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

 

[Timorous]

If we take the 6950XT number for this average and multiply by 1.54x which was the perf/w increase claim then you get 131 FPS for the 7900XTX. That puts the 4090 10% ahead in raster but it costs 60% more.

The 4080 16GB is the real competition and well at just 20% ahead of the 3090Ti in raster it does not even look like it will compete with the 7900XT let alone the XTX.

In RT the tables change and there it looks like the perf/$ for RT performance will be about the same between AMD and NV assuming the 4080 16GB has the same dropoff in performance as the 4090. If it is a bit worse then even in RT the perf/$ tilts towards AMD as well.

 

[IntelUser2000]

I'm glad to see the leakers like MLID/Adored/RGT be knocked down several notches.

Also it was unrealistic to expect 50% perf/watt improvement and move to 5nm would result in a 50%+5nm. Actually, the process gains are included in the 50% perf/watt gain. So the total is 50%.

People also expected Ada to get a bigger improvement. One "law" in complex semiconductors is that 9 out of 10 times, it disappoints the hype.

I'm also in agreement that AMD should not have repeated the RV770 strategy and have a big chip to take the crown. Halo matters, otherwise you're trying to sell many chips with low profit. And in reality you won't even get marketshare.

You lose in three ways:
-Low volume
-Lower costs to compete with competitive product of the same class
-More money lost due to not being able to sell a halo product

 

[Karnak]

The pricing suggests it won't hold up in practice. I think they absolutely would have gone higher on pricing if it was only 10% slower than the 4090.

Because RT exists (especially for enthusiasts) and with subpar RT performance you can't justify a higher price tag. If they can't even beat a 3090 across the board in RT (which they won't according to their own benchmarks) then you can't go higher where the 4080 16GB already sits at $1199. Which will be way faster with RT.

AMD knows that and that's why the price is what it is.

 

[IntelUser2000]

Because RT exists (especially for enthusiasts) and with subpar RT performance you can't justify a higher price tag.

AMD absolutely cannot charge higher than Nvidia, actually even if the product was identical, they cannot charge the same, because Nvidia has far better brand recognition. In order to get out of this cycle, they need leading performance product for several years that offers better perf/$ and then the perception will slowly change.

 

[Timorous]

Because RT exists (especially for enthusiasts) and with subpar RT performance you can't justify a higher price tag. If they can't even beat a 3090 across the board in RT (which they won't according to their own benchmarks) then you can't go higher where the 4080 16GB already sits at $1199. Which will be way faster with RT.

AMD knows that and that's why the price is what it is.

Assuming the 4080 has the same raster : RT ratio as the 4090 (and given how cut down it is it may actually take more of a hit in RT than the 4090 does) it looks as though the perf/$ for RT is going to be similar between AMD and NV so pricing seems to be based on RT performance with AMD having the advantage in raster, by the looks of it. Lets wait for actual benchmarks to see if that is accurate.

 

[GodisanAtheist]

I think the chiplet design allows AMD to be really flexible. I can see a mid-gen ~425mm2 GCD get dropped into the N31 package and just absolutely lay waste (In terms of raster, anyway).

AMD's engineers don't have to worry so much about the MCD design and focus their attention on tweaking the actual logic die.

This is the type of die an AT deep dive was meant for. SOOOO many unknowns. A chiplet design obviously had tradeoffs vs monolithic, but where and how? Its sort of pointless to speculate without a better understanding of how AMD managed to pull off this engineering feat.

 

[DeathReborn]

I think the chiplet design allows AMD to be really flexible. I can see a mid-gen ~425mm2 GCD get dropped into the N31 package and just absolutely lay waste (In terms of raster, anyway).

AMD's engineers don't have to worry so much about the MCD design and focus their attention on tweaking the actual logic die.

This is the type of die an AT deep dive was meant for. SOOOO many unknowns. A chiplet design obviously had tradeoffs vs monolithic, but where and how? Its sort of pointless to speculate without a better understanding of how AMD managed to pull off this engineering feat.

