Quote Originally Posted by Leeghoofd View Post
I'm getting very close now to 2000mhz (test 7 passes more than 5 runs, just test 5 still spews errors) , but I lended out the REX for some PI runs on an E8600 435 batch cpu. I hope to have it back within 2 weeks. I'm using very negative values for the NB, Rams were around +12.5-25 ref volts
Lee, i guess you're on 1:2 - RE maybe running too tight between 490-498fsb... on 266 and 333 straps, Read and vNB are the ~ same @ 490fsb (490 boot) as 500fsb (500 boot), with the same clocktwister and tRD.
so between 490-498 min vNB is higher than @ 500. min vDIMM may also be higher.

booting @ 500, 2x2 Blades ok for Spi/Hpi @ 1000mhz.. skew / vREF / NBGTL all Auto. vNB scales fairly evenly with Read up to ~12.8-13K on 1:2 and 5:8.

the result booting @ 500fsb is not much better than 475fsb for Read, but it's the same with 2x1GB D9GTR, which needs higher vNB than 2x2 elpida for same clocktwister + tRD, 450-500fsb/1:2.

500 (x9) / 1:2 / 7-7-6-20 tRFC 60 (vNB for 2x2 elpida)

333 strap / clocktwister moderate / tRD7 = 12K read 43.5ns = 1.49V or less vNB

333 strap / clocktwister Strong / tRD7 = 12.4k read 41.5ns = ~ 1.59V vNB

333 strap / CT Strong / tRD7 + CHA 1+2 = 12.5k read 39.5ns = 1.59V vNB (HyperPi)

266 strap / clocktwister Strong / tRD6 = 12.8K read 39.5ns = 1.69V vNB (SuperPi, Hpi not tested)