Mar
28
Mar
21
Some useful links to explore the world of 3D inside J2ME. It looks to be pretty good since it does not overload the hardware too much (you do not get hardware acceleration obviously) and still you get pretty good results in handled hardware.
I will post more on this.
Mar
09
This post will be growing up in few days:
Vector normalization done fast
The point looks to be in doing as much operations as possible shuffling the four components you have loaded in xmmN registers:
Mar
04
I was reading a paper about Multi-core (several processors in the same package) vs single packed processors and I rembered why I didn’t enjoyed very much IT papers.
They tried to proof a multi-core architecure perform better than its older brother. They assume that a second processor in other package perform 80% and one in the same 70%.
Let have a processor of 3Ghz and dual-core of 2.2GHz each, you know they are made with lower speed to reduce power and heat:
- If you have two processors in differents packages, they perform as a single machine of: 3000Hz * 1.8 = 5400Hz
- If you have dual-core package they perform : 2200Hz * 1.7 = 3740Hz, so if we add another package you get: 3740Hz * 1.8 = 6732Hz
Oh! Yeah! You get much better performance with dual-core systems, nifty/cool!!!!
Let assume each extra package you add you loose a 20% of performe respect the previous package:
- If you have 4 processor of 3GHz each you get : 3000Hz * 1.8 * 4 = 21600Hz
Sigh! Who is right??!!
That’s why I love papers from biology journals, they almost never leave their proofs uncompleted.
