cycyh
Bro, why do you need 4x oversampling of SSAA (Supersample anti-aliasing) in FullHD? Enjoy graphene. =) Two-fold anti-aliasing on your monitor is more than enough.
cycyh It is
not enough to have a gtx 980, it also needs a CPU of the same level and a power supply unit with a margin. What is your general config?
cycyh
metro is the first game in my memory where the anti-aliasing graph becomes clearer
This is really so. I noticed that too.
cycyh
Of course this is the norm, because in SSAA 4x mode the video card renders many times more pixels than without sampling, for example, without anti-aliasing at all or with FXAA anti-aliasing, that is, with post-processing on a frame already rendered in the "native" resolution of 1920x1080. With SSAA 2x, each pixel is split into four parts - subpixels for greater detail, and when rendering with ROP blocks, a gradient is created from these subpixels (average color in the size of one pixel), after which all this huge number of pixels is compressed into a tiny (relatively rendered) "native" monitor resolution. As a result, on the contrasting boundaries of objects, pixels have a smooth transition color, and not torn squares, which are called "ladders".
And in SSAA 4x mode, each pixel of the "native" resolution is divided into EIGHT subpixels (4 horizontal + 4 vertical) and imagine what loads the chip and the video card memory bus are experiencing during rendering. The video card will have to process eight times more pixels than there is in the monitor itself, at a glance it is 16 megapixels versus 2 megapixels without SSAA at a resolution of 1920x1080.
As for DSR, everything is the same here - crushing a pixel and replacing it with a gradient when rendering. But with a significant difference - the DSR does not have a clear reference to the resolution ratio, that is, there is no dependence on the linearity of 2x2x2x2 ... DSR algorithms apparently allow you to skip pixel crushing through one solid one, that is, not all pixels are fragmented if the DSR resolution exceeds " native "resolution is less than doubled. Here I may be wrong, since these are my personal conclusions but based on old knowledge, since I have not read the details about DSR.
DSR seems to me like viewing a simple jpg picture at a reduced scale, that is, when the number of pixels of the picture does not fit into the number of pixels that is on the monitor. Here the same gradient is created, but not from subpixels, but from real pixels of the image, some of which are "blended" with neighboring ones when the scale is reduced.
Something like this, in short. Perhaps everything I wrote above is the agony of my drunken brain. I already drink on Mondays ... things are bad along the way.
By the way, MSAA (multisampling) differs from SSAA (supersampling) in that with MSAA, multiple sampling of samples and the creation of subpixels occurs only from the boundaries of polygons (roughly speaking, from the edges of the 3D model where the ladders are most striking), all other samples (for example on textures) are in a simple non-multiple sampling (in "native" resolution) without subpixels and without creating a gradient. Therefore, MSAA removes only ladders, and SSAA slightly improves the overall clarity of the rendered frame due to full multiple sampling of all scene samples, in some games even textures are slightly improved when using SSAA 4X or the same DSR in 4K resolution.
ssaa can be roughly compared to rendering in multiple resolution and subsequent subpixel compression before actual rendering.
Thus, when full HD, the game with anti-aliasing x4 works like on a 4k monitor.
for minja at 1600x1200 and 690 zhifors it gives just some unhurried 30 fps. and somehow it looks strange as if the picture and the course of the game becomes very unhurried as a lul is broadcast from Mars.
MunchkiN 616
"at full HD, the game with anti-aliasing x4 works like on a 4k monitor"
Not correct. At 1080p, the video card will render in 4k resolution with SSAA 2x, and with SSAA 4x, it will already be 7k (16: 9 of course).
Let me explain why:
SSAA 4x is a four-fold oversampling of subpixels from one sample of the "native" resolution vertically and horizontally, namely:
1920x4 = 7680 subpixels (ROPs render them as full pixels before creating gradients (before "shrinking")) to the horizontal (this is already 7k, not 4).
Accordingly, the vertical will be as follows:
1080x4 = 4320
Total number of pixels (subpixels) of the rendered frame before "shrinking":
7680x4320 = 33 megapixels versus 2 megapixels "compressed" in the "native" resolution of the monitor or in the resolution set using DSR (or even without SSAA).
That is why SSAA bends any single-chip top so easily at 1080p, even with 96 ROPs and a bus with a bandwidth of 300 GB / s or more. Yes, as the resolution increases, the load on the bus grows exponentially, because to fetch a sample, the chip has to access memory, the more samples, the more often it accesses it. This is probably why at 1080p my 980 with a high-frequency 256-bit bus keeps up with Fury's with 4096 bits with a low-frequency multilayer HBM, and at 4k it merges completely, because despite the huge difference in frequencies, Fury's bandwidth is twice as high.
Or maybe the reason is something else, since there are a lot of variable factors affecting the overall performance of a video card.
White sea channel
for 4 sub pixels. - I also described it. each pixel will have 4 pixels in x2. 1p = 2x2p (4p) so let's say 1000x1000 will be 2000x2000, respectively. at x4, the sub-pixel block increases to x4, thus 1 p = 4x4 (16 pixels),
then the GPU procedure for the mathematical addition of all color vectors to an n-fold monitor, such as bilinear filtering, creates additional work for pixel conveyors or what modern vidyu have them replaces.
therefore, we see the arithmetic mean of all pixels in 1 final as the total half-color and do not see soap as such. well, they see only color soap.
MunchkiN 616
"for 4 sub pixels. - I described it."
You, dear, did not write such a thing.
You wrote the following:
"when full chd, the game with anti-aliasing x4 works like on a 4k monitor"
What is not true. And why - I have stated above. Typo?
Belomorkanal
a, I understood what the mistake is. 4 times lower than 1080p resolution
MunchkiN 616
That's it. At 1080p in SSAA x4 mode, it turns out 7.5k (horizontal) or, if you like, 4320p (vertical). I can imagine a 24 "monitor with such a resolution. And I can imagine its price too. It has already become bad.