OCing the Athlon64 3000+

With expensive 939-pin CPUs dousing many enthusiast dreams, the 754-pin CPUs are coming back into focus with the help of ABIT and VIA's new K8T800Pro chipset.

continued...

Gaming Benchmarks

We thought it would be best to show you what this enthusiast adventure accomplished by comparing our 3000+ at 2.45 to the Athlon64 3700+ that was just launched by AMD. The 3700+ will currently cost you $700 for the CPU alone. Also we have thrown the $500 939-pin Athlon64 3500+ into the mix as well.

Keep in mind these are canned benchmarks used to isolate CPU and Memory performance and do not show actual video gaming experience. As we all know the gaming experience now days is very graphics dependent, but on the other hand we are currently seeing CPU bottlenecks as the top end video cards are getting faster and faster. With new AI and physics in our games, CPU cycles are just as needed today as they have been in the past even though sometimes it may not be a clear as it used to.

Article Image

Article Image

Article Image

Article Image

I am not going to waste your time analyzing what the above graphs obviously show. Our overclocked 3000+ CPU slides into a performance realm somewhere between the 3500+ and the 3700+. Logic would dictate that our $100 ABIT KV8Pro motherboard and its VIA K8T800Pro chipset just turned our $200 CPU in a $600 CPU.

Again this system is running 100% stable. In fact it is being used by Brent currently in an upcoming CPU scaling article we are doing focused on video cards. At that time we will show you what real gaming benefits are seen from this setup.

The Finer Points

There are some fine points to this project that need to be mentioned.

First off, we found that we had to bring our DDR400 ram back down to stock on this particular board. Even with Kingston and Corsair DDR500 samples, we could not get the memory bus to scale 1:1 as I had wished. That said, the tweaked out stock DDR400 speeds did not hold us back at all on the single memory controller as is pointed out by our benchmarks.

Locking the PCI/AGP bus was a must as mentioned earlier. Also, the HyperTransport bus scales with the FSB, and there is no "lock" for it. You can however change the multiplier of the HT bus. We set the HT bus at a "3" giving us a 3*245. This is a little below stock speeds but you notice for gaming it did not impact us. Note also that the actual BIOS setting we used shows "600" and not "3" (3 x stock 200 = 600). Just be aware as it might be a bit confusing at first as you are not truly setting the HT bus speed, but rather the multiplier the HT bus uses to get that speed.

CPU core voltage, or VCore, also needed some adjusting. We were running ours at 1.75v as set in the BIOS, but our voltage monitors were showing VCore more in the neighborhood of 1.8v. While we did not have to go this high with this particular CPU, we decided to run all of our benchmarks and stability testing at 1.8v in order to see if it caused us any issues. After weeks of testing at this voltage, we have not had any issues and the CPU has been running under 130F at full load.

Lastly, cooling has to be taken into consideration. Certainly when running voltages this high over the spec 1.5v, you want to make sure you are dumping as much heat as possible off your CPU. We used both water and air cooling. A stock Koolance Exos system was used as well as an Akasa Silver Mountain HSF from Plycon along with a high CFM fan. Keep in mind that mounting our heatsink was not stock, but we wanted to show you that "left over" AthlonXP cooling was plenty sufficient for what we were doing. Quite frankly, I think we could have used a lot less cooling and still be just fine with our OC, but I did not want to burn up our setup in the middle of testing. Better safe than sorry.