Digital Foundry Article Technical Discussion [2023]

[TopSpoiler]

couldn't Remedy have used the same "Primitive Shader" solution on PC, allowing the 5700XT to perform much better?

There's no API or programming model for primitive shader on PC. I think only PS5 SDK has it, which allows devs to directly program the geometry pipeline using the primitive shader.

 

[techuse]

So the PS5 GPU does have "Mesh Shaders", after all?
Alex says that Remedy uses the "similar" Primitive Shaders" on PS5. Going by the performance on PC the PS5 is closer to the 6700XT than the 5700XT. "Primitive Shaders" is just another name to the same thing, but it makes me think... couldn't Remedy have used the same "Primitive Shader" solution on PC, allowing the 5700XT to perform much better?

Primitive shaders aren't exposed to developers on any AMD GPU AFAIK. They are simply a driver side pathway AMD uses in whatever scenarios they have deemed suitable.

 

[Nesh]

It probably does deserve its own thread now so unless it's split off I won't comment on it further, but the development of the discussion from the IMO far oversimplified viewpoint that "consoles are better value because it costs more than twice as much to achieve twice the framerate on PC" to include the many other, often individualised factors that make up the real world value proposition is perfectly valid.

Sure you can divide frame rate by full system price in one game to give a result but the result of that calculation doesn't tell you anything useful about actual value or represent how most people purchase their hardware.

My initial statement was the performance of the console being pretty good for this game considering the console hardware's time of release.

The counter argument with price came by someone else mentioning simply the price of a single PC GPU.

I pointed out that when we are talking about hardware we are talking about a complete system able to run the game.

There is no point comparing the console with a single GPU. Of course you can buy a GPU at the price of a whole console that performs better than just the console's GPU. But that was never the point of the discussion or this thread.

The discussion was derailed by posters who didn't want to talk about the technical aspects of hardware, the optimization of the game and how they run the game but the value proposition who assumed the point was implying how much better consoles are. This was never the subject.

 

[mr magoo]

Good to see that remedy isn’t affraid of pushing technology even at the price of excluding older GPUs. Capcom new REX engine will be also using mesh shaders.
We deserve dense, high fidelity geometry. Pop ins and aggressive lod transitions should be relics of the past.

 

[Dictator]

I will make A 2070S Shot later in the Same area for you all - 2060S as well, Just so we can see how it scales

That would put the PS5 right around a 2080 either a bit above it and but below a 2080 Super most likely.

 

[davis.anthony]

That would put the PS5 right around a 2080 either a bit above it.

So roughly the same as the 6600XT then like my post above.

So PS5 is not actually performing any better or worse than expected as RDNA2 was always better at raster then their equivalent Nvidia GPU so it makes sense that it is where it is.

 

[Below2D]

That would put the PS5 right around a 2080 either a bit above it and but below a 2080 Super most likely.

Thank you. Right around where I expected but a bit better. Is this going to be included in your next video?

 

[Dampf]

That's pretty close to the PS5. I wonder if using DLSS instead of FSR2 could lead to enough performance to reach the PS5 target.

 

[DavidGraham]

We should alsp applaud Remedy for their stellar release quality on PC, the game suffers no crashes, no bugs, no PSO stutters, and the menu updates the visual settings live. These quality if life points are a rarity among PC releases.

 

[Below2D]

That's pretty close to the PS5. I wonder if using DLSS instead of FSR2 could lead to enough performance to reach the PS5 target.

Usually, the performance delta between DLSS and FSR is non-existent. HU did a few tests and in select cases, one may perform a bit better than the other but this is really rare. They're within 1-2fps of one another generally.

 

[troyan]

That would put the PS5 right around a 2080 either a bit above it and but below a 2080 Super most likely.

More like the 3070 is overperforming here. +57% over the 2070 Super isnt seen very offen.

 

[davis.anthony]

More like the 3070 is overperforming here. +57% over the 2070 Super isnt seen very offen.

Did Nvidia make any meaningful changes to how the 3000 series handled mesh shaders over the 2000 series?

 

[Below2D]

More like the 3070 is overperforming here. +57% over the 2070 Super isnt seen very offen.

Might be a Turing vs Ampere thing. I've seen charts where the 3070 outperforms the 2080 Ti by up to 15% in this game.

 

[Below2D]

Anyone knows where @Dictator 's comparison point is? I thought it was in the starting area following the prologue but this doesn't seem to be the case. I wanna see how a 2080 Ti performs. So far, with PS5 settings, it's around 80-85fps in this area.

 

[Dictator]

Anyone knows where @Dictator 's comparison point is? I thought it was in the starting area following the prologue but this doesn't seem to be the case. I wanna see how a 2080 Ti performs. So far, with PS5 settings, it's around 80-85fps in this area.

I have a 2080 Ti i can drop in later

 

[ILikeFeet]

 

[Below2D]

I have a 2080 Ti i can drop in later

It’s more work for you but if you don’t mind.

 

[pjbliverpool]

That would put the PS5 right around a 2080 either a bit above it and but below a 2080 Super most likely.

Awesome! I love seeing these kinds of perfectly matched comparisons.

The take away here for me is that the game is performing very well on PS5 but not outside the range of usual expectations for its hardware capabilities. The Ampere vs Turing example is interesting too. Based on the following benchmarks the
3070 is only 52% faster than the 3060 which would put the 3060 at a little faster than the 2070S (where is normally compares to the regular 2070).

So a 3060 should only be very marginally slower than the PS5 here and the 3060Ti comfortably faster, and even the 4060 being marginally faster.

 

[davis.anthony]

6600XT has nearly identical tflop performance to PS5.

So PS5 is where it would be expected to be in that data from TPU.

 

[Globalisateur]

That would put the PS5 right around a 2080 either a bit above it and but below a 2080 Super most likely.

Nice comparison. You have enabled v-sync on PC, right? Because PS5 uses V-sync and we know it costs performance.