What makes a GPU faster/better?

So here are two graphics cards:
http://www.newegg.com/Product/Produ...30-590^14-130-590-TS,14-127-565^14-127-565-TS

Now my question is this, what makes one card better than the other? The cheaper card has a much faster clock speed all around, but with less memory. Why would a newer more expensive card have a slower clock speed, is that not what gives it its speed? What is it that makes the newer card so much faster, even though it has a slower clock?

The difference is the gpu chip, and memory interface, the more processing cores gpu has the more powerful it is, the larger memory interface is the faster textures can be processed, clocks are only the speed at which those work, saying that weaker card with higher clocks should be faster than stronger card with lower clocks is like saying that dual core cpu @ 3.5Ghz has more processing power than quad core @ 3.0Ghz

Ok I see, thanks

no problem, welcome to overclockers.com

As i understand it as well, per generation the architecture is also better, the example given which i like was this,

if you asked a chip to compute 5x10

Way older cards would do 5+5+5+5+5+5+5+5+5+5=50
Thats 10 steps to do the calculation

A new card would be smart enough to do it this way
10+10+10+10+10=50
This uses only 5 steps for the calculation

The latest greatest would just go
10x5=50
Thats 1 step do the calulation,

so even if the first card was clocked 4 times as fast its still slower then the newest card, which is why going from one generation of GPU or CPU to the next, clock speed is largely irrelevant (like a phemon Vs a phenom II or say a GT285 vs a GT 560 Ti), as the chips are inherently smarter and more efficient, now on the same chip architecture clock speed would mean something, but you also have to consider stream processors, mem bandwidth, memory size or evn driect x version supported or open GL for a GPU

On a CPU you need to consider L2, l3 cache and # of cores, hyperthreading, tubo setting for single core performance...


In the specific case you posted though, the "better card" is technically ~10% lower clock speed BUT it has ~33% more stream processors, as well the memory interface is 50% larger too, and total memory is larger albeit slightly slower memory clock, so even with slightly slower speeds, it does not matter cause overall its way better,

for another analogy, is a 4 cylinder at 7,000 rpm redline inherently better than a V8 with only a 6,000 rpm redline.