Thursday, August 7th 2025

Intel to Drop 16x MSAA with Xe3 "Celestial," Says XeSS and TAA Make Legacy AA Methods Obsolete

Intel has signalled that it will drop support for 16x MSAA with its upcoming Xe3 "Celestial" graphics architecture. The architecture will support MSAA, but at 2x, 4x, and 8x. MSAA x16 support will be reserved for certain models of Xe3 GPUs, presumably the higher end discrete GPUs, however the lower-end chips and iGPU solutions are set to lose it. Intel Graphics engineer Kenneth Grauke made a commit to Intel's Iris Gallium3D and ANV Vulkan driver, and announced in his note: "16x MSAA isn't supported at all on certain Xe3 variants, and on its way out on the rest. Most vendors choose not to support it, and many apps offer more modern multisampling and upscaling techniques these days. Only 2/4/8x are supported going forward."

Intel's rationale behind retiring MSAA checks out, as nearly every modern game supports TAA, and most games support at least one upscaling technology, such as DLSS, FSR, or Intel's own XeSS. Intel has a growing collection of games that have implemented XeSS and the newer XeSS 2 technology, which features an AI ML-based upscaling model. Upscaler super-sampling methods tend to do a good job at antialiasing, with whatever minimal performance cost made up for with upscaling. Even native-resolution modes, such as DLAA see negligible performance costs compared to MSAA. GPU manufacturers would still want to retain MSAA for its utility in older gaming titles and 3D productivity apps.
Sources: Phoronix, VideoCardz
Add your own comment

61 Comments on Intel to Drop 16x MSAA with Xe3 "Celestial," Says XeSS and TAA Make Legacy AA Methods Obsolete

#51
Prima.Vera
To be honest, unless you play on 640x480 resolution, on an old CRT, there is no point of having more than 8xMSAA anyways. There is also the option of running 8xSSAA, which is miles better. Nowadays we do have DLAA, however is still way to blurry and crappy to be honest, even if is best on performance.
Btw, nGreedia has some good AA options for MSSAA, such a 8xEQ MSAA, which basically gives you 16xMSAA for the same performance impact as 8x, however a little more blurry also.
Posted on Reply
#52
ZeDestructor
Space Lynxumm, some of us want to play older games from time to time...
Sfine - get the resolution unlimiters and crank it to to 4x (linear, 16x the pixel count) of native res and get the equivalent of 4x FSAA/SSAA. Or go even higher if you want even more sampling, but IME at 100ppi 2.5K (1440p) 4x is enough, and now I've moved up to a 200ppi 5K (2880p) monitor for *even more* detail. The wait for 300ppi (8K) on desktop continues unabatedly...
OnasiYou can SORT OF make it work on transparencies by rendering stuff like alphas with an offset and AAing it (that is what the Transparency setting for AA in NVCP does), but it’s really performance heavy and, from my understanding, works only under forward rendering. Which, contrary to popular belief, is still theoretically possible on modern engines - UE4 (and I assume 5, but don’t quote me on that) do have a forward rendering configuration as an option, though good luck finding any games running THAT. Realistically, it’s a DX9/10 era games setting.
Very much this. Sad as it is, MSAA as a whole is pretty dead in modern deferred-rendering games.
Prima.VeraTo be honest, unless you play on 640x480 resolution, on an old CRT, there is no point of having more than 8xMSAA anyways. There is also the option of running 8xSSAA, which is miles better. Nowadays we do have DLAA, however is still way to blurry and crappy to be honest, even if is best on performance.
Btw, nGreedia has some good AA options for MSSAA, such a 8xEQ MSAA, which basically gives you 16xMSAA for the same performance impact as 8x, however a little more blurry also.
DLAA is absolute garbage based on the little testing I've done with Path of Exile 2 (particle effects? never 'eard of 'er). Better than the upscalers, but I legitimately prefer no AA to DLAA.

Overall, I have to admit that I find FXAA/MLAA to be the best balance between performance and image quality, very comparable to 4x MSAA/SSAA/FSAA once you get the resolution up into the 2.5K (1440p) and beyond resolutions but at a fraction of the performance requirement. In a world where 4K is pretty much the default at the high end, and 5K and 6K monitors are becoming a bit of a thing (including some 120Hz and OLED options being announced).
Posted on Reply
#53
Dwarden
ZoneDymoThis is just a silly post, you even acknowledge how silly the first remark is yourself.
Its like people being against homosexual marriage because "whats next? a man marrying a chair?!!?!"

Everyone will stop supporting MSAA at some point, its an ancient tech, its crazy you specifically mention "Cease to innovate" when sticking to old tech is exactly that....
oh so you want to tell me temporal fake AA, imaginary edges and random in-between frames hallucination is innovation ? 🤣
Event HorizonDid you read the body of the OP? They're maintaining x16 MSAA support on the higher end discrete GPUs.
and you read my reasoning ? because if you start to chunk down what was standard specification into 'layered' more expensive variants

then you end with hardware as service features like buying unlocking more cache / speed coupons from Intel ... {that already did happened}
so now it will be unlocking MSAA or whatever ...

the fact that you just shrug off that standard feature is now 'pay for premium' shows how twisted the market is ...

if the problem is performance on too high resolutions, they could just easily limit it for those or include warning in papers that ie 4 or 8k with 16x MSAA may end as slideshow
but for me this just sounds more like excuse how gate default feature into more sales of 'whatever is more expensive'
ZeDestructorSfine - get the resolution unlimiters and crank it to to 4x (linear, 16x the pixel count) of native res and get the equivalent of 4x FSAA/SSAA. Or go even higher if you want even more sampling, but IME at 100ppi 2.5K (1440p) 4x is enough, and now I've moved up to a 200ppi 5K (2880p) monitor for *even more* detail. The wait for 300ppi (8K) on desktop continues unabatedly...



Very much this. Sad as it is, MSAA as a whole is pretty dead in modern deferred-rendering games.



DLAA is absolute garbage based on the little testing I've done with Path of Exile 2 (particle effects? never 'eard of 'er). Better than the upscalers, but I legitimately prefer no AA to DLAA.

Overall, I have to admit that I find FXAA/MLAA to be the best balance between performance and image quality, very comparable to 4x MSAA/SSAA/FSAA once you get the resolution up into the 2.5K (1440p) and beyond resolutions but at a fraction of the performance requirement. In a world where 4K is pretty much the default at the high end, and 5K and 6K monitors are becoming a bit of a thing (including some 120Hz and OLED options being announced).
also ironically i still consider properly implement and tuned SMAA 2.x (especially to avoid UI and blur issues) to be better than FXAA and MLAA ...
Posted on Reply
#54
ZeDestructor
Dwardenoh so you want to tell me temporal fake AA, imaginary edges and random in-between frames hallucination is innovation ? 🤣


and you read my reasoning ? because if you start to chunk down what was standard specification into 'layered' more expensive variants

then you end with hardware as service features like buying unlocking more cache / speed coupons from Intel ... {that already did happened}
so now it will be unlocking MSAA or whatever ...

the fact that you just shrug off that standard feature is now 'pay for premium' shows how twisted the market is
I certainly do, but I also remember that 16x-64x AAs were introduced at a time where 1024×768 (16" viewable) to 1600×1200 (18-19" viewable) was the norm. These days I have a 5120×2880 27" monitor as my main monitor, with an eye on 6144×3456 to 7680×4320 32" panels as potential upgrades.

Compared to the 2560×1440 100ppi "base resolution" equivalents, I'm already doing 4x FSAA/SSAA by just running native 5120×2880, and adding 4x MSAA/FSAA/SSAA on top would be equivalent to 16x. When (not if!) I do manage to find a 7680×4320 27-32" monitor that's not too compromised (the Dell UP3218K has it's issues), I'd be looking at 9x FSAA/SSAA by just running native (compared to a 2560×1440 baseline), and 4x MSAA/SSAA/FSAA would bumped that to 36x equivalent.

Overall, I say as long as the 4x modes stay. we're good, and it seems that all the graphics engineers agree (even if they don't explicitly say so).
Posted on Reply
#55
Dwarden
windwhirlThey're not removing MSAA. They're dropping only the 16x MSAA.

I haven't played every game on Earth, but I'm pretty sure is fairly rare for most games to offer beyond 4x or 8x.

Also, do not mistake with Anisotropic filtering, that is very commonly offered as an option going up to 16x in almost any game that has an AF setting.
i mentioned AF as example which got stuck on innovation while 'options to improvement' were tried and suggested for over decade
Posted on Reply
#56
ZeDestructor
Dwardenalso ironically i still consider properly implement and tuned SMAA 2.x (especially to avoid UI and blur issues) to be better than FXAA and MLAA ...
Agreed, but there is a somewhat higher performance penalty with SMAA than FXAA and MLAA, all while the visual difference becomes smaller and smaller as you increase the resolution.
Posted on Reply
#57
Dwarden
ZeDestructorI certainly do, but I also remember that 16x-64x AAs were introduced at a time where 1024×768 (16" viewable) to 1600×1200 (18-19" viewable) was the norm. These days I have a 5120×2880 27" monitor as my main monitor, with an eye on 6144×3456 to 7680×4320 32" panels as potential upgrades.

