- Joined
- Mar 7, 2008
There is a back story to this but I'll skip to my current situation.
Mobo: MSI Z170A Gaming Pro, bios 1.5
Ram: Ripjaws 4 F4-3333-GRRD 4x4GB kit
CPU: i5-6600k air cooled
4 sticks boot at SPD 2133, no boot at XMP 3333. Tried randomly increasing ram, IO, SA voltages, manual entering of XMP timings without help.
Tried using only 2 sticks in slots 2 and 4 as indicated in manual, it boots at 3333!
Had some errors in memetest86+ 5. Same with other pair of the 4.
No errors in memtest86+ when I removed manual settings I was playing with and reverted to XMP defaults, clear for ~2 hours consisting of over 2 full passes
Boot to Windows and tried Prime95 stress test. Errors in ~10 minutes, even with CPU forced down to 3.5 GHz
Moved ram to slots 1 and 3, that gave errors instantly, so moved ram back to 2 and 4. I didn't investigate that further.
Reduced ram to 3000, seems Prime95 stable with CPU at 3.5 GHz. Also stable at 3.9 GHz 1.25v.
I've left it like that for about a week doing PrimeGrid LLR tasks with no errors. For practical purposes, that's like running Prime95 continuously.
Today I thought I'd try pushing harder.
CPU still at 3.9GHz 1.25v.
Ram set to XMP at 3333. I tried moving to memtest86 (no plus) v6, where I found test 9 gave a lot of errors. I concentrated on test 9.
Increase ram to 1.37v set. Much reduced errors
ram back to 1.35v set. Lower VCC SA from measured auto 1.288v to 1.05 set. Few errors
Reduce VCC IO from 1.272 measured auto, to 0.95v set. No boot
VCC IO up to 1.00 set, lots of errors
VCC IO up to 1.05 set, lots of errors
VCC IO up to 1.10 set, 1 error after about 90s for test 9
VCC IO up to 1.15 set few errors
So it looks like 1.10 gave minimum errors. How sensitive are the voltage levels? Do I need to test every 0.01 here?
VCC IO back down to 1.10 set. SA up to 1.10 set. 1 error after 14 minutes. Best yet!
VCC SA up to 1.15 set. 2 errors in 5 mins
VCC SA down to 1.10 set. Ram to 1.37 set. 1 error after 3 mins
Back to 1.35v. Slack timing from 16-18-18 to 18-20-20. Error after 1 min. At least we know it isn't those timings.
Manually enter XMP timings. 2 errors in 1 min (tRAS 38, tRFC 434, tRRD(_L) 7, tFAW 36 - auto had set tRFC and tFAW slightly lower)
Back to previous best settings, but drop to ram speed to 3200. Passes test 9 still, but on doing a full set I got errors on test 6.
Drop to 3100. Clear running complete memtest86 for >2h.
So to recap, I'm now running ram at 3100, XMP timings on the mobo, VCC SA and VCC IO 1.10v set. Ram 1.35v set (typically measures 1.36).
Tiered of tinkering with the ram, but any suggestions welcome.
Decided to overclock the CPU some more for fun and now running at 4 GHz core and cache along with the ram at 3100 and no problems in Prime95 after ~20 minutes so far.
Mobo: MSI Z170A Gaming Pro, bios 1.5
Ram: Ripjaws 4 F4-3333-GRRD 4x4GB kit
CPU: i5-6600k air cooled
4 sticks boot at SPD 2133, no boot at XMP 3333. Tried randomly increasing ram, IO, SA voltages, manual entering of XMP timings without help.
Tried using only 2 sticks in slots 2 and 4 as indicated in manual, it boots at 3333!
Had some errors in memetest86+ 5. Same with other pair of the 4.
No errors in memtest86+ when I removed manual settings I was playing with and reverted to XMP defaults, clear for ~2 hours consisting of over 2 full passes
Boot to Windows and tried Prime95 stress test. Errors in ~10 minutes, even with CPU forced down to 3.5 GHz
Moved ram to slots 1 and 3, that gave errors instantly, so moved ram back to 2 and 4. I didn't investigate that further.
Reduced ram to 3000, seems Prime95 stable with CPU at 3.5 GHz. Also stable at 3.9 GHz 1.25v.
I've left it like that for about a week doing PrimeGrid LLR tasks with no errors. For practical purposes, that's like running Prime95 continuously.
Today I thought I'd try pushing harder.
CPU still at 3.9GHz 1.25v.
Ram set to XMP at 3333. I tried moving to memtest86 (no plus) v6, where I found test 9 gave a lot of errors. I concentrated on test 9.
Increase ram to 1.37v set. Much reduced errors
ram back to 1.35v set. Lower VCC SA from measured auto 1.288v to 1.05 set. Few errors
Reduce VCC IO from 1.272 measured auto, to 0.95v set. No boot
VCC IO up to 1.00 set, lots of errors
VCC IO up to 1.05 set, lots of errors
VCC IO up to 1.10 set, 1 error after about 90s for test 9
VCC IO up to 1.15 set few errors
So it looks like 1.10 gave minimum errors. How sensitive are the voltage levels? Do I need to test every 0.01 here?
VCC IO back down to 1.10 set. SA up to 1.10 set. 1 error after 14 minutes. Best yet!
VCC SA up to 1.15 set. 2 errors in 5 mins
VCC SA down to 1.10 set. Ram to 1.37 set. 1 error after 3 mins
Back to 1.35v. Slack timing from 16-18-18 to 18-20-20. Error after 1 min. At least we know it isn't those timings.
Manually enter XMP timings. 2 errors in 1 min (tRAS 38, tRFC 434, tRRD(_L) 7, tFAW 36 - auto had set tRFC and tFAW slightly lower)
Back to previous best settings, but drop to ram speed to 3200. Passes test 9 still, but on doing a full set I got errors on test 6.
Drop to 3100. Clear running complete memtest86 for >2h.
So to recap, I'm now running ram at 3100, XMP timings on the mobo, VCC SA and VCC IO 1.10v set. Ram 1.35v set (typically measures 1.36).
Tiered of tinkering with the ram, but any suggestions welcome.
Decided to overclock the CPU some more for fun and now running at 4 GHz core and cache along with the ram at 3100 and no problems in Prime95 after ~20 minutes so far.
