Linpak vs. Prime 95

I've done some research on which programs you can use to test your overclock or stock cooling to determine if you have a stable system. Presently, the winner is Linkpak. You can get a copy with GUI front end here.

Setup:

1) BlackIce Xtreme 480
2) Laing DDC 3.25 pump
3) DTek Fusion waterblock
4) Delta AFB1212VHE 120x38mm fan

The fans were run at maximum speed.

CPU at stock was using 123W (measured via current clamp) at it's reported maximum. HWMonitor reported ~119-121W. CPU is rated for 136W.

Using a Striker II Extreme, 4GB DDR2 and a QX9770 the following overclocks were achieved:

MHZ X FSB RAM CPU V
4200 10.5 1600 1333 1.5250V

CPU at this stage was consuming 143W (via current clamp) and 139W via HWMonitor.

This passed Prime95 quad-core for over two hours.

Running Linpak with 1/4 memory used over 5 iterations resulted in a mismatch (failure).

Increasing the voltage to 1.650V resulted in a pass on Linpak using the 5 iterations, 1/4 memory. CPU was now consuming 156W via current clamp (HWMonitor ~150-153W). The temperature, however, was sitting around 90C when under load per core. Idle was ~40C.

In an effort to see where Prime95 and Linpak diverged from their pass/fail I started decreasing the voltage till Linpak failed and then continued till Prime95 failed.

VOLTAGE PRIME95 LINPAK
1.650V PASS PASS
1.625V PASS PASS
1.600V PASS PASS
1.575V PASS FAIL
1.550V PASS FAIL
1.525V PASS FAIL
1.500V FAIL FAIL

Each pass of Prime was run for 2 hours. Linpak took less than a few minutes. In an effort to "increase" Linpaks sensitivity I increased it's memory consumption to 1/2 and iterations to 10. This did show a more sensitive failure point.

VOLTAGE PRIME95 LINPAK
1.650V PASS PASS
1.625V PASS PASS
1.600V PASS FAIL
1.575V PASS FAIL
1.550V PASS FAIL
1.525V PASS FAIL
1.500V FAIL FAIL

Linpak's time to run each completed set of tests more than doubled but was still less than 10mins.

During this time Linpak created more heat then Prime95 as well, although Intel still provides tools that actually generate *more* heat then Linpak can do (about 5-10W more).

In the end, Linpak is a great tool and signficantly more sensitive and faster at finding stable clocks then Prime95.