-
Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
PS5 Pro devkits arrive at third-party studios, Sony expects Pro specs to leak
- Thread starter Bernoulli
- Start date
A 14TF Pro is too low of an upgrade, that's only 2 more TFLOPS vs the Series X.
See, IMO comparing it to Xbox would be the first problem.
I would compare it to the PS5.
I don't know how to calculate TOPS but the FP16 numbers suggests 2.2 GHz if the Pro has 30 WGP or 60 CUs.
The only thing that makes sense is if there's a NPU+ the GPU int8 performance.
The Pro could only have 28 or 32 total WGPS with 2 WPGs disabled for yields.
ChiefDada
Gold Member
Depends on perspective, I guess. For example, would you consider running CP Overdrive at Circa 4k30 DLSS Quality Mode on a PS5 Pro groundbreaking (I would)? If so, I think you'll be happy; I don't see any reason why these specs couldn't accomplish this.
Embearded
Member
Yes i understand that.
I've been reading though that for the last year or so, Nvidia and (?) AMD are showing TF with dual issue shaders in mind.
So these 33.75TF for the Pro, are single or dual issue?
Not sure where you are getting the lol confidence from...the 7800xt is right about on par with 4070/3080 in raster benchmarks and it has 37tflops of fp32 compute compared to the pros 32.5, it has more bandwidth courtesy of dedicated ram and bus and all of those gpus provide significantly more performance than the 45% performance uplift stated for the pro in standard workloads. Now onto the comparison with the 3080/4070 not sure about you but I would be hesitant to claim that amd will be able to match let alone surpass nvidias current rt/ml architecture in their first try with dedicated hardware blocks.
I would be extremely happy for the pro to prove me wrong and be amazing but from the information given to us so far its inferior to the 3080/4070 in GPU tech.
Last edited:
That's only in terms of FLOPS, the various fill rates would be higher but the issue would be the numbers of ROPs,TMU and WGPs/CUs.
The Pro would either have higher frequencies or more GPU hardware to achieve a meaningful upgrade.
HeisenbergFX4
Gold Member
Last edited:
playsaves3
Member
tbf I don't his rasterization improvement seems way too low compared to other leaks and the rt improvement seems extreme
Yeah, exactly. Going forward we'll be lucky to get 1080p in performance modes, and generally less. Fuck that.
If the upscaling is DLSS standard then that's like doubling base resolution. I'd take 45% more tflops plus DLSS over 100% more tflops and staying with shit like FSR2.
Also, not sure if it's been mentioned, but the RT uplift could maybe relieve a lot of the impact that RT is having on PS5 frame rates. So that we may get close to 60fps on something like GTA6 just by reducing the impact of RT on the CPU.
sncvsrtoip
Member
its not 14tf, its 16.74tf if clocks are 2.18ghz, and 18.2tf if 2.45ghz (300tops info). Teraflop doesnt scale idealy and 45% improvement in raw perf. vs ps5 can be result of bandwidth limitation, cpu limitation (if they talking about fps in games not raw gpu power) etc
Last edited:
ChiefDada
Gold Member
TheRedRiders
Member
How good is 300 tops anyway?
Gaiff
SBI’s Resident Gaslighter
Dual-issue, of course.
Crispy Gamer
Member
3x-4x raytracing performance is massive I'm glad that's the focus and just like 4k with the ps4pro this will be a reason to upgrade once people see the difference especially combined with this new PSSR antialiasing tech hopefully Sony can show off something with pathtracing.
Zathalus
Member
PT overdrive at 4k/30 DLSS Quality equivalent settings? I don't see that happening, not even a 4080 could do that. Maybe 1440p/30 with Quality upscaling.
Topher
Gold Member
Was that said in the MLID video or just a logical assumption?
Gaiff
SBI’s Resident Gaslighter
Logical assumption. Single-issue would make 0 sense.
TheRedRiders
Member
I don't see too many games taking advantage of dual issue compute, and if they do they'll quickly run into memory bandwidth bottlenecks.
Skifi28
Member
TF is a meaningless number right now.
Rocco Schiavone
Member
And?? If I'm building a gaming PC next year or in two years It will demolish the PS6 easily.
ChiefDada
Gold Member
Yep my mistake. Meant 1440p DLSS
NeonDelta
Member
Where have you seen September 2025?
playsaves3
Member
Sounds more like cheating out while still making a 600+ console
playsaves3
Member
I’ve been saying they will start making premium consoles which is awesome for us console gamers
Not sure what youre trying to argue here....the 3080/4070/7800xt are literally within 3-7% of each other in standard workloads, only the pro has less compute than the 7800xt and less bandwidth via shared bus. Not to mention its performance uplift in standard workloads is stated to be 45% which is again a sign its bandwidth starved.
