I need to find some benchmarking programs to document the differences.
Obviously they need to be free and available on the net.
Failing this, I will have to write one in BASIC, which probably won't give conclusive results (plus I'm out of practise!)

Any advice on this is appreciated.
Cheers!

I have now overclocked the chip (dead simple!) to 266 and 280 mhz (Flat out)
The chip didn't seem to get any hotter to the touch, though I haven't actually measured it yet.

It seemed stable at both speeds.

However, running the Iron dignity demo didn't seem to show any improvements.
This is probably due to the 16mhz memory bus limit.

I'd love to try this with a Kinetic card!
Any donaters?

I've seen PC users with long faces when they can't grasp why changing from a 400MHz PIII to a 600MHz one only gives a few percent improvement (if that) generally a faster harddisk or faster or more RAM generally is the way to get the best speed improvements for the least cost.
[Edited by 144 at 19:30, 14/06/2000]

[Edited by 144 at 19:37, 14/06/2000]

The issue of speed improvement can only be judged with respect to running the same software on the identical same hardware but run with a different processor clock. If you clocked a 700MHz Athlon on a genuine PC board (the SAME one) but at a higher clock rate you would get a SLIGHT performance improvement for tasks that mix I/O (Input/Output) tasks and processor intensive tasks (the latter would improve almost exactly with the increased processor clock while the former would not).

As a rule of thumb if the tasks you run do 80% I/O and 20% processor and you increase the performance of the processor by 10% (say you clock your athlon at 766 rather than 700MHz) the overall performance improvement you'd get is a scant 2%. The I/O would largely remain unaltered (except if it was dependant on processor activity (e.g., 3D graphics set up)) as it accounts for 4/5ths of the overall performance short of doubling the processor speed you won't notice much change. A 1000MHz P-III would not be twice as fast as a 500MHz one (if 20% of the time is processor based the best improvement you'd get overall is 10%). If on the other hand most of your tasks were purely processor based with little or no I/O (like RC5 decryption or the like) then the performance would be (more or less) in line with the higher clock.

And here's the nub... I've seen people spend double the dosh on a P-III at 1GHz rather than a 700MHz one yet they'd get a BIGGER performance improvement by using a 7200RPM disk drive rather than a 5400RPM one (and for a lot less money) or by using 133MHz SDRAM rather than 100MHz SDRAM (a lot lot less money) and then opting for the cheaper (slower) processor.

Fitting a RiscPC as a card on a PC is silly. The PCI bus (32bit at 33 MHz) runs at only half the speed even the current SA-110 can do, updating the memory on a PC AGP video card would involve a long circuitous route from the SA card through the PCI bus to call a driver running under windows (NT) on the PC side the loss in performance would be depressing. You'd probably wind up with a performance somewhere between a RiscStation and the SA, though I/O would be a lot slower than either for most tasks.

It would require extensive re-writes of the RiscOS kernel, the cooperation of PCI card vendors to produce ARM specific drivers and all of this would cost. The boot up times would be that of windows (30-60 seconds) PLUS RiscOS and the overall stability would be that of Windows (rather than RiscOS).

As to Marks other points, the ARM-10 is faster (per megahertz) than the PC or SA-1, it has a vector FP co-processor. It would be less work to adjust RiscOS to support ARM-10 or SA-2 than it would be to get it working on a PC card and also a lot more acceptible to most Acorn users to boot.

However, I agree the amount of work involved in getting RiscOS onto a PC is immense and won't happen.
If only we had faster processors . . . .

The speed the ARM is clocked at is LESS IMPORTANT than the efficiency of the processor. If a processor manages 1.5 times the MIPS at the same clock rate I think most of us would live with it.

The quoted clock speed is 300MHz (but I suspect next year higher speeds will be possible and I am pretty certain its performance would be similar to an SA-110 (StrongARM) at the equivilent of 400MHz).

As far as I know it will require a RiscOS re-write of some of the kernel for 32 bit mode. The voltages used also are lower (1.5V) but this should not be a problem as ARMs are generally on plug in cards and voltage buffering should be possible. Once 32 bit arrives there is no technical reason that ARM10 couldn't be slotted into an existing RPC but to get the best benefit a new motherboard like the Imago would be required.

As an aside the ARM9 is here already and (apparently) has some on board DSP functionality (it does MP3 in realtime) another thing to bear in mind at 190MHz it outperforms a 233MHz SA-110.


[Edited by 144 at 20:50, 1/8/2000]

While reading some of the details on the site pointed to by Mad Nurse I was strangely reminded of the Kinetic card... funny that

For bench-marking software, try Matthias Seiffert's !SICK, which you can get from here