AMD Radeon 9070 XT Rumored to Outpace RTX 5070 Ti by Almost 15%

[chrcoluk]

AMD for years are behind on the software side, they may have had a eye candy control panel, but things like SGSSAA lacking, FSR inferior to DLSS, and not so great DX9/11 performance.
Rumours are FSR4 will be exclusive to the new cards, whilst Nvidia are pushing out a DLSS update for 7 year old cards, AMD has to address these sort of things, in the mean time they have to do what they did with Zen1, heavily subsidise price for market share. 5-10% here and there, is not enough if they serious about it. A 20-30% under cut at minimum.

 

[3valatzy]

AMD for years are behind on the software side, they may have had a eye candy control panel, but things like SGSSAA lacking, FSR inferior to DLSS, and not so great DX9/11 performance.
Rumours are FSR4 will be exclusive to the new cards, whilst Nvidia are pushing out a DLSS update for 7 year old cards, AMD has to address these sort of things, in the mean time they have to do what they did with Zen1, heavily subsidise price for market share. 5-10% here and there, is not enough if they serious about it. A 20-30% under cut at minimum.

And the computing performance of Radeons is extremely low.

This is greed in its poorest form.

I don't think they have time to save the graphics business. Even today the Radeon department is on an artificial life-saving breathing by other niches which are included in the financial sheets in order to be not so obvious that Radeon will be dead soon.

Next news soon - AMD to stop producing Radeon cards and exit the market altogether!

 

[Vya Domus]

No one in their right mind would buy a slow 16GB card for so much. It's 2025 already. Give us 24 or 32 GB VRAM !

Nvidia when they charge 2000$ for more VRAM : wholesome chungus

AMD should give us 32GB for 2$

 

[AusWolf]

For the people who care amount numbers its an improvement. I imagine for people who have use cases outside of gaming (per se) it's also an improvement.

Improvement... I keep asking: what is? You can't keep saying that "it's an improvement" without saying what aspect of it is.

"If AMD does manage to undercut its competitors from NVIDIA by a reasonable margin, there is no doubt that RDNA 4 will be the better choice for most people."

That's just the typical "AMD is cheap shit and should act like it" crap I don't agree with. If AMD produces a solid product, I won't mind paying a reasonable amount for it, even if it's just a little under Nvidia - just enough to call it sane. I'm not expecting them to give anything away for free and to beg for me to buy their card.

Also, didn't Nvidia get kudos for making DLSS 4 was backwards compatible?!

Yes, kudos for that (it should have been like that with all previous versions), but not for the 5080.

And the computing performance of Radeons is extremely low.

View attachment 382676

View attachment 382677

View attachment 382678
This is greed in its poorest form.

How about this?

AMD Details DeepSeek R1 Performance on Radeon RX 7900 XTX, Confirms Ryzen AI Max Memory Sizes

AMD today put out detailed guides on how to get DeepSeek R1 distilled reasoning models to run on Radeon RX graphics cards and Ryzen AI processors. The guide confirms that the new Ryzen AI Max "Strix Halo" processors come in hardwired to LPCAMM2 memory configurations of 32 GB, 64 GB, and 128 GB...

www.techpowerup.com

I don't think they have time to save the graphics business. Even today the Radeon department is on an artificial life-saving breathing by other niches which are included in the financial sheets in order to be not so obvious that Radeon will be dead soon.

Next news soon - AMD to stop producing Radeon cards and exit the market altogether!

How about consoles?

 

[3valatzy]

AMD should give us 32GB for 2$

32GB for 500$ is called revolution and progress. Natural selection - you adapt or disappear.

 

[rattlehead99]

Doesn't help AMD anything at all on having 15% more performance here when AMD lacks all of the great features the NVIDIA GPU have. I would gladly have 10-15% lower performance in trade for the awesome features the NVIDIA GPUs comes with.

The awesome feature set is one of the reasons why NVIDIA sells like hot cakes.

According to rumors RDNA4 should rival Blackwell in both Ray Tracing and upscaling.

