CrazzyMan said:
Taken from beyond3d:
In Lost Planet, each character is 10-20K polys. A VS robot is 30-40K polys. A background is about 500K. With shadows and other hidden rendering cost, it's about 3 million polys per frame.
Different platforms require different care, would not be surprised if ppl working on 360 and suddenly dropping their datasets onto RSX would not observe good numbers (and viceversa)
Now..I can't see how RSX, if used in the right way, should be so limited at vertex processing: in HS we easily render 2-2.5 MTriangles per frame at 30 fps without being VS limited and without making any use of CELL to speed up vertex shading and I know for sure that being more clever we could even go faster..(just using the GPU)
The numbers quoted about HS are slighty wrong..cause a few days a go I realized there's a bug in the code that computes the triangle count: in some cases it's more close to 3M triangles per frame mark than 2M.
Regarding cpu, yes its totally different arguments. The 360's is a very standard setup, a bunch of cores that can all see main memory. The advantage to this is that lots of thread coded written over the years can be ported to it easily. For example, Valve (the Half Life guys) are writing an entirely threaded engine for the PC. When it's done, porting it to the 360 will be easy. How about porting it to the PS3? Nope, won't work, it's a totally different setup. The PS3's main core is basically the same, but its spu's can't see main memory. Each one has 256k that it can work with, so you need to rewrite your code to pack things into 256k chunks, feed it to the spu's to process it, then copy it all back to main memory. In other words, you've gotta rewrite your code.
Now before I get jumped on here let me briefly mention the downsides. The 360's cpu setup is painfully easy to code for. But, they really aren't all that fast. Worse yet, all three cores use the same memory controller so the three cores are not three times the power of the one core on the PS3 due to some overhead. The PS3's spu's on the other hand are monstrously fast. You need to setup your data correctly to work with them, but once thats done then yes, the 360's cpu setup is crap compared to fully working spu's. I say fully working because most current games out there barely even touch on them, it will take time to re-write everything. I haven't done spu coding....yet. I will be though in 2007 which will be pretty cool.
RSX
5 shader ALU operations per pipeline per cycle (2 vector4 and 2 scalar (dual/co-issue) and fog ALU)
27 FLOPS per pipeline per cycle
68 billion shader operations per second theoretical maximum ( ((5 ALU x 24 pixel pipelines) + (2 ALU x 8 vetrex pipelines)) x 500 MHz )
364 GFLOPS ( ((27 FLOPS x 24 pixel pipelines) + (10 FLOPS x 8 vertex pipelines)) x 500 MHz )
24 filtered and 32 unfiltered texture samples per clock
Xenos
2 shader ALU operations per pipeline per cycle (1 vector4 and 1 scalar, co-issued)
10 FLOPS per pipeline per cycle
48 billion shader operations per second theoretical maximum (2 ALU x 48 shader pipelines x 500 MHz)
240 GFLOPS (10 FLOPS x 48 shader pipelines x 500 MHz)
16 filtered and 16 unfiltered texture samples per clock |
For the top text part You're just qouting a dev that works on PS3, in fact he's PS3 second party, on Heavenly Sword. He doesn't even work on Xenos, how would Marco/nao/nostromo know what Xenos can do? He doesn't. He talks a ton of crap about 360 on a couple forums, and that's fine, but he doesn't know what 360 is capable of. He has no experience with it.
Now as for your specs, they're bogus. You can make up all kinds of spec like "gigaflops" that are meaningless. I can tell you Nvidia is especially good at this sort of BS. They set RSX as some kind of monster with inflated specs early on, but the real world throughput is another story. I dont know where you get RSX has 5 ALU per pipe, it has 2. Even then for example, 1 of the 2 must do double duty as texture fetchers when needed, which means they are less useful to shaders during that time. Xenos on the other hand has 16 dedicated TMU's, all it's shader ALU's are constantly available unlike RSX that may have many ALU's tied up fetching textures at any given time. See how you cannot compare stats? It just doesn't work that way in real life.
Things like a "fog" alu can up some artificial flop count, but that sort of flop counting is useless. Comparing like to like, you're looking at 56 (Vertex pipes have one ALU) 32 bit precision ALU's in RSX versus 48 in 360. But considering 360 has less framebuffer bandwidth pressure, has dedicated TMU's, has a arguably more efficient unified architecture, oh and I think Xenos can do an extra component per ALU. Xenos alu's are vec4+1, RSX's are vec3+1. A small 8 ALU deficit on paper for Xenos can easily be made up and surpassed.
