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

 

[TESKATLIPOKA]

Yes, full die but maybe worse binned die? The good dies that can reach higher speeds are possibly being saved because both AMD and Nvidia are keeping their cards close to their chest? Putting out their best dies right now when the market situation is not ideal would only mean having to sell them cheap on sale later. So first put out the slightly slower versions. When sufficient inventory is sold, you show your best card and suddenly, all the people with the big wallets start selling their existing cards to get the newer ones. They know the psychology of people who want the best more than anyone else.

N31 can't compete with Ada102. Saving chips capable of higher clocks is pretty pointless for AMD.
I also don't believe AMD will release N31 with more cache.

 

[igor_kavinski]

Noo way, you want to resurrect the leaks again?
Hype train derailed, now you want to put it back on the tracks again to be derailed once more on December? Very sadistic.

Nvidia doesn't have a 4090 Ti version out yet. AMD wants to see how that performs and then they might respond accordingly (maybe slightly slower but still good enough like 6900 XT was to 3090). If they were to put their best foot forward now, Nvidia would easily deflate their balloon by launching something crazy like a 900W card.

 

[PJVol]

If they could save up to 25% of power just by downclocking shaders 2.5->2.3GHz, then I have to question If It's possible to clock N31 anywhere close to 3GHz without attacking 600W.

We'll find out soon, unless the upper clock limit is psp-firmware restricted )

 

[TESKATLIPOKA]

Nvidia doesn't have a 4090 Ti version out yet. AMD wants to see how that performs and then they might respond accordingly (maybe slightly slower but still good enough like 6900 XT was to 3090). If they were to put their best foot forward now, Nvidia would easily deflate their balloon by launching something crazy like a 900W card.

You didn't see N31's performance? N31 is not fighting against Ada102, but Ada103.
AMD is clearly behind. RDNa2 put a lot better fight than RDNA3.

 

[leoneazzurro]

Well, I think we can expect a 7950XTX sometimes in the future, just like we had the 6950XT. Probably better binned dies, coupled with a substantial increase in power limit. This, is what we saw with this generation. The rumors about stacked cache, I think we can dismiss as fake.

 

[Tup3x]

Yes, full die but maybe worse binned die? The good dies that can reach higher speeds are possibly being saved because both AMD and Nvidia are keeping their cards close to their chest? Putting out their best dies right now when the market situation is not ideal would only mean having to sell them cheap on sale later. So first put out the slightly slower versions. When sufficient inventory is sold, you show your best card and suddenly, all the people with the big wallets start selling their existing cards to get the newer ones. They know the psychology of people who want the best more than anyone else.

Well, maybe they could add third power connector, up the power limit to 450W and call it 7950 XTX or something but I wouldn't say that is worth it. They would get closer to 4090 but NVIDIA can always release the 4090 Ti and the gap would be even wider...



Regarding rumours about extra MCDs... I'm almost 100% sure that it it not going to happen.

Imagine how things would look like with extra MCDs. I would be rather shocked if six is not the maximum. Also then again, I think the main die is the problem and the current bandwidth & cache are plenty.

 

[Kaluan]

Well, we'll know all about Phoenix Point iGPU in about 2 months. However, I'm wondering about 1 year from that point, what GFX1150/RDNA3+ is and if it is iGPU/APU only.

But it's funny seeing people claim PP won't be a big jump from Rembrandt.
It's not like N6 to N4 is a big jump, right? Or that AMD managed a 60%+ jump with just 32% higher WGP+clocks... on a high end die (where scaling likely slows to a crawl), the opposite of what iGPUs are. Yep. 😅

 

[TESKATLIPOKA]

If what Locuza wrote is correct, then If AMD released a ~410mm2 GCD, It would look like this.
It would have 10SE, 160CU(20,480SP), 640TMU, 256ROPs.
96->160CU: 124.8mm2 -> 208 mm2
192->256ROPs: 40.2 -> 53.6 mm2
L2 6M ->10M: 16.8 -> 28 mm2

 

[Yosar]

To me it seems they design this architecture for something else.

But what it means for Mobile, is that
For Phoenix, CPU Cores + Multimedia + GPU + PCIE + DDR5 MC etc. can be done within 140-150 mm2 compared to 180+ mm2 on Ryzen 5000 Mobile and 210+ mm2 on Ryzen 6000 Mobile.

Bean counters likely decided the fate of N31

Well.. who knows.. maybe. One of the earliest rumors were claiming that this time Navi 33 will be first launched. So no high end cards at first. Maybe something changed on the way and they decided after all to launch high end cards first (for unknown reasons).
Fact is APU with the latest graphic architecture RDNA3 will be this time around very fast launched if CES launch would happen.

