The specification of Sony "PS5 Pro" game console CPU has been exposed, which is expected to be announced in September 2024.
CTOnews.com According to @ Tom Henderson, full specifications for the PS5 Pro (unnamed), code-named Trinity, are expected to leak this month as Sony has distributed development kits to third-party studios.
It is reported that this host has 30WGP, 18Gbps GDDR6 and a target release date of November 2024.
According to@Kepler_L2's July revelations, the console comes with a new AMD SoC, code-named "Viola," and its CPU upgrade is "quite small."
CTOnews.com Note: The words Trinity and Viola come from Shakespeare, where Viola is the main character in the play Twelfth Night and Trinity refers to Holy Trinity Church in Stratford-upon-Avon.
Now a member of the ResetEra forum has released some SoC data about this console and said that the machine will be announced in September 2024 (not equal to release).
According to his description, this SoC (Viola) is based on TSMC N4P manufacturing, belongs to GFX1115, maintains Zen2 architecture to maintain compatibility with the current PS5, but the peak frequency can reach 4.4GHz. Each core has 64KB of L1 cache, 512KB of L2 cache, and 8MB of L3 shared cache (4MB per CCX).
When these cores are fully enabled, each Die will have 30WGP, but the chips shipped in the retail PS5 Pro only have 28WGP enabled, or 56 CU.
Trinity, he points out, is the culmination of three key technologies in the mainframe space, including fast storage, accelerated ray tracing, and upscaling.
The GPU architecture used in this SoC is RDNA3, but it has improved ray tracing features in RDNA4, so BVH traversal will be handled by dedicated light chasing hardware without relying entirely on shaders. In addition, it will support thread reordering to reduce data and execution divergence, similar to NVIDIA Ada Lovelace SER and Intel Arc TSU.
Elsewhere, its GPU targets around 2.0 GHz and will have 3584 shaders, 224 TMUs and 96 ROPs, with 16GB of GDDR6 memory and speeds of up to 18 Gbps.(256bit bus width, 576 GB/s memory bandwidth), which makes GPU dual-threaded floating point computing power up to 28.67 TFLOPs, which is 224 (TMU)* 2 (dual-threaded computing)* 2 (GHz frequency), or 14.33 TFLOPs if the dual-threaded factor is ignored.
He also mentioned that this SoC has improved rasterization performance by 50-60% compared to Oberon and Oberon Plus, and the raw ray tracing performance has more than doubled.
On the AI side, this SoC has built-in XDNA2 NPU, which is designed to accelerate Sony's proprietary "time machine learning upgrade technology," which will also be one of the core focuses of PS5 Pro, aiming to achieve stable 4K output capability above 30 FPS.
