*** The AMD RDNA 4 Rumour Mill ***

[adidan]

I wonder if they are just getting efficient with the PCBs or if the GPUs are just overbuild for what they are.
From some reviews I see some silly low temperatures at load in the 50s and 60s.

My 9070 Pulse has maxed out at 55 during Cyberpunk play throughs (low 70's hotspot) and low 80's for the memory.

It is a cool card. Plus it seems to be constantly boosting to 2750Mhz instead of 2500 (even up to 2900/3000Mhz today - how long for I don't know).

Cracking little card, very lucky to snag one at £525.

Hopefully more stock comes in because at the price I got it it's great value.

 

[ICDP]

Sapphire Pulse 9070 XT
Under load with Star Wars Outlaws all RT max (no PT or RTDXi)
55c GPU
82c Hotspot
86c VRAM (88c at 2800 OC)
Fans 45%
304w

I had to replace a front 140mm fan on my Meshify C with a thin profile 15mm one. This gives enough room for the Sapphire Pulse by about 5mm to spare.

The GPU on these runs very cool. VRAM runs a touch warm but well within spec and using a custom fan profile at 55% can drop those 3 - 4c if you are worried. I just left it alone and even OCd to 2800 and let it hit 88c.

My older 3080 FE would be running mid 90s to low 100s on the VRAM.

My latest experiments are to find out how Low I can set the power limit while maintaining stock performance.

 

[adidan]

@ICDP IIRC I don't think my fans have gone above mid-40% either

I've only tried CP benchmark and gaming for real world results and that's only been at 1440p at ultra without any ray or frame gen.

Certainly feel like they're cards with legs though.

 

[michty_me]

Sapphire Pulse 9070 XT
Under load with Star Wars Outlaws all RT max (no PT or RTDXi)
55c GPU
82c Hotspot
86c VRAM (88c at 2800 OC)
Fans 45%
304w

I had to replace a front 140mm fan on my Meshify C with a thin profile 15mm one. This gives enough room for the Sapphire Pulse by about 5mm to spare.

The GPU on these runs very cool. VRAM runs a touch warm but well within spec and using a custom fan profile at 55% can drop those 3 - 4c if you are worried. I just left it alone and even OCd to 2800 and let it hit 88c.

My older 3080 FE would be running mid 90s to low 100s on the VRAM.

My latest experiments are to find out how Low I can set the power limit while maintaining stock performance.

How are you feeling about the upgrade replacing the 3080FE. A worthwhile exercise to you?

 

[ICDP]

@ICDP IIRC I don't think my fans have gone above mid-40% either

I've only tried CP benchmark and gaming for real world results and that's only been at 1440p at ultra without any ray or frame gen.

Certainly feel like they're cards with legs though.

About 8% - 10% overclock is achievable but that if you throw power at them. By contrast I can get 5% extra performance by setting VRAM to 2700 and an undervolt of even - .50mV as this gives the boost clocks room to boost within TDP (304w).

So getting that extra 5% pushes the card into incredibly inefficient territory and is really only for the epeen 3DMark scores. It’s clear AMD tweaked the 9070 XT a lot, yet still has some headroom.

For real world gaming it’s about undervolting and getting lower power usage while not losing performance.

 

[ICDP]

How are you feeling about the upgrade replacing the 3080FE. A worthwhile exercise to you?

My 3080 FE was replaced a while back by a used 4080 FE and I felt it a worthwhile upgrade for 4K but only in performance, never in price. The 9070 XT is so close in raster to a 4080 that the 4080 only wins on a technicality. RT is about 20% faster in extreme cases and about 10% in the majority of RT games that actually have playable FPS at 4k or 1440p.

So yes, if you can get a 9070 XT for £650 absolute max cost, then yes it’s a meaningful upgrade giving about 40% in raster and 30% in RT. Plus the 16GB VRAM helps a lot.

This is going to sound controversial, but all those 3000 series gamers who declared “but what choice do we have” and then went and bought 5080 for £1300 - £1400 as if they had no options. The 9070 XT gives you 85% of that performance for half the cost and that’s the scalped prices for both.

