AMD RX 9070 XT & RX 9070 non-XT thread (OC, undervolt, benchmarks, ...)

[LittleBro]

Stock Voltage / PL +5% → Roughly -7% performance. This made me suspect from a high stock voltage curve on my unit.

Whaaaaat? For real?

Undervolt tuning (PL stock)

• -50 to -75 mV → Gains in the 2–4% range.
• -75 to -95 mV → Gains ramped up to ~5–6%.
• (Note: only Superposition used here, as Heaven started crashing below -100 mV)
• -110 mV → Sweet spot. Best performance (~+6.63% over stock).
• -125 mV → Performance slightly dropped (+6.28%)

For now I’m running -90 mV, which gives me a solid +5.5% boost while staying on the safe side.

Also checked "real-world performance" and there's indeed a consistent 4–5% gain. Tested in Forza Horizon 5 (High/Ultra with RT ON) and Shadow of the Tomb Raider (High/Ultra) at 3440x1440 native, confirming that the undervolt improvements reflect in actual gameplay too. During Forza 1h non-stop, max GPU clock reached 2914 MHz, with peak temps of 56°C (GPU), 84°C (VRAM) and 71°C (Hot Spot) // Room (ambient): 22-23ºC

I still plan to run a few more tests to confirm if the gains/stability hold across different workloads. Worth noting that only below -110/-125 mV I started to see a slight drop in actual GPU Power Max, around 5-6% lower than usual. Even with undervolt set at -90 mV, the card still reaches ~221.5W TBP and ~311W GPU Power Max

Next step will probably be playing around with Power Limit while keeping the undervolt, just to see if there’s an even better balance point between performance, efficiency, temps and power consumption.

Looking forward to seeing results from your playing with PL + undervolt. I'm aiming to undervolt + lower PL to increase efficiency. Many thanks for the info.

 

[TSiAhmat]

Is it normal that a non OC 9070 XT (also not manually) card hits boost frequency's of 3179 MHz? Thought it should have a maximum of 2970MHz.

 

[BoggledBeagle]

There is a very distinct difference in clock speeds of Sapphire Pulse and Nitro+ models. It is up to the manufacturer how they distinguish their models.

 

[TSiAhmat]

hmm interesting... Mine is this one:

XFX Quicksilver RX 9070 XT White Magnetic Air Edition (RX-97TMAQKW9)​

Seems to go easily beyond the 2970 base Clock, it also has only 2x8 Power-Connectors (instead of the 3 mentioned here)

 

[Acesbong]

Is it normal that a non OC 9070 XT (also not manually) card hits boost frequency's of 3179 MHz? Thought it should have a maximum of 2970MHz.

Depends on the load/game. But the 9070 does seem to benefit from Overclocking than the xt, but were talking margins the average person wouldn't even notice for the most part.

 

[AusWolf]

Is it normal that a non OC 9070 XT (also not manually) card hits boost frequency's of 3179 MHz? Thought it should have a maximum of 2970MHz.

Boost is not a maximum number. Different loads will stress different parts of the GPU, so your clock speed may be lower or higher than advertised.

 

[GrimReaper85]

I did something, is that what you want?

I'm not sure why you need such a heavy undervolt but if you want really low consumption you better go for the 4070 ti, it's efficient and can drop to really low watts, the Radeon is not your way into this.

How much Watt at 100% PL?

My 9070 Gigabyte Gaming OC:
245W - 6136
245W +UV -100 mV - 6578 (+7%)
270W +UV -100 mV - 6737 (+10%)

 

[Zach_01]

Well, that sucks.

What about extreme undervolt, then?

I noticed that power limiter on most 9070XTs only goes down to around 30%. Question is, what if someone wants to reduce power consumption by ~50% or so?

Are they out of luck?

I don’t own a 9000 but it shouldn’t be different from 7000.
I ve tested my 7900XTX down to <200W while it was on perf VBios (366W, PL -10%) by lowering max clock. Voltage is automatically reduced as clock decreases down to the minimum of 600-650mV.

 

[RandumbAnonymous]

Amazon Red Devil RX 9070 XT $869 after taxes
My other rig is a 14600k w/ 4080 I think it does better

This rig is Ryzen 9 9950X (3D Boost setting in bios for better game performance

)
X870 AORUS ELITE WIFI7 ICE
CORSAIR VENGEANCE RGB DDR5 RAM 32GB (2x16GB) 6000MHz CL36

 

[Genobee]

Paid $699 for the Powercolor 9070 XT Reaper model from Amazon. Odd little card. True 2 slot card that also isn't wider than a PCIE slot. Perfect for cases with tight spacing for the GPU. Just don't expect to show off this card. It looks like an OEM model!

