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

 

[DAPUNISHER]

Ray tracing is a useless eye candy? Because It eats up performance?
Will you set everything to low quality, because higher settings also eat up performance?
It looks like even that useless 4080 16GB will have higher RT performance than 7900XTX.

That is only one ray traced title, and not indicative of how other engines will perform. Cyberpunk'd is poorly optimized for AMD hardware, CPU and GPU. With Spiderman the bleeding isn't nearly as bad.

The game engines AMD announced yesterday, are going to improve AMD's approach to RT even more. Consequently, I don't think 2077 can be used as the poster child for future RT performance in every title. I am not suggesting AMD will outright win any benchmarks with RT on, but that beatdown in 2077 is by no means a good magic 8 ball for what to expect.

 

[RnR_au]

I want 4 slot monsters to end with the 4090. No more. Please.

Sorry to be the bearer of bad tidings...

https://twitter.com/x/status/1588449243404042240

In addition, there is this interesting report of RDNA 3 arch being able to reach 3Ghz...

At the 46:26 mark.

Will the AIBs be able to create 450+W sku's that can accelerate a fair way past the reference sku's?

Take salt for now I guess.

 

[Timorous]

Sorry to be the bearer of bad tidings...

https://twitter.com/x/status/1588449243404042240

In addition, there is this interesting report of RDNA 3 arch being able to reach 3Ghz...

At the 46:26 mark.

Will the AIBs be able to create 450+W sku's that can accelerate a fair way past the reference sku's?

Take salt for now I guess.

The AIB 6950XTs were around 5% faster than the reference design and had more OC headroom on top to extend the gap to about 10% when both were OCd.

So there is potential but it will cost a lot of extra power.

So having a look at this, if the 7900XTX comes in at 54% faster than the 6950XT in this chart it gets 131 FPS. The 4080 being 20% faster than the 3090Ti comes in at 109 FPS. That gives the following

4090 cost / frame $11.11
4080 cost / frame $11.01
7900XTX cost / frame $7.63

That gives the 7900XTX a very large cost / frame advantage vs the 4000 series and vs any other current GPU even at current prices.

In RT using 0.46x scaling for the 4000 series (number from that review), 0.31x for the 7900XTX (Same as the 6950XT perf loss) and 0.42x for the 3090Ti we get the following.

4090 RT cost / frame $24.24
4080 RT cost / frame $24.00
7900XTX RT cost / frame $24.37
3090Ti RT cost / frame $28.95

This shows the RT cost / frame is essentially a tie for these products and the 3090Ti is a little more expensive at $1,100 and would need to be around $925 to match these cost / frames in RT.

Now there is no guarantee the 7900XTX will match the 0.31x scaling factor, it may actually do worse but there is also no guarantee the 4080 will match the 4090 scaling factor given how cut down it is.

So as stated already. Pricing is scaling with RT performance by the looks of it but you get a lot more raster performance for your money. Just of note the 4080 12GB @ 899 would also be around $24.30 per RT frame so I wonder what NV will do with that when they launch it as a 4070. Will they cut it a bit further? Will they release the exact same spec but at a lower price point? Interested to find out because it could offer the best RT bang for your buck if priced right.

 

[Geddagod]

They 've done it on MI200 (https://www.amd.com/en/graphics/instinct-server-accelerators) and I think that's their great advantage (scalability).

It's not recognized as one GPU though. They are just two separate GPUs with a really fast interconnect.

 

[insertcarehere]

Jesus man, MLID blaming viewers and fanboys for misconstruing his leaks on RDNA3
Now he is leaking RDNA4 already.
Good lord, I pity the poor souls relishing those RDNA4 leaks.

All of these attention-seakers who missed the mark hard on RDNA3 need to be shown with clown profiles for a good while...

 

[Joe NYC]

If was him i will go AMD CPU. Intel CPU 13xxx is too hot. 3DX AMD CPU is best bang.

I agree.

But after quite a bad availability and prices in last 2 years, some people may just finally want to get an upgrade for the Christmas. And Zen 4 V-Cache is missing it. In which case 13600 or 5800x3d are good substitutes.

For gaming system, getting Zen 4 now and not wait for Zen 4 V-Cache around the corner seems like a sub-optimal choice.

