AMD Vega 10, Vega 11, Vega 12 and Vega 20 Rumors and Discussion

I've been reading up on the basics of how MSAA works, and as far as I can understand this as a beginner, doesn't MSAA involve rasterizing at a higher resolution than the intended display resolution?

That's SSAA - Super Sampling.
 
That's SSAA - Super Sampling.
As far as I understand the introduction to MSAA that I read, SSAA involves doing the complete render at a higher resolution than the intended display resolution, not just the rasterization stage, as in MSAA. Here's the relevant passage I was basing my understanding on:
In terms of rasterization, MSAA works in a similar manner to supersampling. The coverage and occlusion tests are both performed at higher-than-normal resolution, which is typically 2x through 8x. For coverage, the hardware implements this by having N sample points within a pixel, where N is the multisample rate.
.
Is this not correct?
 
As far as I understand the introduction to MSAA that I read, SSAA involves doing the complete render at a higher resolution than the intended display resolution, not just the rasterization stage, as in MSAA. Here's the relevant passage I was basing my understanding on:
.

You said "rasterizing at a higher resolution" which is SSAA.
 
As far as I understand the introduction to MSAA that I read, SSAA involves doing the complete render at a higher resolution than the intended display resolution, not just the rasterization stage, as in MSAA. Here's the relevant passage I was basing my understanding on:
.
Is this not correct?

MSAA still only produces 1 shaded sample (pixel shader), the 1 sample is then masked at a higher resolution (coverage), it really is only about contribution.

MSAA is a game setting. It's pretty straightforward as the antialiasing setting lights up as e. g. "MSAA 2x".

Seems not to work then.
 
I've been reading up on the basics of how MSAA works, and as far as I can understand this as a beginner, doesn't MSAA involve rasterizing at a higher resolution than the intended display resolution? In which case, wouldn't MSAA performance be limited by rasterization rate? Doesn't that mean that Vega's presently being limited to four triangles per clock would negatively impact MSAA performance?
As Ethatron said with MSAA depth is rendered at a higher sample rate than the pixel rate. Sample rate is different than the final screen resolution. If anything a higher sample rate makes primitive rate be less of a factor.
 
I've been reading up on the basics of how MSAA works, and as far as I can understand this as a beginner, doesn't MSAA involve rasterizing at a higher resolution than the intended display resolution? In which case, wouldn't MSAA performance be limited by rasterization rate? Doesn't that mean that Vega's presently being limited to four triangles per clock would negatively impact MSAA performance?
MSAA would be limited to edges and transparency (chain link fence, vegetation) in practice while SSAA does the entire scene, essentially downsampling the entire frame. The latter obviously entails far more shading work.

The coverage and samples would be evaluated within a CU(s) per triangle/workgroup and not significantly affect triangle rate. With three vertices it's just a matter of interpolation to establish coverage.
 
To be honest in this game it's hard to tell. With everything cranked up to max the post effects hide the aliasing very well and TAA eliminates the little that is left. The difference between turning MSAA off or 8x is pretty much negligible in those conditions.
I looked around for a bit in the game's last level, but I couldn't find any transparent texture by myself (which is where MSAA would be more noticeable).

This is the best that I could do by turning off TAA, DoF bloom and lens flares:

I'll be happy to follow instructions on what to look for, or what graphics options to turn on/off to see this more clearly.
Just an advice -- screenshots are best evaluated in PNG format, there's too much compression artifacts in yours. But nonetheless, the MSAA does some work here, but certainly not worth the performance hit.

Vega just got R600-ed and good thing MSAA is less of importance now than ten years ago. UHD displays and a decent post-process AA (SMAA) is good-enough for the masses.
Supersampling for the rest.
 
Does Deus Ex MD support setting MSAA in game or is this forced in the driver?
Does AMD driver support MSAA for DirectX 12 games from control panel? Direct3D 12 does not support MSAA swap-chain like previous version, which made driver MSAA settings more difficult to properly implement, especially with the control of the frame buffers Microsoft give to developers now...
 
