First OC Results - Mobile 2400+

Well, after months of anticipation, I finally have most of my system set up. My case needs a lot of mod work to improve the air flow (the front section has three fans, but they're all behind rather bulky built-in grills, which are in turn behind an air filter of sorts, which in turn are behind a grill with much smaller holes, and all of them are off center. So I took out the first grill and air filter, and am going to cut out the front of the case and replace it with a sheet of aluminum and cut my own fan holes so they're going straight down the middle. But I digresss ...).

Just thought I would share what I've got so far. I still want to lap my SP-97, as it looked like it could use it. I'm at 2.9V for the vdimm in the bios, and 1.65V for the vcore, though CPU-Z reports a vcore of 1.616V and MBM5 reports it at 1.62 - 1.63V. I have yet to bust out the multimeter, so if someone wants to point me to some voltage read points for an NF7-S v2, that'd be great.





Sorry for the filesize ... I don't have PhotoShop installed on this new rig yet, so I had to use MS Paint.

I haven't run a Prime95 test for this yet. I tested Prime stable 12+ hours at 2300 Mhz (11.5 x 200), and today was the first day I bumped it up to 2400. I ran 10 passes of MemTest86 error free today with my ram set to 218. I tried booting at 8 x 220 the other day to max out my FSB, and it wouldn't boot at times, other times hung on the bootup screen, and made it to windows once. I haven't tried 219 yet with a low multiplier, but if it can take it, it looks like that's gonna be my limit with a vdimm mod (which I'm waiting on my soldering skills to improve before trying).

My max load temp, using Prime's In-Place Large FFT's for max heat and power consumption, was 44°C during the warmest day we've had in our house ... normally Prime gives me 42 - 43°C max, and in the evenings, around 40 - 41°C. The heating and cooling distribution in our house is terrible, though. When it's 69°F in the back bedroom, the kitchen can get to 75°F. So I think I've still got some more room to play with it. I'm willing to take it to 50°C load, so we'll see.

Anywho, I've been hanging out with you guys and sharing my sob stories about getting parts together since February, so I thought it was time to post some results.

Take care,
Dave

PS - Oh, and for reference ... that PIII 500 comparison score is there because that's the CPU I've been running for the past 5 years. Quite a noticeable difference after the switchover.

Litte question about Sandra results while I'm at it ... in the Whetstone benchmarks, what does the FPU/iSSE2 score mean, when it has two numbers as a result instead of just the FPU?

sounds sweet man. 2.4 ghz on stock voltage - can't argue with that. i gotta ask you what kind of psu are you using?

Antec TrueControl 550W

Question on this:

At 218 (multi at 11), my RAM makes 10 passes of Memtest86 error free. At 11x218, Prime95 fails within 10 minutes. However, at 12x200, Prime is fine (I'm at 1.65V or a little lower according to two programs, have yet to test with a multimeter, and my Bios is actually set to 1.675V). So it seems my cpu is stable at 2400Mhz according to Prime, but Prime has a problem with my higher FSB, while Memtest says the RAM is stable there.

What's the deal? Should I trust Prime's response more than Memtest's? Would upping the VDD to 1.7 from 1.6 help stabilize it at all? (and I've read that at higher FSB speeds, it actually helps to drop it, so I'm gonna try the L12 mod this week and see what I can get after that).

Any feedback would be appreciated.

- Dave

it sounds like your mainboard may not be able to pull those high fsb's. the memory is very very important to getting those high fsb, but if your mainboard can't pull it, then it won't be stable. the highest i've consistently seen is abit an7's getting in the 230's. i mean, i have pc4000 ram, which is a bit of overkill, but if i used your way, i'd be at fsb 250. that's not going to happen (at least not yet).

Very nice... that OC is actually better than my 35W 2400+

goheels - the mobo handles 218 pretty nicely, actually. I play BF:V on 11x218, and I didn't have any problems at all. And like I said, Memtest made it through 10 passes (and I hear the 10th pass is the one the typically starts to error out suddenly, I believe) at 11 x 218. So what I'm wondering is, why is Prime hating that FSB setting, but Memtest is cool with it (when I know Prime can go fine at 12 x 200).

I'm curious what the true voltages are, too ... like I said, 1.675 in Bios, CPU-Z is showing 1.648, and MBM5 is showing 1.63, so I'm assuming my board is undervolting some, but I don't really trust software. I suppose as long as my temps stay good, I don't have to worry too much about what the true voltage is.