Last edited:
ChiefDada
Gold Member
Gaiff
SBI’s Resident Gaslighter
It wasn't just some gaffers, Shawn Layden himself said as much.
Last edited:
HeisenbergFX4
Gold Member
It was actually Andrew House who said it
But yeah it wasn't some Gaffers who started that argument
One of the reasons why Sony is coming out with the PlayStation 4 Pro is to help keep gamers inside the PlayStation ecosystem instead of moving over to PC to find the high-end experience they're looking for. This is according to PlayStation president Andrew House, who shared his thoughts in a new interview.
"I saw some data that really influenced me," House told The Guardian. "It suggested that there's a dip mid-console lifecycle where the players who want the very best graphical experience will start to migrate to PC, because that's obviously where it's to be had. We wanted to keep those people within our ecosystem by giving them the very best and very highest [performance quality]. So the net result of those thoughts was PlayStation 4 Pro--and, by and large, a graphical approach to game improvement."
https://www.gamespot.com/articles/sony-explains-how-ps4-pro-may-help-keep-people-fro/1100-6443431/
Last edited:
ChiefDada
Gold Member
1. Why are you conveniently referencing raster only workload when you claimed this:
PS5 Pro RT is cleary much more performant than 3080 RT tech maybe even a smidge degree above Lovelace with talks of up to 4x PS5 RT
Codeblew
Member
I am doing both. PS5 Pro on launch, then build a PC 1+ year later when GTA6 finally releases on PC.
Gaiff
SBI’s Resident Gaslighter
Ah, yeah, my bad. Mixed them up.
Know what this makes me wonder though, what is the Pro for in that case? What is Sony's objective with it? With the PS4 Pro, it was quite clear, to stop too many players from moving to PC for their premium gaming needs and also 4K support.
With the PS5 Pro though? It's clear that their customers moving to PC is no longer a concern because they said that they would come to PC instead of trying to bring PC players to PlayStation. Xbox is a non-factor and the next-gen Nintendo console obviously won't be as powerful as even the base PS5. So what's the goal behind the Pro? It won't attract new customers or keep existing ones. It won't have high-profit margins. It won't drive subscriptions. So what's the objective?
I would hold my horses if I were you. There's absolutely no guarantee it'll be as performant as the 3080 in pure ray-traced workloads, let alone just below Lovelace. Up to 4x is the absolute max and we have to be very hopeful it'll ever translate to real-world performance. The figure said 2-3x and further (I heard up to 4x) or something along those lines. It's better to be conservative than to take the absolute maximum numbers. It's also difficult to know exactly how the ray-tracing numbers compare because we don't actually have pure ray-traced benchmarks on the regular PS5 to use as a baseline. So it's 2-4x from a relatively unknown number that I used the 6700 XT for comparison.
Again, the figures are a bit muddy because every game is a hybrid workload and it's all dependent on how much RT there is and how much the performance is impacted. It could be as little as on the level of a 2080 Ti to almost as much as a 7900 XTX in a path-traced workload. That's quite a gap.
Last edited:
sncvsrtoip
Member
quite good, rtx 3090 has 285tops tough rtx4070ti 641tops
It'll varry because of architecture but that's a little under the RTX 4060 but more that the RX 6900XT .
Killjoy-NL
Member
This is literally the point I was making.
THE:MILKMAN
Member
It's leak week.
GDC starts Monday and that is when those peeps sitting on info will empty their sacks........Be that devs or journos.
James Sawyer Ford
Member
1) It will have higher margins than the base PS5
2) It allows them to recoup R&D sooner on technology that will be integrated into future hardware like the PS6
3) it allows PS6 BC to have better quality versions than just PS5
4) Their internal data has found that Pro enables Quality modes at 60 fps so you don’t have that compromise
5) Gives their enthusiast class of gamers a better option and keeps them satisfied in the ecosystem
Those are good reasons to release the Pro, and I’m not seeing any negative reasons why you wouldn’t release the Pro
Last edited:
Topher
Gold Member
The PS5? Not really. There was something leaked to github that said PS5 had a 9tf GPU, but obviously that was wrong.
Audiophile
Member
Still think a lower-clocked Zen 4C + higher clocked 54CU setup would've made more sense, but based on more up to date rumour this is where I'm coming down at the moment..
Not by much, assuming the two shade engine thing is real. 60/64 or 52/56 CUs is the likely amount of CUs we are getting,the issue would be clock rate and the minimum frequency is looking like it might be closed to the PS5.
Celcius
°Temp. member
I feel like increasing the RAM would be a much better option than increasing the storage.Still think a lower-clocked Zen 4C + higher clocked 54CU setup would've made more sense, but based on more up to date rumour this is where I'm coming down at the moment..