I am fairly sure that would require a new Interposer itself as there is not enough room for a bigger GCD on the one N31 uses.

 

[blckgrffn]

I think this does put downward pressure on the 3090s and 3080tis of the world, hopefully.

Probably.

We will see the 6950xt drop a bit further? Probably? But given there is a RDNA3 part coming soon with the same performance at 200w or less that doesn’t tickle my fancy much.

The real kicker for me is that I honestly don’t need more performance in any game I play at 1440p with my 6800. At least not enough to try to sell it and then drop some $600 on a 7900xtx.

My real question with regards to these is how they are going to convince current generation owners to upgrade. I don’t want to drop hundreds of dollars on a high refresh rate 32” 4K monitor anytime soon, either. If I do it will need to be oled or micro led and be awesome.

I know a least a couple of people who would like to have maximum raster performance (think CoD) and have their office be less of a hotbox. These seem like a good fit for them, I suppose.

 

[eek2121]

AMD CPU + AMD GPU <<< Show your love for AMD AND Mother Earth. The greener alternative to Intel+Nvidia.

Agreed 100%. I'm in the US, so I have access to cheap power, however I've been complaining nonstop recently about high powered parts. There is not a single country in the world (that I am aware of) that uses 100% green energy. We are fast approaching the breaking point of (drastic, non stop) climate change. Those of us reading this forum will be lucky if we don't have to live through the worst of it, but our kids will...

If we take the 6950XT number for this average and multiply by 1.54x which was the perf/w increase claim then you get 131 FPS for the 7900XTX. That puts the 4090 10% ahead in raster but it costs 60% more.

The 4080 16GB is the real competition and well at just 20% ahead of the 3090Ti in raster it does not even look like it will compete with the 7900XT let alone the XTX.

In RT the tables change and there it looks like the perf/$ for RT performance will be about the same between AMD and NV assuming the 4080 16GB has the same dropoff in performance as the 4090. If it is a bit worse then even in RT the perf/$ tilts towards AMD as well.

Agreed, that is what I stated a couple hours ago. The 4080 is in trouble, and NVIDIA will have no choice except charge much lower prices for the 4070/4070ti. AMD just butchered NVIDIA's plans, regardless of other outcomes.

I'm glad to see the leakers like MLID/Adored/RGT be knocked down several notches.

Also it was unrealistic to expect 50% perf/watt improvement and move to 5nm would result in a 50%+5nm. Actually, the process gains are included in the 50% perf/watt gain. So the total is 50%.

People also expected Ada to get a bigger improvement. One "law" in complex semiconductors is that 9 out of 10 times, it disappoints the hype.

I'm also in agreement that AMD should not have repeated the RV770 strategy and have a big chip to take the crown. Halo matters, otherwise you're trying to sell many chips with low profit. And in reality you won't even get marketshare.

You lose in three ways:
-Low volume
-Lower costs to compete with competitive product of the same class
-More money lost due to not being able to sell a halo product

A couple reliable leakers (who were not wrong with predictions about tonight) have stated that AMD plans to launch a dual chiplet version next year. We will see. If so, they may end up dominating the top of the stack after all.

The RT performance is definitely below what most expected, however to be fair, between leakers at twitter and folks here, it wasn't really unexpected. Disappointing? yes. Unexpected? no.

I am waiting for reviews. I have an EVGA 3090 which has a 450W TGP (silently downgraded from a 500W TGP in a later vBIOS due to bugs related to transient spikes, the press missed that one I guess). I would love to be able to plop in a 7900XTX and have it run quieter. My PC is barely larger than an Xbox Series X, so every watt counts.

 

[Saylick]

I think this does put downward pressure on the 3090s and 3080tis of the world, hopefully.

Probably.

We will see the 6950xt drop a bit further? Probably? But given there is a RDNA3 part coming soon with the same performance at 200w or less that doesn’t tickle my fancy much.

The real kicker for me is that I honestly don’t need more performance in any game I play at 1440p with my 6800. At least not enough to try to sell it and then drop some $600 on a 7900xtx.

My real question with regards to these is how they are going to convince current generation owners to upgrade. I don’t want to drop hundreds of dollars on a high refresh rate 32” 4K monitor anytime soon, either. If I do it will need to be oled or micro led and be awesome.

I know a least a couple of people who would like to have maximum raster performance (think CoD) and have their office be less of a hotbox. These seem like a good fit for them, I suppose.

*Meekly raises hand*

As someone who recently bought a used 6800, I totally agree with not feeling a need to upgrade for 1440p. The 6800, along with a new 5700X, will go to my younger brother who plays at 1080p anyways, so it will keep him satisfied for years to come.

Now, I also have an older brother who suffered through multiple mediocre PCs over the last two decades. Being the financially comfortable middle brother that I am, I decided to build him a 4K120 rig. His games are mostly raster based, fortunately, and he's not crazy about needing the best graphics anyways, so even with the underwhelming RT performance of RDNA 3, I will gladly take the good raster perf/$. The 7900XTX is an easy Day 1 purchase for me, and I'll pair it with the 7700X bought with the free DDR5 RAM. This rig should last him a good number of years as well. I'll just tell him never to turn on heavy RT in the settings lol.

 

[moinmoin]

I expect RDNA 4 to arrive faster. It was 16 months between RDNA 1 and 2, but 25 months between 2 and 3.

 

[insertcarehere]

That is a good point. N23 is about half N21 or less with some clock speeds making up for the or less and that is about half the performance.

So yea N33 is looking to be about 6800 at 1080p and maybe 6750XT at 4k, something like that. Still a good uplift vs 6650 but not as great as hoped but not terrible, especially if it manages that at 120W or so.

Angstronomics, who was the one behind the (looks like accurate) leak, was quoted as saying Navi 33 will be:

As an aside, Navi33 outperforms Intel’s top end Alchemist GPU while being less than half the cost to make and pulling less power.

Intel themselves wouldn't say that the A770 competes with Navi 21/AD102 so the whole 6900XT competitor thing was a misnomer based on theoretical FLOPs and not on anything else.

 

[crisium]

Disappointed in that the RT performance doesn't appear improved. Also the 7900XT makes no sense to buy unless there are stock shortages of the 7900XTX. But like Nvidia, AMD wants to continuing sell thru their old stock so they gotta price above the 6950XT at least, while in another other generation it'd be $600-$800.

I suspect this will be a rasterization winner though and should outpace the more expensive 4080.

 

[Saylick]

Don't want to be a Debbie Downer but...

https://twitter.com/x/status/1588329053530251265

 

[rommelrommel]

Don't want to be a Debbie Downer but...

https://twitter.com/x/status/1588329053530251265

If accurate that’s unfortunate.

 

[In2Photos]

After the Ryzen 7000 series release I wouldn't think AMD had any plans to leave performance on the table. Those days seem to be going away. Even the 4090 had very little room.

What concerns me about the announcement is the lack of performance numbers. Ryzen 7000 had much more comparison data. I don't feel that AMD is very confident in this release.

 

[Leeea]

Disappointing

These are good chips that amd will be able to sell at a profit in tough market conditions. 1/2 the size of Nvidia on a cheaper process sets AMD up well for economic contraction.

The design is a harbinger of things to come.

A double chip version of this design would be quite something.

 

[zinfamous]

I have AMD as a stock you want to short in the next 4-6 months.

dude, they've already dropped 2/3 of their 52 week high...you pretty much lost out on the big short play for them if you ask me, lol. Sure, it's going to dip a bit like all tech stocks have, but they've been relatively OK.

This isn't really their market anyway, and all they need to do is sell a chunk of units it looks like, and they are priced to do it.