 

[gdansk]

It was possible for AMD to have a 450 mm2 die (and AMD still may add one), for $1,500 market.

But the Navi 31 seems like a good fit for the current market conditions. And Navi 32 even better.

But for some reason they don't ever bother to go big. Even Big Navi 21 wasn't Nvidia sized. Maybe some day AMD will wake up and realize the only thing internet commenters care about is halo products.

Of course, they don't do it because AMD knows no one wants to buy Radeon. They just want discount GeForce.

 

[TESKATLIPOKA]

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.

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.

 

[TESKATLIPOKA]

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.

Don't tell me what to do or that I should be quiet. It's not my fault you had bad info, but I will acknowledge many things you said were correct, and maybe I was too critical to you, the problem is that the most important ones about performance or frequency were wrong.

Now that I think about It, It's questionable If greymon55's leak about mobile N32 will be as fast as RX6950XT is true.
Doesn't matter, greymon55 deleted his account, and we will find out later.

 

[DisEnchantment]

Wonder if what "botched" it was the use of high density libraries?

Density is all very well for bean counters, but high clocks + density is probably not easy on the first spin.

RDNA3 might be great in any APU though.

That can't be. N33 should be not be affected in this case, but apparently it is. So it is contradictory.

Compared to CPUs where a respin is usually done in case of HW bugs, GPUs are launched if they can work around those bugs in software.
So we have NGG not functioning for 2 generations until RDNA3 where the legacy geometry got removed.

RDNA1 has a ton of bugs like 7 or 8 bugs with software work around which impacted performance a lot because of issuing dummy instructions, reordering instructions, SDMA flushes etc.
With RDNA3, N31/32/33 has some SGPR bug. While Phoenix not.
One significant known performance bug is with the export conflict related to the new OREO/ROOE. They have work arounds for it in mesa.

 

[GodisanAtheist]

What can they say? It's not close to a 4090, which itself is just a cut AD102. Given the performance, this is the only realistic solution. Companies don't usually say "yeah, we messed up this was supposed to have 25% better clocks and destroy the 4090".

- Shoulda reworked the stack naming and had N31 launch as the 7800XTX instead. Would nip a lot of perception issues in the bud right there. As it stands, AMD's 9 series part is competing with NV's 8 series part, which again just reinforces the "value brand" perception of AMD.

They were smart to launch N10 as the 5700XT when it rightly competed with NV's 7 series Turing cards.

 

[uzzi38]

Weren't they one of the originals claiming high clocks? Videocardz had posted some of the slide pictures before the presentation even started so for angstromnomics to have gotten eyes on the press package the day before and then leaving a little random sourceless quip at the end of his blog that says he was actually wrong is evidence of literally nothing at all.

How can you say nobody wanted to leak it when they literally didn't get this information until nov 3? I really doubt a random little blog trying to make a name for itself would sit on leaks like that.

Check through their site. They haven't updated any of the articles after the fact. Try and find one where they said anything about extremely high clocks.

Although I can't provide proof because the Discord messages have been deleted, of the people I know of Skyjuice is the second person that knew of Navi31 sporting 96CUs - preceeding public rumours by several months. The first being over a year and a half ago.

And FYI, I've known Skyjuice for over two years. He's only started posting anything publicly as of late. So the idea that there's no way he's sitting on leaks he could be using to make a name for himself is fundamentally flawed. If that's all he cared about he could have been doing this months, if not years ago.

 

[Kaluan]

Great articles from Phoronix and Tom's on the new RDNA3 info dump:

AMD Reveals More Details Around The Radeon RX 7900 Series / RDNA3 - Phoronix

AMD RDNA 3 GPU Architecture Deep Dive: The Ryzen Moment for GPUs | Tom's Hardware (tomshardware.com)



Looks like Lacouza was wrong about quite a few things.

Anyway, looks like a MCD is roughly 2.05MTr

 

[TESKATLIPOKA]

I expect AMD will have more full N32 chips than they can sell as $700 7800XTs so they will be selling perfectly good dies as cut versions in both laptop and desktop skus.

You need to remember here that N32 is just 200mm². It is smaller than N23 and while 4 MCDs is around 150mm² the combined area for full N32 is comparable to N22 and far smaller than N21. This means margin in the $700 segment is a lot higher and margin in the $500 7700XT segment is probably similar to last year.

