First, here’s my system:
- Chaintech 6 BTM (slot 1, i440BX) mobo Rev. N with a correct voltage regulator
- DIMM 64 Mb PC-100 Infineon + DIMM 64 Mb PC-100 NCP
- 20.4 Gb Seagate Barracuda ATA-2, 2.1 Gb Seagate Medalist in mobile rack
- ASUS V-3800 Ultra TVR Deluxe
- Opti933 SC
- Rockwell 56k modem
- x24 CD-ROM
- 235W Power Master source
- no-name midi-ATX case with plenty of extra coolers
- Win 95 OSR 2.5 PE/ BeOS R5 Pro
My 466 Mendocino didn’t make me happy any more, so, thinking of some upgrade, I decided that I must grant myself access to the “1 GHz club”!
I got a 600 CuMine Celeron cC0, this upgrade was just $30 for me. Here in St-Petersburg, Russia, money back for CPUs is not easy to find, but I found a firm where people told me those 600’s boot into Windows at 100FSB (900) at the default core voltage of 1.7v. Well, I figured: “This can’t be called stable if it just boots up, but if I add 0.1 or 0.2 v, I’ll make it.” So I got it.
At that time, I had a huge aluminum heatsink with a no-name but fast and big cooler – this was good for a Mendocino CPUs, ‘cuz Orbs do not completely contact their covers – the corners remain visible. The heatsink’s bottom was polished and a specific thermal grease was used; here we use the so-called “Silico-organic Thermal Grease – 8” (“Arctic Silver” is better, but we can’t get it here).
Sure, there are extra fans in the case! So I installed this supercooler and my new 600 and one by one ran at 75, 83, 100 and 103 FSB (925 MHz) with DEFAULT VOLTAGE! “Well”, says I to myself, “just 70 more MHz left, sure I’ll get it!” But that’s where trouble began.
It didn’t run my next step of 112 FSB (1008 MHz) at 1.7 V. Okay, have a cake – I give it 1.8v on my MSI adaptor. POSTs all right, then freezes. I try 1.9v – boots into Windows. Good, but not stable.
Now, a few words about testing: Running 3DMark for a couple of hours is not enough. Methods I use:
- Create a RAR-archive 3 or more times as big as your amount of RAM. Create it in a non-overclocked system to make sure there are no errors. Then write a batch-file:
rar t pack.rar
if errorlevel 1 goto exit
if errorlevel 0 goto cont
Put your archive (here – “pack.rar”), this batch-file and rar.exe into the same folder. This batch runs continuous testing and if CRC fails due to CPU, RAM or IDE error – it stops. It can show your problems very quickly – more often it is an L2 cache error due to overheating or low voltage. It can be run overnight simultaneously with a deep virus search or left to work in the background with your office applications or games.
- Use a “Burnp6.exe” program – enormously heats up CPU, can lock it up in 1-10 minutes if something’s wrong. It can be found on many websites.
- Run a 3D game for hrs or overnight – involves VGA and RAM to make sure they are stable at this FSB.
- Monitor everything with MBM 5 (thanks, Alex VanKaam!!!)
So, at 1.9v it wasn’t stable at 1008 MHz, though it could run 3DMark Demo for a couple of hours. It gave me CRC errors on archive testing. How ’bout 1.95v? Almost the same – little bit better, but Burnp6 locked it up in 1 minute. Not good, though Need For Speed 5 (we call it Nephrospeed) could be played for about half an hour.
Now, I consider a 15 % increase in core voltage safe, but further increases may not burn the CPU, of course, but it will reduce its life. Still, I put 2v, as my mobo a little bit lowers this voltage, so I had 1.97 – 1.98 v. Same picture. Here I must say that CT-6BTM doesn’t show an internal diode temperature, it has a sensor on a small “arm” that touches the heatsink. So at this moment the temp was 32-36 C, which certainly was far from the core temp but lower than on Mendocino – and it didn’t increase much with a voltage rise.
Well, I started thinking here. An adaptor doesn’t allow more than 2v, so I’d need another one. But further increases are risky with just air cooling, so I began dreaming of Peltier’s and Watercoolers. But these systems are not safe themselves – and expensive – so I rejected the idea.
The symptoms reminded me of one of my past cases, when a CPU could only run 83 MHz FSB with an L2 cache disabled. This, of course, decreases performance in 3D games but still worth trying to find out the reason. Disabling L2 cache didn’t help. Neither did setting CAS latency from 2 to 3 – my RAM is PC-100 – it didn’t promise me to be stable at 112 with CAS-2, right? But – no effect!
Damn, what is wrong? I am just trying to get over some 70 MHz, I gave it almost 20% more voltage and the temp is low – why? Just a limit for the technology? Remember, Mendocinos were not likely to run over 500 MHz, at least they needed some SuperResearch to improve cooling? And they say the limit for CeleMines is somewhere 900-1000. “Looks like that”, I thought.
Then I made angry with myself. What kind of a silly and experienced O/Cer is that!? I didn’t overclock it to 930, it did itself, no effort from me!
So I took off the heatsink, suspecting maybe there was just some uneven surface or some defect that gives uneven cooling of the CPU’s surface. I found a small scratch and polished it some more. Then, at installation, I noticed that the sink is not completely flat on the CPU. Somehow, I made a small “hill” while polishing (straight hands are very important at overclocking); it was invisible and couldn’t be detected when testing the surface on a mirror or alike method, but it wasn’t completely flat and gave some rocking amplitude while lying on a processor.
Here I figured: “Why don’t I get myself a Chrome Orb?” In St-Petersburg, we don’t have things like Swiftech or Alpha on sale – mostly some cheap no-name stuff. So a Chrome Orb seemed to be the best choice. I got one and put it on my CPU with default thermal grease – a yellow patch (in one of the reviews I read this is a Thermagone interface and it is good).
Nope, it isn’t; didn’t even run Windows. I couldn’t even warm it up to melt that stuff! So I cleaned it off and put my “# 8” grease. I also put small rubber rings under the screws of the fan to prevent some vibration I noticed. Also, some people paid attention that this “elevation” of the fan in the Chrome Orb improves aerodynamics and enhances cooling by 1 – 2 degrees C.
And of course, the +12 V wire of the cooler was connected to a straight +12 V output of a power source, not to the mobo connector; you cannot get full speed out of a cooler from a mobo connector, it lowers power a little. After this surgery (actually, I am a GP and an immunologist, not a surgeon) with shaking hands, desperate eyes and hope in my heart, I started it.
Yiii-haa! The Chrome Orb really works! It runs stable on all my tests and games for hours; only locked up once when I closed an inflow to my extra case fan to see what happens – I’ll leave it open, there’s not much noise through it. It is rock stable at 2v now. After testing, I installed a software-cooler (AMN Refrigerator rulezzz!) to prolong my CPUs life when I work with office applications. It doesn’t slow 3D, so there’s no need to turn it on/off all the time.
So, here is a 1.68 times increase and I am at 1 GHz!!! My stupid dream came true. I think I’ll stay here for a while and go back to 930 – until I get a GTS, there’s not much difference. But I enjoyed the ride!
And you should see my desk! Lots of cigarette butts in the ashtray, plenty of coffee cups, screwdrivers, thermal grease tubes, jumper caps, a dental mirror (to see if the thermal sensor is positioned okay), a stethoscope (to identify noises), red eyes and dizzy head.
The Author is not responsible for any damage you may do to your hardware, software or central nervous system under the influence of this text.