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

[iroboto]

60 CU's, RDNA 3+, perhaps higher clocks, you can get roughly double raster and perhaps far more RT, but like the PS4 Pro, the memory bandwidth will be a bottleneck. Still, that, and with FSR 3 frame gen (quality a big question mark), I can see this as a viable product category.

I think the biggest concern I have with this is what existing games will have to do in order to take advantage of those extra CU's. Automatically, or will every game require a patch?



The only problem with this is those 30fps games are limited in no small part due to the CPU. Perhaps the cores here would be upgraded to Zen3, and FSR3 could help that bottleneck as well, but yeah I agree - I always found their takes on this to be somewhat contradictory.

This is a very awkward configuration.
I was thinking the pro would be a butterfly of the existing setup for effortless emulation, but 60 CUs puts it very close to XSX architecture.
Keeping the same amount of FF hardware, same bus size, but now with 16 DWGP per Shader Engine, 2 shader engines, with 1 DWGP disabled for yield, this is the only way to get to 60CU layout. Or it's 15 DWGP per SE, and 1 disabled netting 56CUs.

It's effectively XSX, but with a smaller memory bus, less cache, and likely much faster clock speeds. The only raster improvements I'm expecting is going to be on the clockside of things.

Really interested to see how this one turns out. Based on everyone else's previous commentary about XSX having the worst amount of cache or bandwidth per CU, this is going to take the cake.

 

[Deleted member 11852]

AFAIK that 20% figure which was put out only half a year after its launch is the only data we have. We have no idea if that trend continued or they quickly burned through the niche audience interested in a higher spec console. Do you have a link to sales data far into its lifecycle?

I also only recall the 20% Pro:Launch PS4 ratio number from 2017. Ergo whatever the actual ratio was before PS5's launch window , be-it 15% or 30%, it was enough for Sony to repeat the strategy with PS5.

 

[Panino Manino]

I can't believe this thing will use RDNA 3+, it's too much transistors.
Instead I think Sony will get a custom RDNA 2+ from AMD. Same base RDNA 2 CUs but with RDNA 3 RT improvements.

 

[DavidGraham]

So no different to the PC, where almost no devs targets higher-end hardware. ¯\_(ツ)_/¯

On PC, Ray Tracing and Path Tracing changed the equation, there are dozens of PC games with next gen graphics relying on these techniques now.

 

[mpg1]

I think they would keep it simple like the PS4 Pro. Using the same architecture but slightly increased CPU clocks, slightly increased memory bandwidth, significantly increased GPU grunt.

They basically just want a pro console that just going to boost the res, fps, ray tracing of PS5 games.

Big question about BC though. Like if you move away from the "butterfly design" how does still work. They had to have 4 modes in total now wouldn't they?

 

[Saidur_Ali]

I think it is very important to release a mid-gen refresh especially for Sony given there will be more competition in the PC space especially from Intel next year regarding the budget GPU market. AMD need to make a decent RT solution so they remain competitive against the likes of Intel and Nvidia and hopefully PS5 "Pro" has new RT cores that might be used in RDNA 4 as it will be a bit meh if it is just the same as RDNA 3 but still serviceable. For Microsoft, it makes more sense to try and boost Series consoles clock speeds (So SS and XS) (As they probably don't have enough demand for a more powerful console to make it justifiable given Series X is probably the lowest selling console on the market today even though it is likely the most subsidised) if they move to a newer process in a future revision and then maybe start the next generation early but with support still for the older generation consoles (Hence why they named the consoles "Series").

I honestly don't get the pushback given the synergies all lining up perfectly to create one that is not only cost-effective but also a substantial improvement. It makes even more sense this generation than last given the CPU even if they did not update the architecture, the clock speeds can be boosted higher for a lot more performance than what could be done for the Jaguar cores.

I personally was expecting potentially something like 54 compute units (3456 shaders) and sticking with the 256 bit bus and using 18Gbps GDDR6 which is very cost-effective given how popular it is in GPUs at the moment giving total bandwidth of 576.0 GB/s. Hopefully it is on 4nm process which might allow for better clock speeds, they need to hit 2.64GHz if they want to try and market it more than 3x the Teraflops of the Series X although in terms of similar Teraflop rating as RDNA 2, it will be likely less than twice the Teraflops of the PS5 but effective performance will likely be more than double the performance of the Series X and even more so in RT. It will likely be still smaller in die size than the Series X and maybe even the launch PS5 so they could produce it in large quantities if there was demand for it and sell it for current PS5 price if they wanted to. If they went for a larger GPU in the here and now, not only does cost go up significantly, the potential to run higher clocks gets lower if they are targeting the same power limits and would not make sense for 10% improvement for 50% higher console cost. It would be good if in future generations, they also offer a super high end option for the enthusiast like an Ultimate console version with higher power limits as hardware improvements get more stagnant due to not being to cram much more performance at the same power amount.

