How does it look?

[paradisiac]

how does this compare to other i7 920's?
reason i ask is im using a x19 multiplier and others are using x20/x21 so wondered if it made a difference.

CPU Clok ratio 19x
QPI Link speed x36
Baseclock 211
Performance Enhance - standard
XMP disabled
System memory multiplyer SPD - 8.0
intel turbo boost - enabled
Cpu cores enabled - all
CPU multithreading - enabled
C1e , C3/c6/c7, thermal, EIST and PROCHOT all disabled
Virtualization - enabled.
Uncore freqency x16
Isochronous supprot - enabled
PCI frequency - 100
CIAZ -disabled
CPU clockdrive 800Mv
PIC Express clock drive 900Mc
CPU cliock skew 100ps
IOH clock skew 0ps

Local-line calibration - disabled
CPU Vcore 1.300V
QPI/VTT coltage 1.295v
IOH Core 1.16V
DRAM Voltage 1.640v

everthing else is as at default / auto

Ram : 8-8-8-20-1
Max Temp Under Load : 78

Spec : i7 920 D0 @ 4.0GHz, Arctic Freezer 7, GA-X58-UD3R, TEAM XTREEM 6GB DDR3-1866 TRI CHANNEL, 9800GTX+, Samsung SpinPoint F1 HD103UJ 1TB, Hitatchi HDT721050SLA360 500GB, Maxtor STM3500418AS 500GB, NZXT BETA Case, Windows 7 Ultimate 64 bit, 42" LCD TV

Res : 1800x1018

CPU-Z

3d mark 06

3d mark vantage

passmark

fritz chess benchmark

super pi

cinebench

any comments welcome.

 

[paradisiac]

no one got any ideas if the x19 multiplier makes a better/worse overclock?

 

[paradisiac]

i have done plenty of stress tests with OCCT to make sure its stable, but OCCT doesnt really give anything useful at then end that i understand, just some graphs. i know its stable, as its been tested and i have been using it for what i normally do and have found no issues at all with it.

 

[paradisiac]

Why did you start a new thread instead of continuing here?

because in this thread i was hoping to find out how the different multipliers effected the performance, not asking why mine is only stable with a x19 multipier and why i cant get it to work with higher ones, which is what that thread was. sorry if that was the wrong thing to do.

 

[paradisiac]

*bangs head against wall*

so you are saying OCCT is not appropriate for stress testing stability??

 

[paradisiac]

ok fair enough, dont see why not as it puts 100% load on the cpu, so if there was instability surely it would pick it up, but heyho. i have had it running fine now for days without issues, so im happy its stable, even if i dont have a prime95 screenshot to tell me it is.

 

[paradisiac]

but if it works for everyday tasks does it matter?
im a bit confused, surely there is no way of saying any pc is stable then, if different tests fail whilst others pass, but if it works for what you need it for, does that not then make it stable for your needs?
i have used it to encode video, i have played games, i have watched films and listened to music and so far i have had no signs of a problem, so does this not therefore mean, in my situation, its stable for what i do?

 

[paradisiac]

so you are suggesting 24hours with prime, intelburn and OCCT, aswell as memtest86+ added with a week or so of everyday use, would constitute a stable pc?

 

[paradisiac]

i dont want to go over 4GHz, and am pretty sure im stable, dont quite get prime as never used it, doing a blend torture test, will a few hours be ok or does it need to be 24hours?
also what is considered good for OCCT and intelburn, i have been doing 8-10hours OCCT, intelburn i have only briefly used before, but was a while ago.

 

[paradisiac]

ok will give it a go. if you do get errors, how do you start to fault find whats causing it? from what i have read, most people assume its the vcore, but as i changed sooo many settings it could be anything.

 

[paradisiac]

Since you've gone from stock to 4ghz without stability testing, you have no method besides blind guesswork with which to work out what needs to be changed. You'll need to slightly alter several settings, and some may need to go up whilst others down. This is why it is recommended to overclock a bit, get it stable, then overclock more, as you will need to change fewer settings each time the closer the intervals.