Maybe they overestimated some capabilities of RDNA3 when put on high end chips, maybe something didn't work out as planned on silicon level (clock seems really low), or maybe this was all planned from the start. In a few years we will know the details in
some interview probably. Surely not now.

It's not like it was the the first time. They scratched RDNA high end parts seemingly due to problems with efficiency (noone still confirmed anything why). They focused on RDNA2 instead and it worked out great. They seem to have problems with maximizing
profits of rebuilt architecture with first try. And the rebuilt in RDNA3 (rearchitectured CU and going chiplets) was probably even bigger than with RDNA.
I doubt there will be any refreshes, if history learns us and something really didn't work out as planned (I mean really 'if') they already focused on RDNA4 and rightfully so.

For now we have to wait for reviews. And if 7900 XTX is really 85%-90% of 4090 (in raster) I will be not disappointed and will be in line to buy this card at price point they announced (I was never in line for the card above 1000$, no matter how good it would be, no
matter nVidia or AMD, a no go personally for me). Who knows maybe there will be some surprises in reviews (knowing AMD I doubt drivers are ready at this moment, it's almost month and half till launch).

For gamers a good news is they went with price rather aggressively (compared to nVidia, because 1000$ card is nowhere cheap of course), so there is a hope with future weaker models they will keep this trend. If so now the ball will be in players garden. If they
choose again nVidia there will be no mercy from nVidia with prices next time. Simply as that. And AMD won't be cheaper alternative. They will be as expensive as nVidia.

Personally I'm a little disappointed they didn't tease X3D models of Zen 4 (if not now then when?). It seems they are not ready yet and quite a while unfortunately before they appear.

All in all from 'leaks' expected more, but if it was more I'm sure the price would reflect it and it would be no buy from me (I would wait for Navi 32). I got a feeling that some now underestimate AMD (maybe it's better) because RDNA3 is the powerful groundwork
laid if you look closely what they did with it.

 

[H T C]

Maybe something changed on the way and they decided after all to launch high end cards first (for unknown reasons)

Something did: crypto crashed.

 

[TESKATLIPOKA]

....
But it's funny seeing people claim PP won't be a big jump from Rembrandt.
It's not like N6 to N4 is a big jump, right? Or that AMD managed a 60%+ jump with just 32% higher WGP+clocks... on a high end die (where scaling likely slows to a crawl), the opposite of what iGPUs are. Yep. 😅

Are you are talking about me?
What I said was that It won't be as big of jump as Cezanne->Rembrandt was. Rembrandt used RDNA2 instead of Vega and also the number of CU was increased by 50%, clocks also increased.

Is moving from N7 to N5 not a big jump? It didn't help RDNA3 that much.
Even that 54% increase in perf/W is by comparing 6900XT vs 7900XTX both at 300W, yet they managed a similar increase on the same process with RDNA2.

RX-816 – Based on AMD internal analysis November 2022, on a system configured with a Radeon RX 7900 XTX GPU, driver 31.0.14000.24040, AMD Ryzen 9 5900X CPU, 32 GBDDR4-7200MHz, ROG CROSSHAIR VIII HERO (WI-FI) motherboard, set to 300W TBP, on Win10 Pro, versus a similarly configured test system with a 300W Radeon 6900 XT GPU and driver 31.0.12019.16007, measuring FPS performance in select titles. Performance per watt calculated using the total board power (TBP) of the AMD GPUs listed herein. System manufacturers may vary configurations, yielding different results.

 

[Karnak]

Well tbf it's not easy to achieve a 50% Perf/Watt jump on the same node. That's like a 'once in 10 years' thing I'd say.

 

[DisEnchantment]

Well.. who knows.. maybe. One of the earliest rumors were claiming that this time Navi 33 will be first launched. So no high end cards at first. Maybe something changed on the way and they decided after all to launch high end cards first (for unknown reasons).
Fact is APU with the latest graphic architecture RDNA3 will be this time around very fast launched if CES launch would happen.

Maybe they overestimated some capabilities of RDNA3 when put on high end chips, maybe something didn't work out as planned on silicon level (clock seems really low), or maybe this was all planned from the start. In a few years we will know the details in
some interview probably. Surely not now.

