Originally posted by: JZilla
Well thank you for that answer chizow, very well explained. But it still seems wrong to me, it's like the 9500Pro is faster than they meant it to be.
Oh well, guess I just have to put on my super sneakers and run to the nearest store when easter is over, so I can grab my 9500Pro before the rest of the heard.
NP, actually, the 9500pro is
slower than it potentially could be. Using your original analogy, its like owning a Ferrarri but restricting it to the Capitol Beltway during rush hour. Its got the horse power to really burn some tread, but its going to be no faster than the Saturn next to it because traffic is the bottleneck. The only time you'd benefit from the horse power is accelerating and weaving from lane to lane if you happened to have a little space in front of you to maneuver. The same holds true for the 9500pro and 9600pro, where the 128-bit memory interface is the bottleneck for the 9500pro. The penalty of having 1/2 the rendering pipeline (or horse power as you will) is less significant because the bottleneck isn't as severe as with the 9500pro.
The 9500pro was a stop-gap part to help ATi compete in the mid-performance market. They didn't have the time (or resources probably) to redesign the core so they made change external to the core to manage performance (128-bit bus). The RV350 on the other hand was designed from the ground up to be a cheaper solution in the mid-range market. Instead of making changes external to the core, they cut the rendering pipe in half (4 x 1 instead of 8 x 1, which makes up the majority of the ~50 million transistor difference) and shrinking the die, which leads to considerably cheaper fab costs as you not only get similar performance for 1/2 the transistor count, but you also get more chips per wafer. As yields per wafer improve, the cost savings improve as well.
Oh and yah, I'd get a 9500pro ASAP if you were thinking about it but were waiting on the 9600pro results.
Chiz
Facebook
Twitter