There used to be an interesting thread on cubeowner.com about changing the bus speed on the Cube from its stock 100MHz. There is a SG500 timing chip on the Cube and by altering a few jumpers you can change the bus speed. I think there have been similar posts about doing this with the 100MHz towers that use the same chip.
I wanted to give this another try so I put some 133MHz ram in the cube, changed the bus speed and changed the CPU multiplier. Alas, my cube was not stable at 133MHz bus speed. The next step down is 120MHz, so I set things up for that and it seems to be pretty happy (stable) at this speed even with some pretty warm ambient temperatures this summer.
Using the dual 7448 CPU from powerlogix/newer technology it was easy to adjust the bus/CPU multiplier and at 1.8GHz I could do some testing at the same CPU speed but with different bus speeds. i.e. 1.8GHz is 18x100MHz or 15x120MHz.
As expected most CPU benchmarks showed the same numbers, but more complicated programs like javascript in TenFourFox start to show the benefit of a faster bus:
Other benchmarks like xbench and geekbench showed the improved memory bandwidth in some of their sub-tests.
I also ran some tests with an event stepped simulation and got 48.2k vs 54.2k events/s with everything the same except the bus speed.
I wanted to give this another try so I put some 133MHz ram in the cube, changed the bus speed and changed the CPU multiplier. Alas, my cube was not stable at 133MHz bus speed. The next step down is 120MHz, so I set things up for that and it seems to be pretty happy (stable) at this speed even with some pretty warm ambient temperatures this summer.
Using the dual 7448 CPU from powerlogix/newer technology it was easy to adjust the bus/CPU multiplier and at 1.8GHz I could do some testing at the same CPU speed but with different bus speeds. i.e. 1.8GHz is 18x100MHz or 15x120MHz.
As expected most CPU benchmarks showed the same numbers, but more complicated programs like javascript in TenFourFox start to show the benefit of a faster bus:
Code:
Richards: 5624 Richards: 5786
DeltaBlue: 5640 DeltaBlue: 6334
Crypto: 3723 Crypto: 3930
RayTrace: 7121 RayTrace: 8058
EarleyBoyer: 2476 EarleyBoyer: 3018
RegExp: 462 RegExp: 530
Splay: 1482 Splay: 1671
NavierStokes: 3815 NavierStokes: 3944
Score (version 7): 2931 Score (version 7): 3242
Other benchmarks like xbench and geekbench showed the improved memory bandwidth in some of their sub-tests.
I also ran some tests with an event stepped simulation and got 48.2k vs 54.2k events/s with everything the same except the bus speed.
Last edited: