Thats the point even going by the GCN specs we can speculate what will be the performance limitation for NAVI, but still a huge part of the development for Navi is still unknown the only leak I can find on the internet regarding the development is that AMD was looking to improve Geometry engine, which has been problematic area for GCN.Â
Indeed. The Geometry Engines have been a bit of sticking point for AMD... The irony is, they beat nVidia to the Tessellation game by 9-10 years and yet nVidia's Polymorph engines have consistently beaten AMD to the punch.
In saying that, Graphics Core Next has some serious limitations, so I doubt AMD will go past 4 Geometry Engines, but will likely increase each individual Geometry Engines capabilities, which AMD has been doing consistently with almost every Graphics Core Next update.
And I don't remember the article or the year but it was i think 2018 interview where AMD CEO Lisa Su revealed that more than half of the staff from RTG were redirected from Vega on to improving Navi and that it was particularly being designed for Sony, It could be a possibility that Sony might have jumped in mid Navi development and suggested some changes of their own (IBM,Cell) , Now what those changes or secret sauce could be is completely unknown.
I think people are looking into it to much.
Yes, Sony likely made some architectural suggestions, but AMD still needs to work within the confines of Graphics Core Next.
The reason for the amount of staff that is dumped on Navi is simple though, Semi-Custom chip designs, aka. Multiple chips for Microsoft, Sony, PC APU's (Desktop and Notebook!), PC GPU's. - There is hundreds of millions of dollars up for grabs.
It's not a small job to pull off all at once... But the financial gains for AMD is absolutely massive, hence the urgency in dumping as much staff onto the project.
P.S. Wont be Surprised if Ps5 makes an appearance at Navi Reveal on July 7
I agree. Also can't wait!
Hey I know you and CGI were being conservative and mindfull on your expectations at the console releasing @399 without massive losses, thus I said I was joking on seeing Sony troll both of you.
About the Stadia comparison, HW being more than twice as powerfull because of efficiency I understand, but also on real world we have to consider the internet infrastructure for most customers, and in this scenario I would say I see Stadia performing worse than PS4Pro to most customers.
Also happy to be proven wrong. At the end of the day, I am a tech enthusiast, so it's still a win.
I made the question to CGI above, but I also would like to see your input on they having 8GB of HBM2 and 16GB of DDR4. How much impact such different type of RAM and their bandwidhts would have on the balancing and use on the system? Could they use it seemless to have the slower RAM working on the parts that need more memory but lower speed and HBM for the more urgent tasks? Or is that a good way to take on the ineficiency of GCN on needing plenty of memory and bandwidth to keep CUs feed? Also if Sony accepted slightly slower multitasking the 8Gb of RAM for the OS could be increased using the other 16Gb for game right, or perhaps OS doesn't change the RAM but the other apps when initializing get some of the gaming RAM, just like on PS4 some apps will put game in suspension and others will have other apps closed before running.
The main impact will be on the development side of the equation, developers will need to allocate workloads to get the most out of the differing bandwidths and capacities.
They will also need to spend more transistors on additional memory controllers to drive it all, which will also cost power.
In short, it's not impossible, but I don't see it being likely... They would probably be best served going with a single large pool of GDDR6.
Remember... It will also be Ryzen powered... And we all know how much Ryzen loves it's bandwidth!
No way the OS uses 8GBytes. The streaming buffer currently uses a good chunk of the os, hence the large reserves in the PS4/X1. They crammed the PS3/X360 os into a few megabytes, and that is the target for any console, make the os as small as possible.
Given that the SoC will be largehere will be two different memory bus architectures in the SoC, it's just too big a mess to design and synchronise (the cpus already have the synchrinise penalty built in so you would avoid any additional penalty as much as you can). Maybe even only a 256bit gddr6 interface and some cheap 4-8GByte streaming buffer somewhere downstreams (getting rid of the memory reserves we have now) would be enough for a PS5.
The Xbox 360 managed to get away with 32Mb of Ram for it's OS... The Xbox One wanted 3072MB... And each time the consoles were simply doing more.
More data is being cached to speed up various operations, things like higher quality voice chat, video streaming, multi-tasking were all implemented.
In short, the consoles are becoming more PC-like, which means they need more memory.
I don't think there is an immediate need for more than 4GB for an OS yet though, you can do a surprising amount with that... But let's just wait and see how the cards fall.
But if there is going to be targeted 8k User-Interfaces, then I don't expect the memory usage to be small. - Microsoft will likely use a UI that scales up well, but uses minimal memory, but they are a software/OS engineering giant...