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

 

[Saylick]

I'm guessing that "3+19= " Tweet refers to an actual retail launch of RDNA 3 on November 22.

 

[amenx]

How many monitors currently ship with a DP2.0 connection?

None yet. I doubt its a priority display manufacturers will scrambling to implement with HDMI 2.1 already filling the need for ultra high res, hi refresh rate monitors, TVs.

 

[biostud]

None yet. I doubt its a priority display manufacturers will scrambling to implement with HDMI 2.1 already filling the need for ultra high res, hi refresh rate monitors, TVs.

So the problem is pretty academic...

 

[blckgrffn]

So the problem is pretty academic...

I think that the idea that you are buying a new $1600 GPU that has at least a 5 year lifespan, it would be nice if it had the most modern ports possible. That's my $.02. I am not likely to buy it, so that's a big shrug from me, but I can understand why some folks are a bit miffed about it. But that's more of a 4090 thread topic.

It certainly won't hurt these cards to have cutting edge display specification ports. Like when cards finally started coming with HDMI 2, it wasn't right at launch when that was a nice to have, it was a year or two later and certainly its nice now (GTX 9 series comes to mind).

 

[eek2121]

This looks like all those 'legimite leaks' last time that RDNA 2 will only compete with RTX 3070. And we all know how it turned out. Not only competing with Ampere but beating it in 1080p and 1440p in rasterization.
Some idiots were even counting frames in demo from AMD Zen 3 presentation to prove their point.

RDNA3 would have to be really a stumble if it can't compete with Ada. After all according to TPU 4090 is 52% faster than 3090 in 4K in rasterization (and 53% faster than 6950 XT). Far from amazing as some are trying to present it.
Ray-tracing I can believe would be an uphill battle. But rasterization should be in a ballpark.

December probably is right. I suppose it's info from AIBs and that they should know.

The 4090 is actually 70-80% faster vs. the 3090. I agree with your statement, however. Even if AMD simply shrunk Navi 21 they would get 40-50% more performance. They have clearly done more then that. The bonehead that runs WCCFTech clearly fails at basic math. AMD wouldn't be putting in a bunch of engineering effort towards a chiplet design unless it is going to pay off in terms of perf/watt. I suspect the gains will be similar to 3090 -> 4090, possibly even higher gains. We will see.

 

[Yosar]

TPU screwed up. They use a 5800X CPU.
Most reviews show 70~80% uplift over 3090 (Full Ada vs Full Ampere will probably be the same)

In 4K? Let me be very doubtful. It's because they actually choose big set of games (25), and not like most of reviewers 7-8, including 4 sponsored by nVidia.
And they were always very favorable to nVidia (I mean TPU, not games).

They have now probably review with most games at least until HU will do their big test with 55 games. Until then there is nothing amazing in 4090 numbers. There are games with 70-80% increase, and games with only 30-40%. That's how average works.

 

[biostud]

I think that the idea that you are buying a new $1600 GPU that has at least a 5 year lifespan, it would be nice if it had the most modern ports possible. That's my $.02. I am not likely to buy it, so that's a big shrug from me, but I can understand why some folks are a bit miffed about it. But that's more of a 4090 thread topic.

It certainly won't hurt these cards to have cutting edge display specification ports. Like when cards finally started coming with HDMI 2, it wasn't right at launch when that was a nice to have, it was a year or two later and certainly its nice now (GTX 9 series comes to mind).

My guess is that most people who buy the 4090 will also buy the 5090 in two years and sell or find other uses for the 4090. But, yes I also wonder why they don't support dp2.0.

 

[blckgrffn]

My guess is that most people who buy the 4090 will also buy the 5090 in two years and sell or find other uses for the 4090. But, yes I also wonder why they don't support dp2.0.

Maybe, I think the 5090 is going to be very unappealing without some big tech changes as the process technology is basically going to stagnate and I think it will look Turing vs Pascal at best but without the addition of even RTX tech. DLSS 4 where it imagines what a game might look like 2 frames ahead? Ha. But that's thoughts for that thread.

And it's likely the 4090s will find new homes for years, much longer than something like 4060s, for example.

