Do you think there will be a mid gen refresh console from Sony and Microsoft?

3 shader engines. Each one 10WGP. 1 disabled. 54 active CUs in total.

2 shader engines disabled for PS4 BC. 1 shader engine disabled for PS4 Pro and PS5 BC.

54 CU x 2500mhz x 64 flops per clock per cu x 4 (dual issue accumulative) =~ 35 teraflops.

35 TF RDNA3 should be roughly equivalent to ~22 teraflops RDNA2.

So double the performance on paper but should be much better in RT workloads.

@snc Anything to add?
You “sure” about this? Or are we just discussing?
 
Last edited:
3 shader engines. Each one 10WGP. 1 disabled. 54 active CUs in total.

2 shader engines disabled for PS4 BC. 1 shader engine disabled for PS4 Pro and PS5 BC.

54 CU x 2500mhz x 64 flops per clock per cu x 4 (dual issue accumulative) =~ 35 teraflops.

35 TF RDNA3 should be roughly equivalent to ~22 teraflops RDNA2.

So double the performance on paper but should be much better in RT workloads.

@snc Anything to add?
That if its has rdna3 rt capabilities it will be underwhelming, but taking into account 2022 patent about hw traversal support I think it will be better in this matter
 
In the end whats matter is perf, it looks like close to 2x in raster and more with rt, imo nothing surprising here
How are we getting 2x raster though? You’d need to double geometry, rops, etc. There’s no details on the FF hardware provided.
 
Just very simple projection based on rx7900xt with 84cu
I’m not really concerned here with the actual performance. But if you’re modelling after 7900 series, then you’re assuming they’re moving to a chiplet model, and I’ve not seen a chiplet with CPU integration yet. The latest laptop APUs today with RDNA3 don’t used MCDs from what I can see.

A lot of conjecture here is that it operates like 7900 series, which is wild, because that’s a behemoth of a massive chip without the CPU.

It’s a change I’m expecting with a next Gen chip, and not a mid gen refresh.
 
DF might find it acceptable, but I'm a 60fps convert and I am not willing to go back. Whilst I can adapt to the visuals, the latency for many 30fps games is just too much.

This. I use to be a firm believer in graphics over everything else (i.e., 30fps being acceptable), but now, framerate takes priority. 60fps is the bare minimum that I can accept, with 120fps being my primary target in gaming (which 90% of my Steam library games are). Playing at 30fps just feels outdated and unacceptable in this day and age, but, the console space can only do so much within the given budget and thermal constraints placed by consumers and console manufacturers.
 
You “sure” about this? Or are we just discussing?

Considering that we've had 2 consoles so far that required the CUs to be multiples of 18 to maintain BC, I'll believe anything different when I see it. Sony doesn't have a HAL equivalent to abstract the HW layer like MS did.
 
All they have to do is offer something that will offer better performance akin to the PS4 Pro. Even if it's "just" 20-25Tflops.
 
People are forgetting how poorly RDNA 3 scales. A 7900XTX is only 50% faster than 6800XT in raster despite having 3x the compute and double the bandwidth. It’s a huge dud of an architecture.

One of RDNA 3s biggest flaws was that it failed to reach its much higher clock speed targets. I expect that will be the major change with RDNA 3.5. I.e. it will clock like RDNA 3 was originally supposed to.
 
3 shader engines. Each one 10WGP. 1 disabled. 54 active CUs in total.

2 shader engines disabled for PS4 BC. 1 shader engine disabled for PS4 Pro and PS5 BC.

54 CU x 2500mhz x 64 flops per clock per cu x 4 (dual issue accumulative) =~ 35 teraflops.

35 TF RDNA3 should be roughly equivalent to ~22 teraflops RDNA2.

So double the performance on paper but should be much better in RT workloads.

@snc Anything to add?
Improved RT acceleration (new RDNA4 RT tech). I accelerates structure traversal on top of intersection. It should greatly accelerate RT even more so than the tflops would suggest. Supposedly.

 
Improved RT acceleration (new RDNA4 RT tech). I accelerates structure traversal on top of intersection. It should greatly accelerate RT even more so than the tflops would suggest. Supposedly.


I just don't see where a Sony patent fits into this. Nvidia has been doing what that patent describes (as far as my limited understand goes anyway) since Turing. AMD must surely already have it on their roadmap to implement something similar. Most people were expecting it in RDNA3. I don't see why they would need to incorporate custom Sony tech for a function that they absolutely need to be implementing natively in their GPU's next gen anyway.
 
People are forgetting how poorly RDNA 3 scales. A 7900XTX is only 50% faster than 6800XT in raster despite having 3x the compute and double the bandwidth. It’s a huge dud of an architecture.

Is it really all that bad? I mean yeah, it's disappointing, but I'd probably stop short of 'dud'. It's only got 33% more CUs than the 6800XT, and 50% more Shader Engines. Still, it's a lot of extra transistors (58b vs 27b) with not a lot to show for it.

The 7800XT might be a better balanced version of the architecture, with fewer but possibly wider shader engines to shed a bit of bloat. And hopefully they sorted the clocks out.

I just don't see where a Sony patent fits into this. Nvidia has been doing what that patent describes (as far as my limited understand goes anyway) since Turing. AMD must surely already have it on their roadmap to implement something similar. Most people were expecting it in RDNA3. I don't see why they would need to incorporate custom Sony tech for a function that they absolutely need to be implementing natively in their GPU's next gen anyway.

Yeah, RT is deeply integrated into a GPU and I don't think the level of work needed to add such custom hardware into a console would be practical or time or cost efficient. We've never seen anything like that in a console before - it's way beyond adding a custom block for sound or decompression. I think that's why Globalisateur added "supposedly" at the end.

I too was hoping for something like this in RDNA 3, but it turned out to be a relatively minor change from RDNA 2, at least in terms of performance (the multi chip package is a pretty big change). If AMD don't have something big to drop WRT ray tracing for RDNA 4 I don't know where they can really go. I don't think some reviewers out there did anyone any favours by downplaying RT and DLSS for so long ... though you'd hope that wouldn't influence AMD engineering path.
 
I'd agree that there's a low probability of that patent surfacing in actual PS5Pro HW. I do find the idea really intriguing though.

If the Pro's capable of running a 60fps UE5 title in HW RT mode vs software on PS5/XSX, that would be a pretty cool visual difference.
 
Last edited:
It's so rare for any of these patents to manifest for real, I wouldn't place any faith in them.
Sony patents a lot of technology that goes nowhere, but Mark Cerny does not and this patent is credited to him. Cerny's patented checkerboard tech appeared in PS4 Pro. I'd say there was more than an even chance that if AMD don't have something better ready to deploy, and Cerny's approach can yield a good performance boost for little silicon, then you'll likely see a version of it in PS5 Pro.
 
Considering that we've had 2 consoles so far that required the CUs to be multiples of 18 to maintain BC, I'll believe anything different when I see it. Sony doesn't have a HAL equivalent to abstract the HW layer like MS did.
So you’re thinking 60 CUs before redundancy. So effectively 54CU total, 1WGP redundancy per SE.

Yea I guess that would make sense
 
With hints in Tom Henderson's article and Cerny's RT patent that about updated RT capabilities, GPU will most likely move beyond RDNA2.

But I'm more interested to see if they will stick to Zen2. Zen4 would iron out so many CPU-related issues on current games.


PS4 Pro was 20% of all sold PS4 consoles after it released in late 2016, and Sony was happy with that.
PS5 Pro will have same or even smaller ratio of sales [there is no big resolution update push in TV market], but sales will still satisfy new customers "who want the best" and even more importantly core "enthusiast" consumers who spend most on new games. Plus, those who upgrade from base to midgen console usually gift their old console to a family member, thus providing Sony with a new customer who can purchase games/addons.

IMO there are no downsides as long as games made later for base console are without issues. And we saw that PS4 got great game support with stable framerates and advanced tech solutions not only by the time PS5 came, but also after.

I guess as long as Sony can make up the initial development costs with higher profit margins and total sales that's a win for them.

Still think an updated main console would be smarter. We're not seeing a PS6 until 2028 or later, putting out an "upgraded" base console would be a new (mostly) tactic. Include the new RDNA3 video encoder, wifi6e or 7, include a "boost" mode for higher clockspeeds (say, 20% faster), replace the custom SSD with a 1tb 980 pro, and slim down the design. You've suddenly got a package that's visibly better all around, that you can charge just as little as the current PS5 for and doesn't add any costs besides initial development, doesn't require any real support from developers at all, and gives you a PR boost over your competitor.

But if they're really going for a PS5 Pro, then something like an RDNA4 (yay all chiplets!) and a Zen 4c (or Zen 5? 4c can go into chiplet SOCs like on ML300 right?) on TSMC 4p design makes the most sense. RDNA4 upgrades the RT the most since RDNA2, and allows disaggregating everything into chiplets which keeps manufacturing costs down a lot, and Zen 4c is simultaneously an upgrade from the Zen 2 CPU and also incredibly tiny and cheap.
 
Arent any details on RDNA4 just YouTube guesswork for views at this point?

Everything is just guess work.


The issue for me with the whole conversation is that last time Sony created a pro , they kept the same cpu and gcn architecture. I am not sure if they will be able to take advantage of newer zen and rdna architecture if AMD is even still on rdna.

If sony goes with zen2+ rdna 2 but more of everything the system is going to struggle drastically. These current consoles are completely under powered for ray tracing and I would argue traditional rasterization. FF16 struggles to maintain 40+fps at 720p and DF has the footage of Spiderman 2 at 1200p 30fps. On MS's side starfield looks to be 1440p@30fps. The hardware just simply isn't there.

If they stay with RDNA2/Zen2 for a PS5 pro and it launches in 2024 they are likely to run head first into RDNA 4 and the 50x0 series from NVidia. Now I wont claim to know what these companies have on their hands. But even if the next gen cards simply move 7900/4080 performance down to sub $600 price points and 4090 performance down to $1000 price points the pro will look horrible in comparsion. This would also go for a series x pro or whatever using zen2/rdna 2

Now it be great if Sony or MS break from last gen and makes a zen4/5 with rdna4 and rdna4 turns out to be a worth while upgrade over rdna 2/3. Some times like Zen5 rumored to have a big/little setup based around zen2/zen5 cores could allow them to have 4 zen 2 cores for the os and backround tasks and 8 zen 5 cores for gaming. There is also rumors that xillian tech will first show up in zen 5.

But honestly looking at the offerings from both Sony and Ms and the pathetic performance it will take more than just doubling units here to get back to 60fps unless we are all happy with 720p resolutions
 
Back
Top