Question Speculation: RDNA3 + CDNA2 Architectures Thread

[uzzi38]

Man I have been dying to make this one for a while now.

First rumours for RDNA3 are here so new thread time!

Just going to start off with this one for now: kopite7kimi on Twitter: "@VideoCardz Ah, I mean a simple mcm design with 10240 cores is not enough. Because the lift from RDNA2 to RDNA3 is much bigger than from RDNA1 to RDNA2. We should expect many big improvements in GFX11. 🤔" / Twitter

 

[SolidQ]

Why buy a PS5 over 12 games?

FF rebirth and Stellar blade timed exclusives. So 10 left.

Looks cool, but seems short life, before RDNA4

XFX launches Radeon RX 7900 XTX "Phoenix Nirvana" custom graphics card - VideoCardz.com

Phoenix Nirvana, a new Radeon RX 7900 XTX graphics card from XFX One of the latest releases is based on flagship RDNA3 desktop model. At this point, one would expect leaks on Radeon RX 8900 XTX graphics cards, but XFX seems to be aware of the high-end RDNA4 cancellation plans. Their latest...

videocardz.com

 

[Abwx]

how is this not?

195W average and 244W max according to HVinfo.

 

[SolidQ]

​

AMD releases Micro Engine Scheduler (MES) firmware documentation for RDNA3 GPUs​

https://gpuopen.com/download/documentation/micro_engine_scheduler.pdf

for anyone who want read

 

[igor_kavinski]

AMD RDNA5 is reportedly entirely new architecture design, RDNA4 "merely a bug fix for RDNA3" - VideoCardz.com

AMD Radeon GPUs based on RDNA5 architecture might feature new naming scheme First rumors on RDNA5. Hardware leaker ‘wjm47196’, who has not been very active recently, has shared some new details on the next-gen AMD graphics architecture. The last time we heard from ‘wjm’ was during the Radeon RX...

videocardz.com

Since AMD won't admit to anything, that's the best we've got regarding the issues plaguing RDNA3. 192MB IF cache, yowza! One has to wonder if that was forcing AMD to use three 8-pin connectors as standard and FOUR connectors for third party designs? That would've been a sight!

 

[gdansk]

People keep saying Radeon hasn't been relevant in a decade or more (on certain sites).
But wasn't RDNA2 competitive? Are they gaslighting me or was I wrong and it wasn't worth considering? The only reason I ended up with a RTX 3080 10GB at the time (Nov 2020) was because it was in stock first, I was planning for any of the RTX 3080, 6800 XT or 6900 XT.

 

[blckgrffn]

People keep saying Radeon hasn't been relevant in a decade or more (on certain sites).
But wasn't RDNA2 competitive? Are they gaslighting me or was I wrong and it wasn't worth considering? The only reason I ended up with a RTX 3080 10GB at the time (Nov 2020) was because it was in stock first, I was planning for any of the RTX 3080, 6800 XT or 6900 XT.

RDNA 2 was pretty competitive top to bottom - except against arguably the 3090.

The 6600 still hasn't been beat in the value segment, and the 6700/6750 XT is having hot deals posts still because it is still a very well rounded card at $300.

Availability was the issue, yeah.

Any RDNA 2 GPU was a reasonable choice for gaming when in stock, unless you wanted to run Blender or something else very specific.

 

[Aapje]

In my opinion, people are going overboard with their emotions. RDNA 2 was very solid and RDNA 3 had some significant issues that made it underperform a lot, but if they get that missed performance back with RDNA 4, plus a regular improvement, then it can again be a perfectly fine release.

With the current AI boom, there is zero reason for AMD to give up on GPUs.

 

[adroc_thurston]

In my opinion, people are going overboard with their emotions

it's reddit.

RDNA 2 was very solid

reddit hates Radeon, the thing was instantly memoryholed.

RDNA 3 had some significant issues that made it underperform a lot,

not in the market! Outsold RDNA2 by some margin.

 

[Aapje]

not in the market! Outsold RDNA2 by some margin.

Must be because they ordered way too little of RDNA2 and couldn't get them out during the mining boom, because it it makes little sense otherwise.

