Why is AMD better for Gaming?

Well most AMDs are better in direct reference to the Pentium 4 the Pentium D and the Pentium EE Processers which are the old 90nm(same as all AMD64s now) .
But the new 65nm cores with are the Core DUO the Core 2 DUO and most ones that come out are which are Mid range speed of the High end 90nm are almost Clock for clock and Workload for workload(thats what Intels claims and benchmarks are showing) although there is a generation gap in their make and technology *talking in virtual world time that is. (and Core 2.aka.Conroe/Woodcrest is'nt even out) So until the Summer AMD still Kicks all Intel ass

In addition, AMD CPUS have fewer native instructions/instruction sets. As I recall it was 58 less for athlonXP vs PIII and 90-ish less vs P4 (Its been awhile, so I dont recall the exact P4 comparison)

Fewer instructions translate directly to less time spent processing code and conditioning data, so you get faster throughput, however, there are also limitations involved. Though Ive never seen any factual data detailing the disadvantages, I would stick my neck out and say the extra instruction sets are probably why the intels are faster at data compression, encoding converting etc. Purely a guess and quite possibly completely wrong

In regards to gaming performance, there was actually an article that talked about one aspect of Netburst's poor performance not long ago.

http://www.extremetech.com/article [...] 906,00.asp

Essentially, while K8 has great FPU performance, Netburst's is comparatively poor. However, Netburst was specifically designed with SSE in mind so SIMD instructions are done very well. The problem is that many game developers don't even integrate SSE even though it's been around since the Pentium III, much less SSE2 or SSE3. The main reason why there isn't mandatory support is that it increases the amount of time required to debug a version of the game with SSE and a version without although it really shouldn't be much of a problem nowadays since all modern processors (even VIA) have at least SSE. Using more SSE instructions shouldn't put K8 at a disadvantage, but would allow Netburst to catch up. The article mentions that even AMD is pushing for more SIMD integration in games.

12 for integer and 17 for floating? Are they independant pipelines (one for the integer units, one for the floating-point unit)?

It would indeed be sensible to have a longer pipeline on the FPU, since I can't well see conditional branching being heavily used in floating point computing...

Right. I remembered it/them having been lenghtened for the K8, both had 2 stages added according to your graph.

Don't get confused. Jack is a fking computer genius. That's why we pay him the big bucks. Sexbomb is the fking genius, but we don't pay him the big bucks because he spends all his time surfing porn instead of posting clues on Toms. Comprendre?

Well, so far the posts on this thread have been very informative and well-reasoned. But I have a problem with the original post. AMD processors are NOT

AMD CPUs are not only good for gaming, folks. Let's not forget about audio encoding, compression/decompression algorithms and even some simple video encoding tasks (CloneDVD springs to mind) This is just an example of what AMD single core CPUs can do better than Intel.
When you factor in dual cores then all bets are off. AMD X2s are quite capable of destroying Pentium D chips even in complex video encoding, which was once secure Intel territory. As a matter of fact, X2s have levelled the playing field in performance between the two companies in everything except gaming, in which AMD is still the undisputed king. (Conroe may change this. We'll see.)
Check out these links for some comparisons of modern chips in various apps:
http://www.tomshardware.com/2005/1 [...] age39.html

http://www23.tomshardware.com/cpu.html

Just wanted to clear that up a little.