Seems not to work then.
It's definitely working.
Its usefulness is just not there, though. As I wrote before, I tweaked the IQ settings as much as possible to bring the jaggies out. With TAA + bloom + DoF these practically go away.



Just an advice -- screenshots are best evaluated in PNG format, there's too much compression artifacts in yours.
I know, but at this resolution the original PNG screenshots are 11MB big, which I wasn't able to upload to imgur nor B3D. I can't see much of a difference in detail between the screenshots between the PNG and JPG versions, though.
Regardless, here's the left half of the screenshot that I could upload to imgur:

No AA
lYJnwmL.jpg



MSSA 8x
BphchWg.jpg



Vega just got R600-ed and good thing MSAA is less of importance now than ten years ago. UHD displays and a decent post-process AA (SMAA) is good-enough for the masses.
Supersampling for the rest.
According to some (at least Valve), MSAA is pretty important for VR.
I don't know if this problem is permanent or it'll be solved in a driver update, but considering all the stuff that's still not working with Vega I think it's the latter.
There's a multitude of features that are more important to enable before fixing MSAA though, so I don't think this is on their priority list.
 
Last edited by a moderator:
good thing MSAA is less of importance now than ten years ago.
It's still somehow relevant to this day. Many DX12 titles -for example- have MSAA support: Forza 6, Forza Horizon 3, Civilization 6, Warhammer, Ashes Of Singularity, Deus Ex Mankind Divided and the upcoming Forza 7 as well. In addition to many DX11 titles.
 
It's definitely working.
Its usefulness is just not there, though. As I wrote before, I tweaked the IQ settings as much as possible to bring the jaggies out. With TAA + bloom + DoF these practically go away.




I know, but at this resolution the original PNG screenshots are 11MB big, which I wasn't able to upload to imgur nor B3D. I can't see much of a difference in detail between the screenshots between the PNG and JPG versions, though.
Regardless, here's the left half of the screenshot that I could upload to imgur:

No AA
lYJnwmL.jpg



MSSA 8x
BphchWg.jpg




According to some (at least Valve), MSAA is pretty important for VR.
I don't know if this problem is permanent or it'll be solved in a driver update, but considering all the stuff that's still not working with Vega I think it's the latter.
There's a multitude of features that are more important to enable before fixing MSAA though, so I don't think this is on their priority list.

I see just as much aliasing on both pictures; doesn't seem to be working at all.
 
Just an advice -- screenshots are best evaluated in PNG format, there's too much compression artifacts in yours. But nonetheless, the MSAA does some work here, but certainly not worth the performance hit.

Vega just got R600-ed and good thing MSAA is less of importance now than ten years ago. UHD displays and a decent post-process AA (SMAA) is good-enough for the masses.
Supersampling for the rest.
i think its better to wait and see the october driver and then we can all bash or not amd i havent seen a gcn card that could cope with high levels of msaa in general so nothing really have changed so far
 
There is MSAA (or something that looks like MSAA). The woman with the lightly coloured legs has strong aliasing on the edges of her legs in the top screenie and much less in the second screenie.

The large scale jaggies seen on the bright parts of the giant letters where they meet the dark parts of the scene are almost certainly a bug in the way that the game renders "HDR" and performs tonemapping.

Tonemapping and anti-aliasing disagree with each other. It's only correct to tonemap first then perform the anti-alias resolve and to do all these operations in linear space, not gamma space.

A good approximation can be determined:

http://theagentd.blogspot.co.uk/2013/01/hdr-inverse-tone-mapping-msaa-resolve.html

See the end, which links back to a very old discussion here on B3D.
 
MSAA still only produces 1 shaded sample (pixel shader), the 1 sample is then masked at a higher resolution (coverage), it really is only about contribution.
As Ethatron said with MSAA depth is rendered at a higher sample rate than the pixel rate. Sample rate is different than the final screen resolution. If anything a higher sample rate makes primitive rate be less of a factor.
The coverage and samples would be evaluated within a CU(s) per triangle/workgroup and not significantly affect triangle rate. With three vertices it's just a matter of interpolation to establish coverage.