In any case, not really a proper comparison with the mining boom and other issues during most of RDNA2's life.

 

[adroc_thurston]

Must be because they ordered way too little of RDNA2 and couldn't get them out during the mining boom

RDNA2 wasn't good for mining.

not really a proper comparison with the mining boom and other issues during most of RDNA2's life.

It kinda is, ASPs in particular are up due to probably N32.

 

[Mahboi]

People keep saying Radeon hasn't been relevant in a decade or more (on certain sites).
But wasn't RDNA2 competitive? Are they gaslighting me or was I wrong and it wasn't worth considering? The only reason I ended up with a RTX 3080 10GB at the time (Nov 2020) was because it was in stock first, I was planning for any of the RTX 3080, 6800 XT or 6900 XT.

RDNA 2 was fine, but it didn't win. It didn't have any real strong advantages, more like a collection of really good points.

AMD could claim rough parity with NV, not victory. You need a hammer into Jensen's face to claim victory from the King.

not in the market! Outsold RDNA2 by some margin.

RDNA 3 sold because it ran on the tail of RDNA 2's good reputation and because Nvidia literally priced Lovelace to not sell. AI dies first.

 

[Mopetar]

RDNA2 wasn't good for mining.

Didn't really matter as long as it was profitable for the miner, which it was.

I think it was mostly supply constraints at the time. TSMC didn't have spare wafers, everyone was trying to get substrate which was apparently in very short supply for a few months, AMD had to fulfill agreements with both Sony and Microsoft for console SOCs, and AMD would make more money making CPU chiplets and selling for Epyc CPUs than they would selling Radeon GPUs at MSRP.

Also, people act like RDNA3 was a massive flop when it was just underwhelming, largely as a result of RDNA2 being so good and AMD hype train going off the rails. The 7800 XT in particular was a great card and AMD sold a lot of those when it first launched because NVidia really didn't have a good competitor card at $400 - $600.

 

[adroc_thurston]

Also, people act like RDNA3 was a massive flop when it was just underwhelming, largely as a result of RDNA2 being so good and AMD hype train going off the rails. The 7800 XT in particular was a great card and AMD sold a lot of those when it first launched because NVidia really didn't have a good competitor card at $400 - $600.

It is bad.
Remember, major execution slips automatically disqualify the product from being good.

 

[PJVol]

Anything within the rated 350W TDP is not overclocking

What? ))

 

[linkgoron]

Also, people act like RDNA3 was a massive flop when it was just underwhelming, largely as a result of RDNA2 being so good and AMD hype train going off the rails. The 7800 XT in particular was a great card and AMD sold a lot of those when it first launched because NVidia really didn't have a good competitor card at $400 - $600.

It was so bad that AMD lied about Navi 31's performance during the launch presentation.

They could have salvaged some of it They could have released Navi 31 as a 7800xt, and priced it slightly better. Instead, the 7900XTX barely edges the 4080, all the while having worse perf/watt and a very complex build. It's difficult to compare die sizes because of chiplets, but It's clear that Navi 31 is much larger and also has about 12 billion more transistors. They should have put it at $899, but they really wanted that $999 price point.

Navi 33 was also pretty bad, showing relatively little improvement over RDNA2 (on 6nm). However, I don't think people were expecting much.

7800XT is pretty decent, mostly thanks to pricing (as is the 7900GRE). It launched at $499 which is 4060TI 16GB MSRP whilst completely obliterating it at 1440p (~43%) and it kind of edges out the 12GB 4070 which had a $599 MSRP. Note that It's barely any faster than the 6800XT, while having a bit more transistors.

 

[gdansk]

The XTX only got away with the $1000 price point because Nvidia was at $1200 at the time. Not that it sold well but it looked OK in charts:

Spoiler

Like this

The XT's price did not fly at all. And it's really hard to explain how they messed up their own presentation and claims.

 

[Mopetar]

It is bad.
Remember, major execution slips automatically disqualify the product from being good.

I disagree. If someone says they'll make something that's 3x better than anything else on the market, but only make something that turns out to be 2x as good, it's not a good product? It's still 2x better than anything on the market. What if they never told you it was supposed to be 3x better and you never know about the execution slip? What if it was only supposed to be 1.5x as good, but some good circumstances resulted in it being 2x as good, but the Internet hype train turned initial reports of 1.5x into 3x?

