snc
Veteran
yeah maybe but suggestion was that we saw performance of ps5 in this 5700 benchmark, false logic
Digital Foundry Article Technical Discussion [2021]
- Thread starter BRiT
- Start date
- Status
All things considered, the 6800XT being atleast twice as capable isnt all that hard of a logic to understand. Its all theres to it, i hope you didnt have any higher expectations. The 6800XT will get its optimizations the consoles have gotten, its twice as fast just as the specifcations imply. Dedicated BW, infinity cache, 16GB and full set rdna2 features will help this great AMD product aswell
Its a impressive GPU from AMD for UE5 for sure.
And yes, the PS5 gpu isnt far off from 5700XT rendering performance, their close. RDNA1 to 2 didnt see any or much of IPC improvements.
snc
Veteran
6800xt its around 1.85x faster than 5700xt (techpowerup) so its not that I neggate it but just pointing out that you logic is simple false or just trolling as usual in your post, we don't know how ps5 behave in this benchmark, eotAll things considered, the 6800XT being atleast twice as capable isnt all that hard of a logic to understand. Its all theres to it, i hope you didnt have any higher expectations. The 6800XT will get its optimizations the consoles have gotten, its twice as fast just as the specifcations imply. Dedicated BW, infinity cache, 16GB and full set rdna2 features will help this great AMD product aswell
Its a impressive GPU from AMD for UE5 for sure.
https://www.eurogamer.net/articles/...0wF94gLkBtkKW1DT7lbQYJm6g0fvHzlu65LD7uKR3VLCk
6800XT 20.7TF, PS5 10.2TF. Scales very well in UE5 as shown.
6800XT 20.7TF, PS5 10.2TF. Scales very well in UE5 as shown.
5700XT is not supporting Direct X12 feature set. It means on Nanite side it will not improve when they will implement mesh shader on PC. PS5 GPU is not as the same feature set than 5700 XT and like always PS5 API is lower level than PC one. It means the PS5 GPU is above 5700 XT out of effect demanding high bandwidth memory where the dev need to use lower res effect. The memory bandwidth is shared between CPU, GPU and tempest engine.
For people who don't belive it, Quantic Dreams told even Direct X 12 or Vulkan aren't at same level than console APU.
https://gpuopen.com/learn/porting-detroit-1/
snc
Veteran
thats not the point :d we just cant compare if we dont know consoles results with uncapped frame in this benchmark, also we should compare with similar cpu, simple as that, is ps5 behave in ue5 as 5700xt?possible, also possible it has better optimisation
Again This is impossible 5700XT will perform as good as PS5 GPU because 5700XT does not support at least officially mesh shader. This is one optimization they will port from consoles mesh/primitive shader to PC and this optimization doesn't apply to 5700 XT, RDNA2 and NVIDA 2000 and 3000 series will improve but not the AMD 5000 series.
After we don't know uncapped framerate of consoles I agree and same we need to have a comparable CPU too.
snc
Veteran
Wouldnt be so sure but again it wasnt point of my posts
I have heard from a little bird that internally, Sony calls them mesh shaders, but as a part of their documentation lists out in very plain text the areas where it is deficient vs. the full spec mesh shading provided by later architectures (aka what you see on PC ampere, turing or RDNA 2 like in XSX or RX 6XXX)
D
Deleted member 11852
Guest
What's Sony's designation, NetShadersEXTREME(tm). ?
It is actually "Mesh Shaders"
Wouldnt be so sure but again it wasnt point of my posts
Mesh/primitive shader is faster than vertex shader and for the hardware path 5700 XT will use vertex shader. This is 100% sure. Nanite will be slower on 5700XT because Unreal Engine 5 use Mesh/Primitive shader. If you do your comparison using current PC game like if the PS5 GPU was a 10.28 Tflops 5700 XT with raytracing, this will not be the case for title using new features.
Geometry improve a lot compare to RDNA 1 GPU and it has raytracing and maybe other custom stuff.
Ah so it has not the full mesh shading specifications as opposed to Turing and upwards, RDNA2 and XSX. Thanks for the headsup!
Hence no one should buy a 2 year old mid range GPU whenever their available again. 6800XT twice the PS5s performance in UE5, its a very good contender to NV stuff in UE5.
On Navi 10, there's a hardware bug where the lowest and highest vertex index for each primitive has to be within 32 of each other. Unknown if this was ever fixed in other Navi1x revisions. Also Navi 10, is missing the capability to output per-primitive attributes for pixel shaders. It would be difficult for Navi 10 to respect the minimum D3D implementations limits ...
If it makes the others feel any better "mesh shaders" on other APIs like OpenGL (and likely Vulkan too) are incompatible with D3D implementation limits as well. "MAX_TASK/MESH_WORK_GROUP_SIZE_NV" specifies that the dimensions of components x, y, or z must not exceed the value of 32. D3D doesn't have this limitation at all where the dimensions for components x, y, or z can be above 32 with [numthreads(X, Y, Z)] so this means that [numthreads(1, 1, 128)] would be a valid configuration on D3D while it would invalid on other APIs. On other APIs the total groupshared memory (MAX_TASK/MESH_TOTAL_MEMORY_SIZE_NV) consumed by the task or mesh shaders must strictly be less than 16kb for each while on D3D it can be upto 32kb for amplification shaders or upto 28kb for mesh shaders provided that both of their combined memory doesn't go above 47kb ...
Heck, if we looked more into the Vulkan extension it was so close to being given the moniker "VK_NV_primitive_group_shading" which coincidentally sounds awfully similar to a feature offered by a competing vendor ...
If it makes the others feel any better "mesh shaders" on other APIs like OpenGL (and likely Vulkan too) are incompatible with D3D implementation limits as well. "MAX_TASK/MESH_WORK_GROUP_SIZE_NV" specifies that the dimensions of components x, y, or z must not exceed the value of 32. D3D doesn't have this limitation at all where the dimensions for components x, y, or z can be above 32 with [numthreads(X, Y, Z)] so this means that [numthreads(1, 1, 128)] would be a valid configuration on D3D while it would invalid on other APIs. On other APIs the total groupshared memory (MAX_TASK/MESH_TOTAL_MEMORY_SIZE_NV) consumed by the task or mesh shaders must strictly be less than 16kb for each while on D3D it can be upto 32kb for amplification shaders or upto 28kb for mesh shaders provided that both of their combined memory doesn't go above 47kb ...
Heck, if we looked more into the Vulkan extension it was so close to being given the moniker "VK_NV_primitive_group_shading" which coincidentally sounds awfully similar to a feature offered by a competing vendor ...
Still there on PS5 GPU TMK - presumably the other as well.
I think you're missing the point I made which is that some features that you find in one API can be somewhat "loosely defined" in general so that other APIs can take liberties in making different abstractions to expose a similar capability ...
I need to find the tweet but I think a dev told he thinks a nearly similar functionnality than Mesh shader can exist on RDNA 1 supporting some Vulkan extension.
Last edited:
Interesting thanks.
I think we've surmised that PS5 is lacking hardware VRS and differences in mesh shader capabilities.
What would be nice to know is what the difference are, in the end may make no difference in real world or even development considerations, or it could....
Don't suppose birdy gave you any indications?
- Status
Similar threads
