The Radeon RX9070XT / RX9070 Owners Thread

[michty_me]

Regarding this..it’s VERY impressive for those settings. Only 79pts behind mine, and, you’re running less of an undervolt AND light VRAM overclock.

I’d be intrigued to see what you’d get with an UV of around -80 and a VRAM OC of around 2800 fast timings. You’d be pushing the very highest scores.

That’s some monster silicon.

This is my first proper outing of trying to set something up with UV/OC myself so pointers are welcome.
I'm not even sure if I'm using Memtest Vulkan correctly. If someone can explain in laymans terms what I should be looking for to get the max out the VRAM then I'm all ears.
From a brief test yesterday, I think -70mV crashed but I shall test again for sure.

 

[matty007]

I was under impression steel nomad was GPU only, although have seen talk about CPU cache.
I'm like 500 points short of that score, that's a heap of score

Well, it mostly is, but, the CPU still has to communicate data to the GPU and vice versa, so anyone saying it’s 100% GPU affected is wrong lol. Even totally GPU focused tests will still be influenced by the CPU, and especially, RAM.

If you’re 500 pts shy that can mostly be explained by the CPU (good, but no X3D cache) and DDR4. But, you also have to factor in software and OS.

3DMark etc can easily by influenced by background running tasks, services and a vast array of other settings/variables. I gained 150pts alone just by disabling RGB software (OpenRGB). Disabling Core isolation etc will also net a higher score.

It’s surprising how much lower the score will be on a ‘old’ OS install with background tasks and bloat vs a fresh install with anything unnecessary disabled.

In your case though, it’s mostly the CPU and RAM. Even totally GPU orientated tests still have to utilise CPU and RAM lol

 

[matty007]

This is my first proper outing of trying to set something up with UV/OC myself so pointers are welcome.
I'm not even sure if I'm using Memtest Vulkan correctly. If someone can explain in laymans terms what I should be looking for to get the max out the VRAM then I'm all ears.
From a brief test yesterday, I think -70mV crashed but I shall test again for sure.

Have previously explained it in here, but -

Keep increasing the VRAM overclock by 10mhz at a time, on the fly, until you land on the highest average written and checked speeds. Going higher should keep the score the same, or worse, due to error correction.

Set it on that overclock that yielded the highest speeds, and leave it there forever. That’s it.

 

[michty_me]

Have previously explained it in here, but -

Keep increasing the VRAM overclock by 10mhz at a time, on the fly, until you land on the highest average written and checked speeds. Going higher should keep the score the same, or worse, due to error correction.

Set it on that overclock that yielded the highest speeds, and leave it there forever. That’s it.

Thats what I thought but thought I best double check before getting stuck in further.
Is there a set procedure i.e. VRAM first, then voltage offset etc etc?

 

[matty007]

Oh, and make sure to use fast timings. You will see that if you disable it, the speeds will drop significantly.

Does it make a big difference in gaming vs default timings? A little, but not much. In synthetic tasks? It makes a big difference.

 

[matty007]

Thats what I thought but thought I best double check before getting stuck in further.
Is there a set procedure i.e. VRAM first, then voltage offset etc etc?

I would personally do the VRAM first as it’s the easiest and is simply set and forget once you get the best value for your individual card.

Just leave frequency offset on 0. In the most extreme cases (3400+ clocks), it can make a slight difference, but, rarely.

Then just keep increasing the undervolt by -10mv at a time until you crash. For score benching, you obviously want the highest UV possible that lets you complete one run stable. You’re just looking for the highest score. Set 100% fan when trying to push for the highest scores.

For daily settings/game tuning, do it on a game by game basis as every game is different and some will tolerate vastly more aggressive undervolts. However, the Steel Nomad UV value lessened by about 5 or 10mv is a pretty good stable daily baseline.

I’m intrigued to see what score you get once pushed to the max on this. It looks like you have very good silicon.

