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

If we talk about bandwitdh then yeah. But amount, no, somes games uses more than 8gb, even under 4k. Not just allocating...

And I guess in the end it's easier for amd to have one memory config to build ? And of course a nice 16gb sticker vs nVidia products

 

https://www.anandtech.com/show/1620...a2-starts-at-the-highend-coming-november-18th

anand has the preview up

 

VideocardZ also ran the benchmark and got a similar result.
https://videocardz.com/newz/amd-ray...dia-rt-core-in-this-dxr-ray-tracing-benchmark

 

General architecture overview page:

https://www.amd.com/en/technologies/rdna-2

I can't find a whitepaper link.

 

Very good. If we see desktop 6700 and 6600 in Q1/Q2 '21, there will likely be refreshed laptop designs later on in 2021 with Zen3 based mobile processors and RDNA2 mobile GPUs.
Back to school, aka early Q3 best case. Late Q3/early Q4 worst case.

 

CapFrameX's 2080Ti got 470 FPS, maybe others changed the window size from default 1280*720, idk. Usually 2080Ti is twice as fast as 2060, so 470FPS seems to be correct compared to my 240FPS

 

L1 is part of the Render Backends
edit: or next to them, but not visualized

 

The L0 caches would be what I think you're talking about.

 

Isn't L1 paired to the CUs with RDNA1?
Page 17
https://www.amd.com/system/files/documents/rdna-whitepaper.pdf

Edit- Guess it is part of the shader array-
In contrast, the RDNA graphics L1 centralizes all caching functions within each shader array.

 

Ye at first, but I corrected myself with edit there. L1 is at / next to Render backends, L0 is at WGP

 

Yes that makes a lot more sense. I figured it would be a case of different settings but I'm not familiar with the benchmark.

Thats also exactly in line with the previous leak. So 2080Ti level performance certainly isn't too shabby. As has been mentioned before, the synergy with the consoles may close a lot of that gap once games start to optimize RT for that architecture.

 

I never believed in the infinity cache, have to give to to AMD and that british youtuber who was so adamant that it exists. I wonder how this stuff affects potential GPGPU applications, but I guess there's CDNA for that so AMD won't have to do both at the same time in the same chip anymore. Hopefully it's more BW-constrained in pure compute tasks so the miners won't buy out all the cards (or at least there'll be some lag between mining algos running well on RDNA2 like it was with Vega which was more or less available at msrp till late october 2018)

 

I didn't believe in it either, I kinda still don't. It just doesn't scale well in terms of resolution. What happens when dedicated next gen games come out, does the cache get thrashed to hell at 4k?

Anyway, I'm not worried about gpu performance. They have cdna, which is their vega derived arch, coming out this year too.

 

I do expect there to be some edge cases where the infinity cache falls apart as well. I suspect it'll work great most of the time, but there will be some situations where you can't fake your way out of having the real memory bandwidth. Heavy simultaneous RT + rasterization might be one of those cases. It'll be interesting to see how much if any manual tuning AMD will need to do on the driver side to avoid thrashing the cache in heavy RT games with large working sets.

 

Real question is whether the 6900xt can beat the 3080 in ray tracing. I have a feeling it can’t. Means for professional applications that deal with RT the $1000 price tag may not be a bargain.

 

Why would it? You aren't storing raw textures in there, you are storing data that might be needed again from the other caches/units and supposedly RT stuff.
If the 60-70% utilization out of the box is a realistic number and one of the consoles has something similar, we may see more devs than just the AMD partnered ones optimize for that other 20-30%.

 

Ou

It’s questionable if Ampere’s increased RT performance brings any value to actual games. The bottlenecks seem to be elsewhere.

 

All of the odds.
Infinity Cache, as originally mentioned by RGT when they leaked it for the first time, was developed as a way to significantly increase performance on mobile GPUs by decreasing VRAM channels (hence easier to scale down), and most of all to be able to increase gaming performance on the PC APUs that are stuck on standard DDRx or LPDDRx.

 

https://www.anandtech.com/show/1620...-starts-at-the-highend-coming-november-18th/2

 

Interesting. I can only hope for RDNA2 Radeon Pro in laptops. (then again knowing Dell and Lelnovo they'll stick to Nvidia for their Precisions/Pseries)