 

[jesdals]

Is that Nvidia 15% value or AMD 15% value? Or perhaps real world value 15% - nah thats impossible - will wait for TPU review

 

[Denver]

And the computing performance of Radeons is extremely low.

View attachment 382676

View attachment 382677

View attachment 382678
This is greed in its poorest form.

I don't think they have time to save the graphics business. Even today the Radeon department is on an artificial life-saving breathing by other niches which are included in the financial sheets in order to be not so obvious that Radeon will be dead soon.

Next news soon - AMD to stop producing Radeon cards and exit the market altogether!

Nah.. in reality, AMD consistently outperforms its direct competitors across all segments in terms of raw computing power. While CUDA optimization does offer an advantage by enhancing weaker hardware, the tests referenced above were conducted under suboptimal conditions.
TPU utilized outdated software and unrealistic scenarios, such as employing small models that run efficiently even on iGPUs to measure the performance of high-end GPUs. Furthermore, the larger, more realistic models typically used to evaluate GPUs' capabilities do not fit within the VRAM limits of most NVIDIA GPUs, with the exception of the RTX 4090 and RTX 3090. Therefore, it is evident that these tests present a biased perspective.(Even if it's not intentional)

**Constructive criticism:** It would be more effective to use larger LLMs. KoboldCPP (YellowRoseCx/koboldcpp-rocm) could be utilized for the image generation test, providing a straightforward way to ensure that the API operates under the same conditions. Additionally, Blender could be updated to the latest version to leverage better optimization and, maybe explore EEVEE Next, offering a more balanced comparison.

 

[Hiner101]

The other thing I noticed is in the nvidia 5xxx reviews TPU had that strange, unexpected and frankly unnecessary line about not being sure if AMD will be in the GPU space in a couple of years. Didn't really expect it from them as it was...idk something the trashy rumor sites would post and i'll stop at that

Don’t worry, you’re not alone!
It’s the first thing I noticed, and I’m sure everyone else did too. A totally unnecessary personal comment from a fanboy not a reviewer. It stood out so much while reading that it felt completely out of place, like it was just stuck there on purpose. The review, lacking impartiality, was completely ruined. It could have said anything, and I’m sure every negative aspect was avoided or even removed.

If I were the editor-in-chief, there would definitely be a serious talking-to! In your free time, you’re free to express whatever you want on leak and rumor sites, but when you work for us, since our job is to provide reviews, there are "rules" to follow. Otherwise, our credibility is worth zero. People need to see us as credible and unbiased or they go elsewhere. What a fall TPU!

 

[Hiner101]

Talking about raw power from AMD see this published last week, raw power is the one thing that has never been lacking in AMD architectures:

https://chipsandcheese.com/p/sizing-up-mi300as-gpu

 

[AusWolf]

Don’t worry, you’re not alone!
It’s the first thing I noticed, and I’m sure everyone else did too. A totally unnecessary personal comment from a fanboy not a reviewer. It stood out so much while reading that it felt completely out of place, like it was just stuck there on purpose. The review, lacking impartiality, was completely ruined. It could have said anything, and I’m sure every negative aspect was avoided or even removed.

If I were the editor-in-chief, there would definitely be a serious talking-to! In your free time, you’re free to express whatever you want on leak and rumor sites, but when you work for us, since our job is to provide reviews, there are "rules" to follow. Otherwise, our credibility is worth zero. People need to see us as credible and unbiased or they go elsewhere. What a fall TPU!

The reviewer is the editor-in-chief in this instance.

 

[Jtuck9]

Don’t worry, you’re not alone!
It’s the first thing I noticed, and I’m sure everyone else did too. A totally unnecessary personal comment from a fanboy not a reviewer. It stood out so much while reading that it felt completely out of place, like it was just stuck there on purpose. The review, lacking impartiality, was completely ruined. It could have said anything, and I’m sure every negative aspect was avoided or even removed.

If I were the editor-in-chief, there would definitely be a serious talking-to! In your free time, you’re free to express whatever you want on leak and rumor sites, but when you work for us, since our job is to provide reviews, there are "rules" to follow. Otherwise, our credibility is worth zero. People need to see us as credible and unbiased or they go elsewhere. What a fall TPU!

