Xoj said:
selnor said:
hobbit said:
selnor said: God damn. Why 5 years into 360's life, are developers not taking advantage of 360's multicore properly. She points to blaming the ease of coding on 360. The fact that because it'seasy to get great results, She hasn't seen codes for 360 optimized to new ventures in synchronization for example. It's sad in a way. But hopefully, Remedy among others have shown what benefits come from looking at new ways to programme. Rather than stuck in what is essentially familiar single core processing. |
yeah, that was a shock to me. You'd think with all the cell spu vector optimizations that developers would be using the VMX units as well.
|
I thought exactly that. In a way with the PS3 having to rely on the CPU to help graphics, First Party PS3 devs have been forced to really crunch down to new programming techniques for multi core programming. Whereas the 360 has had next to no optimization for it's multi thread abilities. Hopefully We will start to see some nice advanced multicore programming on 360 over the next few years. Crytek seem to be doing that, if each week PS3 team make Crysis 2 look better and then the 360 team trump them the week after.
Come on M$ give us a first party game fully taking advantage of 360!
|
oh no how wrong can someone can be? Cell was different from nay platforms developers were used to work with 1-4 cores,
the ps3 though have 1 PPU and 6 SPUS, 360 its not easier because multithreading but because they are libraries already done it was easier to harrass it power.
and PPU can be threaded too. as far floating point calculation and threading u should know which cpu it's better. since all spus are 3.2ghz, while u thread 360 u get 6 1.6ghz threads
|
I don't think you deserve a response, but here goes.
Sure cell is different. Fine there
Sure the cell has 1 ppu and 6 spu, so what?
No 360 is easier because it has full cores with normal caches.
Sure 360 has better tools.
PPU can be threaded too? Sure the same way the 360 cores can be. Care to complain about that?
Floating Point performance? Sure if you have perfect data structures; sure if you have perfect vectorization; sure if you have perfect pipelining; sure if you have perfect DMA queuing; sure if you are doing simple floating point calculations, The spu's don't have a dot product, cross product, or fnma instructions. See how much work can go into getting the Cell's performance up?
That last part is just a joke right?