EXCLUSIVE – More PlayStation 5 Pro Specs Detailed

Following this week’s leaks on the PlayStation 5 Pro GPU specs and performance targets, Insider Gaming has learned more PlayStation 5 Pro specs. For comparison sake, we’ll also include the specifications of the Standard PlayStation 5.

System Memory

Standard PlayStation 5 – 448 GB/s (14 GT/s)
PlayStation 5 Pro – 576 GB/s (18GT/s) – A 28% increase over the standard console.

Also outlined is that the PlayStation 5 Pro’s system memory is more efficient than the standard console, so the bandwidth gain may increase by over 28%.


The CPU is identical to the standard PlayStation 5, however, the Pro has a ‘High CPU Frequency Mode”, which takes the CPU to 3.85GHz – A 10% increase over the standard console.

In High CPU Frequency Mode, more power is allocated to the CPU and will downclock the GPU by around 1.5%, resulting in roughly 1% lower GPU performance.


The ACV in the PlayStation 5 Pro runs at a higher clock speed over the standard PlayStation 5, resulting in the ACM library having 35% more performance.

  • More convolution reverbs can be processed
  • More FFT or IFFT can be processed

GPU (Previously revealed)

  • 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

In addition: 30 WGPs running specialised BVH8 traversal shaders vs 18 WGPs running BVH4 tranversal shaders on the standard PlayStation 5.

It’s also understood that as a means to make the PlayStation 5 Pro as “competitive” (not my phrasing here) as possible, it will have a detachable disc drive which will be identical to the latest iteration of the standard PlayStation 5, and 1TB of storage space.

Currently, the PlayStation 5 Pro is running on SKD 9.00, and SKD 10.00 is expected in Fall 2024, which is the current targeted release date of the console.

Header Image Credit – https://twitter.com/technizoconcept

For more from Insider Gaming, check out our exclusive news on the next Battlefield game, which will have a free-to-play Battle Royale connected to it

  1. Will it have a built in disc drive or will it be like the slim model with a wifi drive?

  2. “AI Accelerator, supporting 300 TOPS of 8 bit computation / 67 TFLOPS of 16-bit floating point”

    300 / 67 = 4.47

    You can’t divide 300 / 67 without fraction. It is not possible to build hardware where single 16-bit operation is converted to 4.47 8-bit operations

    1. It’s “per second”. The 16-bit operations don’t map perfectly to 8-bit operations. It’s not doing 2×8-bit=16-bit. They have additional hardware in the 16bit operations, resulting in additional signal latency and non-linear scaling. That’s also why 16-bit is so much slower and independent from 8-bit speed.

      1. “The 16-bit operations don’t map perfectly to 8-bit operations. It’s not doing 2×8-bit=16-bit”

        No, this is atomic operation. Compute unit have 32-bit register where you can “insert” single 32-bit number, two 16-bit numbers, four 8-bit numbers or eight 4-bit number. Whole 32-bit register will be full of 0/1 bits. Then based on command GPU will interpret those 32-bits as single number, two numbers, 4 numbers or 8 numbers. This is atomic operation that will return 1-8 results based on selected command

    2. Not it likely means that it also has a separate NPU to accelerate AI with 300 tops of performance and the 67 tflops is what the GPU itself can do in 16 bit mode

    3. why do you keep thinking this is exclusively on the GPU? this is an APU, a custom one that can have any mix of IP blocks each running at any frequency

      i can easily imagine a Xilinx block that can only do int8 supplementing the RDNA block, so (67*2=)135 plus a separate 165, and that’s assuming no change to clockspeed

      phoenix already did this a year ago, it’s called RyzenAI for a reason, not RadeonAI

Comments are closed.