I've cross checked the numbers with different monitors and they all agree.

I thought the boost clock was ~2.5GHz.

I'm running the stock clocks with a -75mV voltage offset and a 10% power limit increase.

Send it back.

 

[Fizz]

how do you check for vram errors , i like this /memtest_vulkan v0.5.0 by GpuZelenograd but i dont think it always detects errors , what i try to do is i pause on a scene and follow when the fps start dropping

I think you can run tests in OCCT.

OCBASE/OCCT : Free, all-in-one stability, stress test, benchmark and monitoring tool for your PC

Ocbase is the home of OCCT, the most popular all-in-one stability / stress testing / benchmarking / monitoring tool available for PC

www.ocbase.com

 

[Fizz]

Send it back.

I wasn't complaining, I swear!!!

 

[Alexrose1uk]

how do you check for vram errors , i like this /memtest_vulkan v0.5.0 by GpuZelenograd but i dont think it always detects errors , what i try to do is i pause on a scene and follow when the fps start dropping

These cards have memory error correction/detection, so you need to monitor for errors, but also notable drops in performance. When you start seeing performance drops, you've pushed too hard

You essentially want to push as far as you can, without drops, and then back it off a bit.

 

[adidan]

So my 9070 (Sapphire Pulse) frequently boosts beyond 3GHz, not just in the stress tests but in games too. I've cross checked the numbers with different monitors and they all agree.

I thought the boost clock was ~2.5GHz.

I'm running the stock clocks with a -75mV voltage offset and a 10% power limit increase.

They're cracking little over engineered cards.

I've done absolutely nothing to mine as I just wanted to get it running and boot up Cyberpunk for the first time (finally!)

Every round of gaming it's hitting 2750MHz and several times it's reaching 3000.

And that's with not touching it. At all. Really cool card to boot.

 

[The Asgard]

Nitro arrived but its a proper screamer. Its louder than the GPU fans under load. Tried setting VSYNCH to even a low 60 FPS but it screams. My case is at desk level with 3 Rads and tuned for silence so the thing stands out. Even the wife and daughter said they wouldn't be able to put up with that. So it will be going back. Luckily from the forest so there will be zero hassle.

How are people finding the coil whine on the Nitro’s? Is it worth trying another?

My 3080 does exhibit some coil whine but not to this extent.

 

[MCFC_ANDY]

so everyone is posting crazy results but none of these are stable , you can easily pass the benchmarks but try to do few different games session and you will get constant crashes

does it really matter, called having fun pushing a gpu like i have past 30yrs plus, games i tune for efficiency, totally different , cyberpunk atm is just -50mv 2800mhz fast timing -20 power core untouched , does me fine. so your comparing apples n oranges.

 

[Tythus]

so everyone is posting crazy results but none of these are stable , you can easily pass the benchmarks but try to do few different games session and you will get constant crashes

Mine has been stable so far, running Zero Dawn, Fallout London, and other games maxed out with some overclocks. I've probably logged at least 30 hours of gameplay without any issues.

 

[iMacMart]

GM all

Any decent under bolting guides I can watch?

I fancy having a tinker this weekend

Ta

 

[ICDP]

No it's a really bad way to check stability, i could pass 3dmark with a 200mv undervolt but it wouldnt work in games. I'm saying I've used all the games I knew to make my 7900 XTX unstable to test my clocks to get the settings I scored in 3dmark. Clocks that aren't stable in games are pointless.

Seems like you really lost the silicon lottery with -45mv.

Not really worried if I did because does it really matter? It does exactly what I wanted it to do and that was use sub 275w and remain at stock speeds. It actually uses about 240w for stock speeds at those settings.

The difference between -100 mV and -50 mV undervolt is maybe margin of error type numbers in actual gaming for hours on end when temperatures have normalised. When the difference between “winning” the silicone lottery is 2% - 3% OC headroom at best you are talking 2 FPS in the majority of games.

 

[Troezar]

I keep seeing it a lot, but I suspect the issue with some of the cards is due to them having silicone in them. Time to RMA for a silicon one

 

[ICDP]

GM all

Any decent under bolting guides I can watch?

