Support NeoGAF

[vkbest]

But this comes only from mild leak, I don't think Cerny or Sony ever confirmed it. Official info is "45% more power".

45% is a theoretical number. You can get lower if the game sucks or its a older game previous to PS5 pro release, and you can get more than that theoretical 45% in some procedural task or shaders where they don't stress the bandwidth. That assuming the game is not using RT

 

[Bojji]

You conviently don't agree with it and the 45% is a moddest raster number from Sony that has been already out done in some cases.
Not looking up the examples so don't bother.

Most raster games show ~30% difference, 45% it "up to" number for raster.

45% is a theoretical number. You can get lower if the game sucks or its a older game previous to PS5 pro release, and you can get more than that theoretical 45% in some procedural task or shaders where they don't stress the bandwidth. That assuming the game is not using RT

Theoretical number is 67%, 45% is closer to real life results.

 

[winjer]

The question of PS5 vs Pro ray tracing performance is not just about one metric. A lot of things have improved, some of which were significant bottlenecks for RT in the base PS5. Consider Amdahl's law for this.
The PS5 can do 4 Bounding Box tests, CU, per clock. The Pro can do 8. But the Pro uses the RDNA4, which supports Oriented Bounding Boxes, so supposedly the Pro also has this, which means fewer false positives.
So the real world BVH test performance of the Pro will be significantly higher than what the theoretical numbers suggest.
The PS5 can do only one ray-triangle intersection test, per clock cycle. The Pro can do 2 tests. This means the PS5 can theoretically do 80 billion tests per second. While the Pro can do 280 Billion.
A nice improvement, but then there is the problem with shader occupancy, which craters while doing RT. The PS5 has 128kb of L1 cache. The Pro has 256kb. This means higher L1 cache hit rates, that can allow for switching to other task faster, having fewer stalls in the shader unit.
And then there is the issue with the PS5 having the RT units coupled in the TMUs. Meaning contention between work for textures and work for ray-tracing.
There is also the problem that the PS5 doesn't have dedicated hardware to traverse a BVH tree. So it has to do it in shaders or the CPU. But the Pro has it, so it doesn't have to waste compute power on this.
The Pro doesn't have this issue. So again, it's able to reach closer to it's theoretical limit.
These are just a few examples, of how the PS5 will hit limits much sooner than the Pro.

 

[Bojji]

The question of PS5 vs Pro ray tracing performance is not just about one metric. A lot of things have improved, some of which were significant bottlenecks for RT in the base PS5. Consider Amdahl's law for this.
The PS5 can do 4 Bounding Box tests, CU, per clock. The Pro can do 8. But the Pro uses the RDNA4, which supports Oriented Bounding Boxes, so supposedly the Pro also has this, which means fewer false positives.
So the real world BVH test performance of the Pro will be significantly higher than what the theoretical numbers suggest.
The PS5 can do only one ray-triangle intersection test, per clock cycle. The Pro can do 2 tests. This means the PS5 can theoretically do 80 billion tests per second. While the Pro can do 280 Billion.
A nice improvement, but then there is the problem with shader occupancy, which craters while doing RT. The PS5 has 128kb of L1 cache. The Pro has 256kb. This means higher L1 cache hit rates, that can allow for switching to other task faster, having fewer stalls in the shader unit.
And then there is the issue with the PS5 having the RT units coupled in the TMUs. Meaning contention between work for textures and work for ray-tracing.
There is also the problem that the PS5 doesn't have dedicated hardware to traverse a BVH tree. So it has to do it in shaders or the CPU. But the Pro has it, so it doesn't have to waste compute power on this.
The Pro doesn't have this issue. So again, it's able to reach closer to it's theoretical limit.
These are just a few examples, of how the PS5 will hit limits much sooner than the Pro.

But all this is about RT calculations. With hybrid RT just part of the calculations are RT based, rest is standard raster stuff and Pro has no benefits here over standard PS5 (other than bigger GPU and higher memory BW of course). So even with 3x-4x better RT hardware you will never get 3x-4x better final results, unless your game is done in pure path tracing with no raster.

This is the same kind of nonsense MLID repeats when he talks about PS6 being 10x better in RT than PS5 - he thinks that would mean PS6 games will look/perform 10x better overall.