- Status
isnt 8k 4x the pixels of 4k?
iirc 8K is 7680x4320 = 33.17M pixels.
from my understanding the xK refers to the number of pixels on the larger axis, on 4k its 3840 pixels, on 8k its 7680, on 1080p its 1920 (as 1080p is 2K and not 1440p which is more like 2.5K)
No. You made a claim, please elaborate on the the issues you think is wrong.Here:
Coretek video about AMD strategy and the console
There resume of video is courtesy of nujabeans https://www.resetera.com/threads/playstation-5-system-architecture-deep-dive-ot-secret-agent-cerny.175780/post-30298317 This channel Coreteks focuses on PC gaming/GPU industry and doesn’t have any sort of “allegiance” to Microsoft or Sony. He...www.resetera.com
Why's it seem like we're gonna skip 6K and go straight for 8K?
If the SX and RDNA2 follows the same structure of RDNA1 (I don't see why not) the SX could end up having up to 96 ROP. I guess we'll have to wait and see how MS decided to arrange the dual CUs. We know RDNA1 has a max of 5 dual CUs per Shader Array, and we know we can have less than 5 but we don't know if it can be more than 5. If more is posible then then I guess they could have 7 dual CUs per array with with 2 shader engines for 64 ROP with 56 CUs. If more than 5 is not possible, we know less is possible so 2 shader engines with 5 dual CUs per srray and 1 with 4 dual CUs per array could be possible giving you 96 ROP with 56 CUs. It's all speculation until we know more about RDNA2 or SX SoC, but it's possible.
Last edited:
I'm not sure, but if I had to guess I'll say the display technology would be easier to scale to 8k than 6k.
NAVI is fundamentally different from GCN is a lot of key areas, but we're approaching NDA territory here.
But the original point of the discussion stays. Different parts of the GPU can be scaled differently, especially so in NAVI that let go of some old GCN limitations.
PS5 rasteriser is *not* faster than Xbox from the released informations so far, and it's wrong to assume that it is just because the clock is higher.
Also, all modern GPUs are greatly oversized in ROP output anyway compared to what's actually needed. Pushing pixels out at the theoretical peak is a dream that in a real world AAA is extremely rare (and not for the entire game anyway) and it probably means something else is wrong.
Paradoxically, while one of the main benefits of 4k and other 4-multiples of 1080p is perfect integer scaling, very few monitors and televisions do this.
Since my words don't good, here's a link:
https://tanalin.com/en/articles/integer-scaling/
Your input is greatly appreciated, not only in ths thread but generally!
True, my napkin was folded lol.
Corrected 8K = ~33M Pixels x 60Hz = ~2Gigapixel/s
Really appreciate the input here, thanks for clarifying things.
We understand that.
I'm very curious about how any UMA-style consumer system handles virtual memory. I wonder if the new systems will let GPU read assets, say, sampling a texture, in memory mapped file style.
PS5 Patent Hints at Faster Game Booting with Custom Templates
A Sony patent that was just lately made public explains how the PlayStation 5 might let customers bypass menus and loading screens besides video games exactly the place they need to begin enjoying. Together with the PS5’s upgraded SSD, the brand new know-how may very well be yet one more method...
Doom Eternal seems to load much faster on SSD and there this time a difference between NVME SSD and SATA SSD probably a window to the future.
SATA SSD is 10 seconds.
If all games are optimized like this the less than 1 second look like possible on PS5.
So this will possibly translate into Series X loading in a few seconds and PS5 loading a second or so quicker. Such a massive leap over those old HDD's. 30+ seconds for loading is beyond crazy but so many games do.
If this is true then the PS5's SSD's performance lead is vastly overstated.
Exactly, XSX and PS5 will have the advantage of realtime decompression data without demanding too much to the CPU but the assets will grow a lot too.
EDIT: Open world games will probably take longer to load too...
because of Covid-19 I bought some games to play together with my girlfriend (who is a non-gamer). When we were playing some Crash Team Racing; one of the first things she said was "do you always have to wait this long before the race starts"?
I think the amount of hidden loading screens is enough for me. The last third of GoW was too much on that front.
If GTA can stopped spawning the same 8 cars in a zone, it is great too.
SATA SSD is less than 10 seconds it is friends playing on PC but there is at least a difference maybe 2-3 seconds.
That's a really interesting quote, thank you for pointing this aspect.
If I may ask, what's in your opinion the real output and TF peak that the consoles will push out on triple As?
Yeah, those Activision remasters (Crash / Spyro) have insane loading times.
Not at all, loading times are just a small benefit of raw speed being faster.
If anything it has been a lot of downplaying of the difference.
The idea that the significant difference in dedicated I/O hardware and raw SSD speed will only benefit 1st party games or not have any noticeable benefits at all is based on the assumption that Sony made wrong architectural investments and we already have hints from the reputable sources that the opposite is true.
I'm sorry but by modern games do you mean games made during the cartridge era? You're right, the Pokemon games on the old gameboys were really great and did a masterful job at that.
I know you're not talking about most titles released during the past few years because no one who's played a recent game would make a claim like that. One of the biggest AAA games right now -Destiny- has had it's entire game world design gimped because of slow HDD speeds. Even linear shooter games like CoD have fairly long loading screens if you die or when you first load up a level. Those fluff loading areas in Uncharted and GoW feel incredibly slow at times and sometimes break the flow of the game. And of course, I'm sure everyone enjoys the minute long loading screens in titles like GTAV and RDR2. Who doesn't enjoy periodically staring at a mostly static empty screen for 50-70 seconds with nothing else to do? Yeah, bring on that immersion baby!