 

[RBImGuy]

About 8% - 10% overclock is achievable but that if you throw power at them. By contrast I can get 5% extra performance by setting VRAM to 2700 and an undervolt of even - .50mV as this gives the boost clocks room to boost within TDP (304w).

So getting that extra 5% pushes the card into incredibly inefficient territory and is really only for the epeen 3DMark scores. It’s clear AMD tweaked the 9070 XT a lot, yet still has some headroom.

For real world gaming it’s about undervolting and getting lower power usage while not losing performance.

I dropped 70fps in one of my games in 4k.
No difference in gameplay
but card is silent also and 100w less power draw.
I would adjust for gameplay not check fps difference

 

[ICDP]

I dropped 70fps in one of my games in 4k.
No difference in gameplay
but card is silent also and 100w less power draw.
I would adjust for gameplay not check fps difference

This mirrors my own experience at 4K. As long as I can stay a few FPS over my VRR floor I am happy. So the difference between a 9070 XT getting 60 FPS and a 5080 getting 72 FPS is meaningless at 4K on a VRR screen. It won’t even make much difference to the latency if you turn on frame gen.

 

[michty_me]

My 3080 FE was replaced a while back by a used 4080 FE and I felt it a worthwhile upgrade for 4K but only in performance, never in price. The 9070 XT is so close in raster to a 4080 that the 4080 only wins on a technicality. RT is about 20% faster in extreme cases and about 10% in the majority of RT games that actually have playable FPS at 4k or 1440p.

So yes, if you can get a 9070 XT for £650 absolute max cost, then yes it’s a meaningful upgrade giving about 40% in raster and 30% in RT. Plus the 16GB VRAM helps a lot.

This is going to sound controversial, but all those 3000 series gamers who declared “but what choice do we have” and then went and bought 5080 for £1300 - £1400 as if they had no options. The 9070 XT gives you 85% of that performance for half the cost and that’s the scalped prices for both.

Thanks for your insight.
I do have a Nitro+ on order that was ordered at £680 so I'll keep going with this order unless there is a drop in prices and stock availability which would see prices dip a little further. I can't really see it happening at the minute or in the next few months.
My target was to spend similar or less than what I got my 3080FE for (RRP)

 

[BS Dave]

Sapphire Pulse 9070 XT
Under load with Star Wars Outlaws all RT max (no PT or RTDXi)
55c GPU
82c Hotspot
86c VRAM (88c at 2800 OC)
Fans 45%
304w

I had to replace a front 140mm fan on my Meshify C with a thin profile 15mm one. This gives enough room for the Sapphire Pulse by about 5mm to spare.

The GPU on these runs very cool. VRAM runs a touch warm but well within spec and using a custom fan profile at 55% can drop those 3 - 4c if you are worried. I just left it alone and even OCd to 2800 and let it hit 88c.

My older 3080 FE would be running mid 90s to low 100s on the VRAM.

My latest experiments are to find out how Low I can set the power limit while maintaining stock performance.

I wonder how much some top end pads would help the memory? but then you'd proabably have to replace the (very good) TIM on the core (it's the honeywell stuff isn't it?)

 

[Alexrose1uk]

For those toying with VRAM speeds, I'd heartily recommend this tool:

GitHub - GpuZelenograd/memtest_vulkan: Vulkan compute tool for testing video memory stability

Vulkan compute tool for testing video memory stability - GpuZelenograd/memtest_vulkan

github.com

It's very useful because it slams the VRAM, so is good to test heat, but it will not only tell you if it spots any errors, but also a computation speed, so you can see if you've clocked too high and are triggering errors, or a performance regression due to error correction kicking in.
Was very useful when I was tinkering with prior cards with similarly hot memory like the 3080 as could see if it was hitting into the throttle zone easily, even if it was stable.

 

[ICDP]

I wonder how much some top end pads would help the memory? but then you'd proabably have to replace the (very good) TIM on the core (it's the honeywell stuff isn't it?)