 

[TESKATLIPOKA]

In 4K? Let me be very doubtful. It's because they actually choose big set of games (25), and not like most of reviewers 7-8, including 4 sponsored by nVidia.
And they were always very favorable to nVidia (I mean TPU, not games).

They have now probably review with most games at least until HU will do their big test with 55 games. Until then there is nothing amazing in 4090 numbers. There are games with 70-80% increase, and games with only 30-40%. That's how average works.

You are right.
TPU also tests CPUs and there is only 2.2% difference in performance between 5800x vs 12900ks at 4K, but they test only 10 games and RTX 3080 was the GPU.

 

[Leeea]

So the problem is pretty academic...

Until you have a display that uses it and said $1600 card doesn't. It is not like this card has a lifespan of 1 year ...

 

[Panino Manino]

Like Hardware Unboxed said, the 4090 seems to be the first GPU to be held back by current CPUs at 4K.

 

[Leeea]

AMD wouldn't be putting in a bunch of engineering effort towards a chiplet design unless it is going to pay off in terms of perf/watt.

Not necessarily. I think it is all about profit.

A defect in a large monolithic bleeding edge 4n die is going to hurt, and if it is in the wrong place invalidate the entire die. Expensive. The entire 608mm rx4090 die is expensive.


AMD goes with chiplets, 1x 308mm last gen 5n die, + chiplets made on the 6n process. 5n = bug free and paid for by Apple. 6n = well understood and cheap.


It is just a cheap design. AMD may be able to crank these things out at less then 1/2 the bill of materials Nvidia deals with.


Is it going to have the performance nvidia has? probably not
Is it going to bury nvidia in price to performance? if AMD needs to. But odds are AMD will get close enough and pocket the $$$.

 

[TESKATLIPOKA]

The 4090 is actually 70-80% faster vs. the 3090. I agree with your statement, however. Even if AMD simply shrunk Navi 21 they would get 40-50% more performance. They have clearly done more then that. The bonehead that runs WCCFTech clearly fails at basic math. AMD wouldn't be putting in a bunch of engineering effort towards a chiplet design unless it is going to pay off in terms of perf/watt. I suspect the gains will be similar to 3090 -> 4090, possibly even higher gains. We will see.

I am pretty skeptical about N21 gaining 40-50% more performance by simply moving to N5. With the unrealistic linear scaling, It would already mean 3234-3465MHz, and you would also need 40-50% higher clocked memory to feed It(24gbps from Samsung).
Let's say It could be clocked so high, but TBP also wouldn't be the same but a lot higher because of those clocks and needed voltage.

 

[jpiniero]

Maybe, I think the 5090 is going to be very unappealing without some big tech changes as the process technology is basically going to stagnate

Not process technology. The cost... which is completely different. With Moore's Law being dead, you're going to have to either get very frugal with transistors, continue to raise prices or jack up power. Or do all 3.

AMD goes with chiplets, 1x 308mm last gen 5n die, + two gen old 6n chiplets. Mature tech gets better yields off the bat, and even when there is a fatal defect AMD just tosses a cheap little part in the trash.

TSMC has been saying that yields are extremely good on all the nodes you mention. I'd say the wafer cost difference between an N5 and the "N4" that nVidia is using is probably not much. It's still going to be cheaper overall but that's because of the MCD chiplets being on an older node. I'm assuming that any cost savings will just result in better margins for AMD than any sort of price war.

 

[DeathReborn]

Its official!

https://twitter.com/x/status/1580534804696551425

So now they are re-tweeting the tweet they already put out a while back. How many times are they going to announce an announcement of an announcement of an announcement...

 

[maddie]

Not process technology. The cost... which is completely different. With Moore's Law being dead, you're going to have to either get very frugal with transistors, continue to raise prices or jack up power. Or do all 3.



TSMC has been saying that yields are extremely good on all the nodes you mention. I'd say the wafer cost difference between an N5 and the "N4" that nVidia is using is probably not much. It's still going to be cheaper overall but that's because of the MCD chiplets being on an older node. I'm assuming that any cost savings will just result in better margins for AMD than any sort of price war.

How do you suggest anyone, either AMD or NVIDIA, sell enough to justify a design, when overall sales might be starting to decline? Do you really believe that sales this gen will not be less than the previous one, at least at present prices?

