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

 

[PJVol]

AMD may be the only one to have cards for gamers with a budget below $250.

I'm pretty sure that's not gonna happen, except may be the salvaged N33 pats.

 

[jpiniero]

so AMD may be the only one to have cards for gamers with a budget below $250.

The original MSRP of the N23 models was $329 and $379. I assume the MSRP is going to be that or more.

If you say that people will be mad because that's what N22 is selling for about right now.. I'd agree. Which makes selling it with N22 still out there very questionable.

 

[Aapje]

@jpiniero

I do expect them to ask too much at first, but if it's cheaper to make, then it makes sense that the price will drop to below what N23 is selling for. But it may take a while and in the meantime, it might make more sense to buy the 6600 (XT) if you want a relatively cheap 8 GB card.

 

[Aapje]

I'm pretty sure that's not gonna happen, except may be the salvaged N33 pats.

The 6600 is already selling for far less than 250, so the 7600 should end up below it too, but probably not at first.

 

[igor_kavinski]

The 6600 is already selling for far less than 250, so the 7600 should end up below it too, but probably not at first.

$300, to account for the slightly better raytracing performance.

 

[PJVol]

$300, to account for the slightly better raytracing performance.

$330 and $280 for the 32cu and 28cu parts respectively, i.e. $10/CU my guess)

 

[Kaluan]

IMHO MLID and his friend put it pretty well, AMD can't ask much more than $299 for a 7600XT that's probably close to but not quite a 6700XT in raster and maybe beyond a 6750XT in ray tracing, at a quite bit lower power envelope, BUT just 8GB of VRAM, considering there's already the 10GB 6700 available at $299 and the previously mentioned 2 for not much above that.

A good initial price would make a splash and likely will give it a great uptick in initial sales, but like most people here said, even with a more unstomachable MSRP, it will likely come down in mere weeks after launch. The only problem for AMD is how much more stock of N23 chips/cards do they have and how would their sale of those be affected.

I much rather prefer great initial MSRPs than post-launch "market readjustments" tho, as the market where I am is always incredibly slow in following suit with lowering prices.

 

[Aapje]

I don't understand why Igor and Kaluan think that the 7600 XT would be a viable RT card. It won't be and if AMD thinks that they can ask a significant premium because it is less bad at RT than the competition, then that seems silly. They can't even really advertise with it, as it would effectively be advertising for Nvidia.

Of course, I do expect AMD to think that they can ask a significant premium, only to be met with a huge shrug from buyers,

 

[igor_kavinski]

if AMD thinks that they can ask a significant premium because it is less bad at RT than the competition, then that seems silly.

They will use some Radeon optimized game to show off and promote how capable the 7600 XT is in RT, telling people that they don't have to turn RT off on an affordable card. That's exactly what any marketing department would do. Make the older gen look unattractive to force users to upgrade.

 

[Aapje]

They will use some Radeon optimized game to show off and promote how capable the 7600 XT is in RT, telling people that they don't have to turn RT off on an affordable card. That's exactly what any marketing department would do. Make the older gen look unattractive to force users to upgrade.

That assumes that customers are dumb and don't understand that Nvidia is better at RT. For AMD, any mention of RT is at best a defensive maneuver, to convince customers that the card is not as bad as people may think. Any attempt to convince customers that the 7600 XT than the 6600 XT is better due to its raytracing ability, automatically pushes customers to cards that are better at raytracing, which are Nvidia's cards. After all, if RT is a reason to pay more for a 7600 XT over a 6600 XT, then RT is also a reason to pay more for a 4060, because the card will be better at that than a 7600 XT.

 

[Joe NYC]

IMO, AMD really did a smart, optimal first step for GPU chiplets. Too ambitions would have been dividing up compute.

The problem multi-chip compute GPUs was really going to be needing massive BW, and super low latency comparable to being on the same chip. If not you then need to resort to horrible SLI/Xfire software solutions.

