Discussion Pixel Fillrate problem about Turing/Pascal Architecture?

eraser666

Junior Member
Feb 23, 2019
3
0
6
First, how about this formula?
I think, turing/pascal's pixel fillrate=minimum(ROP, GPC*16, TPC*4)*frequency/1000

For example: they have the same pixel fillrate in test.
2070:3GPC/18TPC/64ROP 2060:3GPC/15TPC/48ROP 1660Ti:3GPC/12TPC/48ROP
1080:4GPC/20TPC/64ROP 1070Ti:4GPC/19TPC/64ROP

they have different pixel fillrate in test.
1080:4GPC/20TPC/64ROP 1070:3GPC/15TPC/64ROP
1060:2GPC/10TPC/48ROP 1050Ti:2GPC/6TPC/32ROP

In my opinion, it is inaccurate to use ROP only to calculate the theoretical filling rate of pixels.

105062.png
QQ图片20190220143540.png
2070.JPG
1060.jpg
 

eraser666

Junior Member
Feb 23, 2019
3
0
6
For example:
they have the same pixel fillrate in test.
2070:3GPC/18TPC/64ROP 2060:3GPC/15TPC/48ROP 1660Ti:3GPC/12TPC/48ROP
1080:4GPC/20TPC/64ROP 1070Ti:4GPC/19TPC/64ROP

they have different pixel fillrate in test.
1080:4GPC/20TPC/64ROP 1070:3GPC/15TPC/64ROP
1060:2GPC/10TPC/48ROP 1050Ti:2GPC/6TPC/32ROP 1050:2GPC/5TPC/32ROP
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
In my opinion, it is inaccurate to use ROP only to calculate the theoretical filling rate of pixels.

Well, yea. Did you ever think a single set of numbers would have represented performance in games?

I think, Maxwell, Pascal and Turing have the same problem.

It's not a problem.

Fillrate results are affected by:
-Quality of the test. They are not perfect and lot of them are affected by memory bandwidth
-What its testing. There's a lot of different metrics.
-The frequency the ROP is running at. The designer of the GPU may choose to keep the clocks of the ROP separate from clocks of the shader core.