There is also the possibility that the 7800XT may also be a high bin for N32 meaning not all complete dies will pass.

N7: $8,000; N5: $16,000; Defect Density : 0.05 Link

	Dies per wafer	Good dies per wafer	Cost per chip	Cost per good chip
N21	98	76	$82	$105.3
N22	160	136	$50	$58.8
N23	233	208	$34.3	$38.5
N32	280	254	$57.1 + 4* $6.2 for MCD = $81.9	$63 + 4* $6.2 for MCD = $87.8
N32 Cutdown	280	254	$57.1 + 3* $6.2 for MCD = $75.7	$63 + 3* $6.2 for MCD = $81.6

The cost of N32 dies is a lot higher than N22, and It's not that much lower than N21. If we include the higher packaging cost, then they should be pretty close to each other.
BOM of the whole N32 card will be also pretty close to N21, considering It also uses 16GB Vram at higher speed, you will save a few $ on the cheaper cooler and weaker power delivery.
Full N32 will have better margin than RX6800($579) or RX6800XT($649) depending on the final price, but have worse margin than RX6900XT($999) or Rx6950XT($1099).
Cutdown N32 will need to have higher MSRP than RX6750XT($549) to end up more profitable.

 

[Panino Manino]

In other news, Gameworks is back: https://www.techpowerup.com/review/portal-with-rtx/3.html

AMD will never catch up, because catching up is useless in a Nvidia world.

 

[coercitiv]

Also, the reviews on this are hilarious: https://store.steampowered.com/app/2012840/Portal_with_RTX/

Top review delivers everything with one short punch.

 

[Kaluan]

Enable DLSS and enjoy 16 FPS experience on the rtx3080.


Well, 16 FPS is not single digit, so that is a massive improvement.


. . .
Even more interesting, on the 4000 series is that DLSS3 they are turning on? With its fake frames and latency issues? In a First person shooter game?


edit:
It appears that it is DLSS3! In a first person game!


It did not take long for journalists to abandon all pretension of integrity and jump onto nvidia's fake frame bandwagon. What a pathetic way to destroy TPUs reputation. Nvidia shills, nothing more.

Well, they also have the 8GB(!) 3070 as a "ideal" card for 4K up on their GPU database. But the 16GB 6800 somehow is just "acceptable" at 4K.

We're also getting closer and closer to 2 years since FSR was launched, but they still can't be bothered to add a FSR sub-section to their ray tracing section, as they have with DLSS when they review nVidia cards... since pretty much ever since DLSS was a thing.

Oh and don't get me started, they bent themselves backwards to give DLSS3 it's own section with every 4000 series review (so far) now.

Pretty obvious nVidia has a say in their review methodology.

 

[TESKATLIPOKA]

From AGF (@XpeaGPU) on twitter.

N31 miss clock target by 700~900MHz. But problem is not only engineering. Current AMD bean counters at top management, ie focusing on extreme PPA for max profit, set impossible goals, thus too much cut corners, and extreme design complexity for the timeframe...

Misunderstanding. AMD management wanted perf crown AND beat NV on price (small dies + chiplet). Too aggressive target that engineers could not achieve without tons of complex tricks but it finally failed. Proof is in RDNA3 ISA, so much compromises everywhere to cut # of xtors

Problem in your analysis is that N31 was designed to compete with AD102 not AD103...

But initial AMD plan was to slightly undercut 4090 price with "competitive" N31 (at least in raster). From what I heard last year, top N31 was aimed to launch around $1.5k N31 should have been a milking cow for Green team...

That's 3200-3400Mhz or 28-36% higher clocks. If this is true, then AMD really *ucked up big time.

 

[TESKATLIPOKA]

On similar but also serious note. I'm willing to bet if and when they eventually launch a fixed lineup, top SKU will be called 7970 something. Cashing in on HD 7970 fame.


Maybe it's just me, but ever since we got confirmation RDNA3+ is a thing, some of us should've gotten the hint that they're not happy with something about RDNA3 that they're willing to launch a mid-cycle redesign.
But as some pointed out in other places I follow, AMD didn't want to be THAT brand (*glances at Intel*) who can't keep schedules so they had to have something ready by end of 2022 to show the investors.

