[MasterPooBaa]

heh
funny you should ask. i DID do some personalised benchmarks.
got some decent results, posted them in various forums.
here is the condensed version of them for your viewing pleasure.

4 settings were tested, same cpu Mhz, all memory timings were absolute maxxed:
1 - KT133A, 133/133 (10x133)
2 - KT333, 133/133 (10x133)
3 - KT333, 133/166 (10x133)
4 - KT333, 166/166 (8x166)

A. SuperPI to 1 Million decimal places

1. 103 sec
2. 92 sec
3. 87 sec
4. 83 sec

superPi uses a 8mb dataset, so lots of memory fetching.
all results were also numerically correct also, indicating no data errors.

B. Sandra 2002

1. 1010/1009
2. 1935/1884
3. 1929/1894
4. 2410/2332

note: sandra didnt recognise the mem running asynchronous, nor the 1/5 divider at 166fsb.

C. UD_task (cure for cancer) randomly selected chunk of a job.

1. 138 sec
2. 136 sec
3. 136 sec
4. 136 sec

ud is obviously fully bandwidth happy with the switch to DDR.

D. MemLatency:

1. 4096, 15.2, 183.6
2. 4096, 11.5, 144.1
3. 4096, 11.4, 143.1
4. 4096, 10.2, 115.2

see here that 166/166 has a big impact on mem efficiency.

E. DIVX!!! [img]smile.gif[/img]

1. 336 sec for 6000frames = 17.86 fps.
2. 306 sec for 6000frames = 19.60 fps.
3. 299 sec for 6000frames = 20.07 fps.
4. 293 sec for 6000frames = 20.48 fps.

big jump to DDR with divx. smaller jumps to 166/166, but still significant considering the time it takes to encode a movie at high res.

there ya go. those are my benchmarks.