by With the rumor that Sony is planning to release their own upgrade technology, with the upcoming PlayStation 5 Pro, comes plenty of tidbits that purport to be what the console’s next generation hardware will be. The CPU will still be the same ol’ Zen 2 core, and the RAM a bit faster, but the GPU is being replaced by a completely new design. And it seems to be a funky mix of RDNA 3 and RDNA 4.
The leaks in question come from Inside Gaming and while they are unfounded claims, the details seem pretty believable to me. For example, the PS5 Pro’s custom AMD APU is said to retain the same Zen 2-powered CPU as the original PlayStation 5. So that’s still eight cores and 16 threads, but that’s fine for most games.
It won’t run faster, but it looks like there will be an additional operating mode, which will allow it to draw more power to achieve a 10% higher clock speed. That will come at a cost to the GPU, which would get less power in that mode, although it would only lose about 1% of its performance. Why wouldn’t Sony go with a faster CPU? The answer to that is simple: Backward compatibility. All PS5 games, and older ones that run on the platform as well, expect a CPU running at 3.5 GHz at best. A lot of things could go wrong if that were changed.
The same is true for the system RAM. Inside each PS5 is 16GB of mid-speed GDDR6, and access to it is shared by the CPU and GPU. Running at 14 Gbps on a 256-bit aggregate memory bus, a modest 448 GB/s of total bandwidth is available. It seems that the PS5 Pro will offer 576 GB / s of bandwidth and assuming that the width of the bus has changed, that corresponds to a speed of 18 Gbps.
To put those figures in perspective, the Radeon RX 6800 and RX 7800 XT have 256-bit buses, but the first one uses 16 Gbps GDDR6, but the chip is running at 19.5 Gbps for 624 GB / s of bandwidth. On paper, that would make the PS5 Pro look very nice, as the RX 7800 XT is nothing, but that GPU also packs 64 MB of Level 3 cache (aka Infinity Cache) to ease the load on the VRAM. reduction.
There is no sign of the PS5 Pro’s APU getting Infinity Cache, as AMD and Sony need the chip to be as small as possible, to keep manufacturing costs down. But since the CPU and GPU is using the same pool of RAM, that 576 GB/s of bandwidth will be hit hard. Especially when you read about the required changes to the graphics processor.
In the original PS5, the GPU is almost identical to the RDNA 2 processor, with 36 Compute Units (CUs) paired in 18 Work Group Processors (WGPs). Inside each CU are two banks of 32 ALUs, which handle all the shaders. The rumors for the PS5 Pro claim that the new GPU sports 30 WGP, the same as the RX 7800 XT, for a peak FP32 throughput of 33.5 TFLOPS.
If those two figures are correct, the only way this could happen is if the CUs are based on the RDNA 3 architecture, which means they have two banks of 64 ALUs. A quick calculation on a piece of paper puts the boost clock at 2,180 MHz, which is lower than the PS5 but the extra shaders more than make up for it.
However, we are also being told that Ray Accelerators in the CUs support BVH8 traversal shaders. BVHs are data structures, or bounded volume hierarchies used to work out which object a light ray is interacting with and in RDNA 3, the shaders work on four BVH children from each node (aka BVH4).
Increasing this to eight means that, in theory, the PS5 Pro GPU should be able to work through the interpolation process much faster than before. It could, at least, as it is not as simple as ‘the number is doubled and the performance is also better’. While there is no indication that AMD has moved these operations away from the CUs and onto dedicated hardware units (as Nvidia does with its GPUs), the change in the intersecting shaders tells us that the new GPU there. no one pure RDNA 3 but instead a hybrid RDNA 3/4.
It’s unlikely to be a full RDNA 4 design because it has no Infinity Cache, but if all the figures are to be believed, it’s more RDNA 4 than 3. The key sign for this is demand for the AI Accelerators , which first appeared in RDNA 3. In the PS5 Pro chip, they are supposed to reach 300 TOPS in INT8 mode and 67 TFLOPS in FP16 mode.
Radeon RX 7800 XT 3-power RDNA, for example, can perform 512 INT8 / FP16 operations per CU per clock cycle, on its AI accelerators, which means peaks of 75 TOPS and 75 TFLOPS respectively. For comparison, the Tensor cores in Nvidia’s GeForce RTX 3090 Ti have a peak INT8 throughput of 320 TOPS.
What would be the use of all that? AI, be it in the form of upscaling or any other task where machine learning can be applied.
Your next upgrade
Best CPU for games: The top chips from Intel and AMD.
The best gaming motherboard: The right boards.
Best graphics card: Your perfect pixel crusher is waiting.
Best SSD for gaming: Get into the game ahead of the rest.
At this stage, it is all if yes and boots, and there is no guarantee that any of this is true. But machine learning and ray tracing performance have been RDNA’s Achilles heel before, so it makes a lot of sense that AMD would improve them significantly for RDNA 4, given how important machine learning workloads are these days .
I’m unlikely to go out and buy a PlayStation 5 Pro when it launches, since I already have a perfectly decent PS5 in use in my house and a lot of gaming PCs, but from a tech writer’s point of view, it looks like the – promised him.
For GPU enthusiasts, though, RDNA 4 looks interesting: It might not top the performance charts, but it looks like it’ll have all the features you could want.
