RDNA 4 Fine Wine? (HUB Vid)

[3DVCash]

Didn't see anybody talking about this:. HUB seems to have found some pretty significant performance gains for the 9070XT since their launch review and attributes it to driver improvements:

Any chance at a TPU deep dive, @W1zzard?

 

[Thimblewad]

We just talked about it in the RX9000 series owners club

HUB(as always) did a great video comparing games that haven't received any updates since the launch of RX9000 series, meaning any improvement in performance can be linked to driver improvements.

Across the board, 1440p sees a 9% uplift, 4k a bit less. Still very important for people using FSR at 4k, since the mative resolution then becomes 1440p. Not saying the 9070XT now beats the 5070 Ti, but it gets pretty damn close.
The price difference seems to be + 150-250 €/$ for the Nvidia card.

 

[Dr. Dro]

A "Fine Wine" argument for a 3 month old product feels so forced. It'd be better to just admit launch drivers were not performing to expectation to begin with. Still, happy to see improvements. It's not in anyone's interest to see AMD permanently wallow in mediocrity while the other company keeps jacking up their prices.

 

[LastDudeALive]

The big question is are these actually "unlocking" the full potential of RDNA 4, or does this just fix specific games that had issues? Looking at Space Marine 2 and CS2, for instance, the 9070 XT under performed in both of those. CS2 specifically has a huge issue for all RDNA 4 GPUs thus far. AMD has done an extremely impressive job with improving the RDNA architecture thus far, so I'm totally open to the idea that there's still more performance sitting on the table. We should just be sure that's actually what's happening, instead of just fixing mistakes with launch drivers.

 

[biffzinker]

Videocardz has it mentioned on their website since I looked earlier.

https://videocardz.com/newz/amd-rx-9070-xt-now-faster-than-rtx-5070-ti-with-9-gain-at-1440p-from-driver-updates-since-launch

 

[BlueBee]

Hopefully these later drivers give a nice uplift for both 9060 XT and 9070 /XT models. I do recall seeing not the best performance in games that ran certain engines, CS 2 being one that stood out and seems to have really improved. Strange how they only did 1440p and 4K though and not 1080p, hopefully that`ll come in due course!
Be even better if the next WHQL drivers can improve this a bit more

 

[Macro Device]

Says the person who has bought a 5090

Because there's nobody else making a GPU that fast.

And yeah, 3 months after release RDNA2 also got some bumps here and there, along the lines of 7 to 10 percent if I ain't tripping. This isn't fine wine, this is bug fixes. Don't confuse these. We all know RDNA2 and 3 aged like milk, the same might happen to RDNA4 as well. But who knows.

 

[biffzinker]

Yeah, what do you want me to buy? Where's the AMD GPU that performs like a 5090? Can you please let me know where it is?

9070 XT in Crossfire could have been the answer.

 

[Dr. Dro]

9070 XT in Crossfire could have been the answer.

Biff, you know better than anyone that CF is dead... and next to no games support multi GPU anymore. So many asterisks, ifs, and setbacks involved.

 

[DemonicRyzen666]

9070 XT in Crossfire could have been the answer.

They support mGPU, not Crossfire. There's like 10 legitimate games that use mGPU. It's like mGPU supports slightly more games than what have path tracing by comparision.

Chernobylite should work for a test, it also has Raytracing on it.

I swear DX12 is just utter trash. it's not supposed to heavily driver dependant, while it's 10x more driver reliant that DX11.

the RTX 2080 ti got like 20% increase over it's lifetime.

 

[biffzinker]

I know Crossfire is dead I was speculating if two 9070 XT cards paired together would have matched your 5090.

 

[Grotesque Subhuman]

Bug fixes, which was predictable after some games struggled on early RDNA 4 drivers.

Meanwhile, recent NVidia drivers boosted performance 5-8% in most synthetics, which I'm sure AMD Unboxed will cover...

 

[Dr. Dro]

I know Crossfire is dead I was speculating if two 9070 XT cards paired together would have matched your 5090.

If they scaled to a perfect 2x, something that has never been achieved by either NV or AMD in the prime days of scalable multi GPU. The 5090's performance advantage is just off the charts. And there's also the memory bandwidth consideration, which at high resolutions can matter a lot - won't even go on the 16 GB merit, since VRAM is never doubled in this scenario.

 

[blinnbanir]

If they scaled to a perfect 2x, something that has never been achieved by either NV or AMD in the prime days of scalable multi GPU. The 5090's performance advantage is just off the charts. And there's also the memory bandwidth consideration, which at high resolutions can matter a lot - won't even go on the 16 GB merit, since VRAM is never doubled in this scenario.

Don't forget 4x the cost

 

[biffzinker]

They support mGPU, not Crossfire. There's like 10 legitimate games that use mGPU. It's like slightly more games that have path tracing by comparision.

Chernobylite should work for a test, it also has Raytracing on it.

I swear DX12 is just utter trash. it's not supposed to heavily driver dependant, while it's 10x more driver reliant that DX11.

So much for being a low level API framework if the GPU manufacturers have to keep tweaking, bug fixes, and shader replacements.

 

[LastDudeALive]

If they scaled to a perfect 2x, something that has never been achieved by either NV or AMD in the prime days of scalable multi GPU. The 5090's performance advantage is just off the charts. And there's also the memory bandwidth consideration, which at high resolutions can matter a lot - won't even go on the 16 GB merit, since VRAM is never doubled in this scenario.