But the Memory Controller/Cache chiplets sidestep that issue, since all the compute is still together, plus both memory controller and cache take a lot of space, and don't really respond much to process shrinks anymore so is ideal to put on a cheaper less advanced process node.

But it's mostly a cost savings move, so they don't get more performance out of it, and NVidia made a bigger leap this generation because they were getting a much bigger boost moving away from inferior Samsung Process node, and AMD a smaller upgrade from an already good TSMC process.

Hybrid bond link solves the Bandwidth problem. I doubt another approach can solve the bandwidth problem at acceptable power overhead.

One hybrid bond link gives you vertical communications, and it will take 2 hybrid bond links to get horizontal communications.

Mi300 (CDNA3) will have one approach to it, RDNA4 may take a different approach to the achieve the horizontal link between 2+ compute chiplets... There were some patents showing a thin bridge between 2 chiplets, which may be one of the options being considered.

 

[Joe NYC]

Development costs and economic forecasts as usual. There is basically no point to finish development of N32 unless they are sure that (margin of N32 - margin of N21) * expected number of units > cost of finishing N32.

It's probably true given how similar N32 is to N31. The cost of finishing it should have been very low. But it isn't preordained when the design is such a dud as RDNA3, AMD is selling only about 10% of what Nvidia is selling and the economy is in a downturn. These factors conspire to limit number of units.

But these factors, to me, imply no respin. We're getting unfixed N32 because it was too far along to save any money by cancelling or nothing.

IMO, if AMD was not about to respin N32, it would have been released already. The fact that it is still possibly months away are pointing more and more to respin of the silicon.

IMO, it is in AMD interest for the RDNA3 not to be a lost generation, that it eventually gets fully fixed. There will be many subsequent chips, APUs, maybe gaming chps that will be RDNA3 based.

 

[Joe NYC]

It looks to me like stock of N21 is rapidly dwindling. PC partpicker only shows 7 of 61 (7/43 if you remove the 6900 XT) models of N21 cards available for purchase. My local MC echos this as well. 6 models of N21 cards available with only one showing 25+ stock. There were 2 cards with 25+ availability yesterday when I looked. The other card is now showing 21 in stock.

I was under the impression they have a bunch of 5nm capacity booked and would assume they need to make something besides Zen 4 and N31 to utilize it all. If that's the case all of this speculation about N32 being canceled seems kind of baseless.

I have been doing a highly unscientific survey of as well, and I have come to the same conclusion. That there is no inventory overhang of RDNA2 cards any more...

 

[Joe NYC]

Where does 9350 come from for 7nm? I have seen 17k ballparked for 5n before but 7N was always around the 12k mark when Ian spoke with industry contacts and 6N was a bit cheaper since it goes through fewer process steps.

This means in the very best case laid out above N21 is 15% more expensive than N32 and I think with more realistic N7 and N6 wafer prices that gap widens further.

This is what Locuza came up with for N7 and N5 onwards.

TSMC's 7nm utilization went almost off the cliff, down to 60-70% range. I think big customers, such as AMD are getting far better prices on 7nm than that.

The figures I came across (may be wrong) was in $7,000 - 8,000 range for biggest customers even before the dramatic drop in utilization.

 

[Joe NYC]

The original MSRP of the N23 models was $329 and $379. I assume the MSRP is going to be that or more.

If you say that people will be mad because that's what N22 is selling for about right now.. I'd agree. Which makes selling it with N22 still out there very questionable.

I am not sure why you would make an assumption that if MSRP during shortage was $329 - $379, that now, during the glut of inventory and capacity, the MSRP of a similar product would be higher.

I would say that MSRP will be obviously lower.

But then, AMD marketing is known for shooting itself in the foot, by at first over pricing their product, to assure negative reviews, only to discount the prices shortly after those negative reviews have been published.

