Heaven benchmark MSAA ROP impact study

[BrightCandle]

How to participate

 

I would like it if people with varying graphics cards run heaven 4.0 as a benchmark with custom settings and only varying the Antialiasing setting. At the very least to usefully participate in this study you need to run AA off and 4x and post both results along with what card you are running but I would also like antialiasing at 8x if you have time to run the benchmark a third time. The custom settings I want you to use for Unigine Heaven 4.0 ( https://unigine.com/products/benchmarks/heaven/  ) are:

 

DX11

Quality: Ultra

Tesselation: Normal

Stereo 3D: Disabled

Multi-monito: Disabled

Full Screen: True

Resolution: 1920x1080

 

And then antialiasing of off and 4x as a minimum, 8x would be nice too. I only need the average FPS for this type of information.

 

As an example data post:

970 SLI

Off: 159.9
4x:  127.8
8x: 108.0

 

The theory

 

The big die GPUs typically have 50% more ROPs than the mainstream GPUs and with the Pascal/Polaris chips due I am thinking about the consequences of having mainstream chips. If you don't know the ROPs are a factor in MSAA performance and how much of a frame rate hit you take, all GPUs today have enormously more ROP performance than need otherwise specifically to accommodate anti aliasing but there is a big difference with the highest end GPUs to the mainstream ones. A 1080 has 64, a 980 ti has 96, a 980 has 64 and my 970 has 56. So given that I would expect that a 980 ti would loose less performance from using MSAA than a 980 and indeed the 1080 would  as a percentage. Considering the choices we have in the near future and how often I personally use MSAA (and how often I suspect others do as well) I want to gather some data on the topic to get an impression of how much of a practical impact this has.

 

I have looked at reviews specifically about this I didn't find much of use. I find various articles talking about AMD and Nvidia saying how much MSAA performance should be improved with their new higher throughput ROPs and then no one really testing it. You either find tests of different GPUs using or not using MSAA but never both with nothing else changed or alternatively tests determining the impact of MSAA verses other forms of AA for a singular Nvidia or AMD card but never a range of cards. Neither of which gives me an answer to this particular aspect.

 

Since I don't have a 980 ti or a 980 or indeed a Fury X and a 390(x) with which to compare relative results I need some help. I want some simple to obtain benchmark results to start to look at this aspect of performance. By gathering multiple data points with all else being equal I can eliminate the other factors (CPU etc) that may impact on the performance of the test that make the tests not comparable all that well with each other, because its a relative number. Heaven is a reasonably good benchmark for this aspect so long as tessellation isn't really nutts (that sucks up memory bandwidth and so does MSAA) because it uses MSAA and it doesn't seem to do all that much to optimise its use, its pretty average in how it looses performance with AA. Its also free and easy to run.

 

What I will do is compile the data and then present it back, whether it shows anything interesting or not.

[agent_x007]

ROP's need Memory Bandwidth to operate.
Having all the ROP's in the world is pointless if U don't have enough bandwidth to feed all of them.
Color compresion or not, Memory Bandwidth is the key.

CPU has nothing to do with MSAA since it's 100% GPU thing (ROP thing to be exact). It sure is taxing, so if your GPU is working at 95-99% already, enabling MSAA will make U lose performance.

Also, Unigine Valley is really not a good program to measure this since it's not consistant enough (min drop will make avg. higher/lower depending on run).