Current Generation Games Analysis Technical Discussion [2020-2021] [XBSX|S, PS5, PC]
- Thread starter Shortbread
- Start date
- Status
I thought UE5 demo showed the best way to render very small triangles is with pure compute? and the rest of polygons ideally being renderered by primitive shaders (at least on PS5) ?
When it comes down to actual rendering, isn't that demo not rendering small triangles. I thought that was the point of the virtualized pipeline, to avoid rendering those pixel/subpixel sized polygons without losing much of the detail they would have provided.
These are orthogonal problems.
UE5 still very likely does deferred shading, where all shading happens on full screen quad rather than small triangle fragments like in forward.
Epic also have to calculate derivatives manually since there is no built-in function for them in compute shaders, so even if they had to process fragments in a forward fashion they still might have had some opportunities to merge quads in CS before shading them.
It definitely will, but I think the transition time should be par for the course, ignoring lockdowns related to the pandemic (which seem to be easing up/going away). We had an even larger cumulative install base going from 7th-gen to 8th-gen, and that gen went more or less fully 8th-gen by 2015.
That said I know there are market differences today vs. back then, which could make it harder to fully transition. But that just puts more pressure on Sony and Microsoft in particular to make fuller use of their own hardware with 1P content because 3P uptake could be slower than before.
Now that we talk UE5, just found out that UE5 will/can use doubles for world/scene coordinates.
Meaning it reduces/selects polygon count good for area to get near pixel sized polygons and rasterizes them using compute/software path.
Result will most likely end to G-buffer and then be shaded just like every other opaque object.
As far as we know it's compute software rasterizer with dynamic LoD.
Meaning it reduces/selects polygon count good for area to get near pixel sized polygons and rasterizes them using compute/software path.
Result will most likely end to G-buffer and then be shaded just like every other opaque object.
Last edited:
Well this is not a like for like comparison. Resolution in another but the main point should be that DX11 is really a bit limiting here. It is also funny that games still not use all the GPU memory available . Just ~4GB is used for on the RTX2070.
Loading times are not really surprising. They are already fast on the HDD of the PS4 Pro, the the NVME won't change much.
Btw, what CPU is he using for the PC? A Zen+ (2700) or what kind of CPU. This can also make a huge difference because Zen2 is much, much faster in games.
But the main point would still be DX11.
Last edited:
Are you contending that the difference in load times are attributed to D3D11 shader compilations ? To truly get an idea if that was the case we'd have to know what his testing conditions were such as if he was testing against the game in it's very first startup or subsequent startups.
Some games on startup or during multiple loading sections will issue dummy draw calls to trigger shader compilation. Others, particularly games running on UE4 will trigger shader compilations during runtime when a draw call is issued which will potentially cause frame time spikes in-game. On subsequent runs of the game, none of this matters since most shaders are compiled by the driver and are stored as cache for future reuse so the affect of shader compilation on loading times becomes nonexistent.
Great. We finally have that game benchmarked against PC GPUs. Great results there.
PS5 CPU has reduced clocks (and supposedly dynamic) and 8MB of L3 cache instead of 32MB, which is significant. We know PS5 CPU is about on par with a 1700x (mostly because of the reduced L3 cache). At 120hz the main bottleneck should be the CPU. Why should he use a high end 5ghz CPU to test such a mode? How that should be fair?
Dirextx11, you mean tools are not mature on PC?
PS5 gpu performs where it belongs, around, or a tad above 5700XT performance. Which even nioh 2 showcases. In other titles its below that, abit over etc. Around 2070 performance for non rt rendering. DLSS tech discounted altogether for normal rendering, ofcourse.
No idea why some still are hung up on comparing a late 2020 machine to a 2018 mid range RTX gpu.
No idea why some still are hung up on comparing a late 2020 machine to a 2018 mid range RTX gpu.
D
Deleted member 11852
Guest
How can there ever be a like-for-like comparison between a game on consoles and PC?
snc
Veteran
Probaby because its very popular card and nxgamer have it in his pc :d he could use 2021 rtx3060, results would be very similar
At first you should always try to balance the details so it at least almost looks like the console version. That you should be open about what options you set. And at least you should try to use the same resolution.How can there ever be a like-for-like comparison between a game on consoles and PC?
The CPU with the reduced cache is also used with the Ryzen 4xxx Mobile processors. The cache reduction didn't really make the CPU much worse. It still easily outperforms a Zen+ processor because of the much better IPC especially in games.
And the thing about DX11 is, that it still has e.g. a draw call problem. So you need a much higher clocked CPU to compensate that because most times a single thread is limiting the whole performance.
D
Deleted member 13524
Guest
Consoles at release are subsidized and have a greater economy of scale than most discrete components, so they're always a spectacular bang-for-buck in processing power compared to anything else on the market.
Unless they're consoles from Nintendo.
File sizes for Crash 4, both before and after the nextgen updates. Looks like a 10% difference (2GB) between PS5 and Series X|S. Large differences from last-gen and current-gen though, around 50%.
Details from https://www.futuregamereleases.com/...-size-cut-to-half-on-ps5-and-xbox-series-x-s/
Details from https://www.futuregamereleases.com/...-size-cut-to-half-on-ps5-and-xbox-series-x-s/
Is the PS5 more spectacular consider how the PS4 was against it's PC contemporaries, price to performance? Not sure about the other components but the SSD, at least, seem very good compare to what the PS4 had versus the most common storage on PC back then.
Either to exaggerate the ps5's value or show the currently terrible pc market of jacked up gpus prices, even some of the older cards. Maybe both.
Not considering price, but technically (were at b3d still), its about the same thing really. PS5 can be compared to about the lowest end NV and AMD have to offer now, RX6700 and 3060. Or mid range 2018-19 stuff. CPU wise, their a generation behind, basically a 3700X thats heavily cut down in clocks and cache performance. Ram quantity their sporting what a higher end GPU packs alone in total at lower speeds.
PS5's ray tracing performance is basically lower tier Turing 2060 from 2018. Advanced DLSS tech is currently absent aswell.
Yeah, you can get a PS5 for 3060's money, but you also are stuck with 3060 performance in non-ray traced games then, which is the next gen feature now and in the future. A 3060 doing RT and DLSS will outperform the PS5 at quite great lengths.
- Status
Similar threads
- Locked