Companies will try, as much as possible to maintain sales and margins, as they should, but both won't happen together. if AMD has the mindshare that NVIDIA has, maybe they could, but they don't, and if they start to be too low cost and increase marketshare, NVIDIA will have to respond and so it starts.

Anyway, my take.

 

[Leeea]

when overall sales might be starting to decline?

Are they though?

Because over all sales for AMD includes server chips. Data center is where AMD is doing very well, has very compelling products, and is well positioned.

AMD gaming includes console, and on third 3rd quarter guidance AMD was indicating flat sales.

Embedded revenue was also indicated to be flat.


If AMD server chips make up for the loss of client revenue, AMD may just allocate manufacturing to that and keep prices at MSRP. AMD likely makes more $ on server on a per mm basis.


------------------------------

If a price war is going to happen it is going to be nvidia vs nvidia last gen.

Because lets admit it, outside the $1600 rtx 4090, nvidia's 4080s are not looking compelling. Not with $700 3090s floating around and prices dropping fast.

 

[jpiniero]

How do you suggest anyone, either AMD or NVIDIA, sell enough to justify a design, when overall sales might be starting to decline? Do you really believe that sales this gen will not be less than the previous one, at least at present prices?

At the highest end I think sales will be OK. The people who buy like the 4090 are (or at least they think they are) not going to be affected by any recession. The Money Printer will probably be turned back on eventually and crypto will be back. But that appears to be some time from now.

I could see the dev cycle increasing.

 

[amenx]

I just hope AMD can look beyond goals of quick profits and grab the chance at increasing market share with RDNA 3. And seriously undercut Nvidia in pricing at all tiers. Typically I go with Nvidia but I'm not buying a 4080 16gb at $1200 nor a misnamed 4070 12gb at $900. Even if RDNA 3 equivalents are slower, if cheaper, its bye bye Nvidia this round.

 

[maddie]

Are they though?

Because over all sales for AMD includes server chips. Data center is where AMD is doing very well, has very compelling products, and is well positioned.

AMD gaming includes console, and on third 3rd quarter guidance AMD was indicating flat sales.

Embedded revenue was also indicated to be flat.


If AMD server chips make up for the loss of client revenue, AMD may just allocate manufacturing to that and keep prices at MSRP. AMD likely makes more $ on server on a per mm basis.


------------------------------

If a price war is going to happen it is going to be nvidia vs nvidia last gen.

Because lets admit it, outside the $1600 rtx 4090, nvidia's 4080s are not looking compelling. Not with $700 3090s floating around and prices dropping fast.

OK, I'll take this on.

I think we're only at the start of an extended financial turmoil period, and everything I argue is based on this.

When AMD depended on low margin consoles in the old days to keep the company afloat, there was a very good reason for that. You need revenue to continue operating and maintain the staff for further advances.

Wafers are no longer, it seems, a chokepoint to greater sales, so they no longer need to prioritize higher margin products at the expense of less sales in lower margin ones. Now, they can do both, to as much as the market allows, and in this worsening market, they, and others, will have to drop prices. The memory manufacturers have this boom/bust cycle every few years. Now, it's them and the rest of the industry also.

if AMD no longer has to choose, but can sell all the market wants, why would they limit sales?
Are prices at the optimum point for sales volumes and margin/sale so as to get the highest (not %) profits?

Share prices are also concerned with growth and returning to fiscal reality, earnings/share. Both margins and overall sales affect this. I think, we've been programmed since 2008 to see money printing as the solution, but more are realizing the perils of that seductive trap and where it's ultimately leading us. A corner has been turned by the FED, just witness the screams of the financial & governmental crowd, who are still stuck in the past.

I do not suggest they will sell at a loss, just that competition for fewer sales will force a change, which will lead to lower prices, and let's admit it, the latest and greatest computer stuff is not more important than the necessities of life.

Page 3 - Question - $80 new budget CPU king?

Page 3 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

DAPUNISHER WROTE:
"Older financially established forum goers seem to have trouble understanding how tight a budget many younger gamers have to work with." to which I added "also applies to most of the world".

