I seek advice on the final stages of tweaking required to reach 4GHz
stable on my new i7 860 build. It's fine at 3969 (189x21), but not
quite right at 3990 or 4011, ie. the system boots ok, I can run things
like Firefox & 3DMark06, but it doesn't pass Prime95 or LinX. Note that
using a 20X multiplier didn't work so well, though as I type I'm trying
again (currently at 189x20 & climbing).
Centurion Plus 534 case, Scythe SFF21F intake/exhaust
Thermaltake Toughpower 750W PSU
Asrock P55 Deluxe mbd
Core i7 860 (FPO/Batch L935B730)
Thermalright U120E Rev B (with AS5), Scythe Kaze-Jyuni 1900rpm fan (#1)
GSkill Ripjaw DDR3/2000 4GB kit (2 x 2GB, normally 9-9-9-27 @ 1.6V)
ASUS 8800GT 512MB (Glaciator) PCIe gfx
WD VR 10K SATA system disk, 1TB Samsung Spinpoint F3 data
LG 22X DVDRW
Antec Spot Cooler (for the RAM, just in case)
Win XP Pro SP3, apup/Autopatcher, all the latest drivers, etc.
#1: The 1900rpm gave temps 3C lower compared to using an SFF21F
1600rpm, though the Kaze is definitely a fair bit louder.
BIOS Revision: 2.00
Spread Spectrum: Disabled
XMP Setting: Profile 1 (DDR3/2000, 9-9-9-27 @ 1.60V)
CPU Ratio: 21
QPI Link: 6.048GT/sec (Options: 6.080 or 6.840)
CPU Voltage: 1.38125
DRAM Voltage: 1.629
CPU PLL: 1.895
DRAM Timings: Manually set to 10-10-10-29, 2T, all others Auto.
Real Temp results:
Idle: 34, 33, 35, 32
So, anyone have some suggestions as to the best way to reach 4GHz? Atm
I'm not sure whether I have the voltages particularly balanced - do
they look mismatched in any way? When trying it at 190x21 or 191x21,
what should I be changing first to achieve stability? Or even just some
recommendations for making the 189x21 setting run better? Perhaps I've
been increasing core voltage when it should be VTT I should be boosting?
What about the RAM timings?
I didn't want to proceed with further core voltage increases until I'd
heard some opinions. I've done a moderate amount of overclocking in the
past, but am certainly no expert. I know least about RAM timing issues,
though I did discover that switching from 9-9-9-27 all-Auto to
10-10-10-29 did help at one point.
Also, watching the results in RealTemp as tests crunch away, I notice
there seems to be a consistent inter-core temperature difference of
about 3/-1/3, ie. if the 1st core temp is X, then the other three temps
will normally be: X+3, X-1, X+3. Should I infer anything from this? I
was wondering if it implies the HS isn't attached evenly. I used AS5,
but I could redo the setup using Thermalright's own supplied paste, or
maybe AS5 is fine and I just didn't apply it correctly.
I've tried to find some examples other people have posted around, but
it's not easy to locate such info as most users are posting results for
Anyway, my goal is 4GHz, ie. either 200x20 or 191x21. I would very much
Note that all tests so far have been conducted with the system lying
flat and case panel removed (this pic taken before the Spot Cooler was
Thus, the intake/exhaust fans shouldn't be a factor atm. Once I reach
4GHz stable and fit the panel, if things do get too warm then I have
two more 1900rpm Kaze fans I can use, plus some fan controllers to
perhaps pull them back a bit if the noise is too much.
Besides, atm the testing has the panel off, so this isn't a factor just yet. If it does turn out to be an issue, then I do have
a larger case I can use (would just rather not as the other case biege, old & ugly, former Viglen CX1). If I do have to
use the old case, I'll spray the damn thing black & gold before any modern components go anywhere near it. :}
> 2nd, since you aren't concerned about how long your computer will last, just keep upping voltages. You are way
> over Intel's max for vtt already, since it's 1.21v.
Many, many people are using more than 1.21 with no problems. I'd certainly like to have it lower though, thus my query.
It's the precise interplay between the different options that's not so well covered from all the reviews, etc. I've read. I'm
trying again atm btw, this time with a 20X multiplier. So far I can get 3900 stable.
However, I just found an interesting how-to guide:
which didn't show up on my earlier Google searches, so I'll maybe start again from scratch, perhaps redo the HSF
NB: one thing which seems to be missing in the BIOS is a way of directly reducing the Bclk/RAM ratio, which is currently
2:12, giving a rather high RAM speed (I'm sure the RAM can handle it, but would be nice to be able to run it slower
in the meantime), ie. running DDR3/2240 at 195x20.
That seems very strange to have an OCable board that wouldn't give you the ability to change the memory multiplier.
My Gigabyte board gives the multiplier option but an Asus board I have gives me the option to set the frequency directly, hence setting the multiplier. It should have some way to alter the setting, but i've never had an Asrock board.
It's because I've selected the XMP profile. If I don't use the XMP option, it presents a
selection of RAM speed choices from DDR3/800 to DDR3/2133, which of course then
sets the multiplier accordingly. Just a different way of doing it I suppose.
I've redone the HSF attachment following Artic's instructions and the results are definitely
better. Letting it burn in atm, but now there's virtually no core temp variation under load,
cores showing either 42 or 43C (just defaults for now, ie. 2940MHz running Prime95). Also
most intriguing, just using the OC Tuner allowed me to reach 3900MHz with no change in
core voltage (just an increase in Blck to 150, Turbo enabled). Definitely looks more
promising this time.
Meanwhile, I've received a Kingston DDR3/2000 kit. Should be interesting to see whether
it's any better than the GSkill Ripjaw kit, though that's for later. Ideally I'd use both, but it'd
be surprising if that worked ok, ie. mixing different brands.
CPU Mult: 21
QPI: Auto (6.876GT/sec)
DRAM Timing: all AUTO, except for Command Rate = 2
CPU Voltage: Overdrive Offset = +0.16075
DRAM Voltage: 1.615
DRAM CTRL/DATA REF: AUTO
Onboard LAN #2: Disabled
Note that the DRAM speed menu shows a DDR3/2292 option, but I don't think the
Ripjaw kit can handle it. However, I do also have a Kingston DDR3/2000 HyperX kit
I can try, not done this yet.
I've been accumulating some benchmark info using 3DMark06 and Stalker COP:
The results match or exceed my friend's i7 930 4.1GHz 4890 CF setup, and I get
better CPU scores (probably because I have faster RAM). OTOH, I expect 4890 CF
would be a lot better if we were testing at high res with AA.
Anyway, I know I had a bit of a struggle trying to find P55 overclocking info, so I hope
this helps others in the same boat!
Only thing I've not yet tried is testing with HT disabled. I've read this can drop load
temps by quite a bit. If my main application (video encoding MJPEG data) doesn't
benefit from HT, then this might be a long term option and a route to a slightly
higher clock, assuming power consumption doesn't go crazy.