AMD's cancelled RDNA 4 GPU could have doubled the 7900 XTX's performance

themastergoose said:

This comments section is exactly as I thought it would be:

- RIP AMD, you're finished
- Rebranded 4 year old tech will never sell, so good thing it was cancelled
- The drivers will be trash for years after launch
- AMD can't compete with any of Nvidia's cards
- AMD is going to loose to Intel Arc next generation

Everyone took Nvidia's **** firmly up their butt then. It almost seems fashionable to trash anything AMD do at this point, and I don't get it...

Click to expand...

It's pretty simple. AMD is nvidia's main competitor. We all want AMD to succeed, in part to keep nvidia in check, in part because AMD can do some crazy things when they have an advantage (9800 pro, 5870, 290x anyone?). AMD going under benefits nobody, and while that is not an issue like it used to be, the fact is that AMD is still struggling with the GPU market.

When AMD announces they are abandoning the high end enthusiast market, of course those people will be very animated as a result. They are being left out to dry. AMD always gets SO CLOSE but can never consistently meet expectations, and that wears on people. When AMD constantly squanders goodwill then people are going to get salty and talk about it.

Besides, the comments are not entirely wrong. AMD wont be able to compete with the 5090, 5080, or possibly the 5070 depending on what Nvidia pulls off. Battlemage is a total wild card, if intel pulls off significant improvements AMD could be in trouble in the low and mid range. And frankly, if rDNA4 has no major RT improvements and the same core count, is it not a rebrand? Did anyone really buy the line that the 270x really wasnt a 7870 just because it was GCN 1.1?

redgarl said:

Typical fanboyism...

If AMD provide an XTX for 500$, then Nvidia will need to figure out something to justify the price of a 5070 at 650$, which will be on par with a 4080 at best.

Click to expand...

I agree with emmzo, there's no way in hell AMD is releasing high end performance for $500 on near the same node as said 7000 series. Not a chance in hell.

I find it far more likely they would release 7900xtx performance for $750 at best, 7900 xt at $650.

redgarl said:

If AMD canceled their top GPU, it would be because of CoWoS allocation at TSMC. They are going to make 4B$, at least, on MI300 accelerator alone.

Click to expand...

True but how is Nvidia Making 40 billion usd on ai and planning to have enthusiast level graphics cards with Blackwell. Nvidia made 10.4 billion in USD in 2024 from gaming alone. I guess 4 billion does sound a lot if you don't take that into consideration.

Dang, this is a real bummer! A cancelled AMD monster GPU that could've crushed the competition? Sounds like they prioritized profitability with AI chips over a potential powerhouse. Still, those rumored mid-range cards with killer performance for less sound very tempting! Here's hoping ray tracing gets a bump in RDNA 5 though, gotta keep up with the Joneses (Nvidia).

godrilla said:

True but how is Nvidia Making 40 billion usd on ai and planning to have enthusiast level graphics cards with Blackwell. Nvidia made 10.4 billion in USD in 2024. I guess 4 billion does sound a lot if you don't take that into consideration.

Click to expand...

Nvidia has more capital, and can and has acquired far more fab space at TSMC, then AMD has. AMD is also splitting that workload with Ryzen CPUs as well.

emmzo said:

Yes, most people buy mid range, but not two, three, four years ago mid range. I own a 6700XT. that`s a card released in 2021. Then they released a 6750XT and now a 7700XT, the latter giving me a "whooping" 5 to 12 fps increase if I were to buy it. RT in the dumpster as usual. Not that it matters for a mid range card, but you always want more stuff for your money. Next, they will release a 8700XT that will add nothing, but it will cost... what, 300$? Ok, why should I buy that? For lower electricity bill? Is this AMD`s marketing now? And keep dreaming you`ll get 7900XT performance for less than 500$. It`s 750$ now. They`ll shave off 100 tops for a 8900XT and still be in the loss, considering they started selling at 900. That will make it 650$.

Click to expand...

up to 30% more performance, same MSRP as the RX6700 XT. Excellent gain, and the normal gain expected between generations. The fact that there is leftover RDNA2 stock on the market being sold at a loss is an anomaly, not a rule.

If you're expecting much bigger gains than that, prepare to be disappointed, because on the chip manufacturing side things are getting very slow, which translates into expensive products and minor improvements.

WhiteLeaff said:

up to 30% more performance, same MSRP as the RX6700 XT. Excellent gain, and the normal gain expected between generations. The fact that there is leftover RDNA2 stock on the market being sold at a loss is an anomaly, not a rule.

If you're expecting much bigger gains than that, prepare to be disappointed, because on the chip manufacturing side things are getting very slow, which translates into expensive products and minor improvements.

Click to expand...

There's about 10% more performance not 30% from 6700XT to 7700XT. Look it up on any tech site! And now, 0% performance gain for a 8700XT, but you get to pay 50 bucks less. Great!

emmzo said:

There's about 10% more performance not 30% from 6700XT to 7700XT. Look it up on any tech site! And now, 0% performance gain for a 8700XT, but you get to pay 50 bucks less. Great!

Click to expand...

25% @ 1080p,
27% @ 1440p
More than 30% in some games like cyberbug.

(*An up-coming August announcement from AMD)

Attention Gamers...

AMD Radeons have your back. Whether you are gaming at 1080p, or 1440p AMD will give you professionals frames (ie: 200+ frames) for under $600 bucks & starting at $299.

For those looking to push the edge of gaming and are one of the Gamers who are looking to upgrade to 4k+ gaming... and looking to invest $$ heavily into a 4k Gaming display and are looking for a dGPU to push your new 4k @ 200+ natural frames (no gimmicks needed), then you will have to wait a handful of months for our new High-end Radeon Pro Gaming cards using advance RDNA5 multi-chip architecture.

Stay tuned...!

WhiteLeaff said:

25% @ 1080p,
27% @ 1440p
More than 30% in some games like cyberbug.

Click to expand...

Hmm, depends on the site, some say 10%, some 25%, from a Tehspot chart, comparing an overclocked Sapphire vs stock isn't great. But hey, have it your way, make it 30% to round it up, for 15frames more I couldn't care less anyway. And mind you, this is all you're going to get from next gen too.

Theinsanegamer said:

Nvidia has more capital, and can and has acquired far more fab space at TSMC, then AMD has. AMD is also splitting that workload with Ryzen CPUs as well.

Click to expand...

That's total BS. Nvidia just barely may beat AMD when it comes to money paid for TSMC. However Nvidia does not make anything on somewhat cheap 7nm (AMD does) and Nvidia has at least one custom node that is much more expensive than normal. Also AMD is more loyal than Nvidia.

So AMD surely has much more wafer allocations than Nvida.

HardReset said:

That's total BS. Nvidia just barely may beat AMD when it comes to money paid for TSMC. However Nvidia does not make anything on somewhat cheap 7nm (AMD does) and Nvidia has at least one custom node that is much more expensive than normal. Also AMD is more loyal than Nvidia.

So AMD surely has much more wafer allocations than Nvida.

Click to expand...

AMD was ahead until early 2023. As the AI frenzy progressed, Nvidia became TSMC's second-largest customer last year.

"TSMC's top 3 customers in 2023: Apple with 25%, NVIDIA with 11%, and AMD with 7%
https://www.tweaktown.com/news/9745...-apple-with-25-nvidia-11-and-amd-7/index.html

WhiteLeaff said:

AMD was ahead until early 2023. As the AI frenzy progressed, Nvidia became TSMC's second-largest customer last year.

"TSMC's top 3 customers in 2023: Apple with 25%, NVIDIA with 11%, and AMD with 7%
https://www.tweaktown.com/news/9745...-apple-with-25-nvidia-11-and-amd-7/index.html

Click to expand...

Fab space (ie wafer count) and revenue are different things. I very much doubt Apple has most fab space. They just use most expensive nodes and want them first, both cost money a lot.

Again, Nvidia use custom nodes and mostly use newest and most expensive ones. Nothing indicates Nvidia is bigger customer than AMD when it comes to wafer area ordered.

Theinsanegamer said:

Devil's advocate. Polaris was in a similar situation. It offered performance that matched the old 290x flagship in performance for a cheap $200 price. While they did sell, Nvidia' 1060, 1070, and 1080 ALL outsold the entirety of polaris individually.

It wasn't a "rebrand", but it also wasnt moving the needle forward. The halo effect is real. Even if there is no flagship, stopping at the same performance they already had isnt going to interest as many people.

Click to expand...

They were not better than the 1060 and they weren't cheaper. So it's good and normal that they didn't sell.

I don't get all the hate AMD is getting...
But of course brand loyalty is I strong thing in this world for some reason.

I recently build a new PC, and went with the 7900 XTX... it even felt like a no-brainer for me...

It's slightly lower priced than the 4080 and is on average better performing, and even trades blows with the 4090 once in a while...

My only hesitance was for Machine Learning, as it's a thing I'll occasionally use as a computer engineer...
But AMD is making decent strides these days, and many things kind of just works now...
But Nvidia is ahead with their widely supported and easy to use CUDA "stuff".....

As much hate as the drivers get, I much prefer the AMD software over the Nvidia software also...
Nvidia Control Panel is... borderline useless... Nvidia Experience is the most cluttered driver auto-updater I've ever seen.... maybe except for some motherboard manufacturers "control centers"
And then I still need something like MSI Afterburner to "play" with my card...

The Adrenaline software has all of that stuff in one place... and their overclocking area has everything you need...
You can even make application specific overclocks, for games more sensitive to overclock instability...