These are international companies with a world market. We often forget this. In the end, this is a difference of opinions that only time will truly tell which viewpoint is correct. Entertaining for now though.

I agree on NVIDIA pricing. Lower than top tier is not looking great.

 

[maddie]

At the highest end I think sales will be OK. The people who buy like the 4090 are (or at least they think they are) not going to be affected by any recession. The Money Printer will probably be turned back on eventually and crypto will be back. But that appears to be some time from now.

I could see the dev cycle increasing.

NVIDIA consumer division can't survive on 4090 sales. Can they really fund their driver + game studio dev teams on 4090 sales or even + 4080 ones? I can't believe you're suggesting this.

Eventually might be a lot, lot longer than you seem to think. Some of your posts recently made it seem as if it should already have been restarted. We all see how that's going.

 

[jpiniero]

NVIDIA consumer division can't survive on 4090 sales.

That's correct. But I wouldn't be surprised if nVidia for instance decided that making new Ampere SKUs made more sense than releasing AD106/7 on desktop right away.

 

[Heartbreaker]

How do you suggest anyone, either AMD or NVIDIA, sell enough to justify a design, when overall sales might be starting to decline? Do you really believe that sales this gen will not be less than the previous one, at least at present prices?

Not even sure what you are suggesting here. They don't look at the economy every day and decide whether or not to design products. If you stop designing new product, you are soon dead as a company. You just have to weather the economic downturns like every other company. Stronger companies can increase R&D during tough time to come out further ahead after recessions, than competitors that cut R&D during the tough times.

NVIDIA consumer division can't survive on 4090 sales. Can they really fund their driver + game studio dev teams on 4090 sales or even + 4080 ones? I can't believe you're suggesting this.

I didn't see anyone suggest that. NVidia is going to keep selling 20 series, 30 series and keep introducing new 40 series cards.

During a recession the best business to be in, is high end segment, that caters to the wealthiest consumers because they are largely immune from real recession impacts. 4090 sales won't be impacted significantly.

Lower end cards won't cease to exist, they will have lower sales, that is just something everyone will have to contend with.

I also don't expect the recession to lower prices much because then you just get lower margins. You aren't really going to drive low end sales during a recession. For discretionary spending like GPUs, a lot of these lower end sales just won't happen during a recession because people worrying about money for food/fuel/housing aren't going to waste it on a GPU, even if it's a more attractive price. I think it's more likely companies will be looking to limit production where they can, so they are not caught with an oversupply.

 

[Joe NYC]

OK, I'll take this on.

I think we're only at the start of an extended financial turmoil period, and everything I argue is based on this.

When AMD depended on low margin consoles in the old days to keep the company afloat, there was a very good reason for that. You need revenue to continue operating and maintain the staff for further advances.

Wafers are no longer, it seems, a chokepoint to greater sales, so they no longer need to prioritize higher margin products at the expense of less sales in lower margin ones. Now, they can do both, to as much as the market allows, and in this worsening market, they, and others, will have to drop prices. The memory manufacturers have this boom/bust cycle every few years. Now, it's them and the rest of the industry also.

if AMD no longer has to choose, but can sell all the market wants, why would they limit sales?
Are prices at the optimum point for sales volumes and margin/sale so as to get the highest (not %) profits?

Exactly, I agree with that.

It was Intel's philosophy to either have 50% Gross Margin, and if not, sell or close the division. The financial analysts have badgered Intel over the margins, and Intel just capitulated to the analysts. As result got out of many ventures that would have still been quite profitable.

I don't think AMD shares this philosophy.

But what do I know? It is an opinion. It will be put to test, we will see how AMD prices chips, which, at attractive prices, could bring in high sales, very high revenue.

As it happens, the key ones are both 7800: Ryzen 7800x3d and Radeon 7800 XT (Navi 32 based).

If AMD keeps 7800x3d at $449 and the price of 7800 XT starts under $600, we will know that AMD is serious about trying to increase its market share.

 

[moinmoin]

Not process technology. The cost... which is completely different. With Moore's Law being dead, you're going to have to either get very frugal with transistors, continue to raise prices or jack up power. Or do all 3.

Is there any indication which way away from monolithic chips Nvidia will take? They don't intend to keep them monolithic for the coming gens as well, do they?