"The rumor mill also churns out something on graphics. Depending on how the Radeon RX 9000 series and RDNA 4 fare in the market, AMD could revisit the enthusiast segment with its next generation UDNA architecture that the company will make common to both graphics and compute. The company's next-generation discrete GPUs will be built around the TSMC N3E foundry node."

AMD to Build Zen 6 CCD on TSMC 3nm Process, Next-Gen cIOD and sIOD on 4nm

AMD is rumored to be building its next-generation CCD (core complex die) that implements the "Zen 6" microarchitecture, on the 3 nm TSMC N3E foundry node. This is part of a set of rumors from ChipHell forum, which got past rumors on AMD right. Apparently, AMD will also refresh the I/O dies for...

www.techpowerup.com

 

[mkppo]

Talking about raw power from AMD see this published last week, raw power is the one thing that has never been lacking in AMD architectures:
View attachment 382692
https://chipsandcheese.com/p/sizing-up-mi300as-gpu

Yeah MI300X/A are compute monsters and it's no secret that in many benchmarks it slaps the Nvidia H100 around. Problem is, leveraging the power isn't a small feat so it's mostly the largest hyperscalers and the like who end up utilizing these. Meanwhile if you're anyone smaller, it's just a lot easier to deploy Nvidia clusters because of the support you get.

The 300A is an APU too, but the sheer size is such that 24 Zen 4 cores are sitting in one corner and the complexity is astonishing. Seeing Turin, Instinct etc always makes me think of what AMD can do, and what crumbs they feed the client side. But it is what it is.

 

[Jtuck9]

Improvement... I keep asking: what is? You can't keep saying that "it's an improvement" without saying what aspect of it is.

Go and read some of the reviews with people who are hands on with the card.

Nvidia GeForce RTX 5080 review: "the 4K GPU I'd aim for this generation"

The Nvidia GeForce RTX 5080 boasts fantastic DLSS 4 abilities, and the new Founder's Edition design makes it one of the nicest looking graphics cards yet.

www.gamesradar.com

That's just the typical "AMD is cheap shit and should act like it" crap I don't agree with. If AMD produces a solid product, I won't mind paying a reasonable amount for it, even if it's just a little under Nvidia - just enough to call it sane. I'm not expecting them to give anything away for free and to beg for me to buy their card.

The point I was focusing on was the "for most people". For how long? What advantage does a 50 series card have over the 9000 cards going forward?

Yes, kudos for that (it should have been like that with all previous versions), but not for the 5080.

I did mention that I would be ironic if Nvidia were seen to be more consumer friendly based on the backwards compatibility of DLSS 4

 

[mkppo]

"The rumor mill also churns out something on graphics. Depending on how the Radeon RX 9000 series and RDNA 4 fare in the market, AMD could revisit the enthusiast segment with its next generation UDNA architecture that the company will make common to both graphics and compute. The company's next-generation discrete GPUs will be built around the TSMC N3E foundry node."

AMD to Build Zen 6 CCD on TSMC 3nm Process, Next-Gen cIOD and sIOD on 4nm

AMD is rumored to be building its next-generation CCD (core complex die) that implements the "Zen 6" microarchitecture, on the 3 nm TSMC N3E foundry node. This is part of a set of rumors from ChipHell forum, which got past rumors on AMD right. Apparently, AMD will also refresh the I/O dies for...

www.techpowerup.com

It's known or it should be. Turin previews the new IO die on a larger scale on 4nm and Zen5c is on 3nm already. Zen 6 on client side should follow suit.

That new IO die better be something else. While the current desktop IO die struggles to get 6400 at times on 2 channels, Turin does 6400 officially supported on 12 channels. 12!

 

[Ayhamb99]

Going to wait for reviews of this card and see how the performance actually is before I make a judgement of the card itself, and I hope people don't start overhyping the card based on rumours like this because as previous launches have shown before, overhype will lead up to disappointment, it has happened so many times with AMD specifically with how many times the rumour mill lead to very high excitement and hype for their launches but then it turning into a polarising launch when the reviews come out. Zen 5 is the most recent example, not a bad generation by any means, but the reviews were very polarising due to the performance increase not being as great as expected.