I bought some of those pads for my 3080FE because it was hitting low 100s under load. I never bothered in the ends as going to all that trouble to drop temps to 90c was a faff. 88c is nothing to worry about and it’s more about getting over an irrational worry.

 

[sunn]

AMD's unreleased Radeon RX 9070 XT "reference" design shows up in China - VideoCardz.com

Radeon RX 9070 XT in reference design spotted Remember the low-resolution renders that someone found on Reddit a few weeks ago? We were quick to confirm that those were official renders of soon-to-be-announced Radeon RX 9070 XT cards, despite many claiming they were fake. But to our surprise...

videocardz.com

Wonder what the reasoning is not the release? (if real)

 

[ICDP]

Sapphire Pulse 9070 XT
Under load with Star Wars Outlaws all RT max (no PT or RTDXi)
55c GPU
82c Hotspot
86c VRAM (88c at 2800 OC)
Fans 45%
304w

My latest experiments are to find out how Low I can set the power limit while maintaining stock performance.

Just quoting myself here for reference.

Steel Nomad Demo - Stock. (304w)
6883

Max OC (335w)
7462 - 8.4% over stock

-30% Power Limit (212w)
6070 - 11.8% under stock

-30% Power Limit, 2700 VRAM Fast Timing (212w), -75 mV
6394 - 7.1% under stock

-20% Power Limit, 2700 VRAM Fast Timing (235w) -75 mV
6821 - about stock performance

This show much AMD pushed these outside their efficiency envelope to get close to the 5070Ti performance.

GPU temp 55c
Hot spot 76c
VRAM with OC 84c

 

[xtal191]

So I have a Nito+ XT pre ordered elsewhere for €840 but can get a Gigabyte 5070 TI Windforce OC for €950, is the 5070 TI worth the extra?

 

[t2na]

So I have a Nito+ XT pre ordered elsewhere for €840 but can get a Gigabyte 5070 TI Windforce OC for €950, is the 5070 TI worth the extra?

Depends on if you think the Nvidia feature set is worth 110EUR to you. Reflex is better than Anti Lag, RT is still better on Nvidia but it's significantly better than it has been with the 9070XT.

Performance is usually 5-10% in favour of the 70Ti - but swinging back in other titles that are more suited to AMD (Call of Duty, Horizon etc.) - so I think it's really only you who can decide if you'd feel happy paying that extra cost.

A 70Ti series of card for 950EUR is honestly bonkers to me.

 

[ICDP]

So I have a Nito+ XT pre ordered elsewhere for €840 but can get a Gigabyte 5070 TI Windforce OC for €950, is the 5070 TI worth the extra?

Neither one is good to be brutally honest. But the 5070Ti seems closer to the actual MSRP but that’s a low bar. What is your VAT/Sales tax rate?

 

[Nate--IRL--]

Just quoting myself here for reference.

Steel Nomad Demo - Stock. (304w)
6883

-20% Power Limit, 2700 VRAM Fast Timing (235w) -75 mV
6821 - about stock performance

This show much AMD pushed these outside their efficiency envelope to get close to the 5070Ti performance.

That's insane, basically the same perf for 77% of the power.

Nate

 

[Sumanji]

I'm so triggered right now...

They were real all along!

AMD's unreleased Radeon RX 9070 XT "reference" design shows up in China - VideoCardz.com

Radeon RX 9070 XT in reference design spotted Remember the low-resolution renders that someone found on Reddit a few weeks ago? We were quick to confirm that those were official renders of soon-to-be-announced Radeon RX 9070 XT cards, despite many claiming they were fake. But to our surprise...

videocardz.com

 

[Tythus]

I'm so triggered right now...

They were real all along!

AMD's unreleased Radeon RX 9070 XT "reference" design shows up in China - VideoCardz.com

Radeon RX 9070 XT in reference design spotted Remember the low-resolution renders that someone found on Reddit a few weeks ago? We were quick to confirm that those were official renders of soon-to-be-announced Radeon RX 9070 XT cards, despite many claiming they were fake. But to our surprise...

videocardz.com

I reckon this was just for development purposes.