That being said...I'm now pretty convinced that a sudden deal on 6900/6950XT is pretty much what I need to look for over the next couple of weeks to upgrade my Vega 64, lol. seems like the wise choice in my shoes, to wait for a mid-refresh or something, certainly AIB cards anyway and mature drivers, etc

 

[Thunder 57]

I think Lisa Su has been reading the forums. I posted a video of the Playstation price announcement and @Saylick nailed the price in the next post:

Page 41 - Discussion - Ada/'Lovelace'? Next gen Nvidia gaming architecture speculation

Page 41 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

It's a little bit odd seeing some people complaining about it because of RT or whatever reason they want to come up with. They did (hopefully) what they had to do and priced it much better than NVIDIA. I'm waiting for NVIDA to "unlauch" or "readjust" the price of the 4080 now.

 

[Hans Gruber]

dude, they've already dropped 2/3 of their 52 week high...you pretty much lost out on the big short play for them if you ask me, lol. Sure, it's going to dip a bit like all tech stocks have, but they've been relatively OK.

This isn't really their market anyway, and all they need to do is sell a chunk of units it looks like, and they are priced to do it.

That being said...I'm now pretty convinced that a sudden deal on 6900/6950XT is pretty much what I need to look for over the next couple of weeks to upgrade my Vega 64, lol. seems like the wise choice in my shoes, to wait for a mid-refresh or something, certainly AIB cards anyway and mature drivers, etc

I am still planning on getting a new graphics card. If the price is right on the AMD GPU's, that is what I will get in the next 3 or 4 months. The sky is still falling in the GPU market.

 

[gdansk]

I don't think Nvidia will adjust its price officially. Because I don't think they expected it to sell well at that price. It was always about framing the 4090 as a 'good deal'.

 

[jpiniero]

They did (hopefully) what they had to do and priced it much better than NVIDIA. I'm waiting for NVIDA to "unlauch" or "readjust" the price of the 4080 now.

The 7900 XTX is likely not fast enough in raster for them to care. Maybe the 7900 XT will give nVidia a reason to cut the price of the 4070 Ti versus the 4080 12 GB though.

 

[rommelrommel]

The 7900 XTX is likely not fast enough in raster for them to care. Maybe the 7900 XT will give nVidia a reason to cut the price of the 4070 Ti versus the 4080 12 GB though.

?

The 4080 is only 60% of a 4090 by cuda cores. It sounds like the 7900XT and XTX will beat a 4080 rather handily in raster.

 

[Karnak]

But they'll lose by a pretty big amount in RT. There's a reason why N31 is priced at $999 and below and not like $1099 although the 4080 sits at $1199.

 

[Carfax83]

Most Nvidia buyers (especially the younger ones) simply pay. It's Nvidia! That's enough reason for them. I don't expect any reactive pricing change from Nvidia.

If you have a 4K gaming monitor and want to use RT along with high to maxed out settings then Nvidia is the only game in town at the moment.

Unfortunately, it doesn't seem RDNA3 will change anything in that regard.

AMD has consistently failed to match up with Nvidia on the high end for many GPU cycles now which just further amplifies the mindshare that Nvidia has superior hardware. RDNA2's rasterization performance is good at 1080p and 1440p, but then suffers at 4K presumably due to less efficient memory compression, less bandwidth and less shading power. Add DLSS to that mix and AMD falls behind even more, and with ray tracing, it's game over.

I just bought at Samsung Neo G8 4K 240Hz HDR monitor. I'm basically locked into Nvidia products if I want to enjoy the monitor and play games the way I want to.

Believe me, I would love to buy the much cheaper 7900XTX GPU but what joy would it bring when my games failed to perform the way I expect them to?

This is why we need a 3rd player in the GPU market. Intel has great potential and even at this early stage, XeSS is better than FSR 2.0 and their Arc GPUs have comparatively good ray tracing performance. Once they achieve node supremacy again, I think Intel will have a good shot of robbing Nvidia of the high end performance crown provided they stick with it.