The idea is get x bsck stable. Then increase a tiny bit. Hopefully you will now only need to change one setting to get it stable again, so try things sequentially, setting them back if they didn't help, until it's stable. Then move up a bit more.

Or you can jump in at the desired speed, hope it posts, then spend ages guessing what needs to be changed. In practice a compromise between the two is best, but for a novice, the slow and steady approach is much more reliable. I went from 133 to 200 bsck in steps of 5, and will be going up in jumps of 2mhz next time I try for over 4ghz. I'm considered cautious though.

i was running it at 3.7GHz for a while, but again i only used OCCT and memtest86+ to test it was stable. the thing is, i dont really know what everything is, and as everything piece of hardware is different then its hard to have a generic setup to get to a desired speed, just rough figures that might work, which is what i have been working off, looking at lots of other peoples settings with the same mobo/cpu combo and seeing if there are any figures that pop up regularly. but the thing is, im not to know if they have tested their setups fully.
the thing is, im warey of temps, i dont want to go over 80 if i can help it, and get to as close to 70 as i can.
to achieve this, i have been trying to keep my voltages as low as i can, making it post, boot and do 8hours of OCCT. now im aware that this is not enough testing, so its now going to be a huge learning curve as to what to do, keep it as it is and see if i get crashes, start from scratch but pretty much going in blind, or tweaking what i have and doing lots of stress tests, but again, doing it blind.

 

[paradisiac]

Minimising voltages is a good thing, not so much because high temperatures are dangerous but because it's harder to get it stable the higher the voltages are.

I think you've guessed the flaw in your strategy. Looking at lots of peoples settings and trying to correlate them is never going to give you your settings, as hardware is unique. Well, not all identical anyway. Instead it's a much better idea to find out what all the individual terms are. I personally google everything in the list, one at a time, then experiment for a while and then google the terms again. The better you understand the settings, the better your chance at getting them set correctly.

This approach isn't without fault either though. It's time consuming, and you're still a bit screwed if you come up against an unusual problem.

i see where you are coming from, and it makes a lot of sense, and i tried prime, it bsod after 4hours of blendin but that could be down to anything so im no further forward other than im now sure its not 100% stable.

 

[paradisiac]

in your experience is 4Ghz stable obtainable?

 

[paradisiac]

My experience isn't particularly comparable. I'm using a Gigabyte UD5, so while your bios will look pretty familiar to me, my board is "supposed" to be better at overclocking. I'm also watercooling. However I'm running 12gb of ram instead of 6, and I'm very certain that makes life a lot more difficult.

yeah and im using cheap ram so maybe thats an issue too.

I think I'll get 4.4 with 12gb of ram stable when my board gets back from rma. The number of people on here with 920s at 4ghz certainly suggests its very possible, and at least a few of them are using UD3Rs.

but are they all proven stable, or are they like me, it works ok with what i do so its ok?

I suspect you'll be able to get 4ghz stable by only varying vcore and qpi, personally I'd try 1.3V vcore, 1.3V qpi, 1.66Vdimm, x36 qpi multiplier, x15 uncore, x6 ram, turbo off, bsck = 200, multi = 20. I know that's stable on my board, so it's where I'd be inclined to start with your one. If it's unstable, lower bsck until it's stable then work upwards, playing with the uncore multi and vcore. This is a poor second to researching all the terms and spending ages on trial and error though, as I only really know this would work for my system because I've tried it.

PCI frequency - 100
CIAZ -disabled
CPU clockdrive 800Mv
PIC Express clock drive 900Mc
CPU cliock skew 100ps
IOH clock skew 0ps

what about these, put back to auto?

