Quad Core O/C... shouldn't be this difficult?

Khyris

Junior Member
Mar 1, 2007
4
0
0
I'm going to admit right off the bat I'm a newbie when it comes to overclocking... but I'm learning quick, so anyone who can help me, please feel free to be as technical in your explanation as you like.

Premise: I've just built a machine for statistical simulation and visual modeling, so it's a number cruncher, not really much for gaming. I went with a QX6700 because the software I use lends support for quad core optimization. I'm dealing with HUGE volumes of Data, so I also picked up 8 GB (4 x 2 GB) of Dual Channel PC26400 specifically Kingston KHX6400D2LL_2G. I know that in general more ram = runs slower, but I'm having trouble getting it to work at STOCK speeds. I intend to O/C the CPU to 3.2 ghz and no further, but have not yet tweaked it at all. I have a CoolIT Eliminator water cooler to that end, and due to the nature of my work, the extra clock cycles will make a tangible impact on extended runs of computation, but I don't want to overly-risk shortening the lifespan of the system.

According to the datasheet here
http://www.valueram.com/datasheets/KHX6400D2LL_2G.pdf
"This module has been tested to run at DDR2 800MHz at low latency timing of 4-4-4-12
at 1.9V. The SPD is programmed to JEDEC standard latency 667Mhz timing of 5-5-5-15 at 1.8V."
True enough, The Striker Extreme this is running on (don't laugh, I was already spending too much money on the system and the pretty LED's were a selling point above the P5N32-E SLI) sees the RAM default at 667 5-5-5-15 1.8V
Going into the "Extreme Tweaker" menu on the BIOS, setting the O/C to Manual, unlinking FSB and Memory Bus, Keying in 800 4-4-4-12 for the RAM is easy enough to do... but the system then always blue screens when loading windows. I even set the Voltage to 1.925 just to add leeway. Only by dropping the timings back to 5-5-5-15 could I get the machine to load windows with the RAM at 800mhz. I did not experiment with intermediate values of timings. Anyone have any idea why this is happening? No other values have deviated from default, the CPU is still running at 2.66ghz and peaks at a happy 35 degrees C under heavy load. Does this qualify the RAM is "faulty" and I should seek a replacement under warranty?

I have one other question: what is the inherent value of increasing clock speed via FSB as opposed to multiplier? If the multiplier is unlocked, and given that the RAM bus speed is unlinked and I've hypothetically managed to get it stable, should I see better benchmarks out of the system with 1280 FSB 10X multiplier, or 1066 FSB 12X multiplier?
 

Elfear

Diamond Member
May 30, 2004
7,163
819
126
You'll get more bandwidth with a higher FSB and/or higher memory clocks. That generally translates into higher performance in some applications. There are a few Anandtech articles that show the difference. CPU clock speed reigns supreme though and is more important than a higher FSB.
 

Khyris

Junior Member
Mar 1, 2007
4
0
0
Thanks much, that's good to know

Does anyone have a clue why the RAM wouldn't work at factory specified settings?
 

firewolfsm

Golden Member
Oct 16, 2005
1,848
29
91
Try taking out each stick one at a time and find out which one is limiting you, it could be that you got a bad stick (or even two) that won't run at default settings, RMA that.
 

Shimmishim

Elite Member
Feb 19, 2001
7,504
0
76
usually with more memory, you're going to need more than stock, especially with 8 gigs!

i'd say up the voltage to 1.95 to 2.0 to see if it'll boot. 0.1 above stock shouldn't hurt the memory.

 

aka1nas

Diamond Member
Aug 30, 2001
4,335
1
0
If you look at Asus's QVL list for the striker, they didn't actually validate any 2GB sticks on the board. I'd contact their support FWIW, and see if they have any RAM model they recommend. Unfortunately, when you're dealing with somewhat rare memory configurations like 4x2GB is currently, it can be somewhat hit or miss to get things working.
 

aigomorla

CPU, Cases&Cooling Mod PC Gaming Mod Elite Member
Super Moderator
Sep 28, 2005
21,052
3,534
126
Originally posted by: Khyris
I'm going to admit right off the bat I'm a newbie when it comes to overclocking... but I'm learning quick, so anyone who can help me, please feel free to be as technical in your explanation as you like.

