Why the PS4 Pro won't run Destiny 2 at 60FPS

[WYP]

Many Destiny 2 fans are wondering why modern "upgraded" PS4 Pro will not be able to run the game at 60FPS, let's find out why.



Read more on Destiny 2's console performance.

 

[NeverBackDown]

Or they could you know not increase graphics and instead focus on frame rate. And offloading the physics pipeline to the GPU since it's faster at doing it anyway.

 

[grec]

"Using the same logic and looking at the specifications of Xbox's Project Scorpio, we can also see that the Scorpio will also not have enough power to run Destiny at 60FPS. Microsoft's new console will use a similar AMD Jaguar CPU to the PS4 Pro that runs at identical clock speeds, leaving the CPU once again as a limiting factor for the game. "

This isn't strictly true, it no longer really makes sense to do clock for clock comparisons between "clean" Jaguar and "Evolved" Jaguar. Besides Scorpios clocks being around 10% higher than the PS4's, the Scorpios cores also have to do considerably less work. There's far more baked in logic for unloading things like audio processing, overlays, and now GPU command processing (Often a significant bottleneck and resource hog for the CPU). Meanwhile the lower memory latencies, increased memory bandwidth, and increased memory size should all deliver notably improvements to the efficiency of the CPU. And while it'd require more work from devs, improved support for UMA and HSA means offloading floating point heavy calculations(Like physics calculations) to the GPU would now be a far more efficient and practical process.

Still for competitive/gameplay reasons I'd be surprised if devs allowed Destiny to hit 60fps on Scorpio even if doing so was trivial.

 

[NeverBackDown]

You have to prove him wrong with factually correct information instead of just saying he is with an explanation

 

[SPS]

"And while it'd require more work from devs, improved support for UMA and HSA means offloading floating point heavy calculations(Like physics calculations) to the GPU would now be a far more efficient and practical process.

Faster hardware doesn't make offloading physics to the GPU necessarily any easier. Some fundamental parts of physics don't lend themselves well to massively parallel architectures. Some parts of a physics engine even struggle with multithreading on the CPU, let alone GPU.

 

[grec]

I never said he was wrong(In fact I kinda agreed with him), just pointed out there are holes in his logic, and gave a detailed explanation of why given what we know so far of the hardware difference between "traditional Jaguar" and "Evolved Jaguar". What did I say that wasn't factual? I often see gamer "Techies"

Faster hardware doesn't make offloading physics to the GPU necessarily any easier. Some fundamental parts of physics don't lend themselves well to massively parallel architectures. Some parts of a physics engine even struggle with multithreading on the CPU, let alone GPU.

At no point did I imply faster hardware alone made it any easier. However technologies specifically designed to improve memory cohesion and simplify the creation of software that takes full advantage of the heterogeneous of the APU will. Some parts (Mostly relating to control and flow) of almost every problem are better ran on the CPU, and any interactive physics will inherently require these forms of operations. However anything involving masses of geometry can be treated as an explicitly parallel floating point model, and these lends themselves well to SIMD FP processors. So basically, yeah you're right- The fact that interactive physics operations requires a mix of both of these types of calculations means it's very hard and sometimes(Previously) impossible to do an efficient implementation(Of interactive physics) purely on the GPU, but that's also why these technologies designed to make use of multiple types of core for a single problem with minimal data juggling are inherently beneficial to it.

 

[SPS]

I never said he was wrong(In fact I kinda agreed with him), just pointed out there are holes in his logic, and gave a detailed explanation of why given what we know so far of the hardware difference between "traditional Jaguar" and "Evolved Jaguar". What did I say that wasn't factual? I often see gamer "Techies"



At no point did I imply faster hardware alone made it any easier. However technologies specifically designed to improve memory cohesion and simplify the creation of software that takes full advantage of the heterogeneous of the APU will. Some parts (Mostly relating to control and flow) of almost every problem are better ran on the CPU, and any interactive physics will inherently require these forms of operations. However anything involving masses of geometry can be treated as an explicitly parallel floating point model, and these lends themselves well to SIMD FP processors. So basically, yeah you're right- The fact that interactive physics operations requires a mix of both of these types of calculations means it's very hard and sometimes(Previously) impossible to do an efficient implementation(Of interactive physics) purely on the GPU, but that's also why these technologies designed to make use of multiple types of core for a single problem with minimal data juggling are inherently beneficial to it.

The trouble with GPU physics processing is that each wavefront thread wants to run independent of the other threads but at the same time doesn't want to diverge within its wavefront. These restrictions just make it very hard to do game-level physics. Things like particle simulation/physics are perfect though as particles do not tend to collide with each other.

Generally we're looking to move more and more stuff into GPU compute and there are some really cool things coming about since the latest generation having proper compute support. Things like mesh clustered rendering is really cool that Ubisoft did link. Or the deferred texturing that Red Lynx implemented (though this has been around for a while now). Oxide games also have quite a novel approach to rendering which they call object space lighting.

The ultimate goal is to dispatch your entire scene with once call and do all the culling/occlusion/rendering/texturing/lighting/postprocess without ever having to invoke the CPU.