TeamGroup Cronus Micron - Part TWO!

**ozzimark** · 04-01-2006, 09:18 PM

(sorry if pictures take a bit to load at times. my host seems to be having some slight difficulties)

Ok, to make a long story short, Ben (from TankGuys) and I were talking, and decided that one way or another, something needed to be done about the number of people having trouble with these Micron sticks... Nothing better than a little bit of "normal" testing, is there? That said, a HUGE Thank You goes out to Ben @ the crew at TankGuys for helping me get this done. I hope this thread will make a good supplement to OPB's "real truth" thread, and helps add some additional insight to those of you with picky Micron sticks

So, I present to you a random kit. My last thread showed what micron 5b F is capable of, and this is more "what to expect" if you get a retail kit

Note the date code, 0609M. Apparently, this doesn't matter, as we have word directly from OPB that ALL Team's chips are week 0538

Also, I'd like to point out the voltage rating. I forget when this was asked for, but it definitly didn't take long for them to show up on retail sticks. Though I'm gonna have to ask: Does going over rated voltage void warranty? (I'm assuming not)

NOW! Let's get into testing, shall we? I'm going to focus on clocks for now, performance can come later when I have something to compare it to. Don't ask for 3d benches though, as I'm running on a measly Geforce4MX-440

EDIT:
The board I'm using is a DFI NF4 Ultra-D, R.AD0, bios 11-14

First off, used optimal settings from my first set (including drive @ level 5 and data drive @ level 1 are used until otherwise noted) and set the stock primary timings & mhz. Headed off into memtest at 2.6v. Came out a few minutes later, unscathed

Being the impatient guy I am, I opted out of running 32m at stock mhz, immediately went for....
255mhz @ 3-3-3-8, 2.6v
Take note of the SPD. What's shown in there WILL NOT effect the clocks, assuming that you're NOT overclocking with "auto" settings

but 255mhz is pretty lame, isn't it? Let's see how far I can get without having to mess with too much.
260mhz @ 3-3-3-8, 2.6v

Getting to 265mhz required a little bit of tweaking. Slightly dissapointing, but oh well. Take note, I had to loosen the secondary timings to get 265mhz done without changing the voltage
265mhz @ 3-3-3-8, 2.6v

Took a bit of work to get to 270mhz stable, but I was getting bored with 5mhz increments at this point, so I went for the max speed with a simple voltage bump and lower tras

273.2mhz @ 3-3-3-6, 2.8v

Though, after messing around with 4x1gb, I came across something odd. I had to loosen timings when at 2.6v to get more mhz. Now, I had bumped up voltage a bit to get higher mhz, but as I was fine tuning the settings to get the most mhz, I noticed something strange....
Tighter timings were behaving better than looser

I also took the time to find the optimal drive strengths. To my surprise, it ended up being at dram drive = Level 8 & data drive = Level 3

276.3mhz @ 3-3-3-0, 2.8v
(Ignore the change in SPD, I was doing some preliminary testing with 4x1gb before running this bench. DFI boards require that all SPD's are the same for all the sticks to be used; I quickly flashed the SPD of the new sticks to that of the old ones NOTE: I WOULD NOT ADVISE MESSING WITH THE SPD IF YOU DON'T HAVE A VERY GOOD REASON, AS IT IS QUITE RISKY)

I thought that was a pretty good setting, so I ran prime95 Blend as I went to get some lunch

Stopped it after about 50 min, with no errors. While not rediculously stable, I've found that about 15 min of prime95 is all Micron based sticks need for 3d stability in my rig.

Trying to push further forward, I was met with some erratic results. I'm sure some of you guys have seen things like this before. Makes me go

's
22 loops?

22 loops again, but not enough!

So, I figured I've pushed the loose timings to high enough levels for now. I went back into the bios and set the HTT back to 200mhz and went for the tightest timings. Almost got 2-3-2 error free in memtest, though it took 3.0v, which I wasn't feeling very happy about. Rather than giving the sticks more abuse than they deserve, I raised to cas2.5:
210mhz @ 2.5-3-2-0, 2.8v
They scale pretty well with voltage at tighter timings, I'm guessing ~215mhz is possible with 3v, but again, unneeded abuse.