Sentential - Thanks. I've been following a lot of your threads ever since I got here. Any input you have about the FSB stuff (since I know you're a multiple NF7 man, though sadly multiple in one rig, as goes your luck, it seems ) is greatly appreciated, too. What kind of stock do you put in the different stress tests?

If I ran Large FFTs instead of Blend to stress the CPU, and then ran Memtest to test the RAM instead of letting Prime focus on it, do you think that would give a more reliable reading of stability? (I've heard Prime can be really frickin' picky sometimes).

Take care,
Dave

NF7-S has tendency to undervolt CPU a bit

Well, glad others experience the same thing. I don't mind ... I like setting my vcore to 1.675 and know I'm really only at 1.65 ... kinda like when you set your clock 15 minutes fast, and then when you wake up in the morning, you smile because you remember you actually have 15 more minutes to relax.

However, I must admit I lied (though not intentionally). 12 x 200 isn't Prime stable at 1.65v actual, either. It just took a good deal longer to fail than 11 x 218, and since games seemed to be running fine, I didn't test it longer the first day.

So now I'm back to 11 x 218, but with 1.68v actual, according to CPU-Z and MBM5 (what's weird is CPU-Z keeps jumping between 1.664 and 1.680). So a .03v overvolt for 600Mhz increase in speed ... we'll see if it passes this time around. Hope so. I'd hate to keep going up, because a 1.63v, I'm rock solid at 11.5 x 200, and I'm only going up 100Mhz from there.

I'll update the thread as I get more out of the chip.

- Dave

Well, Prime failed in 10 minutes again at 11 x 218 with the vcore at 1.68. I have the vdimm at 2.9v (though MBM5 shows it at 3.0v actual), and the chipset at 1.6v. I wonder if I should raise that to 1.7?

Currently, I'm dropping the FSB back to 200 and raising the multi to 12 again to see if I can get it stable at this voltage. If not, I'm gonna be kind of bummed, because if I have to up the voltage anymore for this 100Mhz jump above the 2300Mhz I was running stable at 1.63v, I don't know how much it'll be worth it.

I used to be a strong proponent for Prime95 as the ultimate overclocking stability test but I have stopped recommending it due to irregularities I saw when ppushing at the top end of my setups capability. With Prime95 Torture Test Priority 10 running and nothing else, my 11x220 would run for 24 hours error free. The next try, it might only make it 18 minutes while the next try would go for 2 hours, then 45 minutes, 5 hours, etc. No predictable failure period. My setup runs cool with the CPU, NB, GPU all being water cooled, along with copper MOSFET, RAM and SB sinks in a room regulated at 20C. Those of you who know me, know that I'm fastidious about case ventilation and high performance water cooling. Vcore has no effect above (true)1.8V, nor does Vdimm above (true) 2.9V, nor does Vdd above (true) 1.7V

(true) means measured at the chip using an accurate DMM, not relying upon the often inaccurate MBM readings.

I can run 11x210 at (true)1.70Vcore and (true) 2.8Vdimm Prime95 Torture Priority 10 until the "cows come home" so I must assume the erratic behavior at 11x220 is indicative of an unstable overclock by Prime95FanBoy standards. Outside of Prime95 anything I throw at the setup runs stable at 11x220.

The perceived/realized performance difference between 11x200 (2-2-2-11), 11x210 (2-3-2-11) and 11x220 (2-3-3-11) is in all honesty, not there as far as I can tell and I don't run synthetic benchmarks for entertainment purposes, so I don't worry about it.

Hoot

So besides Prime (which I'm a little fed up with), what other stability tests do you run? Something like Sandra's Burn-In wizard, or do you just run it until a program you use that heavily taxes the system crashes?

Well, I'm posting with you now at 2520Mhz (12 x 210), at a vcore of 1.825, a vdimm of 2.9, and a vdd of 1.6.

I enabled CPU Interface, and was actually able to post and boot at 11 x 220 (which was a FSB wall for me before). I ran a couple of Sandra benchmarks with no troubles, so I rebooted, and tried for a higher speed.

At 11.5 x 220, I got a two-tone siren if I use anything less than 1.875 on my vcore, and 1.7 on my vdd. Not wanting to run those high of voltages yet (and I realize my RAM is what's holding me back ... time for the L12 mod, I believe, and then maybe Dirty Punks vdimm mod), I kept the same speed (well, minus 10Mhz), and upped my multi and FSB to their current settings and dropped my voltages back down.

I'm going to start burning in at these settings, and then see if I can post higher FSB speeds 1Mhz at a time. Ideally, FSB-wise, I'm shooting for 245+ so I can run a multi of 10.5 or 11 (grand dreams, I know).

