Excuse me? That most certainly can't be just the source code. There's got to be textures and audio in there too, possibly in an uncompressed format.

So I looked a bit more into Smart Access Memory as that is the most interesting part outside of the performance numbers. Now this is just me looking into things so I could be totally wrong and as these are just theories, take them with a grain of salt.

So firstly, what Smart Access Memory is:

"On all systems (Intel and AMD), the processor has direct access to only part of the VRAM on the graphics card. The maximum value of this area is just below 256 MB. If the processor wants to access the entire graphics memory directly, this is done over 256 MB blocks. Access to this part of memory is fragmented and slow, as it goes through the PCIe bus and also RAM is slower than VRAM."

"However, the size of the direct access area can be changed: In the motherboard BIOS and the graphics card BIOS. AMD apparently did this and will allow the processor to access the VRAM in almost the entire 16 GB capacity. That requires another thing – PCIe 4.0 as the fastest PCIe today. Therefore, there is a limitation on B550 / X570 chipsets where full PCIe 4.0 functionality is guaranteed. This idea is not new, Intel has done the same with the Phi accelerator. It also needed direct CPU access to the entire VRAM for its functionality."

https://tekdeeps.com/amd-smart-access-memory-what-it-is-and-how-it-works/

Secondly on whether or not it could work outside of 5000/X570/6000 combo

"According to Alex Deucher (AMDGPU maintainer):

Smart Access Technology works just fine on Linux. It is resizeable BAR support which Linux has supported for years (AMD actually added support for this), but which is relatively new on windows. You just need a platform with enough MMIO space. On older systems this is enabled via sbios options with names like ">4GB MMIO".

I don't know what windows does exactly, but on Linux at least, it will work on any platform with enough MMIO space. I suspect windows would behave the same way (although I think windows has stricter requirements about BAR resizing compared to Linux so you may need a sbios update for your platform to make windows happy)."

https://www.reddit.com/r/Amd/comments/jk76u9/smart_access_memory_already_works_on_linux_and_is/

Now again, take that with some salt but the more I look into it, the more it sounds like something that can be implemented on Ryzen 3000/Intel/Nvidia devices. One of the big requirements certainly could be PCI-E 4.0 as it really needs that bandwidth to actually be useful. With that being said, it does sound like it's something that might need to be implemented by the developer to have a noticeable impact so having one company do the CPU/Platform/GPU is better than Intel/Nvidias current situation.

^Well, Intel could try it with its GPUs. Nvidia will need more time to try it with ARM.

There's that, and maybe all the documentation associated with the development.

Mercury Research came and published new market share numbers for AMD: https://www.tomshardware.com/news/amd-vs-intel-q3-2020-cpu-market-share-report

The Results are as follows:

• Desktop Market share climbs over 20% for the first time since 2013 with 20.1%
• A new record market share in mobile with 20.2% (which is certainly going to continue to rise over the next months, considering Tiger Lake is again only 4c/8t and Renoir's successor will come out before Alder Lake.
• x86 Client share is also on 20.2%, so all three segments are very close in market share. It's also their highest share in that segment since mid-2011.
• Curiously, overall x86 market share is over those numbers, at 22.4%. That's a whooping 4.1% increase QoQ! It's also a new high since 2007, so we're getting very close to reaching the highest market share AMD ever had, period.

Who would have expected this rebound in 2016?

Considering how terrible the game runs it might actually be just the code.

The studio is paid by Nvidia to make us buy more 3090s. I almost did after reading that article.

Kopite7kimi: NVIDIA GeForce RTX 3080 Ti rumored to feature 10496 CUDA cores and 20GB GDDR6X memory
https://videocardz.com/newz/kopite7kimi-nvidia-geforce-rtx-3080-ti-rumored-to-feature-10496-cuda-cores-and-20gb-gddr6x-memory
Just last a week ago, Kopite reported on the plans to revive the GeForce RTX 3080 Ti SKU. This was quite surprising, even to us, as the SKU has been canceled for at least two months on all the roadmaps that we have seen. Plus, with the arrival of RTX 3090, there was very little reason to launch such an SKU.

However, the situation has changed and the competitor has showcased its Radeon RX 6000 series. AMD even started using NVIDIA high-end GPUs in its comparisons charts. That certainly did not go unnoticed by NVIDIA, who is now expected to release a new SKU – GeForce RTX 3080 Ti  – to counter the AMD Big Navi series.

Before we go into detail, we must say that it was not possible for us to confirm the previous specs that Kopite had reported. Either his information is very fresh or NVIDIA is just fabricating the specs in order to find the source of their leaks. Due to the high credibility of the Kopite’s leaks in the past, we believe that they are worth sharing regardless of us being able to confirm them on time.

According to a new tweet from Kopite, the RTX 3080 Ti would feature the same CUDA core count as RTX 3090, which is 10496 cores. This is quite a surprising move from NVIDIA, which confirms that the Ti model will certainly not be cheap. Another interesting change to the specs that we saw last week is a new memory configuration. The revival of 20GB GDDR6X SKU appears to be on a table, as the RTX 3080 Ti is now rumored to have such memory configuration.

Kopite provides very detailed specs, including board and SKU numbers (PG133 SKU 15) with the GPU codename of GA102-250. This is something that we might later be able to track and it adds more credibility to this leak. Yet, it is worth noting that the previous rumors from Kopite, also included this time of data and suddenly they have become ‘canceled’. Sadly, this has allegedly already happened to the RTX 3070 Ti, as Kopite himself reported a few days ago.

We know that the 3090, with the same core count, is only 10-15% faster than a 3080. Would that be enough to compete with the 6900XT? Will they make it using TSMC's 7nm to get a more efficient card with clocks that can be pushed further to increase its performance? Why did we get an article about the rumored 3070Ti but not another one with its cancellation? Will our Hero be able to arrive in time to save the... sorry, I got carried away.

At this point I no longer know if it's as the article says and Nvidia is playing with us, trying to find the source of the leakers, or if they are really in complete panic mode and don't know what to do.

It would make some sense to launch a card that competes with 6900XT on both price and performance. Ti models were likely in the works all along, but their exact specifications probably hinged somewhat on what AMD brought to the table. A 3080Ti with 20GB memory could go for 1100-1200$ or so, or what do you think? If it could kick AMD's ass on RT, that alone would be a decent argument for enthusiasts, and when talking 1000$+ GPU's, we're most certainly in enthusiast territory. Still, it would put the 3090 in awkward territory.

If they released a 3080Ti at about 1100$ (1200$ for aftermarket editions) with good RT performance and an overall boost to performance similar to past Ti iterations, I'd gladly get one in my build next year. I think I paid around 1000$ for 980Ti back in the day, and I never regretted it.

So basically I'm fucked no matter if I buy a 3080 or 3090.