Rumor: More PS5 Pro Specs Revealed, to Release in Fall 2024 - News
by William D'Angelo , posted on 17 March 2024 / 4,840 Views
Following a report of the PlayStation 5 Pro GPU specifications being leaked online, Insider Gaming's Tom Henderson has reportedly revealed more specifications of the mid-generation console. Henderson leaked the PlayStation 5 'Slim' before it was officially announced, including the fact it would have a detachable disc drive.
The CPU in the PS5 Pro is reportedly the same as in the standard PS5, however, it will have a "High CPU Frequency Mode" that increases the CPU power by 10 percent to 3.85GHz. This mode will downclock the GPU by about 1.5 percent.
The system memory for the PS5 will be increased by 28 percent over the standard PS5 from 448 GB/s (14 GT/s) to 576 GB/s (18GT/s). The system memory is also reportedly more efficient than the standard PS5, so potentially the bandwidth gain will be more than 28 percent.
The ACV in the PS5 Pro will run at a higher clock speed over the standard PS5, which results in the ACM library having 35 percent higher performance. There will be more convolution reverbs that can be processed, as well as more FFT or IFFT.
The the main difference with the PS5 Pro over the standard PS5 is with a more powerful GPU, which had its specs previously leaked:
- Rendering 45% faster than PS5
- 2-3x Ray-tracing (x4 in some cases)
- 33.5 Teraflops
- PSSR (PlayStation Spectral Super Resolution Upscaling) upscaling/antialiasing solution
- Support for resolutions up to 8K is planned for future SDK version
- Custom machine learning architecture
- AI Accelerator, supporting 300 TOPS of 8 bit computation / 67 TFLOPS of 16-bit floating point
The PlayStation 5 Pro is also said to have a detachable disc drive that will be identical to the PlayStation 5 'Slim,' as well as a coming with 1TB of storage. This is to make the PS5 Pro more "competitive."
The PS5 Pro is currently running on SKD 9.00, which SKD 10.0 is expected to in Fall 2024, which is the current target release window for the console.
Thanks, Insider Gaming.
A life-long and avid gamer, William D'Angelo was first introduced to VGChartz in 2007. After years of supporting the site, he was brought on in 2010 as a junior analyst, working his way up to lead analyst in 2012 and taking over the hardware estimates in 2017. He has expanded his involvement in the gaming community by producing content on his own YouTube channel and Twitch channel. You can contact the author on Twitter @TrunksWD.
More Articles
Hopefully for games that have "graphics" and "performance" modes it can run games at "graphics" quality at 60fps.
The increased CPU (even if by only 10%) and Memory bandwidth (28%+) are both things that will help with hitting 60fps and running higher resolutions (at higher fps). I think the DLSS like upscaleing with AI behinde it, and this stuff (along with a raw +45% minimum increase in gpu performance) should lead to games running higher fps.
Definitely getting this for GTA VI. This will be the most money I ever spend for a single game. Probably $600 for the Pro + $1300 for a new LG OLED to replace my 2018 model + $70 for the game. Totally worth it though. GTA games are a once a decade experience now.
45 percent faster is a very small improvement. The GPU is changing from the Radeon 6700 to the 7700 XT. It's all about the PSSR (DLSS2-3 style) upscaling. This actually makes sense since Nintendo Switch 2 will get the same upscaling at the same time. Fall 2024 is all about that upscaling I guess.
It could even be $500 since the chip is not really a lot larger and the rest of the stuff is the same.
Yes and the Xbox Series X is on par with a PS5 in most games and that only has 36 cores so that's all the proof you need that you can't just look at the GPU core count and say something sucks. You can downplay it all you want but it will be a mega hit for Sony
for sure Xbox GPU cores are not used well, if Sony can use the extra 4 years of experience to get good use out of their 60 GPU cores they will be much faster than the Xbox Series X, however I don't think this is like the PS4 Pro, as Sony says, you get 45 percent more performance, but the PS4 Pro was 120+ percent faster for the same money
Its a differnt architecture... this is like RDNA 3.5 (newer tech than the PC cards out now).
While the Xbox series X / PS5 are RDNA 2. Thats why its raytraceing performance is 3-4 times as high.
RDNA 3 has SIMD units (shaders) do 2 instructions per cycle, compaired to RDNA 2's.
So if you wanted to compair RDNA 3 vs RDNA2, its like cut it in half, and compair.