I don't know how damn slow that HDD is, but I never had loading times longer than 10s on PS4 with Doom Eternal .
Mark Cerny told loading time is not the reason they work so much on a custom SSD. The I/O and Audio hybrid CU/SPU take some place on the SOC die too, they could probably have use it for a little bit more GPU power. Time will tell if it was good choice. Like always wait and see on the games.
When it comes to loading it won't be a big difference few secs the most .
The bigger question was always what can the extra speed Sony SSD can do when making the game or helping out with .
Which may take a while to see or might not even matter time will tell .
Also extra few second won't matter to most but when you get accustom to a certain speed sometimes it can end up being annoying if something take longer lol.
Side note the SSD speed will also help with UI and QOL improvements so it's not only for games .
We don't know that, no, the 5500 has 6 WGPs per SA.
That's what their CGI render of the chip layout implies. They could be lying there, though.
Paper spec doesn't show actual performance. Especially with an unreleased GPU. Simple answer seems the most likely here, they'll throw whatever at the wall and see what sticks
Thabks for the informative replies. I have to admit this has been the most condusing console launch in history for me.
Yeah, I wouldn't take that render seriously. I'm 99% certain that not close to accurate.
I never said the PS5 was? not sure what this reponse is in relation to?
Plus you are now agreeming with my earlier statement on ROP count and being bound by it, which was what I said.
Reading your comment I can assume you are a Dev and as such the NDA, if so understood this was a conversation not wanting to force any breaking of that, thanks.
Last edited:
It helps a lot and they have Navi GPU to so some test and use raytracing software. I would say wait and see this maybe not the only game at 4k at 60 fps with raytracing/Global illumination(maybe not full hybrid rendering and from dev with a devkit). The games will show if it is possible or not.
Edit: At least they have a target, Nice to see if they will reach or fail to reach it.
I follow now. I'm inclined to believe it's real based on all their previous shots of such things.
A few different threads got mixed in.
My original statement in the thread was that the idea PS5 was 22% faster than Xbox at rasterisation is incorrect, and I said that's because the rasteriser is not a single block and it scales with the number of CUs (and even then, you're unlikely to be rasteriser bound).
Somehow someone (I don't even remember how) brought the ROP count in the mix, even tho it's unrelated to rasterisation.
I never disagreed with the statement that both won't be ROP bound in most cases or that they both will likely have 64 ROPs.
What I disagree with is that, *in theory*, nothing architecturally is preventing Navi/RDNA to change that number as the design is extremely modular and flexible compared to GCN.
From what we learned PS5 should load most games about twice faster, maybe even faster.
I guess many - including myself - were conflating the geometry processor for the rasteriser. If rasterisation itself if done by the 'prim' units which seem bound to CU count, then that part of the 'top' of the pipeline won't be stronger on PS5.
I think it's fair to say that PS5 could be xx% faster at whatever that pre-rasterisation, centralised, 'geometry processor' is doing, though. Assuming PS5 and XSX follow that template. I assume it's responsible for pre-rasterisation geometry work - culling, 'mesh shading' etc?
I think it's fair to say that PS5 could be xx% faster at whatever that pre-rasterisation, centralised, 'geometry processor' is doing, though. Assuming PS5 and XSX follow that template. I assume it's responsible for pre-rasterisation geometry work - culling, 'mesh shading' etc?
Actually I forgot all about the 5500XT which does have 6 WGPs per SA as you said. So it's possible that they went with 7 WGPs and 2 SEs (2 SA ea) for 64 ROPs and 56CUs.
Yep, I agree with that, they can design around IF a constraint but I think 64 is enough to not become an issue even at the higher CU and thus throughput.
I think mesh shading is linked to CUs but I am not sure.
EDIT:
Mesh Shader Possibilities – Nathan Reed’s coding blog
Pixels and polygons and shaders, oh my!
reedbeta.com
This. Don't overlook what devs have been saying and how they have been reacting very happily.
Good thing its SSD performance lead isn't just about loading times then.
To use Cerny's example, The PS5 SSD and the custom hardware stack built around it to optimize and guarantee its performance means that you can stream in 2GB worth of data in a quarter of a second.
That's 2GB of data in 0.25 seconds.
Now even if the XSXhas the same hardware customizations built into its APU and (which it doesn't) that would mean that the PS5 can still stream in at last 1GB more data (this should actually be more compared the XSXbut keeping it simple for argument sake) in 0.25 seconds than in the XSX.
That's what all the fuss is really about.
Yeah, it makes sense mesh shading would be running atop the cus I guess.
The whitepaper is a bit a mute on what the geometry processor does. From reading around other previews it sounds like it - along with the prim units - collaborate on pre-rasterisation tasks. That the geometry processor can be responsible for some assembly/fixed-function tessellation, and also coordination of the other prim units which also do assembly/tessellation work. A bit confusing! Some element of that work seems to be centralised though.
The other big caveat, as mentioned by others though, is that the resources in each shader array can vary per design. We can surmise PS5 and XSX are 'standard' in this regard, but it's possible their resource ratios may differ.
How we know this is true?
This is true if we assume that both systems have to transfer the same amount of data in assets neaded to load a similar scene. In the case they don't, these straight raw speeds comparisons are meaningless until we see real world performance.
- Status