The 7800 XT wasn't as extreme, but it was solidly better (~30%) than the 4060 Ti 16 GB that had the same $500 MSRP. It's a good product in the way that matters to consumers which is performance per dollar.

 

[adroc_thurston]

The 7800 XT wasn't as extreme, but it was solidly better (~30%) than the 4060 Ti 16 GB that had the same $500 MSRP.

Bigger die for less margin is better value.
Oh no.

I disagree.

Yes it's bad.

 

[Mahboi]

I disagree. If someone says they'll make something that's 3x better than anything else on the market, but only make something that turns out to be 2x as good

Except they said 50% better perf, 50% better efficiency, for 50% better perf, 0% better efficiency (you could get 150% 6950 xt performance if you just pushed your XTX to 750W).
That's a massive failure.

The 7800 XT wasn't as extreme, but it was solidly better (~30%) than the 4060 Ti 16 GB that had the same $500 MSRP. It's a good product in the way that matters to consumers which is performance per dollar.

The 7800 XT is an exceptionally decently priced card in a market full of exceptionally awfully priced cards.
Exception doesn't make the rule.

 

[PJVol]

Except they said 50% better perf, 50% better efficiency, for 50% better perf, 0% better efficiency (you could get 150% 6950 xt performance if you just pushed your XTX to 750W).

A little bit exaggerated, but the idea is stll correct. They promised 50%+ perf/w and get to the 25% at best. And you unlikely get 150% from 6950xt, since there are other limits.

 

[Mahboi]

A little bit exaggerated, but the idea is stll correct. They promised 50%+ perf/w and get to the 25% at best. And you unlikely get 150% from 6950xt, since there are other limits.

https://www.reddit.com/r/Amd/comments/13hygsi/the_true_limits_of_the_7900_xtx_unlimited_power

tl;dr:

36 971 with AMD Radeon RX 7900 XTX(1x) and Intel Core i9-13900KS Processor​

vs

I scored 24 528 in Time Spy

Intel Core i9-13900KS Processor, AMD Radeon RX 6950 XT x 1, 32768 MB, 64-bit Windows 11}

www.3dmark.com

24 528 with AMD Radeon RX 6950 XT(1x) and Intel Core i9-13900KS Processor​

So we are roughly looking at 50% better power for 0% better efficiency. Yes it's 20% better at a "reasonable" point of the V/F curve of course, but that's at a part of the V/F curve that's already insufficient to feed the Mistake.
RDNA 3's entire problem is the power efficiency bust.

Edit: ofc this is considering that RDNA 2 was on N7 and RDNA 3 was on N5. So yeah, it is a pretty gigantic mistake of Russian Execution Squad proportions.

 

[PJVol]

Edit: ofc this is considering that RDNA 2 was on N7 and RDNA 3 was on N5

Oh, I thought you meant you could squeeze 50% out of the 6950XT by oc'ing )

 

[Mopetar]

The 7800 XT is an exceptionally decently priced card in a market full of exceptionally awfully priced cards.
Exception doesn't make the rule.

So the rest are bad, but the 7800XT which is the exception isn't good. I like you would greatly prefer it to only cost $350 and that the rest of the market would shift right along with that, but I don't see that coming to pass anytime soon. If anything we'll be lucky to see the new price tiers remain stable as inflation nibbles at them over the next decade.

The problem with using your subjective valuation is that it would mean there are no good cards. There may not be any for the foreseeable future either. If everything is bad, the word loses meaning.

 

[Mahboi]

So the rest are bad, but the 7800XT which is the exception isn't good. I like you would greatly prefer it to only cost $350 and that the rest of the market would shift right along with that, but I don't see that coming to pass anytime soon.

I wouldn't quite put it to those extremes.
If we're talking general lovelace pricing, I have had one opinion since it was announced: outside of the 4090, everything needs a 25% cut.
4080: 900
4070 Ti: 600
4070: 450
4060 Ti 16Go: 375
4060 Ti: 300
4060: 225.

Now that we know that the 4090 tends to enjoy a nice BBQ time, you can easily say it's entitled to a nice price cut too.

As for RDNA 3, the problem is that once AMD realised it was bad, they seemingly lost any and all ambition whatsoever and priced it overall poorly. An XTX scarcely deserved 1000 but it's a flagship so fine. 7900 XT was ridiculously badly priced compared to the XTX at 900.
The 7600 offers the same VRAM, same general capacity, at a barely lower price than a 4060, except with far worse power draw and no DLSS.

In this horror story of a gen, that Nvidia didn't want to sell because H100 was 1000% the production price and even a 4080 was only 300% at best, where AMD didn't know how to price because...because they're idiots (can anyone seriously defend the 7900 XT at $900???), Navi 32 is an honest offer.
It's a card that they could have sold for $550 and got sold for $500. Sure, the 7700 xt is a bad upsell next to it, but that's fine since yields were probably good and frankly the 7800 xt is just a better card in so many ways.

If anything we'll be lucky to see the new price tiers remain stable as inflation nibbles at them over the next decade.

The problem with using your subjective valuation is that it would mean there are no good cards. There may not be any for the foreseeable future either. If everything is bad, the word loses meaning.

Irrelevant. This isn't inflation, that's the story Nvidia fed people. It's disinterest in the consumer market, plain and simple. In Nvidia's case, the balance they wanted was to sell as little product as possible without losing any marketshare to AMD. AMD didn't want to get aggressive with pricing because A) the product was bad and B) they really have no chance of making more money by lowering prices since 90% of Nvidia buyers think buying AMD is a mistake no matter what the price is. Ultimately a price war just means less margins unless Nvidia initiates it.

Out of a gen that neither side wanted to sell because neither had anything they wanted to sell, we got the worst asking prices in history. If AMD was actively taking marketshare, we may see something different. If Nvidia wasn't making 6x more margins with AI, same.

Now it's about when either of those will happen. RDNA 4 won't be it, that's for sure.

 

[inquiss]

I wouldn't quite put it to those extremes.
If we're talking general lovelace pricing, I have had one opinion since it was announced: outside of the 4090, everything needs a 25% cut.
4080: 900
4070 Ti: 600
4070: 450
4060 Ti 16Go: 375
4060 Ti: 300
4060: 225.

Now that we know that the 4090 tends to enjoy a nice BBQ time, you can easily say it's entitled to a nice price cut too.

As for RDNA 3, the problem is that once AMD realised it was bad, they seemingly lost any and all ambition whatsoever and priced it overall poorly. An XTX scarcely deserved 1000 but it's a flagship so fine. 7900 XT was ridiculously badly priced compared to the XTX at 900.
The 7600 offers the same VRAM, same general capacity, at a barely lower price than a 4060, except with far worse power draw and no DLSS.

In this horror story of a gen, that Nvidia didn't want to sell because H100 was 1000% the production price and even a 4080 was only 300% at best, where AMD didn't know how to price because...because they're idiots (can anyone seriously defend the 7900 XT at $900???), Navi 32 is an honest offer.
It's a card that they could have sold for $550 and got sold for $500. Sure, the 7700 xt is a bad upsell next to it, but that's fine since yields were probably good and frankly the 7800 xt is just a better card in so many ways.

Irrelevant. This isn't inflation, that's the story Nvidia fed people. It's disinterest in the consumer market, plain and simple. In Nvidia's case, the balance they wanted was to sell as little product as possible without losing any marketshare to AMD. AMD didn't want to get aggressive with pricing because A) the product was bad and B) they really have no chance of making more money by lowering prices since 90% of Nvidia buyers think buying AMD is a mistake no matter what the price is. Ultimately a price war just means less margins unless Nvidia initiates it.

Out of a gen that neither side wanted to sell because neither had anything they wanted to sell, we got the worst asking prices in history. If AMD was actively taking marketshare, we may see something different. If Nvidia wasn't making 6x more margins with AI, same.

Now it's about when either of those will happen. RDNA 4 won't be it, that's for sure.

Why so sure? AMD has a downturn in console demand and a vertical with falling revenue to prop up