PS5pro at 33,5 Tflops fp32, is like 16,75 Tflops on rdna2 (on a ps5/xbox series x).
GPU compute is like 40% above the series x (12 vs 16,75).
First of all, you need a compiler that actually can make use of the dual shader instructions. If PC games have shown us so far, that magic ompiler simply doesn't exist yet. It may be that companies like Naughty Dog can use them but this requires hand assembler. For third party games, I don't see any improvements incoming, they neither have the time nor the experience.
This 33TFlop nonsense has to stop, plain and simple.
1) Dual-issue instructions does NOT and will never result in a doubling of performance because other bottlenecks in the pipeline come into play.
This was evident when we moved to VLIW5 to VLIW4 to GCN...
That and they become far more compiler dependent.
Remember the Radeon 7850 having almost 1 teraflop less than the 5870, still managed to out perform the 5870.
The 5870 used 5-wide VLIW, the 7850 was single issue.
https://www.anandtech.com/bench/product/511?vs=549
2) AMD's RDNA3 is about 50% better than RDNA2 in Ray Tracing, but gets a few extra tricks like hardware box sorting and traversal which gives it an edge.
It's very clock dependent of course.
3) 33.5 Terfalops is "like" 16.75" teraflops is an odd thing to throw around... It only just reinforces my position that... Unless you know what Teraflops stands for and means... (And you don't!) It's ultimately a bullshit number for spreadsheet wars.
And it really doesn't correspond with the minor performance uplift going from Radeon RX 6000 to 7000 series in the low-end and mid-range.
4) The jump from RDNA2 to RDNA3 in regular rasterization wasn't a massive generational shift, especially in the low-end and mid-range which is where consoles get their technology from.
So while you move from a Radeon 6700XT to a 7700XT with a jump from 11.29 Terafalops to 35.17 teraflops, you would expect a 3x boost to performance.
You don't. It's more like a 15% advantage.
https://www.techspot.com/review/2735-amd-radeon-7700-xt/
Sony will be relying on some clever tricks like upscaling and reconstruction to make up the difference, but that's the reality of AMD's GPU division currently.
both RDNA2 and 3 have dual issue, two instructions per clock, the difference is that RDNA3 can do TWO FP32 instructions if no INT is required, while RDNA2 pumps out INT+FP32 at the same time
the problem is game engines need to solve a lot of algebra and do arithmetic, so the INT is essential, so you can't use the dual FP32 very often (it is dual FP32 or INT+FP32, those are the options, so if you do INT calculations that second FP32 is not available)
dual FP32 is mainly for AI and compute, not rendering, that's why it didn't make much difference for gaming FPS
GPU's for years have done "two instructions per clock".
You are getting yourself confused and confusing dual-issue ALU's and dual-instructions per clock.
Basically with RDNA 3 AMD increased the number of ALU's per Compute Unit from 64 to 128... And they achieved this by widening the ALU's to dual issue instructions rather than doubling the number of ALU's themselves.
Each pipeline is still capable of two instructions per clock, which has been the norm for decades now.
It's the same approach that they did with their Terascale architectures (VLIW5 and VLIW4), but on a much smaller scale to garner extra Instruction Level Parallelism. (ILP).
Let's call it VLIW2 instead.
And obviously you have dependencies and mixed precision which means not all the pipelines in RDNA3 will ever be used as it's very compiler dependent.
You are not getting the full "teraflops" out of RDNA3 as it's simply impossible due to bottlenecks in the design.
As for FP32, it most certainly is used for rendering.
A.I tends to favour low-precision, high throughput like INT8 and FP8. (And there is a shift to INT4 and INT8 happening currently.)
It's what nVidia's Tensor cores are well geared towards.. As is AMD's new NPU "A.I" cores which are VLIW cores design for low precision integer and floats.
Feel free to do some reading:
https://www.techspot.com/article/2570-gpu-architectures-nvidia-intel-amd/
that is completely wrong, jesus
there was already dual issue, but it was limited to INT + FP32 per core
now it can do FP32 + FP32, and if you need INT, you can't do that, which is why it affects AI and compute a lot and not game engines much (which rely on a lot of INT calculations, there is no dual issue INT mode)
you did the classic thing of giving me a link that proves my opinion, not yours
did you read it?
quote: "Just as in AMD's RDNA 3, the SM supports dual-issued instructions, where each partition can concurrently process two threads, one with FP32 instructions and the other with FP32 or INT32 instructions."
that is from your article
it was not configurable before, you could only do int + FP32
now you can do FP32 + FP32, that is what changed with Ampere and with RDNA3 which allows you to claim a doubling of "teraflops" where flops refers to FLOATING POINT
if you give RDNA3 and Ampere a high load of INT + FP32 operations, it reverts back to the same general speed of Turing and RDNA2 as the INT blocks the FP32 option
DLSS doesn't exist on AMD GPU's.