Customers can easily slap in more storage space but the RAM is fixed and more advance ray tracing along with the rumored PSSR will eat more RAM.
Last edited:
Gaiff
SBI’s Resident Gaslighter
I think this was the Gonzalo or Oberon thing? Forgot the name but as I recall, it was accurate except for the clocks.
Audiophile
Member
My only thinking is that while more RAM will be welcome and the chips themselves wouldn't be prohibitively expensive, the extra 64-bit (for 18GB over 320-Bit) or 128-Bit (for 20GB over 384-Bit) GDDR6 memory interfaces on the APU would add a lot of complexity and area to the Die. Sony seem to like the simple 256-Bit setup.
Might be better to offload as much of the 2.5-3.5GB OS footprint to cheap DDR ram with a small 16-32-Bit interface and then give that extra GDDR6 RAM back to devs to use.
A usable 16GB @ 576GB/s in the Pro would still be a decent (though not ideal) bump over the useable 12.5-13.5GB @ 448GB/s in the base system.
I doubt framebuffers will get much bigger and that extra RAM would be about right for larger RT BVH structures & AI/ML upscaling workloads.
Last edited:
Kazekage1981
Member
Is the goal of PS5 to have locked 60fpsand 120fps with raytracing and Sony version of FSR using AI accelerators? Sony compares there FSR to FSR 2.0 but NOT FSR 3.0 frame generation and DLSS 3.5 (AI upscaling WITH raytracing)
In other words: 4K 60fps-120fps Sony FSR WITH Raytracing
In other words: 4K 60fps-120fps Sony FSR WITH Raytracing
thicc_girls_are_teh_best
Member
You're probably right. They are doing price cut promos in Europe ATM, but nothing permanent. That'll probably be how Sony operate going forward. Kind of expecting Nintendo to do similar (i.e no permanent price cuts for hardware, just semi-random temporary sales promos).
This is probably the new way platform holders will drum up hype and spur sales when wanted, instead of stock scarcity (the actual scarcities the pandemic caused have probably scared them from using that tactic ever again).
Ah......ok. I misreadthicc_girls_are_teh_best then. So marketing bullshit has finally made teraflops completely meaningless, it would seem.
Kinda. It sounds like we should pay more attention to the TOPs numbers as those are relevant to AI-related performance, which I think most of the Pro's targeted improvements will be leveraging. TFs are just really easy to gin up hype around.
There won't be any 30+ TF systems until the PS6. Or whatever the next Xbox system(s) go for. And both of those'll probably do closer to 40-50 TF standard.
Not like it'll matter much tho; TF will become increasingly useless for measuring game performance, and it's already kind of useless for that.
1: Don't underestimate SIE's hardware engineers, especially wizards like Mark Cerny & his team.
2: TF/raw compute are increasingly meaningless for gauging real-world gaming performance.
3: It's about the whole package, not paper spec parts here and there, that will determine the upgraded performance.
Dunno how the TOPs stuff works, but for the 33.5 TF dual-issue/16.75 TF single-issue perf the GPU only needs a clock of ~ 2.34 GHz. So a very small clock increase over PS5's.
Do you take the FP16 compute and multiply by 4 to get the TOPs (if they're 8-bit)? Because I got 268 TOPs, not 300. However I don't know if other hardware outside of the shaders are handling TOPs. If so, then it's possible they're clocked higher than the shader clock, like how in RNDA3 the shader and frontend clocks can be calibrated separately to different frequencies.
Do you think it's possible Sony increase the decompression bandwidth of the I/O? The current peak decompression rate is 22 GB/s, but what if they increased that by say 50% or even 100% ? Like you were saying, framebuffers probably won't get much bigger, but the PSSR stuff seems like it'd need 250 MB of RAM storage to function, and that's probably in addition to what is already being used by the OS.
So upping the decompression rate of the I/O block could maybe help with managing more data coming through into the framebuffer if the RAM capacity is still locked to 16 GB. It's a potential solution maybe, even if not ideal vs. simply increasing the amount of RAM. But like you and others are saying, increasing RAM capacity could require reworking memory controllers, and result in a bigger die.
GDDR6W would've been a great option if module capacity wasn't doubled at the expense of also doubling the bus width. Otherwise it's mainly only helpful here for shaving off a bit of power consumption and halving the physical footprint of memory modules on the motherboard.
Last edited:
This is just incorrect. There were massive leaks of the PS5 SoC years prior to launch (as early as the beginning 2019). The only real surprise when Sony unveiled the specs was that the GPU clock speeds were ~10% higher than anticipated. Everything else was known about.
Last edited: