AMD Radeon RDNA2 Navi (RX 6500, 6600, 6700, 6800, 6900 XT)

Instead, they chose the much more profitable option.

 

Yeah.
Better luck next year with NV31 then!

 

Radeon 6900XT Liquid Cooled is official.

Looks like 18Gbps GDDR6 modules exist after all:

https://wccftech.com/amd-unleashes-...id-cooled-graphics-card-fastest-big-navi-yet/
https://maingear.com/6900xtlc/

 

Too bad SI-only or hunt for bulk SKUs.

If those are genuine 18Gbps parts then woohoo, only took 3.5 years since the announcement.

 

6800 would've been better for efficiency not so much for absolute performance.

RDNA3 will clock even higher than 3GHz or IPC improvement?

 

Absolute performance too, given it was a 150W part by itself.

Both.
Though the N6 parts may be a bit below that freq-wise.

 

RDNA 3 will also be bundled with free jokes. For maximum funny.

High clocks and big caches go hand in hand so it would make sense. It's also going to be harder to keep wide and slow architectures busy given the regression in target resolutions. Upscaling tech will be all the rage for this generation.

 

This was where my head was going; glad I'm not the only one thinking it. Here's hoping FXR (or whatever) is sufficiently competitive with DLSS.

 

I really doubt that we'll see any kind of resolution regression on PC in the future. 1080p is the most popular resolution on PC even today.

 

The point is, until cards can actaully run RT at these enormous resolutions at 60+ FPS consistently, most users will be running at reduced resolutions with something like DLSS or FXR on top to upscale. 1080p is certainly a significant resolution regression (one quarter) compared to 4k.

 

I guess his point was not a lot of people play in 4k.

 

4K isn't a typical resolution for PC gaming right now, 1080p is. Thus it's not that we'll see a regression of resolution but more of a prolongation of 1080p-1440p range into the future thanks to reconstruction techniques. So a "wide and slow" GPU will work about the same as it does now, probably a bit better due to games which will target native 4K anyway (4K displays should become more widespread on PC over the next years) or use more complex shading.

 

That's true. There's a huge gulf right now between what reviewers are doing and how PC gamers actually play games. So I should rephrase my earlier comment to say there will be a regression in the "internal" resolution that games are benchmarked at.

 

Which brings up an interesting possibility for the RDNA2 refresh. Theoretically 2.64ghz "boost" clock and 20gbps would be doable for a "6950xt", boost the cache clock to 2.6ghz etc.

But around a 20-25% boost would put it right on par with a 3080ti, in raytracing games like Metro EE. Now the caveat is that's only at 1440p, but if you're going to be benchmarking "FSR Ultra" which might run around that resolution/a bit higher, well then you have AMD's high end matching Nvidia's high end in raytracing titles at "4k", at least in some RT titles, though obviously Metro EE is the gold standard right now.

One would expect lower clocks all around from say a "6850 xt" so that wouldn't quite meet a 3080 in Metro EE. But who knows, maybe AMD will try for as high as possible anyway, they seem to be within striking distance unless Nvidia puts out some sort of refresh this year that somehow improves things markedly. And considering the powerdraw they already have it doesn't seem like they have the same headroom AMD has for this.

 

RDNA2 would have to be clocked at around 5GHz and have 30Gbps G6 probably to match Nvidia's performance in RT.

 

Have Cerny take a look

 

No such thing.

 

I'm still not convinced these clocks are 'real' (as in the card actually 100% utilizes it and benefits from it), it just seems that the GPU goes sky high when there is some sort of bottleneck somewhere else apart from CUs themselves. What I mean, XTXH cards are not locked at silly 2150 mhz memclock, but they are not leaps and bounds over the regular 6800xt/6900xt s in 4K even if they are extremely overclocked and fine-tuned both in terms of core clock and mem clock. I wish I could get one at not eye-gouging price, it would certainly be interesting to test, especially with water cooling.

 

They are, I've had me boys torture whole two 6800XTs with +15% PLs and not.
N21 does indeed OC well by modern standards.

Done.
Dude runs 2.6 6800XT waterblocked, draws ~335W.

Memclk does nothing for N21.
At all.

 

So you mean we are basically bound by IC clock or maybe the ROPs? It'd be nice to be able to tweak cache clocks if it is possible.

In games - most probably, but it could be of some use in specific memory bound benches (like FS GT2, Timespy gt2, some gpgpu stuff)