You won't be getting that on the PS5.
You will be getting another solution... AMD is well known to be a generation or three behind on reconstruction and up-scaling compared to nVidia and even Intel in some instances.
I am going to assume Sony is taking checkerboard rendering and making some significant refinements and leveraging FSR to give them that extra edge.
It will be an impressive result either way.
they are not a generation behind, those words have no meaning, they are actually ahead when it comes to upscaling that doesn't use AI (for example FSR2 works great on GTX GPUS from NVidia too, not just AMD ones)
all they have to do is add AI recognition to determine the sampling depth like NVidia, not hard to do, that's all PSSR will be, it will be DLSS-like
AI is a bunch of worthless hype mostly
don't forget, the actual AI upscaler is DLSS1, outside of a few remastered games using it for offline texture upscaling the tech completely failed
here we are more than 5 years later and DLSS1 was never fixed, it still not used
DLSS2 is mostly relying on temporal upscaling and uses the AI to replace the custom code that determines sample depth over how many frames at each point of the image
it is basically using image recognition (is this a dog, is that a cat) to see TAA artifacts and adjust the sample depth at those points
AI is a tiny portion of it, there is no "creation" of information from nothing like most AI projects, like DLSS1 and all the other things AI can do, it is basically just image recognition and sample depth adjustments
It's going to be interesting how the detachable hard drive works.
Hope it’s not Still ugly, it needs to powerful enough for always 60FPS in every game. And I personally would like good quality game capture…which none of the next gen consoles have.
No console ever will give you 60fps in every game. Not even the PS8. Why? Because consoles are fixed platforms. Once theyre launched their specs stay the same, whereas games developers ambitions and consumer expectations with regards to graphics continually increase. Environmental complexity, RT, special effects, physics, resolution all can increase as a generation progresses to the point where those fixed specs are no longer sufficient to run the latest games at the highest framerates. Even on PC, there comes a point where the GPU you own cant handle a new game at the highest possible settings and you have to reduce the settings to make games enjoyable to play, or buy a newer GPU. Gotta manage your expectations.
Will definitely be 60FPS every game until the PS6 comes out. After PS6 there's no guarantees
It doesnt work like that im afraid. It'll be a mid tier GPU when it launches and then its stuck in that fixed format, whilst developer ambition and graphical complexity will continue to increase. So whilst initially it might bring more 60fps experiences to the Sony ecosystem, in more complex titles, it'll also begin to struggle just like the PS4 Pro did. Especially where like the PS4 Pro, the CPU isnt getting anywhere near the upgrade that the GPU is. Its an uneven design that could bottleneck.
It won't be 60 fps on all games, stop dreaming.
Like Sony are stupid enough to release a pro console that can't run every PS5 game at 60fps. Wow some people on here geez
Well, it depends on how the developers optimize their games.
I agree. Dragons dogma 2 is unlocked 30fps on ps5. PS5 PRO might be able to brute force it up to 45 without a patch. But with a patch and using all the technology I'm sure it won't be a prob getting 60
The PS5 Pro will be intended for 4k resolution and ray tracing. Not all games will run at 4k 60fps with ray tracing enabled. That's what I was trying to say.
Digital Foundry have confirmed the authenticity of the spec leak and had the following to say:
"In PlayStation 4 Pro, the CPU gained a 33 percent increase in CPU power with no impact to GPU performance at all - the kind of gain you would expect from a new silicon process node. The fact that the PS5 Pro is compromised in comparison does seem to suggest that Sony is doing the best it can with a 6nm process.
In real terms, those hoping that PS5 Pro will turn CPU-limited 30fps titles into super-smooth 60fps experiences will be disappointed."
https://www.eurogamer.net/digitalfoundry-2024-spec-analysis-playstation-5-pro-the-most-powerful-console-yet
Hope it has a new look, the originals are the ugliest consoles ever made.