It would also be more than 2x power consumption, and performance would be limited by the fact that consumer CPUs can only do 2 PCIe 5.0 x8 GPU slots, so the communication between the GPUs and the CPU is gimped by that.

 

[Dr. Dro]

Don't forget 4x the cost

Cost is not exactly the first limiting factor when building any sort of extreme system with multiple graphics cards. It used to be popular because you could start with one card and then just add another down the road, limiting the up front investment. An SLI/CF rig with multiple high end GPUs was never an affordable option.

The 5090 is expensive, but it's justified on the fact that there is really nothing else like it outside of the professional realm (at 4x its already high cost). There is also no shame in going with something more down to earth, and if you put your obsession with "Dro is an Nvidia fanboy" aside, you'd see I always recommended the 9070 XT to everyone.

It would also be more than 2x power consumption, and performance would be limited by the fact that consumer CPUs can only do 2 PCIe 5.0 x8 GPU slots, so the communication between the GPUs and the CPU is gimped by that.

Yeah, platforms aren't built for this anymore. Though, the power consumption already happens with the single 5090, it's 600 W after all.

 

[Outback Bronze]

Guys, I have already had to LQ (now deleted) two posts. Please try and keep this thread on track.

This is not a 5090 vs AMD driver's thread. Thanks.

On the subject, I did see this vid from hub and it's great to see these drivers improve most games by some margin.

 

[Dr. Dro]

Guys, I have already had to LQ two posts. Please try and keep this thread on track.

This is not a 5090 vs AMD driver's thread. Thanks.

On the subject, I did see this vid from hub and it's great to see these drivers improve most games by some margin.

That is fair, yes. I withdraw my reply, since it does come off a bit as stoking the flames.

So much for being a low level API framework if the GPU manufacturers have to keep tweaking, bug fixes, and shader replacements.

DX12 and Vulkan sought primarily to shift responsibility for the optimization stages from game developers to engine programmers, streamlining the development process. Engine programmers would then have a closer knit collaboration with GPU vendors and driver engineers, to the benefit of all. This is why most modern engines work with a near WYSIWYG editor, but much of the work has been shifted towards the engine vendors themselves (Unity, Unreal, etc.) - the problem with that model is that the GPU drivers become more of a bedrock that software must rely upon to function well than ever.

 

[wolf]

Looks like multiple of the gains come from games which were clearly underperforming at launch relative to where the card was expected to perform. And it's yet more HUB testing where the test suite itself is chosen for this video to purposely test a higher proportion of games that have the bigger gains - well done again HUB constructing the testing with your desired outcome already chosen.

Fine wine is how the cards age after years, not fixing games that underperformed at launch 3 months later, but that's also HUB doing their best to sensationalise/outrage farm the matter because that's another thing Steve loves to do, he knows the term fine wine is hotly debated / contested and. Afresh bucket of fuel on that fire is part of his goal.

Having said that, the card was priced based on launch performance so this is largely a positive for buyers of the card itself.

Well done AMD for fixing the games that underperformed. Well done HUB for another contrived testing scenario to make whatever point you wanted to.

 

[oxrufiioxo]

well done again HUB constructing the testing with your desired outcome already chosen.

I mean 7 out of the 16 games chosen showed 0% gains and 10 showed less than 5%... I'm sure if HUB tried hard enough they could find 15-20 games that showed gains but did not going by your accusation of them fixing the results into a narrative they wanted to push.

These are also the same games from the initial 9070XT review done by them.

 

[LastDudeALive]

Fine wine is how the cards age after years, not fixing games that underperformed at launch 3 months later,

Yep. Getting a ~9% performance boost from bad launch drivers is way different than, say, the RTX 20 series getting the DLSS Transformer model 6 years after their release.

 

[MxPhenom 216]

The big question is are these actually "unlocking" the full potential of RDNA 4, or does this just fix specific games that had issues? Looking at Space Marine 2 and CS2, for instance, the 9070 XT under performed in both of those. CS2 specifically has a huge issue for all RDNA 4 GPUs thus far. AMD has done an extremely impressive job with improving the RDNA architecture thus far, so I'm totally open to the idea that there's still more performance sitting on the table. We should just be sure that's actually what's happening, instead of just fixing mistakes with launch drivers.

View attachment 406317View attachment 406318

I get about 150-200 more fps in CS2 with my current rig specs than Wizzards review benchmarks and thats on the 25.5.1 drivers. Havent tried with the 25.6.3s yet. Granted thats just going off the fps counter during matches I play and not the same custom scene Wizzard uses for reviews.

 

[Hecate91]

Bug fixes, which was predictable after some games struggled on early RDNA 4 drivers.

Meanwhile, recent NVidia drivers boosted performance 5-8% in most synthetics, which I'm sure AMD Unboxed will cover...

Which games were struggling to run? I only recall people whining over a few games not getting a bazillion FPS. An improvement is nice regardless of bug fixes, fine wine or whatever you want to call it RDNA4 users get a nice performance boost.
And meanwhile Nvidia drivers were causing blackscreens and crashes, apparently they still are but the media has remained quiet on it because reasons.
The performance increase is quite significant in some games from HW Unboxed testing. But well done for the mindshare bringing in the flamethrowers.

 

[oxrufiioxo]

People are getting way too hung up over the title of the video that was obviously used for more clicks lol