Card is stable at -65mv. If I go any lower than that then there are instabilities in certain titles, but not all. Making it appear to be okay at first glance when in reality it isn't. From there I dropped the power limit by -15% and haven't observed what I would consider a "noticeable" decline in performance. If I don't drop the power limit and leave it undervolted - then performance goes up a decent bit.

Card performs rather well, and can be pushed surprisingly hard with decent thermals.That is once you consider how tiny this card is in comparison to all the other models out there. Memory does hit 84c under full load which I'm not liking, but is what it is (tempted to replace the thermal pads since Powercolor doesn't mind people doing that). Hot spot seems to climb rapidly on this thing too. Though that appears to be normal across all of them. Card gets a lot hotter if you haven't tweaked any settings.

I'd say I'm satisfied with it. Coming from an RTX 3080 it's been a very solid experience.

Oh, and it already has PTM7950 on it straight from Powercolor. A welcome addition to a "lower end" card.

 

[SirKeldon]

Whaaaaat? For real?

Yes, I think it was some kind of thermal limit, because since I adjusted the fan curve, positive PL increases performance as expected, at least up to a certain point. I actually created a spreadsheet to monitor my 3DMark tests: TimeSpy, Steel Nomad, and Speed Way.

Anyway, I had to dial my undervolt back to -80mV, since some strange behavior happens beyond -90mV. In both 3DMark Steel Nomad and Unigine Heaven, I'm getting a driver crash sometimes. When I check the monitoring software, it seems to follow a similar pattern: core clocks ramping up close to their max limit, followed by a VCore spike and then a crash. I tried limiting the core clock to a lower value, but same behaviour happens. Didn't experience it at Forza or God of War, can't understand. Maybe it's related to P-state transitions, or it could be a driver issue, or it's my card which can't handle that curve anymore, IDK. I'll give a shot to -85mV within some days.

Looking forward to seeing results from your playing with PL + undervolt. I'm aiming to undervolt + lower PL to increase efficiency. Many thanks for the info.

I just played with PL till -5%, and for the same undervolt value, -80mV. Differences are as following, and just for those 3 tests, so basicly DX12 high settings:

• PL ~0% // Undervolt -80mV -> 104,16% gain // 99,18% max draw // 105,02% relative performance (score per watt consumed)
• PL +5% // Undervolt -80mV -> 104,69% gain // 99,73% max draw // 104,97% relative performance (score per watt consumed)
• PL -5% // Undervolt -80mV -> 103,75% gain // 97,52% max draw // 106,39% relative performance (score per watt consumed)

Raising more the PL till +10% lead to worst results with the same undervolt. I need to understand more stuff, this is my real first time trying to tune an AMD card since 2010 or so, lol

edit: adjusted percentages

 

[TSiAhmat]

Did a quick run in the MH Wilds Benchmark [Ultra 1440p, FSR4, Framegen & Raytracing High] (don't own the game yet)

Quite good, I never saw the card pulling the advertised 304 Watts, but might be game dependent.

The card is undervolted with -80 mV, nothing else

 

[AVATARAT]

How much Watt at 100% PL?

My 9070 Gigabyte Gaming OC:
245W - 6136
245W +UV -100 mV - 6578 (+7%)
270W +UV -100 mV - 6737 (+10%)

Total board power:
349w at 110%
317w at 100%
285w at 90%
253w at 80%
222w at 70%

 

[LittleBro]

Raising more the PL till +10% lead to worst results with the same undervolt. I need to understand more stuff, this is my real first time trying to tune an AMD card since 2010 or so, lol

That is strange behavior. Raising PL should definitely not lower performance, as it increases headroom for boost, thus increases performance or stays at same performance level.
Only reason I see would be sort of "boost" of memory clock, which is not real. Memory power draw counts towards TBP, so OCing memory can in fact degrade performance, since
it leaves less power headroom for GPU on the table. How do GPU clocks look like when you set PL +10%? Could you post your voltage frequency curve?

ASUS GPU Tweak III utility supports any brand of RX 9070 (XT) GPU.

 

[SirKeldon]

That is strange behavior. Raising PL should definitely not lower performance, as it increases headroom for boost, thus increases performance or stays at same performance level.
Only reason I see would be sort of "boost" of memory clock, which is not real. Memory power draw counts towards TBP, so OCing memory can in fact degrade performance, since
it leaves less power headroom for GPU on the table. How do GPU clocks look like when you set PL +10%? Could you post your voltage frequency curve?