Premise: I've just built a machine for statistical simulation and visual modeling, so it's a number cruncher, not really much for gaming. I went with a QX6700 because the software I use lends support for quad core optimization. I'm dealing with HUGE volumes of Data, so I also picked up 8 GB (4 x 2 GB) of Dual Channel PC26400 specifically Kingston KHX6400D2LL_2G. I know that in general more ram = runs slower, but I'm having trouble getting it to work at STOCK speeds. I intend to O/C the CPU to 3.2 ghz and no further, but have not yet tweaked it at all. I have a CoolIT Eliminator water cooler to that end, and due to the nature of my work, the extra clock cycles will make a tangible impact on extended runs of computation, but I don't want to overly-risk shortening the lifespan of the system.

According to the datasheet here
http://www.valueram.com/datasheets/KHX6400D2LL_2G.pdf
"This module has been tested to run at DDR2 800MHz at low latency timing of 4-4-4-12
at 1.9V. The SPD is programmed to JEDEC standard latency 667Mhz timing of 5-5-5-15 at 1.8V."
True enough, The Striker Extreme this is running on (don't laugh, I was already spending too much money on the system and the pretty LED's were a selling point above the P5N32-E SLI) sees the RAM default at 667 5-5-5-15 1.8V
Going into the "Extreme Tweaker" menu on the BIOS, setting the O/C to Manual, unlinking FSB and Memory Bus, Keying in 800 4-4-4-12 for the RAM is easy enough to do... but the system then always blue screens when loading windows. I even set the Voltage to 1.925 just to add leeway. Only by dropping the timings back to 5-5-5-15 could I get the machine to load windows with the RAM at 800mhz. I did not experiment with intermediate values of timings. Anyone have any idea why this is happening? No other values have deviated from default, the CPU is still running at 2.66ghz and peaks at a happy 35 degrees C under heavy load. Does this qualify the RAM is "faulty" and I should seek a replacement under warranty?

I have one other question: what is the inherent value of increasing clock speed via FSB as opposed to multiplier? If the multiplier is unlocked, and given that the RAM bus speed is unlinked and I've hypothetically managed to get it stable, should I see better benchmarks out of the system with 1280 FSB 10X multiplier, or 1066 FSB 12X multiplier?

id hate to burst your bubble, but, if its number crunching and doing high intensive simulations, then i dont recomend overclocking at all. I think you should keep it at stock.

Thats why large companys that utilize important servers, dont overclock. And im sure any IT guy would know how to do it. Its just not recomended for a full blown server like that.


Also, yeah the QX is a PITA to OC. Thats why i sold mine.

EDIT: oh shiet you had the same system setup as mine. YES our board is uberly BUGGED, mine kept resetting CMOS everytime the power in the back went out. And not to mention i had TONS of issues with that stupid board.

The WW for QX overclocking was set on a EVGA 680SLI, by fugger on his cherry picked QX. If and when the 45nm die's come out and it proves to be better then barcalona, that is the next board i will go with.

Thank god i sold this rig, im seeing deja vu with you all over again.

If you can return that craptasic uberly expensive board and try to exchange it with the EVGA's. Far less troubles are reported with that board and a QX then with the Striker.

And yes the STUPID LCD got me too, thats why i got that board.
 

Khyris

Junior Member
Mar 1, 2007
4
0
0
Actually I was doing this as a D.I.Y. cheaper alternative to buying an HP 530, which is a QX6700 factory overclocked to 3.2ghz. I figured if HP is willing to put a warranty on that speed setting, it's probably reasonably safe to do.

This isn't a mission critical application server, like something customer facing with an SLA... I'm just hoping that a calculation run that would take me a week at stock speeds will take me 5.8 days overclocked. I'm fairly competent as a system builder, I've just never really had a reason to really look into overclocking before now. It's a good learning experience for me =)

I'll try the higher voltage settings and try hitting up asus as well.

I'm kinda to the point with this machine that... it's kinda hard to comprehend the work that went into building it... definitely not selling it or disassembling it.. just have to get out of it what I can get out of it. Thanks again all for the help, and any further suggestions, keep em coming =)
 

lopri

Elite Member
Jul 27, 2002
13,310
687
126
As firewolfsm says, I'd suggest you first check individual sticks separately. You can also test two sticks at a time in a dual-channel configuration. IMO what you're experiencing seems very natural. Memory overclock is always a co-operation of memory + memory controller. With Intel boards, memory controller resides at northbridge, and you may know how different chipsets handle memory already. (just compare 975X/P965/680i/RD600) I had to raise vDIMM by 0.05V when I went to 4GB. Your 8GB will be quite a stress to the memory controller therefore you will have to either loosen the timing (which you already successfuly did), lower frequencies, raise voltages - or any combination of these. 99% of the time memory overclock will decrease with 4 sticks instead of 2 sticks. Just enjoy the smoothness of 64-bit + 8GB! (Damn.. Now I want 8GB, too!)