 

[AnotherReader]

This exact thought came to my mind lol. Where's that 4870 team that David wang and co led, they were great. There was no shitty marketing, great prices, great cards, no nonsense. They made nvidia look like absolute fools in that generation. IIRC nvidia had like a massive 25% or so price cut a couple of months after launch because of RV770's arrival. Not happening now, but for a card that's supposed to be a stop gap till UDNA, if it can somehow be close to 4080S that's somehow not too far off a 5080 which I would've never expected but that 5080 turned out to be a POS on PCI-E.

The other thing I noticed is in the nvidia 5xxx reviews TPU had that strange, unexpected and frankly unnecessary line about not being sure if AMD will be in the GPU space in a couple of years. Didn't really expect it from them as it was...idk something the trashy rumor sites would post and i'll stop at that. Anyway, what people fail to realize is development cycles and the company's position at the time.

1) GCN was being developed around 2008-2009 when AMD inherited the arch during it's infancy from ATI who were doing quite well. It was a banger, and even though they ran out of money right after launch it served them well for a decade

2) RDNA was developed around 2016-2017 when AMD were deep in debt and putting all their money, hopes and dreams on Ryzen. It turned out okay, but nothing close to what GCN achieved.

3) UDNA is being developed now, when AMD have money, resources, time and a bunch of clowns in their marketing department. Speaking to people at AMD, they're putting a lot of resources into that thing and rightly so - their whole AI money pit depends on it. There's every possibility it's going to be another banger, but let's wait and see. I just can't see it being worse than RDNA on the 'relative to competition' basis.

It's supposed to launch around the same time TPU claims AMD discrete GPU division might not be around so erm..let's wait and see I suppose.

RDNA is pretty good; it improved significantly upon GCN in performance. The issues with its third iteration are due to missing frequency targets and not aiming for a large enough die. 6900 XT had 80 CUs against the 3090's 82 SMXs while the 7900 XTX only brought 96 CUs to the fight against the 4090's 128 SMXs; of course, it was going to be thrashed.

 

[Zazigalka]

If it beats 4080S, then I would consider 699 a fair price that I would personally pay for it

 

[Jtuck9]

It's known or it should be. Turin previews the new IO die on a larger scale on 4nm and Zen5c is on 3nm already. Zen 6 on client side should follow suit.

That new IO die better be something else. While the current desktop IO die struggles to get 6400 at times on 2 channels, Turin does 6400 officially supported on 12 channels. 12!

I'm waiting for you to spill some more of the beans on UDNA

 

[TheinsanegamerN]

Going to wait for reviews of this card and see how the performance actually is before I make a judgement of the card itself, and I hope people don't start overhyping the card based on rumours like this because as previous launches have shown before, overhype will lead up to disappointment, it has happened so many times with AMD specifically with how many times the rumour mill lead to very high excitement and hype for their launches but then it turning into a polarising launch when the reviews come out. Zen 5 is the most recent example, not a bad generation by any means, but the reviews were very polarising due to the performance increase not being as great as expected.

Anyone getting hyped up by Moore's Law Is Dead frankly deserves to be dissapointed. how many times has this guy been wrong?

 

[Zazigalka]

Anyone getting hyped up by Moore's Law Is Dead frankly deserves to be dissapointed. how many times has this guy been wrong?

This is from MLID ? Damn, and here I was hoping for a $700 4080S...

 

[AusWolf]

Go and read some of the reviews with people who are hands on with the card.

Nvidia GeForce RTX 5080 review: "the 4K GPU I'd aim for this generation"

The Nvidia GeForce RTX 5080 boasts fantastic DLSS 4 abilities, and the new Founder's Edition design makes it one of the nicest looking graphics cards yet.

www.gamesradar.com

I want your answer, not a generic tome on Nvidia's glory that I don't have time to read. I want data, not text - data actually is available here on TPU, and tells me that the 5080 is a 4080 Super with a new name.

What advantage does a 50 series card have over the 9000 cards going forward?

Dunno. You tell me.