If anyone has any tips for getting the RAM to run faster (at what speed should I actually start DECREASING my vdd, do you think?) besides the L12 mod for now, feel free to share.

- Dave

maybe relax mem delays a bit...
my mem ran like 2-2-2-5 T1 at 210mhz, 2.5v (but I ran memtest pretty shortly, as it was teste on friend's mobo; old Hyper-X PC3200); so your mem should work similiar... But for 220Mhz+, I'd relax timings

btw, for so high voltage, U need forced air flow for memory; Other thing - it's 1Gig of ram at high voltage - mem voltage regulators on your mobo may have problems with that - try with 1 stick only and then again with 2, then in dual - U'll know what limits speed

Well, I'm not going to relax timings ... I didn't buy BH5 to run at 2-3-x-x. This is a pretty typical wall I'm hitting as far as NF7's are concerned. I have a fan blowing down on my ram (it's actually blowing from the top down just a bit in between my RAM and my CPU), so it's being actively cooled. I think it's time for the L12 mod, most likely. This BH5 usually doesn't really take off until it hits 3.1v+, either.

I'm booted at home right now at 11.5 x 217. At any higher multiplier, Windows will boot but lock up. Right now I'm running the lowest vcore I can with this and still have it boot, and trying Sentential's burn-in process (1.825vcore, 1.6vdd, 2.9vdimm).

One interesting thing I've found is that at a setting of 1.875vcore in the bios (my board undervolts by about .025v, so an actual vcore of about 1.85v, most likely), I get a two-tone siren if my VDD isn't set to 1.7v. Also, if I set my VDD to 1.7v but lower my vcore below 1.875, I also get the two-tone siren. If I leave them both at the two higher settings, everything boots fine.

I'm at work now, but I left Sandra's Burn-in running for the day ... not sure how long it'll go before it crashes. I'll test some more tonight when I get back home.

- Dave

So my recomendation is L12 + OC attempt with 1 stick only (maybe try 2.6-2.8v on mem? - I heard that NF7S overvolts ram by like 0.2v) If problem is not with sticks, but mem powering, lowering voltage should help.

Yeah, I'll pull one stick out. My NF7-S does appear to be overvolting the RAM by about .1v.

But why (and how) would lowering the voltage help with stability at a higher FSB? From everything I've ever read, that's not how BH5 works ... the more juice you can give it, the better. I want to run 3.3v into it, but my soldering skills aren't up to par yet.

I love the low vcore. I may try the mobile.

Hoot said:

I used to be a strong proponent for Prime95 as the ultimate overclocking stability test but I have stopped recommending it due to irregularities I saw when ppushing at the top end of my setups capability. With Prime95 Torture Test Priority 10 running and nothing else, my 11x220 would run for 24 hours error free. The next try, it might only make it 18 minutes while the next try would go for 2 hours, then 45 minutes, 5 hours, etc. No predictable failure period. My setup runs cool with the CPU, NB, GPU all being water cooled, along with copper MOSFET, RAM and SB sinks in a room regulated at 20C. Those of you who know me, know that I'm fastidious about case ventilation and high performance water cooling. Vcore has no effect above (true)1.8V, nor does Vdimm above (true) 2.9V, nor does Vdd above (true) 1.7V

(true) means measured at the chip using an accurate DMM, not relying upon the often inaccurate MBM readings.

I can run 11x210 at (true)1.70Vcore and (true) 2.8Vdimm Prime95 Torture Priority 10 until the "cows come home" so I must assume the erratic behavior at 11x220 is indicative of an unstable overclock by Prime95FanBoy standards. Outside of Prime95 anything I throw at the setup runs stable at 11x220.

The perceived/realized performance difference between 11x200 (2-2-2-11), 11x210 (2-3-2-11) and 11x220 (2-3-3-11) is in all honesty, not there as far as I can tell and I don't run synthetic benchmarks for entertainment purposes, so I don't worry about it.

Hoot

Click to expand...

This is my first post to the OCForums, so Hi to all!

Hoot, I've found a similar thing with Prime with my 2500-M. It ran 8 hours before erroring at 210x11.5, so I upped the Vcore a tad and re-ran the test. This time I got a blue screen error, IRQ_NOT_LESS_OR_EQUAL, which is usually mem or bad drivers AFAIK. Dont understand how it ran so long at lower volts, then BSOD at higher ones.

What do you call stable then? I know most people wont deem a CPU stable until after 12 hours priming. At what point do you call your OC stable?

Cheers

Oli