Yes, the Raytracing is better on Nvidia...
But in my world, Raytracing is still only at a point where it's interesting to turn on for a little bit too see "what it can do", maybe take a few screenshots and then go back to the nicer experience...
The people with expensive enough GPUs to play games with it on, likely also has expensive enough monitors that they'd rather have high refresh rate...

Sure DLSS is the better upscaling algorithm...
And Nvidia even has access to FSR (and XESS) so them being better wouldn't be a real issue...
But the others are still pretty good, and they're all getting better constantly...

And... I don't buy high-end GPUs to turn down my graphics...
I'm not saying I don't use them... but in almost ALL cases, I'd rather have 120fps on ultra native 1440p, than 150 fps in upscaled ultra 1080p...
And only ever use the quality ones... if "balanced" or lower is needed, I'll adjust something else...


And the amount of stupid comments online with "AMD never has the strongest GPU, so they're useless"...
How many people (especially from the ones commenting this) actually have the strongest card on the market...?
I don't know about you guys, but I'd rather get the best performance and the price point I'm shopping at...
Would you decide not to buy a Mazda Mx5 as a decently priced sportscar, because they didn't produce the Koenigsegg Jesko?

Then of course there's the fact that I'm a person driven by principles...
I'd just rather support the nice competitor, instead of the little child throwing a tantrum and not letting the other kids play with their toys...
DLSS is an example of this...
Initial Gsync and FreeSync was an example of this...
Vulcan also comes to mind...

For machine learning, the AMD stuff is largely even build so you can "use CUDA devices" in the code, and they'll convert it to something the AMD GPU understands for you... so you don't have to adapt already written code to something new...
This is of course also to have a better chance of adaptation in the machine learning area, but Nvidia would never do this...... they'd force you to rewrite the code to fit their GPUs

AMD has had some very unfortunate "acts" also... everybody has...
But Nvidia don't seem to do anything good in between... it's all "screw the costumer" and upping prices....

HardReset said:

Fab space (ie wafer count) and revenue are different things. I very much doubt Apple has most fab space. They just use most expensive nodes and want them first, both cost money a lot.

Again, Nvidia use custom nodes and mostly use newest and most expensive ones. Nothing indicates Nvidia is bigger customer than AMD when it comes to wafer area ordered.

Click to expand...

I have no doubt that AMD uses more wafers than Nvidia, as it has a more diversified lineup and still produces chips in 7/6nm. But when it comes to 5nm and derivative processes(4N) at the moment, Nvidia has more contracted capacity, some analysts have mentioned that AMD won't grab a bigger share of the AI market because it's suppressed by TSMC's capacity, which won't keep up with demand until the second half or the end of the year.

(After growing up with 8-bit Ataris, then my parents getting an 8088 PC, then a 386 which I inherited as my first PC, already with Linux on it... so I've been in the PC game for a long time...) I've NEVER seen a market like this.

It's completely extraordinary how card prices have stayed high for so long, how both AMD and Nvidia still have ongoing sales of like 4-5 generations of cards (in the past, it's not like the older cards disappeared off the market, but the card pricing would drop enough that those 3, 4 generation old cards would be like a $75 card to drop into a potato to help it a bit, not something even a budget gamer would consider since they could get a 1-2 generation old card for about what a 4-5 generation old one costs now.)

And it's extraordinary to have the 4090 sit on the top of the heap anywhere near this long, with AMD not even planning to make a card to take away the speed crown; and Nvidia knowing they are in no hurry to release a faster card either (.... they HAVE faster silicon already, but as with AMD, Nvidia is using the faster designs only for AI accelerators.)

To be honest, given the new Nvidia cards (that are usually being used for AI) are in fact a video card that supports vGPU and all that, I'm surprised they don't just make a variant with a monitor output hooked up and sell it for some ridiculous price like $3000-5000.. They wouldn't have enough takers to cut into that precious AI chip production I imagine, but I imagine some of those members of the "PC master race" that paid like $2000 for a 4090 would cough up the $4000 or whatever to say they have the fastest available card. (Edit: never mind, I see the Hopper H100 is selling for like $30,000... so it would indeed be madness to sell a gamer variant even for $5000.)

If AMD don't pull finger I can see them going the way of ATi which did the same thing AMD are doing now even if they don't compete at the very top end of GPU's with nVidia then they atleast need to compete with the high end of midrange which they are currently struggling to do with 7900XT / XTX cards unless all you look at is rasterization speed they struggle with RT so you need software tricks to get better FPS that's not what we want to see AMD
1: Stick with the Rasterization speed increases
2: Add in purpose built RT cores and stop using rasterization CU's to do the work it doesn't work well
3: Keep up with fps/W gains
4: Keep the prices at a point where people will actually buy your GPU's let nVidia be the most expensive with less sales

I'd rather buy a 250~275W GPU that doubles what FPS I get now for less money and less software tomfoolery needed to do it

It's a damn shame as I would definitely have liked to have like to purchase a new top tier AMD Card... I don't want Nvidia cards right now. My 2080ti straight up died on me.