Next up was 2.5-3-3. I was honestly expecting around 240mhz here. No such luck
222.5mhz @ 2.5-3-3-0, 2.9v

Though one set is hardly indicative of what 5b F will do at lower mhz (I honestly don't recall ever seeing anyone else try settings below 250mhz

), it looks like keeping the timings loose and pushing the mhz is where it's at for these IC's.

4x1gb Testing:

4x1gb has been quite an interesting experience thus far. I was initially having a lot of trouble staying stable in windows.
First, I couldn't even get into windows XP 64 with the memory hole enabled in the bios to show all 4gb in the POST. the 32bit version of XP (SP2) worked just fine, but only showed 3.2gb. that wasn't satisfactory

So, after doing some poking about, I came across a switch that appeared to do something that should fix the problems i was having, /NOLOWMEM (moved all the pci devices to >4gb). Worked perfectly, and I was able to get into XP-64 without a problem...
BUT! as stability testing ensued, things would happen like I would be almost done with a 32m run and the computer would lock up. I figured that I was over the limit of my memory controller. I was trying to boot at 270mhz, at 2T command rate. When I backed the speed down some.. 250mhz failed to get past POST, 260mhz would sometimes lock up when booting into windows. Needless to say, I was very confused. On a whim, after trying a ton of settings, I set the HTT to 275mhz, and went into windows.. no problems at all

So, first off, just for kicks, 284.57mhz CPU-Z validated - http://valid.x86-secret.com/show_oc?id=83655
After rebooting, I opened up 4 instances of windows memtest, set each to 875mb to use up most of the free ram, and let it run for about an hour. Still sitting at those settings, so I'd say it's pretty stable.
275.1mhz @ 3-3-3-0, 2.8v (2T)

Quick testing puts the max prime95 stable at 276.3mhz again (I can produce a screen if someone so desires), which I find most curious. NO mhz loss, or gain, even though I'm running at 2T, and incurring a huge load on the mobo and memory controller with 4 sticks...

**~~Onepagebook~~** · 04-01-2006, 09:34 PM

Great works, Mark!! Nice and easy to read,too bad my English just as bad as only S7e9h3n can understand, well I am also happy to see all these results on the normal set of 2gb cronus.

Good luck for 2 4gb bench

**ozzimark** · 04-01-2006, 09:46 PM

thanks Kevin

i'll be sure to try to get some results in with a memory ratio too, since it should work out perfectly with where my cpu tops out

ps, i can understand your english just fine

**Hell-Fire** · 04-01-2006, 10:01 PM

Nice write up ozz.

Well worded and easy to follow. Good to see some tweakage. BTW, are the spreaders there just to be perty, or do the sticks get warm at such low voltage?

**TankGuys** · 04-02-2006, 11:41 AM

Just to point something out to everyone as well, in case they are wondering: we did some special shipping with these kits so that nobody at Team knew who was getting them. That way, he would get a random kit just like any other customer.

So, there is no way this set was cherry picked, just so everyone knows

**ozzimark** · 04-02-2006, 11:53 AM

Originally Posted by Hell-Fire

Well worded and easy to follow. Good to see some tweakage. BTW, are the spreaders there just to be perty, or do the sticks get warm at such low voltage?

thanks

i ended up having to put an 80mm tornado i have sitting around in addition to the 120mm fan i normally use to cool any memory i have in...

these sticks seem hotter than my other set, curiously enough.. though my other set doesn't have spreaders on them anymore, these still do

also, i'm about to edit in the 4x1gb test results now that the forums are back up. refresh in about 5 min

edit: also added in how to get 4x1gb working in xp-64

**Deathspawner** · 04-02-2006, 12:15 PM

Great results Mark!

Out of curiosity, have you run any SANDRA unbuffered tests on any of the overclocks? I am curious to see what kind of bandwidth they push at 4*1GB.

Also thanks for the advice on getting 4GB to work in x64. A while back I had to work with Server 2003 in order to see the full 4GB, but didn't even realize it was also possible with x64.

