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]

Also, on the eve of what might be a leak with clockspeeds and performance targets, does anyone want to make their final guesses? This pump that dumps fuel into the hype train's engine ain't going to prime itself!

Here, I'll start. My guess for full N31 specs:
- Up to 60% perf/W uplift over N21
- 1.8x - 2x raster uplift
- 2.2x to 2.5x RT uplift
- 3.1 GHz game clock
- 375W TDP
- 24 GB GDDR6 RAM

 

[maddie]

Also, on the eve of what might be a leak with clockspeeds and performance targets, does anyone want to make their final guesses? This pump that dumps fuel into the hype train's engine ain't going to prime itself!

Here, I'll start. My guess for full N31 specs:
- Up to 60% perf/W uplift over N21
- 1.8x - 2x raster uplift
- 2.2x to 2.5x RT uplift
- 3.1 GHz game clock
- 375W TDP
- 24 GB GDDR6 RAM

I see >375W & >2X raster. AMD seems no longer reluctant to pushing power limits.

 

[GodisanAtheist]

I fully expect N31 to be competitive, maybe even beat, the 4090 in terms of Raster, and I expect the gap to be closed with RT (albeit the 4090 will still pull ahead) when it comes to RT performance loss.

 

[Timorous]

I expect a 420W TBP and with a 50% perf/watt gain that is 2.1x the 6900XT so about 10-20% faster than a 4090 at 4K.

Same sums were used when estimating N21 performance from 5700XT and that was bang on the money for the 4K advantage the 6900XT had over the 5700XT and I don't think AMD will change how they have done their perf/watt for the last few gens.

 

[exquisitechar]

I said before that AMD probably won’t beat the 4090ti in raster or RT, and it still seems that way now. Given the chiplet design, it seems that they were conservative with Navi31 and they could have built a significantly faster GPU. It competing with AD102 at all is an achievement, and it puts them in a good position, cost-wise. As for RT performance, I would be disappointed if they can’t beat the 4080 16GB. Aside from the massive increase in compute contributing to that, I hope they made specific RT related improvements.

 

[jpiniero]

I said before that AMD probably won’t beat the 4090ti in raster or RT

I am curious about the performance impact of the vcache model. It's possible that might be enough.

 

[Timorous]

I am curious about the performance impact of the vcache model. It's possible that might be enough.

I don't think it will impact at 4K but it might at 8K and it does look like the 4090 is an 8K capable card if you want 8K60.

 

[moinmoin]

With 4090 being bottlenecked by all current CPUs I wonder whether AMD tries to avoid such cases, aligning GPU performance with their plans for CPU performance. That's the only reason I can see for AMD restraining the performance of this round's marquee GPU.

 

[Timorous]

With 4090 being bottlenecked by all current CPUs I wonder whether AMD tries to avoid such cases, aligning GPU performance with their plans for CPU performance. That's the only reason I can see for AMD restraining the performance of this round's marquee GPU.

Launch it with Zen 4 3D.

I can also see AMD doing better in DX12 titles at 1080p / 1440p due to the NV driver overhead that really helped them in DX11 titles but is now hindering them somewhat.

 

[moinmoin]

Launch it with Zen 4 3D.

Yeah, that's what I'm hoping they'd do.

 

[RnR_au]

So 7800X3D available on 22nd Nov! You heard it here first on Anandtech forums!

 

[moinmoin]

So 7800X3D available on 22nd Nov! You heard it here first on Anandtech forums!

Haha, no, the V-Cache variants of Zen 4 seem to be in line to launch around March next year. AMD could launch the V-Cache variants of RDNA3 then as well.

 

[Timorous]

Haha, no, the V-Cache variants of Zen 4 seem to be in line to launch around March next year. AMD could launch the V-Cache variants of RDNA3 then as well.

If we take AMD at their word on stage they said this year for Zen 4 3D parts.

 

[exquisitechar]

If we take AMD at their word on stage they said this year for Zen 4 3D parts.

I think that was an unintentionally misleading quote. Won't N5 stacking only be ready by the end of this year according to TSMC? Like moinmoin said, I wouldn't expect it before March. Possibly later.

 

[Karnak]

Where's today's leak now? Need to know that within the next hour.

 

[RnR_au]

I think that was an unintentionally misleading quote. Won't N5 stacking only be ready by the end of this year according to TSMC?

I had a quick google. Seems N5 on N5 is delayed till 2023. Interestingly there is no mention of N7/N6 family on N5, but there is a mention of N5 on N3... if I'm reading the table right...



Snagged from https://semiwiki.com/semiconductor-...posium-review-advanced-packaging-development/

So yeah seems 7800X3D is definitely 2023. Maybe even late

 

[Kaluan]

If we take AMD at their word on stage they said this year for Zen 4 3D parts.

Maybe that just means "Genoa-X will be sampling to our closest partners and customers", a-la how they did with Milan-X and Azure Cloud or wtv in Q3-Q4 2021.

Either way, stacked RDNA3 doesn't seem to be constrained by any of those considerations since it's stacked dies are not on 5nm-class nodes.

 

[TESKATLIPOKA]

It's possible we won't see any new info about RDNA3 today.

 

[igor_kavinski]

So yeah seems 7800X3D is definitely 2023. Maybe even late

No way. It should be widely available by late January. AMD can't afford to lose mindshare to Intel. I would hate it if I see the 13900KS being paraded everywhere for being the first 6 GHz gaming CPU despite 200+ Watt power consumption in games.

 

[KompuKare]

It's possible we won't see any new info about RDNA3 today.
View attachment 68994

Well the whales have now already bought 4090's so not revealing anything implies they won't compete with the $1,500-$2,000 cards.

 

[naad]

4090 seems to have a bit of issue utilizitng its hardware at under 1440p, probably half due to CPU bottleneck and half to due low shader utilization, probably a good niche for AMD to carve out for themselves, 3D Raphael might actually push high end GPU sales

 

[moinmoin]

If we take AMD at their word on stage they said this year for Zen 4 3D parts.

One of the Twitter leakers (greymon? don't remember) was linked/quoted in another thread claiming last month that mass production just started, which was one month "earlier" than 5800X3D, implying that the launch may be one month earlier as well (which would be in March).

 

[eek2121]

Hoping for a slimmer GPU TBH. 95% of the 4090 performance with significantly lower power draw and a slimmer card would make me buy one. Until then I will stick with my 3090.

 

[igor_kavinski]

I can also see AMD doing better in DX12 titles at 1080p / 1440p due to the NV driver overhead that really helped them in DX11 titles but is now hindering them somewhat.

What if it's just an artificial and intentional attempt at sandbagging? Once AMD releases their halo part, Nvidia could just flip a toggle in a new driver release and voila! Beast unleashed!

 

[Stuka87]

I said before that AMD probably won’t beat the 4090ti in raster or RT, and it still seems that way now.

What makes you think an unannounced AMD product won't beat an unannounced nVidia product?