It's not like it was the the first time. They scratched RDNA high end parts seemingly due to problems with efficiency (noone still confirmed anything why). They focused on RDNA2 instead and it worked out great. They seem to have problems with maximizing
profits of rebuilt architecture with first try. And the rebuilt in RDNA3 (rearchitectured CU and going chiplets) was probably even bigger than with RDNA.
I doubt there will be any refreshes, if history learns us and something really didn't work out as planned (I mean really 'if') they already focused on RDNA4 and rightfully so.

For now we have to wait for reviews. And if 7900 XTX is really 85%-90% of 4090 (in raster) I will be not disappointed and will be in line to buy this card at price point they announced (I was never in line for the card above 1000$, no matter how good it would be, no
matter nVidia or AMD, a no go personally for me). Who knows maybe there will be some surprises in reviews (knowing AMD I doubt drivers are ready at this moment, it's almost month and half till launch).

For gamers a good news is they went with price rather aggressively (compared to nVidia, because 1000$ card is nowhere cheap of course), so there is a hope with future weaker models they will keep this trend. If so now the ball will be in players garden. If they
choose again nVidia there will be no mercy from nVidia with prices next time. Simply as that. And AMD won't be cheaper alternative. They will be as expensive as nVidia.

Personally I'm a little disappointed they didn't tease X3D models of Zen 4 (if not now then when?). It seems they are not ready yet and quite a while unfortunately before they appear.

All in all from 'leaks' expected more, but if it was more I'm sure the price would reflect it and it would be no buy from me (I would wait for Navi 32). I got a feeling that some now underestimate AMD (maybe it's better) because RDNA3 is the powerful groundwork
laid if you look closely what they did with it.

Not sure why you are disappointed, RDNA 3 seems like a good scalable architecture. Suitable for everything.
The fact that they did not, lets say, use 8 SE on N31 does not mean the architecture is weak. The other guy totally ballooned his die size relatively.

Is moving from N7 to N5 not a large jump? It didn't help RDNA3 that much.
Even that 54% increase in perf/W is by comparing 6900XT vs 7900XTX both at 300W, yet the same increase they managed on the same process.

They probably lost the perf/watt gains to MCD interconnects. If you cut out 30-50W due to IF energy usage from 7900XTX the perf/W numbers will be much better.
Now on monolithic it is going to be totally different.

But RDNA 2 CU vs RDNA 3 CU, should be around 1.25x perf/clock gain.
1.25x perf/CU * 1.2x CUs * 1.15x frequency = 1.725x, round about the advertised gains by AMD (You can shave off some percentage gains due to CU scaling issues)

So they made some significant gains all around with a significantly less footprint. This bodes well for the integration of the architecture in all form factors.

From VCZ slides, seems @Kepler_L2 is right about the L0 and L1 sizes getting doubled and the frequency. We have to wait for the partner cards. He will be vindicated after all.

 

[jpiniero]

For now we have to wait for reviews. And if 7900 XTX is really 85%-90% of 4090 (in raster)

See I think in the end it's going to be more than that. And that's reflected in the pricing. It should still be decently faster than the 16 GB in raster but only so much.

I definitely get the impression that AMD doesn't want to go much over a grand. And even then, they are getting burnt a bit because even N6 is getting more expensive over time and not cheaper.

 

[H T C]

Regarding rumours about extra MCDs... I'm almost 100% sure that it it not going to happen.

Imagine how things would look like with extra MCDs. I would be rather shocked if six is not the maximum. Also then again, I think the main die is the problem and the current bandwidth & cache are plenty.

Not unless they increase they size of the green part where the chiplets sit (forget the name ... duh ...).

That said, i wonder how much more performance one would get with such GPU, relative to the XTX.

 

[TESKATLIPOKA]

They probably lost the perf/watt gains to MCD interconnects. If you cut out 30-50W due to IF energy usage from 7900XTX the perf/W numbers will be much better.
Now on monolithic it is going to be totally different.

But RDNA 2 CU vs RDNA 3 CU, should be around 1.25x perf/clock gain.
1.25x perf/CU * 1.2x CUs * 1.15x frequency = 1.725x, round about the advertised gains by AMD (You can shave off some percentage gains due to CU scaling issues)

So they made some significant gains all around with a significantly less footprint. This bodes well for the integration of the architecture in all form factors.

As an architecture RDNA3 is pretty good. Just the released N31 could have been better.
The size is tiny, that is good for IGPs, but sadly PP will have only 6WGPs. Of course, more would be pointless with the available bandwidth.
N32 will be faster in raster than Ada104, RT will be worse, but that's expected.
I am not sure about N33, how It will fare against Ada106.

 

[TESKATLIPOKA]

From VCZ slides, seems @Kepler_L2 is right about the L0 and L1 sizes getting doubled and the frequency. We have to wait for the partner cards. He will be vindicated after all.