I always associated RDNA3+ with Strix Point IGP.
If they want to capitalize on HD 7970 then It should look like this:
RX 7970
RX 7950
I think they can get rid of XT or XTX, I just don't like It.

 

[biostud]

AMD has been quite honest with their performance when they launched their CPU's. So when they say 54% increase performance/watt going from RDNA2 to RDNA3 we no longer believe this to be the case?

 

[Karnak]

AMD has been quite honest with their performance when they launched their CPU's. So when they say 54% increase performance/watt going from RDNA2 to RDNA3 we no longer believe this to be the case?

It def wont be the case for their real products because there wont be a 300W 7900XTX which they did use for the perf/watt claim. Pushing it all the way up to 355W means it'll be less efficient, same goes for the 7900XT since it's obviously slower @300W than a XTX.

 

[exquisitechar]

Nvidia is simply a strong competitor in comparison to Intel.

In addition. AMD being on 7nm when Nvidia was on Samsungs 8nm garbage created an illusion of AMD catching up but since maxwell, AMD has generally been behind.

Add in the high development of finfet chips, AMD generally CPU first spending for R and D and we get the typically AMD being behind in the GPU space.

What kind of magnifies this disappointment however is the hype train that AMD's guerilla marketing team creates for every launch. Fake leakers that mysteriously disappear after launches(the most recent is greymon). The use and then subsequent destruction of youtubers/influencers who use leaks for their platform. Eg. s semiaccurate, adoredtv, MLID, redgamingtech. After each of these hype trains crashes, typically these influencers have faded into obscurity. The next person awaiting that fate is likely MLID. AMD used to do it more transparently with people on forums and AMD representatives, but after similar hype crashes, AMD reps and subsequently own reputation AMD took a hit.

Don't think most of those have anything to do with AMD. Greymon probably disappeared because his leaks were a flop and people were going to stop taking him seriously or would just make fun of him if he kept going. AdoredTV also just had a meltdown after his terrible Zen 2 and Navi leaks. Charlie is just Charlie, he hates Nvidia.

AMD isn't really terribly behind in PPA even with the current N31, only the RT performance is dismal. If there's a bug that affects clock speeds as badly as rumored, the parts without it will be a slam dunk in terms of PPA. We'll see when N32 or the supposed fixed N31 releases.

 

[TESKATLIPOKA]

N31 in this state is a FLOP only the price is at least good.
Power draw during Video and multimedia is embarrassingly bad.

 

[Kaluan]

Not great, not terrible.
At least better than those suspicious 3DMark "leaks" VideoCardz put out (and later double downed on, without a source... again)

Power usage for 7900XT is around reference 6900XT's and for 7900XTX slightly less than reference 3090. Closer to 4090/4080 efficiency than closer to last gen's best (RX 6800). Overall big efficiency gains, tho not as huge as I hoped. There seems to be some bugs with power usage in non-gaming workload tho. Hope they're driver level fixable.

IMO the architecture shows a lot promise, (look at the huge percentile lows advantage it has over 4080 and even 4090 in a lot of the reviews), too bad a lot if it got neutered by this need to rush it before they're properly adressed.

Hopefully AMD's eventual second go at it will properly flesh out RDNA3.

 

[GodisanAtheist]

@tajoh111 I think a 50-60% gen on gen increase at the top end is a totally rational perspective to take, and that was the general perspective I saw on this board as we zoned in on release. A die shrink, an arch bump, similar total silicon to the prior gen top part.

AMD seemed to have found its footing with the RDNA arch, and RDNA2 really knocked things out of the park while staying on the same 7nm process. 5700XT to 6700XT was a 30% performance bump with the same shader layout (much larger die though). The arch scaled fantastically top to bottom. AMD (really any GPU vendor, and AMD also had great CPUs in its corner) was making money hand over fist so there really shouldn't have been a funding problem.

I think it would have been irrationally pessimistic to say that RDNA 3 would at best scrap together 35-40% improvements over the prior arch despite more than doubling the transistor count.

I hope we get a post mortem on this launch, part of the reason AMD is discussed more is the variability of their launches as well. NV is a known quantity, we generally know what they're going to do and there is a good amount of rabble rousing when they miss the mark (2080Ti only managing a 30% raster improvement over the prior gen).