I am interested in buying an amd 2500, but I've been under the impression that I have to have an unlocked cpu to change the multiplier. However I have been hearing that some people have bought cpus that they oced out of the box with no modifications.

Also which is better:
2500-333fsb-512cache-1.8
2600-333fsb-256cache-2.08

the 2600 is better in 90% of workloads. Depends on the workload.

Before the tbred cpu amd only used 4 bits to determine the multiplier, which gave them multipliers 5 through 12.5. When they began using the 5th bit many/most boards had no ability to adjust that bit, but they could adjust the other 4. So today when you buy a 1700+ you may get access to multi's 5-12.5, but if you buy a 2400+ you may get access to multi's 13 and up. Some boards, including most nforce2 boards and many kt400 boards, have figured out how to address the 5th bit and therefore give you access to all multis. The rest just leave the 5th bit as it comes from the factory, so if the bit is on it stays on, off it stays off, which restricts the mutlipliers to only the upper or lower half.

The point is your cpu is unlocked, but your motherboard may restrict your multiplier access due to its own failings.

The 2500 is better because it is a little faster, and easier to overclock.

Here check out the benchmarks, in about everything, especially 3D Gaming, 2500 beats 2600.

http://www6.tomshardware.com/cpu/20030210/barton-14.html

juice that bench mark is for a 2600 running at 133 not 166 I'm interested in the 166

The question is does the barton's double cache make up for the 15% less ghz if a 2500 and a 2600 both are set to 333fsb. See post #10 in the below thread to gauge the value of the cache. It does not come close to making up a 15% ghz deficit no matter the environment. Amd didn't want to stop having a competitor to line up against Intel's top cpu so they overrated the barton and gave also gave it PR point benefit for fsb increases.

http://www.computing.net/cpus/wwwboard/forum/6274.html