LukasTaves
Member
it might be overkill on xb1 yea, but on PS4? You can't just say it is based on one machine. If you could base it on just one machine let's look at other GPU's made by AMD they certainly don't find 32 ROP's overkill when you have a bandwidth of 176gb/s, in fact they use 32 ROP's with much less:
7850 32 ROP's 156gb/s
7870 32 ROP's 156gb/s
Comparing with desktops gpus is not the way to go, because they are designed to at least support a myriad of different resolutions above 1080p.
If you look at 360's ROP count and fillrate numbers it also seemed a little puny compared to gpus of the time, but it did had bandwidth to deliver the fillrate, and since the console was developed with a 720p resolution in mind that fillrate was adequate for most titles.
And again: Xbone is better equipped to do 1080p (in both ROPs and memory buffer size) than 360 was.
The only "evidence" they have is that somebody who works at Microsoft is saying in the XB1 they are only sometimes limited by ROP's.
They could run the math and realize that xbone has proportionally 40% more fillrate and memory for the framebuffer than 360, and realize that this statement is probably true?
Let's look at that Bandwidth example closer shall we?
8 bytes write 4 byte read for a total of 164gb/s which pretty much saturates their esram bandwidth, sounds a little odd to me.
Let's quote Goosens himself speaking about the esram:
http://www.eurogamer.net/articles/digitalfoundry-vs-the-xbox-one-architects
Now bear in mind this information comes from the same interview but in a different article. 164 is bigger than 140-150 In fact, i'd say he's a little off with his "pretty much" saturates the esram bandwidth, but wait, there's more.
You are completely mixing arguments there.
There are two different affirmations:
- The ROPs need 164GB/s to be saturated.
- The real world bandwidth that shipping games are already achieving on esram is 140-150 GB/s.
What is the esram's maximum theoritcal bandwidth in write only? 109gb/s
What are his bandwidth numbers again? 164 with a split of 8:4 between write and read, which would be a total of: 109.333333gb/s write and 54.6666gb/s read what do you notice? It's higher than the 109gb/s maximum theoretical peak
Do you notice that in the same article you are quoting they mention that you can read *and* write at the same time on esram? And even though you can't read and write at the same time all the time that the theoretical cap is 204 GB/s?
Because 140-150GB/s is a real world measured/benchmarked/tested performance from an actual running game. 164 is the theoretical bandwidth all the ROPs could use if they were active all the time. In a real game you are not going to go that far all the time. But if you think about bandwidth per clock instead of seconds, it means that there are portions within a frame were pretty much all the bandwidth of the system is going to the ROPs. If you have more ROPs that time would be lower? Yes, but they also would require more bandwidth, so at a target resolution, having an extra of them will not necessarily make much of a difference.Why Am I the only one doing this math, why isn't Richard Leadbetter? In one interview he has gone from 140-150 to suddenly using 164 and over the maximum theoretical write speed.
That is why this does not pass my smell test.