Overclocking is a complicated topic. Here's how I did it on this new Chaintech VNF3-250. The net is that I got an Athlon 2800+ costing $150 to perform like an Athlon 3700+ costing$400 with a 30% overclock. And also got another 10% out of the ATI Radeon 9800 Pro. Read away:
h4. The Overclocking Tools
OCFaq Article :: KB84 - Overclocking: Tools & Info. A good summary of what you need to know before you overclock. Its a complicated topic to say the least.
Another good tool is "Clockgen":http://www.cpuid.com/clockgen.php. It works with all nVidia nForce3 250 boards like the Chaintech. It is great because you don't have to keep rebooting your machine to try new clocks and tweaks. You can just live in Windows
Well, that's the tools, what about how to change things? It's complicated enough, but "Bleeding Edge":http://www.bleedinedge.com/reviews/chaintech_vnf3250/chaintech_vnf3250_01.html has done a nice review of the board and included photos of how they precisely overclocked it. The main issue is that using a 3200+ results in different overclocking parameters than running 2800+. The 3200+ runs with a 10x multiplier (so a 200MHz FSB interface to memory means the processor is internally running at 10x200MHz=2GHz). The 2800+ runs at 9x multiplier or 1.8GHz at 200MHz FSB.
BTW, there are two strange things if you are using the BIOS. First, don't cache the BIOS into system memory (something I normally do). This is because if you are overclocking, this means the BIOS image in RAM doesn't get read properly and the whole BIOS resets. Second, when you are playing with the cpu multiplier (called the Hammer FID in the BIOS), you don't hit ENTER and see the list of options. If you do that, the BIOS hangs. Instead, you have to hit page-up and page-down to see different multiplier values.
h4. Overclocking results
"MSCS.MS":http://hardware.mcse.ms/message61149.html has someone with the same 2800+ and ZNF3 250 where he is getting these results which are quite similar when you realize he's running Corsair 3200LL at 3-2-2-10 while I'm running Crucial Ballistix more aggressively at 2-2-2-10. He got 225MHz clock at 1.6V core, 2.8V DimmV and 4x HT.
| FSB | Memory | FSB/Mem | Vcore | DimmV | HT | Timings | CPU | Mem | HT |
| 200 | Auto | 1:1 | 1.5V | 2.8V | 4x | 2-2-2-8 | 1.8GHz | DDR400 | 800 |
| 215 | Auto | 1:1 | 1.65V | 2.9V | 4x | 2-2-2-10 | 1.935GHz | DDR430 | 860 |
| 240 | 166 | 6:5 | 1.5V | 2.8V | 4x | 2-2-2-10 | 2.160Ghz | DDR400 | 960 |
| 250 | 166 | 6:5 | 1.65V | 2.9V | 4x | 2-2-2-10 | 2.250Ghz | DDR 416 | 1000 |
| 255 | 166 | 6:5 | 1.5V | 2.9V | 3x | 2-2-2-10 | 2.295GHz | DDR425 | 765 |
These were all done with aggressive timings and all of these were just to get to Windows XP booting, not to getting to 24 hours of stability running Prime95 which is the practical goal.
Now if we back off from 2-2-2-10 to 2.5-2-2-10, the performance gets much better and quite close to Anandtech review levels. Not that these are done running Prime95 and are thus really stable, usable, vs. just the unreliable maximums listed above:
| FSB | Memory | FSB/Mem | Vcore | DimmV | HT | Timings | CPU | Mem | HT |
| 212 | Auto | 1:1 | 1.45V | 2.9V | 4x | 2-2-2-10 | 1.92GHz | DDR224 | 848 |
| 245 | Auto | 1:1 | 1.45V | 2.8V | 4x | 2.5-2-2-10 | 2.2GHz | DDR490 | 980 |
| 265 | 166 | 6:5 | 1.55V | 2.7V | 3x | 2.5-2-2-10 | 2.39GHz | DDR445 | 801 |
I could actually get Windows to boot and look stable at 275MHz or 2.475GHz, but it wouldn't pass Prime95. Most folks when doing the testing don't seem to burn it in, but check to see if Windows boots. Also for the 1:1 maximum speed, this is the burned in version where I could get Prime95 to run for 7 hours before crashing. Note that the voltage is way down. The Newcastle runs fine at lower temperatures as it turns out and this means that it much cooler and more reliable. The rough statistics are that at 1.45V the maximum temperature is 61C, 1.55V (the default), the machine runs at 67C and at 1.65V you get to 73C and it normally shuts down. So, dropping the voltage as low as possible is important.
Not that I couldn't get to 2.5GHz or more because I didn't get an 10x multiplier chip (either a Athlon 3000+ Newcastle or an Athlon 3200+ ClawHammer). Rats.
In terms of actual performance at these stable levels I got these results. All of these were done with the ATI Radeon overclocked to 425/375MHz (DAC and VRam respectively):
| FSB | CPU | Timing | Temp | Sandra 2002 | CPU | Mem | Aquamark3 | CPU | GPU |
| 212 | 1.9GHz | 2-2-2-10 | 58C | 2847 | 11853 | 5740 | 44540 | 8097 | 6144 |
| 245 | 2.2GHz | 2.5-2-2-10 | 62C | 3129 | 13645 | 6773 | 47901 | 9423 | 6423 |
| 265 | 2.4GHz | 2.5-2-2-10 | 68C | 3157 | 14779 | 5961 | 47736 | 9357 | 6408 |
The interesting thing is that the 2.2GHz balanced system has the best overall performance. Going to 2.4GHz makes the CPU fast, but it runs very hot (68C) and because memory has to slow down so much, the faster CPU doesn't compensate for running at DDR445 instead of DDR490. So this is going to be the default setting.