On a side note, I kinda agree with aigomorla on your choice of motherboard. 680i is heavily catering gamers/overclockers. Various signals/timings are programmed to maximize the bandwidth from PCI-E so often times it'll beat competitors in 3D (especially with NV GPUs) but when it comes to pure CPU push it'll come a bit short of Intel chipsets. And probably not many manufacturers will take 8GB into account for their 680i boards, knowing the demographics of their marketing target. Besides, the 680i itself is known to not clock very well with quad-cores.

For usage as yours, I'd have picked Bad Axe 2, P5W64, or similar boards based on 975X that are targeting semi-workstation/power user folks. I would recommend 680i in current state to folks who want SLI or like to fiddle with various BIOS settings to extract more performance (which often times is not worth the hassle).
 

Khyris

Junior Member
Mar 1, 2007
4
0
0
While the system is not "much" for gaming, especially as far as our accounting department is concerned, it will be used for such on occasion... hence the BFG 8800 GTX OC in one of the primarey PCI-E slot. Hehe, I feel I knew what I was doing when I picked 680i as the chipset.... maybe not so much when I was bamboozled by flashy LED's...

I will definitely try pushing the RAM voltages adn timings as suggested and let everyone know my results on stabilizing this particular RAM configuration. Thanks again all for the comments!
 

aigomorla

CPU, Cases&Cooling Mod PC Gaming Mod Elite Member
Super Moderator
Sep 28, 2005
21,052
3,534
126
Originally posted by: Khyris
While the system is not "much" for gaming, especially as far as our accounting department is concerned, it will be used for such on occasion... hence the BFG 8800 GTX OC in one of the primarey PCI-E slot. Hehe, I feel I knew what I was doing when I picked 680i as the chipset.... maybe not so much when I was bamboozled by flashy LED's...

I will definitely try pushing the RAM voltages adn timings as suggested and let everyone know my results on stabilizing this particular RAM configuration. Thanks again all for the comments!

the 680i is a nice chipset. Im sure if you and i had EVGA's instead of the ASUS, we would of been more happily successful with that cpu.

Hehe.. on the side note... yup definitely seeing all the simular issues i ran into again, with you.

Op, that board... gah... i honestly feel for ya, because i got suckered on getting it for the same reason, only to have my hair fall out. :X
 

Dadofamunky

Platinum Member
Jan 4, 2005
2,184
0
0
Originally posted by: aigomorla
Originally posted by: Khyris
While the system is not "much" for gaming, especially as far as our accounting department is concerned, it will be used for such on occasion... hence the BFG 8800 GTX OC in one of the primarey PCI-E slot. Hehe, I feel I knew what I was doing when I picked 680i as the chipset.... maybe not so much when I was bamboozled by flashy LED's...

I will definitely try pushing the RAM voltages adn timings as suggested and let everyone know my results on stabilizing this particular RAM configuration. Thanks again all for the comments!

the 680i is a nice chipset. Im sure if you and i had EVGA's instead of the ASUS, we would of been more happily successful with that cpu.

Hehe.. on the side note... yup definitely seeing all the simular issues i ran into again, with you.

Op, that board... gah... i honestly feel for ya, because i got suckered on getting it for the same reason, only to have my hair fall out. :X

I've had builds that blew up in my face before because of mobos and I know how you feel... but 8 GB is definitely outside the normal envelope. I was thinking about the Stryker Extreme but that is definitely crossed off my list for the future. The EVGA board has been getting excellent reviews across the board.

2 GB sticks can be very finicky and hard to oc at all. Be very careful with them. Particularly with Kingston, which is not widely considered to be effective OCing RAM. OCZ and Corsair are considered to be the kings in this area (and I THINK both have some 2 GB stix). Stock or just above stock may be all you can do unless you want to do another mobo replacement. RMing RAM just because it won't OC is a nonstarter.

Label each stick numerically, and test each of them in pairs. You should be able to isolate whether one stick is defective (which seems likely, given the density and the sensitivity of 2GB stix). But in all cases, be very careful raising voltages with them, as those 2GB stix may not take kindly to higher than rated voltages.