Compared to the 2560×1440 100ppi "base resolution" equivalents, I'm already doing 4x FSAA/SSAA by just running native 5120×2880, and adding 4x MSAA/FSAA/SSAA on top would be equivalent to 16x. When (not if!) I do manage to find a 7680×4320 27-32" monitor that's not too compromised (the Dell UP3218K has it's issues), I'd be looking at 9x FSAA/SSAA by just running native (compared to a 2560×1440 baseline), and 4x MSAA/SSAA/FSAA would bumped that to 36x equivalent.

Overall, I say as long as the 4x modes stay. we're good, and it seems that all the graphics engineers agree (even if they don't explicitly say so).
note i added edit while you were writing your answer ...
like i said the problem is, they removing 'standard' feature (while edge case)
ZeDestructorAgreed, but there is a somewhat higher performance penalty than FXAA and MLAA, all while the visual difference becomes smaller and smaller as you increase the resolution.
i don't like the visual degradation of the FXAA and MLAA ... the performance hit became non-issue over time ...
ironically some of the fake hallucination predictive alg cost more frametime than using SMAA with worse result
Posted on Reply
#58
ZeDestructor
Dwardennote i added edit while you were writing your answer ...
like i said the problem is, they removing 'standard' feature (while edge case)



i don't like the visual degradation of the FXAA and MLAA ... the performance hit became non-issue over time ...
ironically some of the fake hallucination predictive alg cost more frametime than using SMAA with worse result
Like I said, it gets better as resolution goes up. The first time I say FXAA, it was at 720p and it was very much not good. Then I tried it again at 1080p, then again at 1440, then 4K and now 5K. Every step higher, the smaller the difference. At 5K, it's giving 4x MSAA/SSAA/FSAA/DSR/VSR a real good run for the money even, and certainly slapping DLAA silly.

EDIT: regarding the removal of the feature: keeping the feature around is not free - there's a bunch of extra tests you need to not only run, but also bugfix the bugs you find using em, and also maintain the tests themselves. Especially when it's also costing silicon die area. That's why I don't mind dropping the really niche stuff, especially when you can hack arbitrary DSR/VSR factors as you see fit into the modern drivers.
Posted on Reply
#59
Dwarden
nagemeBest AA: Matrox Parhelia. :)
web.archive.org/web/20170221211150/https://www.anandtech.com/show/911/7

(AnandTech's main site / article archive now gone?!)
Parhelia chip AA was another example of last hardware level AA innovation but sadly Matrox ceased to exist for gaming market before iterating
while Matrox as company still exists, they using AMD GPU chips
ZeDestructorLike I said, it gets better as resolution goes up. The first time I say FXAA, it was at 720p and it was very much not good. Then I tried it again at 1080p, then again at 1440, then 4K and now 5K. Every step higher, the smaller the difference. At 5K, it's giving 4x MSAA/SSAA/FSAA/DSR/VSR a real good run for the money even, and certainly slapping DLAA silly.
there is another reason i don't like such removal of standard(s) features ...
let say you run old games which look best with that feature ...

all of sudden you have game which was limited (ie not even supporting 4k) where 16x MSAA was just working great, without 16x MSAA ...

so imho there is something shady weird in the way how this decision was made or why
ZeDestructorLike I said, it gets better as resolution goes up. The first time I say FXAA, it was at 720p and it was very much not good. Then I tried it again at 1080p, then again at 1440, then 4K and now 5K. Every step higher, the smaller the difference. At 5K, it's giving 4x MSAA/SSAA/FSAA/DSR/VSR a real good run for the money even, and certainly slapping DLAA silly.

EDIT: regarding the removal of the feature: keeping the feature around is not free - there's a bunch of extra tests you need to not only run, but also bugfix the bugs you find using em, and also maintain the tests themselves. Especially when it's also costing silicon die area. That's why I don't mind dropping the really niche stuff, especially when you can hack arbitrary DSR/VSR factors as you see fit into the modern drivers.
somehow i doubt 16xMSAA on/off costs any die area if the MSAA 8x is claimed to exist ...
unless there is no MSAA anymore and it's replaced by something else and that something else can't handle MSAA 16x {cough, i totally didn't see this conspiracy coming}

ofc more likely is that whoever was responsible for the MSAA implementation was fired by Intel recently before finishing it
and someone in marketing department (or AI) just seen that 16x is more than 8x and thus it will be sold at premium in future hiend
Posted on Reply
#60
theouto
So intel is just joining everyone else in the market in only supporting up to 16 times MSAA, big nothing burger of an announcement.

Unfortunately for intel, TAA and XeSS does not make "legacy" AA methods obsolete, not when those methods are preferred by many ;)
Posted on Reply
#61
theouto
theouto16 times MSAA
Meant to say 8 times, not 16 times, and it's too late to edit now
Posted on Reply
Add your own comment
Nov 22nd, 2026 07:17 PST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts