Support NeoGAF

[Desprado123]

We have a few clues. one is that the difference between RDNA2 and Ampere, remains similar with and without RT in Spider-man. While on other games using RT, the difference is much bigger.
The other is that the amount of data transferred between CPU and GPU, through the PCIe bus ramps up much more than on other games using RT.
And finally, because Nixxes told so to Digital Foundry, being one of the reasons why this game is so heavy on the CPU. At minute 21:35

RDNA 2 vs Ampere major difference is that FP32 X 2 of Amepre, which is why in most of recently released game it is betting it's equivalent more than 2020, 2021 AAA game release. Expect Call of Duty Modern Warfare 2, Nvidia RTX 3080 has beaten AMD's RX 6800 XT 90% of time in AAA release of 2022 ,however, in 2020, 2021 RTX 3080 was mostly equal to RX 6800. Reason is because Nvidia has more Advance Hardware than AMD expect Vram Disadvantage that is why AMD made changes and RDNA 3 is also FP32 X 2 copying Nvidia method.

 

[winjer]

So Nvidia GPUs have to be handicapped because AMD is shit a doing raytracing.

Not exactly.
It's just that this remaster was made first and foremost for the PS5. Then ported for PC.
Since consoles use RDNA2, it's normal that the console version don't have code to use units that accelerate BVH traversal.
Unfortunately, this isn't a great PC port. And Nixxes hasn't made the proper optimizations to use RT as it should.
So there are issues with asset streaming overusing the CPU and BVH still being done on the CPU.

 

[ToTTenTranz]

We have a few clues. one is that the difference between RDNA2 and Ampere, remains similar with and without RT in Spider-man. While on other games using RT, the difference is much bigger.

There's still a larger performance hit from RT on RDNA2 GPUs than RTX30 ones. The gap will always be narrower in RDNA2 console-first games if the devs follow some optimization rules published by AMD through the open-sourced raytracing analyzer:
https://gpuopen.com/learn/improving-rt-perf-with-rra/


This has nothing to do with blocking hardware accelerated stages on RTX GPUs.

And finally, because Nixxes told so to Digital Foundry, being one of the reasons why this game is so heavy on the CPU. At minute 21:35

From the timestamp you mentioned:

"I asked Nixxes why the game is so heavy on the CPU and we learned that the BVH building there is very expensive, as well as the extra cost on PC for decompressing assets from storage into memory by using the CPU".

BVH building is expensive because the amount of instances of objects coming in and out of the screen/area is larger on Spider Man than it is on other games.
There's nothing here suggesting the game is built with some magic code that block hardware accelerated stages on RTX GPUs.

Since consoles use RDNA2, it's normal that the console version don't have code to use units that accelerate BVH traversal.

There is no such thing as "coding to not use units that accelerate BVH traversal".

 

[Kenpachii]

And here is the vram usage for Spider-man Miles Morales. Using a slightly improved engine.
Just look at the 4090, with 24Gb of vram, only using 9.1Gb, with RT at 4K.
So once again, the RTX 3080 does not have it's performance limited by vram.

Ofcourse v-ram isn't a issue, if the v-ram was a issue u would see single digits, no clue why anybody though it would be.

There's still a larger performance hit from RT on RDNA2 GPUs than RTX30 ones. The gap will always be narrower in RDNA2 console-first games if the devs follow some optimization rules published by AMD through the open-sourced raytracing analyzer:
https://gpuopen.com/learn/improving-rt-perf-with-rra/


This has nothing to do with blocking hardware accelerated stages on RTX GPUs.




From the timestamp you mentioned:

"I asked Nixxes why the game is so heavy on the CPU and we learned that the BVH building there is very expensive, as well as the extra cost on PC for decompressing assets from storage into memory by using the CPU".

BVH building is expensive because the amount of instances of objects coming in and out of the screen/area is larger on Spider Man than it is on other games.
There's nothing here suggesting the game is built with some magic code that block hardware accelerated stages on RTX GPUs.



There is no such thing as "coding to not use units that accelerate BVH traversal".

I could see that, u move through the city pretty rapidly so loads of loading and deloading stuff specially at higher textures. Add RT with that and CPU will be crippled. Still they could have opted to allocate more v-ram/ram towards the game to deal with that problem in some form.

 

[winjer]

There's still a larger performance hit from RT on RDNA2 GPUs than RTX30 ones. The gap will always be narrower in RDNA2 console-first games if the devs follow some optimization rules published by AMD through the open-sourced raytracing analyzer:
https://gpuopen.com/learn/improving-rt-perf-with-rra/

This has nothing to do with blocking hardware accelerated stages on RTX GPUs.

The game without RT runs better on Ampere than on RDNA2.
And with RT that Ampere advantage increase just a bit. But not to the amount it usually does in RT games.
If this game was fully utilizing the BVH units in Ampere, then the performance gains would be much greater.

From the timestamp you mentioned:

"I asked Nixxes why the game is so heavy on the CPU and we learned that the BVH building there is very expensive, as well as the extra cost on PC for decompressing assets from storage into memory by using the CPU".

BVH building is expensive because the amount of instances of objects coming in and out of the screen/area is larger on Spider Man than it is on other games.
There's nothing here suggesting the game is built with some magic code that block hardware accelerated stages on RTX GPUs.

Yes, there is more to render. But that is not just the issue.
Like Nixxes said, while specifically talking about the CPU, "the BVH building there is very expensive"

If it was just a matter of having to render more of the world, for reflections, it would double the world FOV and double the amount of things to render.
But we get 5 times more data transfers with RT enabled, going through the PCIe Bus. It's not just more draw calls and asset streaming. It's also the BVH struture being calculated on the CPU, then sent to the GPU.

There is no such thing as "coding to not use units that accelerate BVH traversal".

The game was coded first for consoles, that don't have units to accelerate BVH traversal. And when they ported the game, they kept it that way.