 

[paradisiac]

Hard to say if they're stable by my standards, by gaming-only standards, or something else entirely. Rjkoneill seems to think some of the ud3r boards struggle to hit 4ghz stable, that you made it 4 hours into blend suggests you're pretty close so I think you'll be alright.

pci-e frequency should be 100, though 102 or 103 will help if you try for over 210 bsck.
clock drive should probably be on auto, though these seem to be system dependent which settings work best. Mine seems to like 800/800.
clock skews on auto or on 0ps I believe, as they shouldn't matter at all for these frequencies

cheap ram shouldn't matter actually, unless trying to go over the specs written on them

ok, well will save my settings and then start tweaking it a bit. is it true that 1.35v is the safest max for the vcore?
im aware my mobo isnt the best, but i was only planning on hitting 3.5-3.7GHz, but the temptation to reach a round 4GHz was too high. but this has now made me wonder how many 4GHz's are actually stable.
as for my ram, its running under what the sticker says in some instances as its 1866MHz and im only just over 1600MHz i think.

 

[paradisiac]

The principle I work on is no more than 1.35V for uncore (QPI/VTT?) and no more than 1.5V for vcore - although currently my vcore is at 1.35 otherwise temps get unmanageable on my current cooling.

im on air cooling so i dont think i will be able to get much over 1.3v is there any way of setting the cpu fan to run faster, or get it to kick in sooner?

 

[paradisiac]

been on gigabytes site and there is a new bios update, but its a BETA. is it worth downloading or shall i wait for the final release?

Version : FF2
Release Date : 2010/01/12
Description :
1. Beta BIOS
2. Support Intel Gulftown CPU

 

[paradisiac]

My experience isn't particularly comparable. I'm using a Gigabyte UD5, so while your bios will look pretty familiar to me, my board is "supposed" to be better at overclocking. I'm also watercooling. However I'm running 12gb of ram instead of 6, and I'm very certain that makes life a lot more difficult.

I think I'll get 4.4 with 12gb of ram stable when my board gets back from rma. The number of people on here with 920s at 4ghz certainly suggests its very possible, and at least a few of them are using UD3Rs.

I suspect you'll be able to get 4ghz stable by only varying vcore and qpi, personally I'd try 1.3V vcore, 1.3V qpi, 1.66Vdimm, x36 qpi multiplier, x15 uncore, x6 ram, turbo off, bsck = 200, multi = 20. I know that's stable on my board, so it's where I'd be inclined to start with your one. If it's unstable, lower bsck until it's stable then work upwards, playing with the uncore multi and vcore. This is a poor second to researching all the terms and spending ages on trial and error though, as I only really know this would work for my system because I've tried it.

it didnt like that at all, failed to post, which goes to show that one persons settings will not be the same for another chip/mobo. back to square one.

 

[paradisiac]

reverted back to my original 4GHz OC and changed a few settings see if that sorts it, if not then will go back to my 3.7GHz and see if thats stable, and if that fails, then start again think my mobo just doesnt like the multi being anything other than x19.


maybe there should be a sticky/how to on checking stability of an OC, as im sure i am not alone with thinking running a few apps will constitute a stable setup. and also how to use the programs and the best settings to get the most accurate results. just a thought, maybe there is one and i missed it

 

[paradisiac]

You need to stress the system. Just opening a few things and seeing that it can do it isnt good enough. I only consider an overclock stable after 8 hours continuous stress testing past. OCCT, IBT or Prime95.

Some people go for longer.

I disagree with the people who just run a game for a while and say its fine.

this is what i did, i ran 8-10hours of OCCT and was told this was not enough, which i then found out doing prime as it failed after about 4hours.
after tweaking it i have run OCCT for 30mins (i find it usually runs fine for the 8 hours if it manages 30mins but will do it overnight when i get chance to make sure) and its done 10 cycles of intelburn on maximum (around half hour) and that passed, going to do a quick prime blend and then if its still ok will continue to use it and then do some more tests overnight when i dont need it.