Exactly. As far as I understand Radeon support hasn’t been stellar to this point, right?
Please correct me if I’m wrong… 
I’d rather spend a little extra from the get go and ride first class than skimp out and ride “economy” (that’s not worth it to me)… 
Do you have the source that shows AMD doesn’t support 4x4 and all the others that are specificed by DX12?
Your link just shows the marketing slide with some example configurations in Dirt 4. And then has another link to the Direct3D spec, showing all the combinations again, including 4x4.
As I understand it, Dirt 4 uses only the 2x1 and 2x2 combinations, because those yield a large performance benefit, without reducing image quality too much. 4x4 is very low resolution and will be noticeable on most scenes, unless you’re already dealing with very low res textures to begin with.
No, I’m assuming that if AMD supported the “by 4” variants they would mention it. If you take a closer look at the DX12 doc (which AMD linked to), you will see that the “by 4” variants are marked “CAP”, which means that they are optionally supported.
I’ve been trying to get an answer on that question for many months now (since it has been obvious that AMD will compete with Nvidia’s next gen flagship).
https://community.openmr.ai/t/latest-source-code-of-dynamic-foveated-rendering/29414/9
So far, I haven’t gotten an answer from Pimax.
It’s a shame, because I’m sure other HMD manufacturers have most likely been working on their AMD implementation for a while already and Pimax is not being very forward looking with their approaches. Instead, we get unqualified comments from Sweviver in the sense of “I don’t understand why people care about AMD GPUs, just get an Nvidia GPU instead”.
That is actually hugely relevant!
If those are the peak framerates those graphs are entirely useless.
And let’s not forget here, this is what AMD has accomplished using just 72 CU and 80 CU (new RDNA2 Compute Units) 4608 and 5120 respectively on RX 6800XT and RX 6900XT to just reach similar performance to the competition !.. Nvidia had to use twice as much of their compute units variant, this speaks volumes about lesser efficiency and wasted transistors !.. because today, just adding more compute units or shaders, even if doubling, don’t really add as much performance, technology needs to be deeply optimized today, and will soon employ really new architectures and materials, still unknown to the masses of the consumers, but ready to be commercialized, like the already available in sampling form magneto-resistive memory modules from Samsung.
AMD certainly did a smart move using a big chunk of ultra fast and low latency cache to the chip, just like they have done on the Ryzens, that sport some of the hugest cache quantity in the entire CPU market, something that has been done until now only on super expensive CPU chips like the IBM Power used on supercomputers, even that they are selling such expensive and complex chips to normal consumers, at popular prices…
Going full AMD ? Sure…why not, it is the smart thing to do if you’re an enthusiast and really understand true innovation, Intel has already lost on all fronts after all… 
It really isn’t that black&white but I do love it when an underdog performs, which AMD certainly has in the last 24 months.
Nvidia did not double compute units for comparable performance, I’m sure you know how they came to that marketing term simpy due to the difference between INT32 and FP32 workloads.
AMD has the advantage of the better 7nm process, I absolutely expect nVidia to go the same route for the GA102 or GA103 once they have the right capacities with TSMC. This was a very good move by AMD, helped by the volume they contracted with Sony and MS.
For VR specifically we will have to wait how third party reviews stack the competing cards; DLSS or something similar together with FFR and DFR via VRSS are of utmost importance to enable meaningful fidelity jumps in the future and nVidia have a considerable, objective advantage over literally any other competitor in this space.
For me the current (great!) position of AMD fully hinges on nVidia’s inability to manufatcure not just the A100 but also the GA102 with TSMC in 7nm. Had they gone this route rather than Samsung’s 10nm+ (“8nm”) they would have had the advantage of higher clocks, less TDP, less heat and incredibly fast GDDR6X resulting in pretty much the same spread as we had seen between competing Turing and Navi cards. They lost that gamble, rushed to push a 7nm arch into 10nm at Samsung, turned up the TDP and this is the result.
For us as consumers this is absolutely fantastic, as it will force competition on a price level.
Yup, This is what id like to know about as well. Im building a new AMD system now with the 5900X and perhaps ther 6900XT and the only thing keeping me from deciding is the lack of information or development for AMD from Pimax. Ive been fine without FFR and such on my 1080 TI but with eye tracking coming along is it only going to be possible with a NVIDIA card?
While talking about the compute units, according to the info from the presentation, the only difference (I am curious if there are others, just not mentioned there) between 6800XT and 6900XT are 8 compute units (6800XT with 72, 6900XT with 80). This alone should indicate that the performance increase from the one to the other will be rather insignificant (<10%).
I guess that is why 6900XT has to run with rage mode and special memory access to get better results, but I do not think the price/perf ratio makes sense for 6900XT (well it does not for 3090 either).
I did not realize that, but this makes sense. AMD could have very good leverage at TSMC thanks to their volumes for consoles. Kind of Apple-like status.
Yeah, not quite Apple status as literally no one but Apple has that but AMD have their entire CPU as well as GPU range at TSMC as well as all custom SOCs for consoles. That does give them priority over nVidia.
Just to reply myself there is a summary about the differences directly from AMD:
https://www.amd.com/en/products/specifications/compare/graphics/10516%2C10521%2C10526
While talking about the compute units, according to the info from the presentation, the only difference (I am curious if there are others, just not mentioned there) between 6800XT and 6900XT are 8 compute units (6800XT with 72, 6900XT with 80). This alone should indicate that the performance increase from the one to the other will be rather insignificant (<10%).
I guess that is why 6900XT has to run with rage mode and special memory access to get better results, but I do not think the price/perf ratio makes sense for 6900XT (well it does not for 3090 either).
Correct. This is a typical from AMD, to have their cut-down versions only cut down slightly while keeping most of their die fully enabled, while Nvidia usually cuts down various parts of the die.
The 6800XT (and 6800 for the most part) get the full 120 ROPs, the full 256 bit memory bus (and as a result the full 16GB of memory) and the full 128MB of ultra fast SRAM.
This should result in relatively close performance to the 6900XT, especially if the 6800XT can also be clocked a bit higher thanks to more thermal headroom, then the 6900XT.
I think the 6800XT will be the winner here, outperforming the RTX 3080, while offering more memory, better power efficiency, etc. all while costing less at the same time.
The ultimate flagship will still be RTX 3090, by a small margin, which is the successor to the RTX Titan. It’s not really a gaming card. The real “gaming” flagship is the 6800XT.
Just remember this chart is showing rage mode enabled on the 6900. That is basically just overclocked vs a stock 3090. The 3090 series has more headroom for overclocking and DLSS. Both cards are absolute beasts though and the price is great for the AMD.
I remember seeing that will be 6900 XTX probably down the line. Not sure what will the changes be if it turns out to be accurate
6900XT is already the fully unlocked die. All they can do is increase clock speeds, but that’s unlikely without optimizations from the foundry.
The RTX 3090 is not fully unlocked yet, so there is still the potential for a Titan branding card, for another 5%-10% or so in performance, but it’s unlikely to happen, because the 3090 already has 24GB memory.
Are You sure You’re not confused by “Navi 21 XTX” (GPU codename for the GPU on 6900 XT)?
Try a Google Search on “Radeon 6900 XTX”…
I saw that. I think i also saw several videos on it, but may be also speculation videos. But who knows. I am no expert on AMD cards.
Neither am I and You might be right. Maybe someone just thought Navi 21 XTX = 6900 XTX?
That said, if they need to, maybe they’ll push it a little bit further?
There seems indications that those AMDs have more overclocking ability. December will be an interesting month.
Sure will…!