Cannot run Kingston HyperX at spec

[thehighroad]

Hello,

I recently had another thread about 3 DIMM, and 3GB, but I chose to buy one 2GB module and changed over to Windows 7 64-bit.

So I mixed the memory a little, hoping it wouldn't be a problem but it seems to be.

My DIMMs are populated as follows in Dual-Channel interleaved mode:
[1 GB GSkill] [1 GB GSkill] [2 GB Kingston HyperX] [Empty]

They are all rated at DDR2 1066 5-5-5-15. However the GSkill is 2.0-2.1V the Kingston HyperX is 2.2-2.3V.

I cannot run them at full speed, my PC won't even post even at 2.3V.
-I tried 7-7-7-21 at 1066 @ 2.3V, BIOS locks up.
-Mobo is Gigabyte GA-EP43-UD3L (P43 Chipset) BIOS is 10d (unreleased, specially for SSDs)

I have a core 2 quad Q8200 overclocked to 2800MHz, and my FSB is thus 400MHz quad-pumped.
Currently, I have the divider at 2.40B which is running them at 960MHz, 5-5-5-15 and it is stable @ 2.10V

I had the divider at 2.50, 1000MHz 5-5-5-15 while I was configuring my 64-bit OS. Through the course of it (~3 hours) Windows crashed 4 times. In event viewer they were all critical kernal-power crashes. There wasn't any BSOD. Yet I performed 8 passes of Memtest x86 with no errors.

Is this the fault of my PSU or what?
Rosewill 530W 80Plus It isn't the best, but I've had NO problems prior to messing with the RAM.

Sorry for the wall of text, I like to provide as much information as possible.

Thanks,
TheHighRoad

 

[dividebyzero]

You're pretty lucky you got them working all together at anything over JEDEC spec (DDR2-800 @ 5-5-5-15 and 1.8v). Pushing the bandwidth harder will likely result in you finding out why the Kingston RAM is only rated 5-5-5-15 when you max out the JEDEC spec (2.3v) for DDR2-1066, while lower operating voltage are more in line with 5-5-5-18 timing.

Memtest is more a test for inherent errors within the IC's. The modules can test good (no faults), but can still fail in "real world" use due to differences in the advanced timings (especially if you have mismatched "Static Read Level" as they are called in the Gigabyte BIOS -also known as "Performance level" or more universally, tRD).
Could pay to download Memset and check that all the advanced values are equal across channel A and B. (Here is an old screenshot of my secondary system from an older thread- post #2)
Memset and the Phoenix-Award BIOS differ in how they report timings
Whatever tWTR and tWR is in the BIOS, add 7 if using 2.00 divider, or 8 is using the 2.40 divider to get Memset value.
Whatever tRTP is in the BIOS, add 2 to get the Memset value.


I doubt you'll get much better than you already have to be honest. Having said that DDR2-960 is pretty good going under the circumstances. You would also get comparable performance (to DDR2-1066 CAS5) using the 2.00 divider (synchonous with core frequency) for DDR2-800 at 4-4-4-12 -assuming you could get the timings that tight- not a given with either of the RAM modules you're using.

Kingston HyperX data sheet (.pdf)

 

[thehighroad]

Thank you for your reply!

Well, the RAM is actually DDR2 800 5-5-5-18, but has to be overclocked to meet the standards of what it is sold as (PC8500). I.e. there is no JEDEC spec for 5-5-5-15 533, nor EPP; my GSkill ram does have an EPP with 5-5-5-15 533. I think you understood this, but I'm just making sure. It makes little to no sense to me why they'd do this...

Just to clarify as well, are you saying the inability to attain stability at any higher clocks, is due to the Northbridge or the RAM itself?


Attached is my memset 4.1 screenshot. It does in fact appear that A & B are running at the same rate.

 

[dividebyzero]

Your Memset screenshot looks fine. The Performance level (tRD) is fairly high, but that is probably a product of the mainstream RAM you're using, so attempting to tighten that timing may give you a dose of the BSoD blues. If this doesn't worry you greatly then you could try to bring that 13 down a bit - I'm betting that the advanced timings are all still set to Auto in your BIOS -they can be a little conservative, but are designed to give you the best chance of stability.
I'd say that the problem would be that to get to 533 (DDR2-1066) you'd need to overvolt the Kingston RAM to 2.3v, while the G.Skill RAM could maintain the same bandwidth/speed, BUT, would then be running way out of specification -since it would also be getting 2.3v, even though 2.1v is enough to run at that speed. The 2.3v (which by the way is only just under the "Do-not-exceed" JEDEC spec of 2.35v) is probably causing instability in the G.Skill.
Conversely, the Kingston RAM is probably too crappy to get to 1000 or 1066 without that extra voltage push. (Sorry to be the bearer of bad tidings. Kingston are not noted for performance RAM, although their modules are fairly bulletproof at the factory spec.)
Welcome to the world of speed-binning of RAM.
If you're determined to get higher bandwidth, you'd probably be best to look at a quality 4Gb (2x2) DDR2-800 4-4-4-12 kit -since most of them will run quite happily at 1000+ 5-5-5-15 (My Ballistix set will do 850 @4-4-4-12 or 1130 @5-5-5-15 and 2.05v. Performance is pretty much equal for both settings). The converse is not always true -a DDR2-1000/1066 kit of say 5-5-5-15/18 may not be capable of 4-4-4-12 timings, especially if the voltage is in the 2.2-2.3 range.

 

[thehighroad]

Hmm, well thanks for the heads up. In that case then I'll probably just leave it as-is unless one day I really need those 1-2FPS which will probably be never.

I guess I just wanted to verify that the way to get the RAM up to spec wasn't an easy fix.

Thanks again for your help on this matter, DivideByZero.

 

[thehighroad]

Update

I had some time to mess around with it today and I got the Performance Level down to 8 (tRD). Which I'm sure is pretty good.

For the record, CL-4 will not boot.

 

[dividebyzero]

That should cut a sizeable degree of latency off your memory performance.
Pretty damn good.