X58 + i7 930 adventures

[koolhandluke]

Hey all-- I just got a 930 on a UD7 and got it to initially run at 4.2 ghz into linux. I did some benchmarking and it froze up. I performed a couple of adjustments such as raising vcore and dram voltages and had trouble booting up for about 10 minutes. I'm now unable to get back into linux at 4.2 ghz no matter what voltage settings or dram timings I set. It boots of course but the kernel crashes shortly after loading.

I was wondering if not being able to replicate a 4.2 clock is due to hardware damage.

I'm using the H50 and temp isn't the problem since the chip is keeping between 40 and 50 C.

I've currently got it somewhat stable at 4ghz with a 20 multiplier and 200BCLK

Anyone have any thoughts, suggestions?

 

[koolhandluke]

o some subtleties I forgot to mention
Vcore 1.33v
Uncore 1.31v
Dram. 1.64 @ 9-9-9-20
Memory: patriot viper ii 1600 4gb
I also changed from single to dual channel after my encounter @ 4.2 so perhaps that's an issue.

 

[n7]

Possibly single channel was easier on the memory controller than dual channel, & as such, you'd need more voltage to achieve the same w/ dual channel?

To make sure these don't hold you back, make sure you set QPI link to the lowest speed (36x), & uncore multi to no more than half of that (17-18x max).

 

[koolhandluke]

Ahh. I'll switch and see. Question is if it works and I'm able to get back to 4.2 or even 4.4 ghz is it worth sacrificing dual channel. I'll have to get a suitable benchmarking suite for linux for at this point I'm basing performance on feel.

 

[koolhandluke]

Indeed it was a single channel issue. I'm back in writing this at 4.2ghz.

 

[Rubycon]

LOL you're not going to "feel" any difference unless you go back to stock or even lower.

 

[Apocalypse23]

HyperThreading on results aren't very exciting as temps hold me back as usual.

Nice even numbers though.

Got the RAM's primary timings tightened up, & w/ a bit more vDIMM, i can actually do DDR3-1600 7-7-5.
Not really worth it though.

4 GHz (20x200) 1.275v BIOS (actual ~1.25v)
DDR3-1600 7-7-6-24-1N 1.6v

Did you try running 21x multi for the 4GHZ with HT on OC? Why 20x? I'm currently playing with my 930 atm , my current OCZ ram is giving me a headache as it won't run past 3.78Ghz with HT on with a 21x multi. Haven't tested with HT off yet. My higher quality Crucial memory is coming in the mail this Thursday.

 

[Kougar]

Howdy n7! I found a retail CPU with this batch# that I may or may not buy... after an hour of looking around, I'd say your results are above average. Most of the newer 300* codes packed the last two months won't even do this without 1.4v or more.

Given your unique memory arrangement and low QPI voltage, I'm very curious if you've conducted any further testing with your chip? Done any testing for a max stable Uncore or higher memory speed? Enabling LLC makes the chip extremely hot so that certainly would be contributing...

I hate gambling, so I figured I'd do research this time around on batches and see if the result justified the time involved.

 

[wiedersehen]

learned a lot, thank you!

 

[n7]

Did you try running 21x multi for the 4GHZ with HT on OC? Why 20x? I'm currently playing with my 930 atm , my current OCZ ram is giving me a headache as it won't run past 3.78Ghz with HT on with a 21x multi. Haven't tested with HT off yet. My higher quality Crucial memory is coming in the mail this Thursday.

Honestly, i've been lazy.

That & in my free time, i've been wanting to play Borderlands, etc, not spend time OCing. :hmm:

Oh, & 20x200 = nice even numbers :sneaky:

HT off will get you MUCH LOWER temps.
HT on = so much hotter.

I personally cannot see one bit of difference between 4 GHz HT on or 4.3 GHz HT off...either is more than fast enough for me.

Howdy n7! I found a retail CPU with this batch# that I may or may not buy... after an hour of looking around, I'd say your results are above average. Most of the newer 300* codes packed the last two months won't even do this without 1.4v or more.