Forgive my confusion, but I'm still very much a beginner at understanding this stuff. Is the rate at which a GPU performs coverage and occlusion tests not limited by its rasterization rate then? What is the limiting factor in the rate at which these tests can be performed by a GPU then?
 
Forgive my confusion, but I'm still very much a beginner at understanding this stuff. Is the rate at which a GPU performs coverage and occlusion tests not limited by its rasterization rate then?
I'm not sure, IIRC rasterization works on tiles. I'm remembering 64x64 tiles, don't remember from where (might be IHV dependent). If a triangle is bigger than 64x64 it takes multiple clocks to rasterize. What I don't know is if it's 64x64 samples or pixels. Meaning if its 64x64 samples then if lets say 4x msaa is enabled then its effective size is 16x16. If its 64x64 pixels there is no loss in tile size. Maybe someone with more knowledge than me can shed some light on this.

edit - maybe it was 8x8, my memory is playing tricks on me.
 
Last edited:
The large scale jaggies seen on the bright parts of the giant letters where they meet the dark parts of the scene are almost certainly a bug in the way that the game renders "HDR" and performs tonemapping.

Tonemapping and anti-aliasing disagree with each other. It's only correct to tonemap first then perform the anti-alias resolve and to do all these operations in linear space, not gamma space.

There's a sharpening filter option in the game. Since it's a post effect it could be introducing noise (that's mistaken by aliasing?) to a properly antialiased scene. Maybe that's part of the cause?
I could disable it and check again how it looks.
 
Does AMD driver support MSAA for DirectX 12 games from control panel? Direct3D 12 does not support MSAA swap-chain like previous version, which made driver MSAA settings more difficult to properly implement, especially with the control of the frame buffers Microsoft give to developers now...
I don't know as I haven't played with MSAA in a while. I knew forcing in the driver could be less efficient though, hence the question.

Forgive my confusion, but I'm still very much a beginner at understanding this stuff. Is the rate at which a GPU performs coverage and occlusion tests not limited by its rasterization rate then? What is the limiting factor in the rate at which these tests can be performed by a GPU then?
The rate at which a GPU performs coverage (depth) testing is limited by its rasterization rate though GPUs have different rates for depth than color. Depth rates are typically 2x or 4x faster than color.

Let's consider Vega10. It has a color rasterization rate of 64 pixels/clock with each shader engine being 16 pixels/clock. Any time a triangle covers > 16 pixels it will take 2 cycles to process in one of the rasterizers or it will be split across rasterizers. Any time either of these happens peak geometry rate is no longer achieved. This situation happens quite often, more so at higher resolutions.

As Infinisearch said the screen is split into tiles with each rasterizer owning some pattern of tiles. This determines if a triangle is sent to multiple rasterizers. Even a 2 pixel triangle can be sent to 2 rasterizers if those 2 pixels happen to cross a tile boundary. It doesn't need to be a large tile. You also can't assume that an IHV always uses the same tile size as this could change based on the chip config.
 
Do we know if there are differences between rx drivers and vega fe drivers (gaming mode) at the moment (since they're at the same version 17.8.2 right now) ?
 
There's a sharpening filter option in the game. Since it's a post effect it could be introducing noise (that's mistaken by aliasing?) to a properly antialiased scene. Maybe that's part of the cause?
I could disable it and check again how it looks.
If you want to find out about MSAA performance, you might want to try another game. In fact, try lots of games. And various resolutions.

This rapidly becomes a hell of a lot of effort with little meaningful payback. Especially if a driver fix arrives with no explanation. A mostly thankless task for the tester.

This is prolly why some journalists decided simply to troll and hope it goes viral on various Vega-related topics: way less effort and it might just result in a meaningful response - and boy look at that advertising revenue. The entitled millennial generation has infected nearly everyone else, it seems.
 
Back
Top