| Michael-5 said: Either way, 9x, 10x, I won't argue that. When CGI says over 15x, then I'll call him out on that. 1) As for last gen, I hear the old XB1 was about 13 GFLOPS, so PS3 is 16-17x more powerful then the XB1, plus I bet the GPU in the PS3 is more efficient then the one in the XB1, like you said, you can't measure things off raw power anymore. 2) However, I think with PS3, because it has was 6 or 7 CPU's, it was hard to program for, hard to reach that theoretical max. ---- Since you know more about Video cards then me, what about the GPU speed, PS4's GPU speed is about 3x as powerful as PS3's. Two people quoted me, and laughed, and CGI even took the opportunity to rub that in my face. I'm not wrong am I? Why is is funny when I mention something like this? 4) |
1) If you compare just PS4's and PS3's GPU, than difference is most certainly more than 10x.
2) You're using FLOPS as only measure of systems capabilities - don't...as I said, shader performance skyrocketed in last gen, but other parameters have not...making overall performance jump around 10x from XBOX1 to 360 (which is insane in just 4 years, and is much more impressive than 10x or slightly more in 8 years)
3) Wiki page states 230 GFLOPS for Cell (that's PS3 CPU) - Cell has 1 PPE and 8 SPEs - each of these performs at 25.6GFLOPS, single precision. 1 SPE is inactive for yield issues, 1 is reserved for OS. So math is (1PPE + 6SPE)*25.6GFLOPS = 179.2GFLOPS available to devs (plus 176GFLOPS from GPU). Of course, unless you're Miyamoto Musashi of PS3 coding, you won't be able to see anyhere near this numbers.
4) Because core speed is completely irrelevant if you don't know other parameters. For example, 360 is running at 550MHz core clock, and it has 48*5 shaders (for total of 240). So in theory it can do 240 * 0.5GHz = 120 Goperations per second. Since each shader can do 1 multiply+1 add at the same time, whole GPU can perform 120*2= 240GFLOPS. That's in theory...360 is built on VLIW5 architecture (just like WiiU, while we're at it), so in most cases only 4 out 5 shaders in group will be working, bringing down that number effectively, and that's just for start.
PS3's GPU (RSX) does not have unified shaders, but instead 24 pixel and 8 vertex shaders, each capable of 10 operations per cycle. RSX runs at 550MHz, so math is: 24 * 0.55GHz * 10 + 8 * 0.55GHz * 10 = 176GFLOPS. This is theory again, because you need perfect alignment of loads on pixel and vertex shaders to achieve theoretical maximum.
PS4's core clock is 800MHz, and it has 1152 unified shaders, so math is 1152 * 0.8GHz * 2 = 1843.2GFLOPS. PS4's GPU is also built on GCN architecture, which has numerous improvements over VLIW5.
Same thing is when calculating theoretical pixel/texel fillrates - clock * number of ROPs (pixel) or clock * number of TMUs (texel) (360 has 16 TMUs and 8 ROPs, PS3 has 24 TMUs and 8 ROPs, PS4 has 72 TMUs and 32 ROPs).
On top of all this, you have memory bandwidth - good example how it affects same GPU at exactly same core clock speed are 3 HD5550 cards with 3 different memory bandwidths (12.8GB/s : 25.6GB/s : 51.2GB/s) - 30fps : 40fps : 49fps
So, as you can see, basing your comparisons on core clock (or GFLOPs for that matter) is...well, wrong. At the end of course, no matter how much theoretical data you have, you need real benchmarks, and that is only real measure of how certain GPUs perform (or compare). In this case...go to 1)