Given your unique memory arrangement and low QPI voltage, I'm very curious if you've conducted any further testing with your chip? Done any testing for a max stable Uncore or higher memory speed? Enabling LLC makes the chip extremely hot so that certainly would be contributing...

I hate gambling, so I figured I'd do research this time around on batches and see if the result justified the time involved.

There is a pretty detailed thread @ XS, & i believe a not bad one @ OCN, both of which have people listing their batches & OCs.
Don't think mine is very good compared to some i've seen, though if i had great cooling, i think i'd be doing a decent bit higher.

I'm presently using LLC1, which doesn't increase voltage under load...it pretty much just maintains it.
If i use no LLC, i get far too large drooping for my liking, which means i need a lot higher idle voltage.

As for uncore/RAM, that pissed me off quite a bit right off the bat w/ X58.

I basically could not get the 10x RAM multi working @ all.

Due to the [not] fun QPI/uncore/RAM speeds all being tied in to each other & the lowest QPI available being higher than it is on P55, i cannot seem to get my DDR3-2000 anywhere close to its stock speeds. D:

I'll have to fight with it more again sometime, but after trying various higher/lower QPI/uncore/RAM speeds combinations in which the majority kept ending in no POST, i gave up & settled on the 8x RAM multi till i feel like corrupting my OS someday. :\

 

[Kougar]

Hm, well then Gigabyte has changed how LLC is handled on their X58A boards. For the standard X58 series (and my old UD5) LLC is either full on or full off. Asus seems to use the "LLC1" by default, but I do get warmer results when I switched to ASUS too so I really don't know. Might have been my mount. I used to use LLC on Gigabyte because of the huge vdroop issue, but on ASUS the vdroop is much less and I've been able to leave it off.

The DDR3-2000Mhz requires a 4Ghz uncore, I'd say more than half of the Core i7's out there can't do this. My own Core i7 920 goes unstable with a 3600GHz or higher uncore. This was one nice thing about Gulftown... the RAM can be 1.5x of the uncore instead of the 2x for Bloomfield.

I run Folding@home and HT is required to get high enough performance to meet the Bigadv work unit deadlines (even at 4.2Ghz ). But I got a nice performance increase switching from DDR3-1333 to DDR3-1600 as well, so I hear you on the RAM...

I did find your threads all over the place (sheesh, you sure get around :biggrin, but Google has been the best for finding batch #s as they index ALL of those forum threads. It's crazy, but even with Google I'd say 9 out of 10 batch # searches yield no results yet. I found two potential batches plus yours, but none of the confirmed special ones yet...

 

[dac7nco]

Yes one can figure this out with experience and above all common sense.

I've used some of the cheapies that fit on a keychain. The thermohawk is one of my favorites. Nothing fancy, no laser, short range. But it's excellent at finding hot VRMs on boards. I'm talking the ones that start to smell - this little devil has saved my skin - literally.

IMO it's an accessory that no techie should be without. Instantly you can read your temps around the cpu socket and the back of gpu boards. May not produce EXACT pinpoint results but certainly beats touching it with a finger and saying yeah it's a little hot!

The other thing touching circuits running at multiple ghz can do bad things for the circuit AND you.

LIR is also nice to see the big picture but can look alarming to those without electronics backgrounds and knowledge that certain areas and components are supposed to run at given temperatures. Comparing pictures taken over time will tell the true story, however.

The Thermo's are the best HW utility I have - the Tyan 8P systems get CRAZY hot. Freakish shit can happen with those boards mixed with a 1680 controller.

Daimon

 

[Rubycon]

The Thermo's are the best HW utility I have - the Tyan 8P systems get CRAZY hot. Freakish shit can happen with those boards mixed with a 1680 controller.

Daimon

I can remember having BIOS address overlapping issues with multiple BOOT ROMs from several HBAs on a particular system. Combined with high end display adapters and it's hair pulling time indeed.