I think people don't really appreciate how cost-effective the PS5 is given Microsoft subsidising the Series consoles by billions of dollars. It has similar BOM to a Series S at launch (NAND and RAM capacity corrected) (The potential slim if on TSMC 4nm process launching potentially towards the end of this year might have lower BOM than Series S) yet over $200 cheaper to produce than Series X from the same technology partner while being able to outperform them by using the bleeding edge of AMD technology like SmartShift and wide high compute GPU design vs the narrow constrained dated approach of fixed clocks and low compute requirements Microsoft used due to their 4x Xbox One and 2x Series S cloud server requirements. There must be a reason why all AMD GPU's designed for PC use the same approach as Sony for the best bang for buck performance.

From the developer point of view, it will be super easy to run on a more powerful system than a lower powerful one and they could run the game as intended given PC ports enable already the detail that exists in the game. It will be like a Ultra PS5 allowing PS5 games to run at like Ultra settings (Soemething around the RTX 4070 / 4070 Ti level and not far from a 4080) without stutter you might experience on PC, it is a product that is needed for smooth gaming without the hassle. Even if it is lower volume than PS5 as I expect the smaller version to sell like hotcakes this holiday with the lower price, it still could be more volume than something like both the Series S and X combined which is good news for AMD as they might be able to sell the product for more margin and given the die size of the original PS5 will already be extremely small, they can get more utilisation on TSMC capacity they have managed to book and it is more in the sweet spot for a high performance console and node they are using in terms of maximising yields like the launch PS5 was.

 

[Deleted member 11852]

On PC, Ray Tracing and Path Tracing changed the equation, there are dozens of PC games with next gen graphics relying on these techniques now.

And that's after five years since the release of the first RTX cards. These mid-gen consoles are going to last 3-4 years at most, because the kind of person who is buying a PS5 Pro is probably dumping it for a PS6 when that becomes available.

In anticipation of Starfield, I upgraded the lounge PC (connected to the TV) to a 4090 and went on the hunt for games that really show off RT and it feels like slim pickings out there. I found myself going back o Cyberpunk, the Witcher 3, Metro Exodus, and Portal.

I assume the reluctant for developers and publishers to fully embrace RT tech to the fullest, and actually rely on it, is that in the PC space there are still few cards where heavy RT doesn't massively tank performance except for simpler games. And consoles.. well. I presume that's why beefier RT is among the expected features of PS5 Pro.

 

[pjbliverpool]

It will be like a Ultra PS5 allowing PS5 games to run at like Ultra settings (Soemething around the RTX 4070 / 4070 Ti level and not far from a 4080) without stutter you might experience on PC,

You're not going to get anywhere near 4080 performance from 60CU's. The 7900XTX which is pretty much a match for the 4080 has 96 CU's and can boost up to 2.5Ghz. It also has two thirds more memory bandwidth on top of infinity cache.

On paper a PS5 Pro with 60CU's and 18000Mbps memory at the current PS5 clock speeds falls a little under two thirds of a 7900XTX if we ignore infinity cache (or assume it comes with it). So that's around 3080/4070/6800XT level performance in raster and somewhere between the 6800XT (~3070) and 4070 in RT depending on how much AMD improve their RT implementation.

 

[Deleted member 11852]

You're not going to get anywhere near 4080 performance from 60CU's. The 7900XTX which is pretty much a match for the 4080 has 96 CU's and can boost up to 2.5Ghz. It also has two thirds more memory bandwidth on top of infinity cache.

I don't disagree, but I would add that consoles have often punched above their weight in terms of what performance the hardware delivers because devs can hone for specific configurations.

Console manufacturers exist in a uniquely privileged position where they can analyse the performance of a large number of games against specific API usage and hardware utilisations with a view to making rapid changes to APIs and hardware to alleviate bottlenecks or accelerate commonly used functions. AMD's public roadmap available today, isn't going to be a reliable basis on which to predict what AMD might produce in a years time. Console APUs are often a mix of today's tech combined with an infusion of architectural improvements to be announced.

 

[mpg1]

PS5 might have been a clever design choice in terms value for performance but you do "pay" in size/aesthetics....