No, he will be forever remembered as the >3GHz Kepler_L2.
But ok, everyone can be mistaken, next time I will just take what he says with a truckload of salt.

 

[exquisitechar]

No, he will be forever remembered as the >3GHz Kepler_L2.
But ok, everyone can be mistaken, next time I will just take what he says with a truckload of salt.

Considering what Videocardz leaked today, he may be slightly vindicated in the future. Or at least, what he said probably wasn’t made up out of thin air. It seems that >3 GHz may have been the target, but it didn’t happen for one reason or another.

 

[Kepler_L2]

I'd say there's another card in the pipeline. People old enough remember amd and before the buy ati used to have multiple high end cards. most consumers aren't getting 4090s and they weren't buying 3090s. The 1600 price tag for founders minus tax is a large cost to foot that most people won't buy unless they need the very best with the cash flow to do it. nvidia's 4090 is plagued with problems at the moment. it is the butt of jokes online. Amd targetting the 4080 16 gb and below is another blow to nvidia.


people like this kepler moron are upset over being deceived when you should never put faith in leaks because they're leaks and they can be wrong even if they've been right in the past. I don't recall amd stating they were targetting 3 ghz clock speeds. the current clocks setup they have is something they would have planned for long ago, before the 3 ghz bs flew out of someone's bum.

If you don't know what happened, you really should keep quiet.

Kepler is saying on Reddit that AMD underestimated Rtx 4090. It is giving same performance as AMDRX 7900 XTX hoped. He said normally generation performance increase 50% max but nvidia did 80% this time and caught AMD of guard. AMD never thought Rtx 4090 will be this fast.

What? I have not said anything like that

AMD expected to be max 50% faster than rtx 3090 but according to Kepler When Reviews came of rtx 4090 than majority of reviews even said it is 100% faster in some game on raster compared to rtx 3090. He also said that Rtx 4090 lack of hype also played big role of convicing Amd that rtx 4090 will be a turd.

Kepler also said some media played big part in downplaying rtx 4090 and convicing Amd to be beat Nvidia this gen.

?????????

Why should we care about what he says? He is no leaker, no insider. He is just a fan, who acts like he knows some insiders, and was one of the guys hyping up RDNA3.
RDNA3 launch just ended up embarrassing him.
His clown tweet is more than enough proof of that.

Pretty much everything I leaked in the past 6 months ended up true, including the 3GHz stuff. Again, if you don't know what happened, keep quiet.

It's funny how he believes everything Kepler is saying.
Kepler lost all credibility with his RDNA3 hype. End of story.

Be. Quiet.

No. These are very big companies and they watchdogs in each companies. These companies does not run assumptions or etc. They run factual data, and research.

Kepler said in June that those rumors are false and AMD is wrong foot to underestimated rtx 4090.

Kepler said this in Jume 2022 on twitter.

Stop spreading lies please.

 

[eek2121]

you should only consider cpu and gpu upgrades every other generation.

Nah dawg, dat ain’t me.

I am addicted to upgrading every year, pls send help.

 

[exquisitechar]

If you don't know what happened, you really should keep quiet.

What? I have not said anything like that

?????????

Pretty much everything I leaked in the past 6 months ended up true, including the 3GHz stuff. Again, if you don't know what happened, keep quiet.

Be. Quiet.

Stop spreading lies please.

I think that guy is confusing you with kopite or Greymon.

Anyway, you got a hint as to what’s happening with the clocks? Is the 7900 botched and another RDNA 3 part will fix it or something?

 

[JayMX]

Why so many people attack the so called leakers is beyond my understanding. I am 99% certain these are the same people that rush on twitter to read the leaks.

I am happy we have people like greymon55, Kepler, MLID and so on because I like technology and instead of reading about war or bs politics I prefer reading about tech - I don't mind even if it is a rumor or speculation. So yeah big thanks to all of them that make my day interesting.

 

[Kepler_L2]

I think that guy is confusing you with kopite or Greymon.

Anyway, you got a hint as to what’s happening with the clocks? Is the 7900 botched and another RDNA 3 part will fix it or something?

Yes, the HW is botched. No, I don't know if we will see a "fixed" Navi31 ever coming out.

 

[igor_kavinski]

Pretty much everything I leaked in the past 6 months ended up true, including the 3GHz stuff. Again, if you don't know what happened, keep quiet.

Enlighten us.

 

[igor_kavinski]

Yes, the HW is botched.

Whose fault? AMD or TSMC?