Gigaflops ?

[pc999]

Can anyone say to me how many gigaflops are capable the PS2 and GC .
XB is capable of 80 G ( acording to Nvidia http://www.nvidia.com/page/console.html )
Just because this is a bit confusing over the net . Some say that that PS2 was 6.2 G . MS says that their console is 2/3 times more powerfull ( PS2 = 20+ G ) sony say that it had 1 G ( PS3 1000 hundred times more powerfull at 1 Teraflop ) .

So if someone can sum up , because I am a bit :?
Thanks.

 

[Guden Oden]

6.2gflops quoted for PS2 is just for the Emotion Engine main processor. The 80gflops figure given by Nvidia is just marketing BS they reach by adding together all the fixed-function operations done to render polygons into pixels on the screen; it's not comparable in the slightest.

A straighter comparison would be to add up FLOPS performed by the 733MHz celeron CPU in the Box. I'm not totally sure of that chip's capabilities, but a guesstimate would be about 2.2gflops (I assume 1 flop/cycle on FPU, 4 SSE ops every 2 cycles, though I could be off here).

Then the Box has twin 233MHz vertex shaders too on the GPU which handles tasks done by the EE CPU in the PS2 so the net sum for the Box will likely be larger in raw ops, with the exception the PS2's EE is fully programmable; the vertex shaders are not. They can ONLY transform/shade already defined vertices. The EE can create/destroy vertices too for example (used to tesselate a higher-order surface into a polygon mesh to create smooth curved surfaces like a landscape for example).

So the Box has higher raw performance, but the PS2 is more flexible, as far as geometry processing is concerned. With pixel processing, the opposite is true.

Sorry if I got anything wrong, it's 2AM right now and I need to go to bed...

 

[pc999]

Cool , thanks

 

[Fox5]

According to SiSoft Sandra, xbox's cpu can do 982.22 MegaFlops. that would be .98 gigaflops then?(sandra lists the megaflops for a 600mhz celeron, so I just calculated what it would be for a 733 celeron)

 

[ShinHoshi]

6.2gflops quoted for PS2 is just for the Emotion Engine main processor. The 80gflops figure given by Nvidia is just marketing BS they reach by adding together all the fixed-function operations done to render polygons into pixels on the screen; it's not comparable in the slightest.

A straighter comparison would be to add up FLOPS performed by the 733MHz celeron CPU in the Box. I'm not totally sure of that chip's capabilities, but a guesstimate would be about 2.2gflops (I assume 1 flop/cycle on FPU, 4 SSE ops every 2 cycles, though I could be off here).

Then the Box has twin 233MHz vertex shaders too on the GPU which handles tasks done by the EE CPU in the PS2 so the net sum for the Box will likely be larger in raw ops, with the exception the PS2's EE is fully programmable; the vertex shaders are not. They can ONLY transform/shade already defined vertices. The EE can create/destroy vertices too for example (used to tesselate a higher-order surface into a polygon mesh to create smooth curved surfaces like a landscape for example).

So the Box has higher raw performance, but the PS2 is more flexible, as far as geometry processing is concerned. With pixel processing, the opposite is true.

Sorry if I got anything wrong, it's 2AM right now and I need to go to bed...

As far as I know, considering that SSE are 128 bit data registers and MADD instructions take more than a cycle we are talking about. (let's consider a simple ADD of two XMM regs)

4 FP/cycle * 733 MHz = 2932 MFlops= 2.9 GFLOPS

I don't know if FPU performance should be added...I am sure that MMX/FPU cannot be used at the same time since they share the data registers but I don't know about SSE-FPU....But let's consider they can since they use different data registers.
Then we should add:
733 MFlops + 2932 = 3665 MFLOPS= 3.6 GFLOPS.

 

[pc999]

Thanks

 

[Farid]

ShinHoshi is right, according to MS specs:

Intel P3 (Yes, i know it's in reality a Celeron Mobile but i'm just quoting/pasting the specs)
733MHz
32-bit INT
80-bit FP
64-bit MMX
128-bit SSE
32KB L1
128KB L2
1980 MIPS
2.93 GFLOPS
1.0 GB/sec Bus

That and remember that the Nvidia Gigaflops are directly coming from the land of Dawn/Dusk.

 

[Guden Oden]

As far as I know, considering that SSE are 128 bit data registers and MADD instructions take more than a cycle we are talking about.

The instruction may take more than a cycle to complete, but since the execution unit is pipelined, you still get the throughput of one instruction/cycle under ideal conditions...