 

[michty_me]

I would personally do the VRAM first as it’s the easiest and is simply set and forget once you get the best value for your individual card.

Just leave frequency offset on 0. In the most extreme cases (3400+ clocks), it can make a slight difference, but, rarely.

Then just keep increasing the undervolt by -10mv at a time until you crash. For score benching, you obviously want the highest UV possible that lets you complete one run stable. You’re just looking for the highest score. Set 100% fan when trying to push for the highest scores.

For daily settings/game tuning, do it on a game by game basis as every game is different and some will tolerate vastly more aggressive undervolts. However, the Steel Nomad UV value lessened by about 5 or 10mv is a pretty good stable daily baseline.

I’m intrigued to see what score you get once pushed to the max on this. It looks like you have very good silicon.

Thanks, I shall go and play about with it all now and see what I can do.

 

[GhostDog1981]

Looks like Steel Nomad likes the large cache on the CPU that the X3D models have. I don't even get close with similar settings, but my CPU frequency is all over the place in Steel Nomad.
7729 was my score at -60mV with faster VRAM settings though 2760.
13700k with 7400 cl34

I'm not fussed though, I've got a good card for my machine.

Maximum achieved with my Pulse is 74fps or 7440 after 48hrs of testing various UV and vram timings. Settled on -90mv UV, -185 on the core offset and vram fast timings 2750mhz with PL +10%.

Put simply the steel nomad crashes when the card boosts to 3400mhz- makes me think the limited tdp isn’t supplying enough power to keep the clocks high.

The card seems happy around 3000-3100mhz

So I guess the Nitro achieving 80fps or 8000 on steel nomad indicates what the extra price to performance you are getting- ie 8% more performance for 10% more cost than a pulse.

 

[matty007]

Maximum achieved with my Pulse is 74fps or 7440 after 48hrs of testing various UV and vram timings. Settled on -90mv UV, -185 on the core offset and vram fast timings 2750mhz with PL +10%.

Put simply the steel nomad crashes when the card boosts to 3400mhz- makes me think the limited tdp isn’t supplying enough power to keep the clocks high.

The card seems happy around 3000-3100mhz

So I guess the Nitro achieving 80fps or 8000 on steel nomad indicates what the extra price to performance you are getting- ie 8% more performance for 10% more cost than a pulse.

I don’t think it’s necessarily down to the brand/SKU. Nitro’s can have bad bins, and Reapers or something like that can have good bins. That being said, the higher power limits of the Nitro/Tachi etc should theoretically allow for better scores overall. I am yet to see a Tachi score badly for example, and it’s one reason I went for it.

It’s not an exact science really. The max undervolt I got on my Tachi in Steel Nomad for 8003 was -90 before crashing, yet I have seen cards undervolted a lot more aggressively, score less. As said, other factors like CPU, RAM and timings and OS bloat/config also come into play.

 

[michty_me]

Best I've managed so far. Managed to grab a little time inbetween meetings to play about.

 

[michty_me]

I seem to have hit a wall, Only way I can get it to pass at -80mV is to start doing a negative frequency offset on the GPU. I'm assuming the lower I go with a mV input, Its possibly boosting higher and creating a driver timeout.

Currently at -80mV, -45Mhz on Max frequency, Fast timings 2744 (Seemed to give the best result) +10% PL.

Now in the top 10 though. That was my goal for this morning. Current score 7986.

 

[GhostDog1981]

I seem to have hit a wall, Only way I can get it to pass at -80mV is to start doing a negative frequency offset on the GPU. I'm assuming the lower I go with a mV input, Its possibly boosting higher and creating a driver timeout.

Currently at -80mV, -45Mhz on Max frequency, Fast timings 2744 (Seemed to give the best result) +10% PL.

Now in the top 10 though. That was my goal for this morning. Current score 7986.