High clocks + unfancy cooling system = large cooling system = large console

PS5 Slim will win some of this back. A PS5 Pro will probably be just as big as the PS5 is now...possibly bigger?

 

[see colon]

AFAIK that 20% figure which was put out only half a year after its launch is the only data we have. We have no idea if that trend continued or they quickly burned through the niche audience interested in a higher spec console. Do you have a link to sales data far into its lifecycle?

That 20% number isn't lifetime of PS4, it was lifetime of Pro up until that point. So it isn't 20% of all PS4's sold were Pro, because that number disregards the 60ish Million units that were sold up until the Pro's launch. They also stopped manufacturing PS4 Pros in 2020, but continues to make and sell PS4 Slims.

 

[metacore]

Refreshes are needed if next gen will arrive in 2028 which is not surprising given very poor transistor/ mm2 announcements from TSMC.

What's the concern about games not scaling? If Sony decided to make the Pro im sure they had taken into account experience with 4pro and dev/publishers ways. It must be done in such way to make easy sliders or even <include> library for easy scaling even if alu utilization is poor like rtx does. I think marketing points will point to frame generation and raytracing and it would be easy to fulfil, especially with main console target still being ps5 and raytracing work and frame gen familiarity on average results should be much better than last pro revisions.

I'm pretty sure Phill is already on phone with amd if he was not bluffing with not making refresh. How would they montage media briefs with confident sarcastic executives talking how everything will be alright in the future to the epic 80 song remixes. Hopefully for consumers current xboxes will not be suddenly kinda half dropped or undedicated like last time. Only scenario im seeing which they are not making the refresh is series S blocking this. I mean scaling between S and 40tflops refresh would be to much work and in practice delta of such size would comically complicate development.


What I would agree with Remji it feels there indeed is some kind of technological atrophy in technology departments. Technology's announced long time ago barrely touched upon, Direct storage barelly touch upon after many years, results are debatable even on console where implementing this is way easier. Same for Sampler feedback, mesh shaders or perhaps SF it is another missfire like geoshaders.

"lazy devs" meme indeed has just a little truth today. Younger generations of programmers raised on java/python and gigs of ram showing up with mentality what's the profiler for? is it third party plug in?. Only handful of devs are/have muscle to be nuts and bolts oriented for tech initiative. Recently i was watching tech interview with some UE based middle size game dev, tech leader he was introduced. Amount of non answears and i don't knows about basic current cornerstone things was baffling, "slider & plugin" mentality all the way.

 

[Inuhanyou]

I don't see the point of PS5 pro. But I hope they say devs should offer an unlocked performance mode option so when ps6 rolls around we can get those games at 60 easier

 

[Deleted member 11852]

I don't see the point of PS5 pro. But I hope they say devs should offer an unlocked performance mode option so when ps6 rolls around we can get those games at 60 easier

I would like to think that Sony, like Microsoft, having done this dance last generation, would gave provided both developers guidance and APIs on PS5 to already make this a lot easier. I.e. APIs that asynchronously divest computational resource to whatever is available rather than X number of free CU.s Likewise any game implementing DRS should just adapt with no further work.

 

[Shifty Geezer]

PS5 might have been a clever design choice in terms value for performance but you do "pay" in size/aesthetics....

High clocks + unfancy cooling system = large cooling system = large console

PS5 is only 5% larger than XBSX going by internal volume. Or do you mean compared to a water-cooled micro PC or something, and count both consoles as large? I don't think there's any way to get affordable 200W consoles smaller.

 

[Seanspeed]

I think people don't really appreciate how cost-effective the PS5 is given Microsoft subsidising the Series consoles by billions of dollars. It has similar BOM to a Series S at launch (NAND and RAM capacity corrected) (The potential slim if on TSMC 4nm process launching potentially towards the end of this year might have lower BOM than Series S) yet over $200 cheaper to produce than Series X from the same technology partner while being able to outperform them by using the bleeding edge of AMD technology like SmartShift and wide high compute GPU design vs the narrow constrained dated approach of fixed clocks and low compute requirements Microsoft used due to their 4x Xbox One and 2x Series S cloud server requirements. There must be a reason why all AMD GPU's designed for PC use the same approach as Sony for the best bang for buck performance.

Where on earth are you getting all these figures? :/

 

[function]

Really interested to see how this one turns out. Based on everyone else's previous commentary about XSX having the worst amount of cache or bandwidth per CU, this is going to take the cake.

Is the Series X really that bad in terms of cache? I know the L1 is a bit small (relative to the CU count), but L0 and LS is as good as any other RDNA 2 GPU, and L2 is large due to having a 320-bit bus. And if accesses are efficiently striped across the bus, I'd presume that means a corresponding increase in L2 bandwidth. L1 isn't used for compute or, iirc, some or all of RT.

I can see the logic behind a PS5 Pro, but I think MS might be better off taking a different path. If you could do a shrink / refresh of the Series X/S and boost GPU clocks, maybe fiddle with the memory configuration (faster memory on 320-bit, or even faster on 256-bit), and voila, you've got a beefier console.

The stock RX 7600 using a pretty lightweight 2-heatpipe cooler, and still only on 6nm (a glorified 7nm), is hanging around 2.46 mhz in Cyberpunk with RT on. Even Furmark is registered as between 1946 and 2008 mhz - nicely above the Series X's 1825 mhz "no throttle" range even with the 7600's cheap cooler.

AMD Radeon RX 7600 Review - For 1080p Gamers

Today AMD is launching their Radeon RX 7600, which is targeted at gamers with 1080p Full HD screens. Our review confirms that thanks to the new RDNA3 graphics architecture, the new card is highly energy efficient.

www.techpowerup.com

At 5nm and using a super conservative, deterministic GPU only boost system (so maintaining the behaviour of launch Series X), there's probably room to get a Series X refresh up to around 2.3 ghz (or higher) in even demanding games like Cyberpunk RT while also reducing cooler costs and complexity. That would get you up to 15+ TF for pretty much the same chip, only with a pretty conservative GPU boost enabled.

I'm looking forward to seeing the TPU reviews of the 7700 and 7800, might give us some insight into what a PS5 Pro or possible XSX shrink might be able to do. A Series S with a bit of a kick up the arse might be a more attractive proposition too!

 

[mpg1]

I don't see the point of PS5 pro. But I hope they say devs should offer an unlocked performance mode option so when ps6 rolls around we can get those games at 60 easier

The point is probably similar to the PS4 Pro....have an option so people to don't switch to PC as the generation goes on.

 

[iroboto]

Is the Series X really that bad in terms of cache? I know the L1 is a bit small (relative to the CU count), but L0 and LS is as good as any other RDNA 2 GPU, and L2 is large due to having a 320-bit bus. And if accesses are efficiently striped across the bus, I'd presume that means a corresponding increase in L2 bandwidth. L1 isn't used for compute or, iirc, some or all of RT.

I can see the logic behind a PS5 Pro, but I think MS might be better off taking a different path. If you could do a shrink / refresh of the Series X/S and boost GPU clocks, maybe fiddle with the memory configuration (faster memory on 320-bit, or even faster on 256-bit), and voila, you've got a beefier console.

The stock RX 7600 using a pretty lightweight 2-heatpipe cooler, and still only on 6nm (a glorified 7nm), is hanging around 2.46 mhz in Cyberpunk with RT on. Even Furmark is registered as between 1946 and 2008 mhz - nicely above the Series X's 1825 mhz "no throttle" range even with the 7600's cheap cooler.

AMD Radeon RX 7600 Review - For 1080p Gamers

Today AMD is launching their Radeon RX 7600, which is targeted at gamers with 1080p Full HD screens. Our review confirms that thanks to the new RDNA3 graphics architecture, the new card is highly energy efficient.

www.techpowerup.com

At 5nm and using a super conservative, deterministic GPU only boost system (so maintaining the behaviour of launch Series X), there's probably room to get a Series X refresh up to around 2.3 ghz (or higher) in even demanding games like Cyberpunk RT while also reducing cooler costs and complexity. That would get you up to 15+ TF for pretty much the same chip, only with a pretty conservative GPU boost enabled.

I'm looking forward to seeing the TPU reviews of the 7700 and 7800, might give us some insight into what a PS5 Pro or possible XSX shrink might be able to do. A Series S with a bit of a kick up the arse might be a more attractive proposition too!

It’s very close to PS5 in terms of compute per cache ratio. But still technically the worst ratio.

I don’t think there’s anything actually wrong with the setup, but with PS5 pro basically just being a super high clocked XSX, would be curious to see how it performs.

 

[snc]

It’s very close to PS5 in terms of compute per cache ratio. But still technically the worst ratio.

I don’t think there’s anything actually wrong with the setup, but with PS5 pro basically just being a super high clocked XSX, would be curious to see how it performs.

you miss point it will be rdna3 (and maybe rt closer to rdna4), rx7900xt has 84cu 5376sp 51tf, keeping same clocks ps5pro will be around 35tf, it should be around rx4070/6800xt performance, close to double in perf