**rick_fx** · 04-02-2006, 12:15 PM

Originally Posted by ozzimark

these sticks seem hotter than my other set, curiously enough.. though my other set doesn't have spreaders on them anymore, these still do

Maybe taking the heatspreaders off will give a few more Mhz's

**ozzimark** · 04-02-2006, 12:20 PM

Originally Posted by rick_fx

Maybe taking the heatspreaders off will give a few more Mhz's

i have ramsinks that these sticks would love, but they must go back to Ben when all is done, i'm not going to put the sticks at more risk than needed

Originally Posted by Deathspawner

Great results Mark!

Out of curiosity, have you run any SANDRA unbuffered tests on any of the overclocks? I am curious to see what kind of bandwidth they push at 4*1GB.

Also thanks for the advice on getting 4GB to work in x64. A while back I had to work with Server 2003 in order to see the full 4GB, but didn't even realize it was also possible with x64.

thanks Rob

I'll get sandra installed and give you some results at 276.3x10, 1:1 with both 2x1gb and 4x1gb. 4x1gb should appear to be slower due to 2T though

**bachus_anonym** · 04-02-2006, 12:30 PM

Thanks for sharing results of your 2nd set, ozzimark

Those 4GB shots are especially nice to look at

At least now we know that average headroom is closer to what was expected by those "with reason". Although, there was no need to prove anythng I appreciate you doing it anyway. Thanks

**Hell-Fire** · 04-02-2006, 12:34 PM

Originally Posted by ozzimark

thanks

i ended up having to put an 80mm tornado i have sitting around in addition to the 120mm fan i normally use to cool any memory i have in...

these sticks seem hotter than my other set, curiously enough.. though my other set doesn't have spreaders on them anymore, these still do

also, i'm about to edit in the 4x1gb test results now that the forums are back up. refresh in about 5 min

edit: also added in how to get 4x1gb working in xp-64

Odd indeed. Its not from voltage, well not directly anyway. Must be from high density or frequency.

**ozzimark** · 04-02-2006, 12:41 PM

i don't know why they're so hot either, but it seems to be a very common trend with micron IC's

Originally Posted by bachus_anonym

Thanks for sharing results of your 2nd set, ozzimark

Those 4GB shots are especially nice to look at

At least now we know that average headroom is closer to what was expected by those "with reason". Although, there was no need to prove anythng I appreciate you doing it anyway. Thanks

thanks bachus. i felt kinda bad about what happened before, so i felt that this must be done in a way..

now, back to what this thread is for..

ran sandra and sciencemark. i wasn't expecting this much of a difference in synthetic benches, but here goes nothing..
2x1gb @1T
4x1gb @2T

each is run at the tightest timings that are still stable at that speed. perhaps i need to run some "real world" tests to dig into this a bit more?

edit: also, just noticed, take a close look at the L2 bandwidth. curious

**sierra_bound** · 04-02-2006, 12:49 PM

Thanks for doing all the extra testing, ozzimark.

**ozzimark** · 04-02-2006, 02:56 PM

i was hoping to be able to match what your sticks do sierra... got a few mhz to go yet

now, i said earlier in OPB's thread that i was going to take a close look at how using a memory ratio effected clocks. luckily, my cpu has just a hair of headroom at the temps i'm at.. like 20mhz or so

so, what i did was lower the memory ratio to 166mhz, and dropped the cpu multiplier to 9x. to get my previous max ram speed, this meant a 338mhz htt speed, 3042mhz on the cpu.

at first, i couldn't even get into memtest, just a blinking cursor... tried fiddling with DRAM response time, and manually changing MAL/RP when that didn't work. slighly worried that what OPB said was true to an extreme degree in my case, i backed down the mhz some until i could get into memtest.. errors ahoy!

long story made short, raising the secondary timings, especially trtw to 3 (even though tighter helped clocks before) helped be get back to about where i was before:
274.9mhz @ 3-3-3-0, 2.6v & 166mhz ratio

so, in retrospect, something strange is definitly going on with the ratios, but it seems that it can be made up for (in my case at least.. i am curious to see what would happen with a dual core though)
also, if dropping the ratio decreases stability a touch.. who wants to see what would happen at >1:1?

**Hell-Fire** · 04-02-2006, 06:42 PM

Considering we are talking 1Gb sticks here (2x1Gb setup), I think those are pretty solid results.

I wonder if people are taking that into consideration when they get upset that their sticks aren't overclocking like 2x256 and 2x512 setups. Whats the old saying: The more links in the chain, the more possible weak links there may be.