Now that's what I'm talking about!
Are you testing w/ small-fft, large-fft, or blend? I recommend running large-fft while pushing the FSB like you are. Once that's stable you can up the multi and run small-fft to stress the CPU.
The P45 is on a smaller process than P35/X38/X48...65nm vs 90nm. As such the P45 needs less vNB than the others. You can push 1.45-1.5v if you want and temps are good, but you really shouldn't need to. 1.3v should take you a long way, and I've found P45 to be particularly picky...too much vNB can actually make you less stable. There is definitely a sweet spot, and vtt also interacts strongly w/ vNB and NB stability.
Blend, I believe. I'll check when I can. I had some pretty good success. Yes, vtt, as well as the references, seem to have a large impact.
ETA: P95 Blend. Passed 4.40G/518/500 for 17.25 hours before I stopped it. Core temps were peaking at about 62C.
-3.0GHz/500FSB passed P95 for 10 hours before stopping.
-4.0GHz/500FSB passed P95 for 12 hours before stopping after upping the CPU core voltage. I had previously thought I could do this a stock voltage, but at stock it would fail P95 in a little over an hour.
-
4.4GHz/518FSB has been passing P95 for the last 13 hours and I want to see if it will make it to 24 hours.
-4.5GHz appears to be a no go. I went up to 1.425V with load line enabled and it still failed P95 rather miserable with very high temps (70C). At 4.4GHz, core voltage is 1.4V without loadline and temps are topping out at about 62C.
The high FSB trick is to sort out the FSB first by running the CPU as slow as possible (smallest mult). Once the FSB is stable, then crank up the mult.
My goal was 4.5GHz/500FSB/DDR2-1000 5-5-5-15 on air. I'll take 4.4GHz/518FSB/DDR2-1036 5-5-5-15 on air thank you very much!
Next steps are to test beyond P95 and start knocking down the voltages. The forum postings here, on Hardocp, and Xtremesystems were invaluable in getting this result.
ETA: I haven't had a result this good since the Celeron 300A. Way back when these were all the rage. On a BX board you could easily get one to 450MHz. I even made a version of the Celery Sandwich. This E8400 cost less than the 300A and the mobo cost the same as a BX did back in 1998. An equivalent OC today to the 300A would be getting an E8400 to 4.5G. Very
.
-Mike