I fancy having a tinker this weekend

Ta

Are you aiming for efficiency or outright performance? Also make sure it runs stable at stock first.

Use the 3DMark demo on steam to get a baseline number to start measuring performance increase. More crucially it gives enough granularity to spot VRAM stability regression.

IMPORTANT: Do a test run between each change, this helps you track and undo for any instability.

Setting a core overclock has little to no effect.

1. Set your desired power limit. Plus 10% if looking outright performance and minus numbers if looking for reduced power draw and efficiency.

2. Find you max stable undervolt. Start at -50mV and keep dropping from there. When it is unstable you will get a driver crash. Then back off the undervolt by -10mV.

3. Once you find a stable undervolt, start overclocking the VRAM. Start with just setting “Fast Timings” and then start increasing the overclock by 100. Once you start seeing a drop in performance, you are seeing EDR (Error Detection and Replay). Back off the overclock by about 50 at this point and test again.

4. Once you have this baseline established, test loads of different and GPU demanding games. You will find that one of them may simply refuse to play ball with what seems like a stable overlock. At this point back off the undervolt and possibly the VRAM overclock until you are stable again.

Don’t be disappointed if you think you’ve lost some mythical “silicone lottery”. My experience the difference between winning and losing is actually margin of error in gaming performance. It just makes for nice benchmark scores.

Have fun

 

[b0uncyfr0]

For those of us who plan to mount it vertically, is there a specific brand we should choose?

As far as I can tell, all 9070xt's use vertical heatpipe so temperatures arent going to be optimal right? Especially with the longer cards. Ive struggled with this on my 7800xt as well.

 

[iMacMart]

Are you aiming for efficiency or outright performance? Also make sure it runs stable at stock first.

Use the 3DMark demo on steam to get a baseline number to start measuring performance increase. More crucially it gives enough granularity to spot VRAM stability regression.

IMPORTANT: Do a test run between each change, this helps you track and undo for any instability.

Setting a core overclock has little to no effect.

1. Set your desired power limit. Plus 10% if looking outright performance and minus numbers if looking for reduced power draw and efficiency.

2. Find you max stable undervolt. Start at -50mV and keep dropping from there. When it is unstable you will get a driver crash. Then back off the undervolt by -10mV.

3. Once you find a stable undervolt, start overclocking the VRAM. Start with just setting “Fast Timings” and then start increasing the overclock by 100. Once you start seeing a drop in performance, you are seeing EDR (Error Detection and Replay). Back off the overclock by about 50 at this point and test again.

4. Once you have this baseline established, test loads of different and GPU demanding games. You will find that one of them may simply refuse to play ball with what seems like a stable overlock. At this point back off the undervolt and possibly the VRAM overclock until you are stable again.

Don’t be disappointed if you think you’ve lost some mythical “silicone lottery”. My experience the difference between winning and losing is actually margin of error in gaming performance. It just makes for nice benchmark scores.

Have fun

Legend! And this is all within the AMD software, no 3rd party tools needed, and thanks for the notes and time it took to write this!

 

[Firegod]

Legend! And this is all within the AMD software, no 3rd party tools needed, and thanks for the notes and time it took to write this!

Should be under Performance > Tuning iirc. Don't forget to save out your settings once happy.

 

[ChosenName]

I find Steel Nomad is a good preliminary test. If you can get through 10 loops of that you're doing ok. But yeah the only way to know for sure is to test lots of games.

For me Neon Noir (1080P) and Heaven / Valley (1440P) were quick to show whether I'd undervolted too much.

 

[ICDP]

Legend! And this is all within the AMD software, no 3rd party tools needed, and thanks for the notes and time it took to write this!

No worries. Some other tips

1. Turn off CSM and Fast Boot in your BIOS boot section

2. Disable PCI power savings in your Windows power settings.

3. Save your stable OC as a profile. You can then set individual games to use their own OC profile. Or set a standard global profile. (Points 1 and 2 above are important to ensure the profile is applied on startup).

4. Disable driver updates in your Windows update settings. This prevents Windows installing older drivers.