ASUS GPU Tweak III utility supports any brand of RX 9070 (XT) GPU.

To be honest, I agree, it's strange. No VRAM OC yet. I see your point, but maybe for PL +10% to actually be effective, I need to undervolt further and properly tune the voltage/frequency curve, not just a simple offset. I can clearly see the GPU clocks going higher, but performance ends up worse, so the card might be choking somewhere else. Maybe the V/F curve is too aggressive on my unit, causing the card to hit TBP more often, triggering more frequent regulation, which then reduces clocks and leads to worse results. Just an speculation. I'll monitor this live with the performance overlay, maybe it'll make more sense with real-time data instead of just looking at max values.

I’m using Adrenalin only for now. I’ve read that mixing it with other GPU tuning tools can cause conflicts, but maybe I'll give one of them a shot, specially one that allows me to set a curve, such as the one you mention or Afterburner.

In any case, do you know a way to read easily the "stock curve" w/o installing further tuning software so I can see that info you asked?

 

[LittleBro]

To be honest, I agree, it's strange. No VRAM OC yet. I see your point, but maybe for PL +10% to actually be effective, I need to undervolt further and properly tune the voltage/frequency curve, not just a simple offset.

Well, power limit extends cap for TBP. It allows a card to draw more power (boost to higher clocks using higher voltage).
This does not mean that card will automatically boost higher. It may have set a hard clock limit.
Undervolt is the opposite approach. You reduce power draw to increase efficiency or to create headroom for OCing.

Which kind of makes me think that the voltage is somewhat capped and even increase of PL will not allow for it to be raised further (automatically by boosting).

In any case, do you know a way to read easily the "stock curve" w/o installing further tuning software so I can see that info you asked?

ASUS GPU Tweak III and default Radeon software suite from AMD should be able to do this AFAIK. Use only one tool at a time, don't combine them.

 

[AusWolf]

MH Wilds Benchmark

That thing is broken beyond measure. Look at how much your clocks and power fluctuate. I got wildly different results in two consecutive runs on the same system (one looking playable, one being a stuttery mess).

 

[AVATARAT]

These cards are dependent on the autoboost frequency, so when the power limit is extended (TBP cap), it will raise the temps, so it's possible that this could lower it instead of overclocking it.
Undervolt is needed to lower temps.

 

[NSR]

If anyone here has the Sapphire Nitro+ edition of the non-XT 9070, please let me know your numbers. It uses the exact same cooler as the XT but I can't find reviews for the non-XT.

 

[TSiAhmat]

That thing is broken beyond measure. Look at how much your clocks and power fluctuate. I got wildly different results in two consecutive runs on the same system (one looking playable, one being a stuttery mess).

Yeah, you are right ^^"

Here is my 3DMark Run:

Power draw was a bit weird, I think it spiked once to that number but stayed at 304 most of the time.

 

[SetsunaFZero]

Anyone tried flashing a XT Bios on a non-XT Card?

 

[Zach_01]

Here we go…

If people want more for less than should’ve follow Jensen’s prompt… lol
Cold joking aside I think multi-hundred $/€ hardware should be left alone, unless people can (potentially) do the time… after committing the crime.

 

[LittleBro]

Anyone tried flashing a XT Bios on a non-XT Card?

Good lord. Who would try such blasphemy to void warranty on brand new card?!

Anyway, don't expect things that hapenned with 6970 VBIOS flashed to 6950. This way of "unclocking potential" is long time fixed.

 

[NSR]

If anyone here has the Sapphire Nitro+ edition of the non-XT 9070, please let me know your numbers. It uses the exact same cooler as the XT but I can't find reviews for the non-XT.

Fine, I'll do it myself.

£614 from Amazon UK. Original MSRP for this variant was £579 if I remember correctly. Feel good about the purchase and I can return it back to Amazon by mid April if I have any regrets. Upgrading from an MSI Z-Trio RX 6800, plan to give that to a family member. Going to play around for a few days and update with some numbers.

Yes, I need to clean my case. I'll get around to it.

 

[PerfectWave]

Fine, I'll do it myself.
View attachment 390593View attachment 390594View attachment 390595
£614 from Amazon UK. Original MSRP for this variant was £579 if I remember correctly. Feel good about the purchase and I can return it back to Amazon by mid April if I have any regrets. Upgrading from an MSI Z-Trio RX 6800, plan to give that to a family member. Going to play around for a few days and update with some numbers.

Yes, I need to clean my case. I'll get around to it.

The gpu is bigger then youyr case