Take it you’re keeping the Nitro then and sending the Pulse back

 

[michty_me]

Take it you’re keeping the Nitro then and sending the Pulse back

I think that was also the plan. I jumped on the pulse initially with it's short date as a mate was after one. Was going to keep it myself then the Nitro shipped so it will go back.

 

[matty007]

I seem to have hit a wall, Only way I can get it to pass at -80mV is to start doing a negative frequency offset on the GPU. I'm assuming the lower I go with a mV input, Its possibly boosting higher and creating a driver timeout.

Currently at -80mV, -45Mhz on Max frequency, Fast timings 2744 (Seemed to give the best result) +10% PL.

Now in the top 10 though. That was my goal for this morning. Current score 7986.

That is a very very good score with a 5800X3D. Although, I was expecting the improvement to more as you were approaching 8000 with a much lower undervolt and VRAM overclock.

Still, nearly 8000 on a 5800X3D is amazing and only just behind my very tuned system on a 9800X3D. What ram and timings are you running?

 

[michty_me]

That is a very very good score with a 5800X3D. Although, I was expecting the improvement to more as you were approaching 8000 with a much lower undervolt and VRAM overclock.

Still, nearly 8000 on a 5800X3D is amazing and only just behind my very tuned system on a 9800X3D. What ram and timings are you running?

To be honest, I've not touched the ram timings in a long time. I actually changed my ram to Corsair dominator a year or so ago which I believe is Bdie. I've just left it in the DOCP settings for about a year as didn't have the time to play about with it.

I did think that may be holding it back a little too so I've been considering getting back in the ram timings train.

Edit: all I recall is it's 3600mhz 14-14-16 or similar. I'll check once I'm home.

 

[michty_me]

RAM timings are:

 

[Kirkatron]

That is a very very good score with a 5800X3D. Although, I was expecting the improvement to more as you were approaching 8000 with a much lower undervolt and VRAM overclock.

Still, nearly 8000 on a 5800X3D is amazing and only just behind my very tuned system on a 9800X3D. What ram and timings are you running?

7900 doesn't even get you in the top 85 scores with a 5800x3d, this benchmark seems to like the x3ds in all varieties. Even the 5700x3d highest score is 8131!!

Edit: Just seen a user who has the top British score and he ran 14600k and 9800x3d and basically has same scores for both. So I have no excuse with my CPU haha.

Are these 7900+ scores being run with 100% fan to achieve maximum cooling? Mine was run at 50%. This could explain my lower GPU clock average and maybe it isn't CPU related.

 

[VonScar]

Just received my RX 9070 XT, sorry no photo's. Not sure how they upload to here. It is a Sapphire RX 9070 XT Nitro Plus. I briefly owned the 7900XT version from Sapphire. So, I like their software for the RGB and I liked the amazing build quality. However, for some reason, I had a tiny mark on the housing on the corner of my RX 7900 XT, out of the box!

I will DDU my drivers and hook it up later...

 

[Kirkatron]

Maximum achieved with my Pulse is 74fps or 7440 after 48hrs of testing various UV and vram timings. Settled on -90mv UV, -185 on the core offset and vram fast timings 2750mhz with PL +10%.

Put simply the steel nomad crashes when the card boosts to 3400mhz- makes me think the limited tdp isn’t supplying enough power to keep the clocks high.

The card seems happy around 3000-3100mhz

So I guess the Nitro achieving 80fps or 8000 on steel nomad indicates what the extra price to performance you are getting- ie 8% more performance for 10% more cost than a pulse.

My current highest score, using my stable settings, is 7851 but I don't think that represents its true performance as the UV was -95mV, 2760 vram and +10% PL. So something is holding the score back slightly, I suspect cooling as I had GPU fan at 50% and case cooling was running silent at around 700 rpm. I don't think I'll get 8000 but I should get into the 7900s.
Going to fix my CPU frequency too as I've noticed it jumps around in Steel Nomad.

 

[MCFC_ANDY]

just did 7412 on steel nomad but im thinking thats the absolute limit of this card tbf