 

[Joe NYC]

I don't understand why Igor and Kaluan think that the 7600 XT would be a viable RT card. It won't be and if AMD thinks that they can ask a significant premium because it is less bad at RT than the competition, then that seems silly. They can't even really advertise with it, as it would effectively be advertising for Nvidia.

Of course, I do expect AMD to think that they can ask a significant premium, only to be met with a huge shrug from buyers,

Agreed. Nobody even remotely knowledgeable would be considering vanity such as ray tracing in the absolute bottom of the budget cards.

 

[maddie]

I am not sure why you would make an assumption that if MSRP during shortage was $329 - $379, that now, during the glut of inventory and capacity, the MSRP of a similar product would be higher.

I would say that MSRP will be obviously lower.

But then, AMD marketing is known for shooting itself in the foot, by at first over pricing their product, to assure negative reviews, only to discount the prices shortly after those negative reviews have been published.

Don't you know the rule that companies raise prices in an economic slowdown?

 

[Aapje]

Don't you know the rule that companies raise prices in an economic slowdown?

Some people who sniff too much leather for their own good seem to think that is a good idea.

 

[jpiniero]

I am not sure why you would make an assumption that if MSRP during shortage was $329 - $379, that now, during the glut of inventory and capacity, the MSRP of a similar product would be higher.

I'm assuming that they wouldn't be releasing this until N22 (at the very least, and maybe N23) has finally gone dry. So while I'm sure people want to compare it to current RDNA2 pricing, that RDNA2 pricing might be gone by then.

Based upon what Igor said it might just be a lowkey release just to have something out there with N22 and N23 (presumably) gone.

 

[Aapje]

I have been doing a highly unscientific survey of as well, and I have come to the same conclusion. That there is no inventory overhang of RDNA2 cards any more...

That is a really strong indication that we won't just see the 7600, but also the 7800 and 7700.

I already thought that it makes a lot of sense for these to arrive very close together, because the 7600 doesn't really compete for the same process node and it's also not a particularly exciting card.

 

[Joe NYC]

I'm assuming that they wouldn't be releasing this until N22 (at the very least, and maybe N23) has finally gone dry. So while I'm sure people want to compare it to current RDNA2 pricing, that RDNA2 pricing might be gone by then.

Based upon what Igor said it might just be a lowkey release just to have something out there with N22 and N23 (presumably) gone.

The AIBs must be making at least some money in $199 to $299 range, where most of the Navi 23 cards are priced currently.

Navi 33 is a smaller die and memory prices are down, I would think Navi 33 would be priced in this range as well for 8 GB cards. Up to $399 if AIBs come out with 16 GB versions.

 

[jpiniero]

The AIBs must be making at least some money in $199 to $299 range, where most of the Navi 23 cards are priced currently.

I'm assuming it's supply from a year+ ago that retailers are still trying to unload. They might be the one losing money.

 

[Joe NYC]

That is a really strong indication that we won't just see the 7600, but also the 7800 and 7700.

I already thought that it makes a lot of sense for these to arrive very close together, because the 7600 doesn't really compete for the same process node and it's also not a particularly exciting card.

I have a feeling that AMD decided to re-spin Navi 32, because otherwise, it would have been released by now. Definitely before Navi 33.

And we have a date of May 25 for Navi 33 release - from MLID

 

[lixlax]

I have a feeling that AMD decided to re-spin Navi 32, because otherwise, it would have been released by now. Definitely before Navi 33.

And we have a date of May 25 for Navi 33 release - from MLID

If they actually do it and get the clocks up, make the dual issue shaders actually give a noticeable boost in performance etc then it would make the current N31 quite meaningless (since the N32 would perform really close to it). In this case they should re-spin N31 as well.
But in the other hand why didn't they wait then to re-spin N33 as well? Because they needed to launch something for mobile? If the rumours are to believed its barely faster and more efficient vs N23.

 

[jpiniero]

See I think AMD is just moving ahead with RDNA 4.