I'm also curious if these also affect the CPUs cache, as Cerny's presentation only block highlighted GPU caches.
Well like I said before
@3dilettante would probably be one of the best people to ask regarding it, they seem to understand a lot about the technical workings on that type of stuff. I remember them bringing up something regarding line flushes, and that in some instances you'd be better off just clearing the cache rather than doing a pinpoint eviction, but I don't recall a lot of specifics mentioned.
Cache is meant to be transparent in hardware, and 'full RDNA2' is just a marketing term. XSX is more a fulfilment of DX12 Ultimate.
I'm not convinced XSX frontend is RDNA2 specification. You can check the Rasteriser Units in the Hotchips block diagram - there are 4 and each Raster Unit is at the Shader Array level. Compare to Navi21, the Raster Units are at the Shader Engine level.
Guess you are referring to the leaks posted by that person on Twitter? I saw those too, they do have a lot worth asking. But I also saw someone else drawing up a comparison between RDNA 1 and RDNA 2 on the frontend, and a distinction between RDNA 1 and RDNA 1.1. That leak could've been pertaining to 1.1, because in the things they listed in frontend comparisons with RDNA 2 virtually all of it is the same.
If that's what that particular leak pertains to then there's not too much difference between RDNA 1.1 and RDNA 2, at least from what I saw. Wish I could find the image that showed what I was talking about, but it could explain the delienation of RDNA 1 and RDNA 2 in that leak, the RDNA 1 could've been referring to 1.1 but not reflect it in what was provided by that leaker.
Also, check Navi21 Lite in the driver leaks (aka XSX GPU), and its driver entries show no change to SIMD waves (CUs) from RDNA1 and Scan Converters/ Packer arrangements from RDNA1 as well.
That might be the case, but again, it could be RDNA 1.1, not RDNA 1. 1.1 made a few notable changes over 1.0 and shares much more design-wise with RDNA 2 than 1.0 does. Seeing how MS got rolling with production (and likely designing) their systems after Sony, I find it a bit hard to believe they have much if any 1.0 frontend features in their systems. 1.1 though? Yeah, that is more possible; but even there it'd be essentially the same to 2.0 going off some of the stuff I caught a look at (hopefully I can find the comparison listing again).
FYI, PowerVR existed on PC before Dreamcast. Smartphones also use/ used TBDR. And a handheld console, PS Vita also used TBDR and a PowerVR GPU.
Interestingly, Mark Cerny was involved with the Vita, and there are hints that PS5 has analogies. I haven't found anything concrete yet of a TBDR architecture.
Isn't there a patent Mark Cerny filed which covers an extension of foveated rendering with a range of resolution scaling among blocks of pixels, basically what would be their implementation of VRS?
And actually while at it, could that possibly just tie into whatever other feature implementation analogous to TBDR Sony happen to use with PS5? I mean at least to what it seems like to me, techniques like VRS and foveated rendering are basically evolutions of TBDR anyway (or at least are rooted in its concepts and adapts them in different ways). Maybe I'm wrong tho
Perhaps a GDC presentation or a dev slide deck will mention what data flows or processes are targeted by the cache scrubbers. Using virtual memory allocation to buffer new SSD data could avoid many flushes, and oversubscribing memory is part of the process for the Series X's SFS (allocates a wide virtual memory range, which demand traffic then dynamically populates).
I hope so; Sony seem more willing to open up discussing their hardware at GDC presentations these days, which beats having their E3 press conferences drowned out with all of that for the first hour (here's hoping to E3 2021 being a thing).
Also it'd seem like even though Series devices don't have cache-scrubbers, if they can utilize virtual memory as a buffer and track the data coming in that way through to memory via SFS and the other parts of XvA, then that should work just as well. It also fits better with the way MS wants to implement their take on restructured I/O data management, a more hardware-agnostic approach that, nonetheless, is still just as suitable, long as the underlying hardware isn't completely out-of-step or deficient (and that's something only people on very low-spec, outdated PC systems or the such are in jeopardy of facing).
Don't know about SRAM to make an estimate, but i'd guess it's a tiny amount or maybe acting as Sonys take at ic?
Been thinking the idea from them (I might have interpreted this wrong) is that Sony's I/O setup as a whole is designed in a way where they are applying concepts analogous to the type of features regarding data management IC does, but applied at the hardware level in a different manner than having a large block of 128 MB L2$ on the GPU.
So in theory, they're doing similar things (improving effective system memory bandwidth flow), but do it differently. Since a PC GPU is "just" a PC GPU, it still has to account for other parts of the system design outside of the scope of that GPU card, outside of its control. A console like PS5 doesn't have that as a factor; every part of the system can be designed explicitly around one another.
Similar goals in improving rate of memory bandwidth, similar concepts, but different means of applying it at the hardware level. That's the main reason Sony don't need IC in the way AMD's RDNA2 GPUs need it. The same can be said with Microsoft's systems; the designs as a whole focus on (among other things) improving the flow of memory through the system and ensuring better bandwidth, that's what the hardware/software features of XvA are built around.
Only big difference between them is MS wants to deploy their implementation onto PC and big server markets as well, so it has to account for different hardware specifications, variations, and scalability. But that's why MS are working with companies like AMD and Nvidia to ensure there's standardization. Once DirectStorage is deployed sometime in 2021, they'll likely be expanding that to motherboard manufacturers as well.
Occasionally I get half of thirds or conversations and I can tell from one side of the conversation it's nuts. It would be neatr if there was also an option to hide posts of people replying to people I've ignored.
