Been reading through a massive number of threads now that I finally have a weekend to start OCing my new rig.
- P6Z68-V GEN3m 3291 BIOS
- 4 x G.Skill D3-12800CL9-2GBXL
- Air cooled, Noctua NH-U12P push-fan
Last (only) time I OC'd was my old E8400, so very rusty. Spent the last night and much of today running the system through steadily increasing voltages and Prime95 1344/1792 tests, to identify at what level I should start to find for running a full on blend test. I'm thinking I don't have all that high-clocking of a 2500k - although still fast - and want to make sure it's not just my lack of experience doing this.
Righto, started with manual voltages for the most of this. The below can be assumed to be constant for this discussion. Also assume that anything I say worked went through at least 20 min of 1344 and 20 min of 1792 at 96% memory usage. In some cases I've tested at 2688 as well.
C3 C6 Auto, EIST Enabled
CPU Current Capability 140%
Phase / Duty controls Extreme
EPU Power Saving Disabled
VRAM Frequency - Manual 350
PLL Overvoltage was enabled, but I confirmed it caused an issue recovering from sleep (thank you for ruining last night's blend test, power saving!) and disabled it. Now I get USB ports back when I return from sleep state!
After some lengthy testing, managed to get a 45x mult at 1.335 VCore (VID 1.3611). LLC was at Ultra-High, PLL 1.7, DRAM 1.5, VCCIO Auto. My actual values were around 1.328 at load.
While getting there, I had just finished reading Shadowfax's post on LLC. Matching up VCore's and VIDs, I established that by switching LLC to high, I needed an extra 0.15 VCore to maintain (since I don't have turbo-boost voltage adjustment in this BIOS version.) Tried playing with that, seemed to work. Mind you, I'm now at 1.35 VCore in BIOS, with 1.344 on idle and 1.328/1.320 on full load.
Wanted to see if I could get anything higher. This is where the fun began.
I *eventually* got a stable x46 multiplier on a BIOS set VCore of 1.39, with VCCIO at 1.09375 and PLL of 1.675. I couldn't get a stable build without the VCCIO and PLL changes, even at 1.39. That's right... I had to raise my VCore 0.04 for one multiplier increase. x47multis were still BSOD magnets. This is when I ran the (interrupted) overnight blend test. I was not happy at the massive VCore jump for one multiplier rise.
In the morning I encountered the USB issue and therefore ended up disabled PLL Overvoltage. This in turn destabilised the above stable x46 multiplier, which couldn't pass a 1792 20min test anymore. I determined through my notes that really, i was suffering from substantial Vdroop. Load voltage was still at 1.368, idle VCore at 1.384.
I went back to Ultra-High LCC, and could pass a 46x test with flying colours. Ultra-high did it's LCC job - idle and load were at 1.392 with mild fluctuations.
I attempted to use that for a 47x multiplier - no luck. Dropped the PLL to 1.6625 here as part of the testing.
Reading that using an offset (which I wanted to get to) would sometimes stabilise an edge case, I calculated my offset (+0.03 for 46x) and switched over.
That is stable at 46x. It is not stable at 47x, still, and 47x pushes loaded VCore up to 1.4/1.408 with it's slightly higher VID (1.3711 for me, whereas 46x is 1.3661). Also, is there a reason the CPU doesn't appear to be stepping down to x16 on idle now that I'm in offset mode?
Anyhow, what are your thoughts? Do I just have a proc that doesn't want to go above 46x? 1.39 VCore seems really high for 46x for me. Keep in mind I tested what appeared to be a stable 45x at 1.335 VCore! If I do my math right, according to my notes that means that at 45x I could run an offset of -0.03!
Offset mode, +0.03 VCore
PLL at 1.6625
VCCIO at 1.09375
DRAM at 1.5
Mult at 46x, giving a VID of 1.3661
Observed VCore at 1.392 under load.
Max temp of 70C on one core.
EDIT: And if it does appear 46x may be my best spot at present, any ideas on reducing VCore? Or is 1.3661 VID for 46x the writing on the wall as far as that goes?
EDIT2: Brain kicked in, answered a question for me. Struck it out, above.Edited by tyranny12 - 4/28/12 at 8:46am