Calling Mediator owners, give me your Quake II benchmark!

[Karlos]

Hi,

redrumloa wrote:
Oops sorry, fixed the link!

Umm.. Would you kindly run Karlos' pixeltest and tell me what your output bandwidth is? That's 68K.

Actually the output bandwidth is measuring conversion speed from one pixel format to another as that is what the program was written for - to help me optimise my low level pixel conversion routines (which convert between different chunky formats on the fly, in this case, direct to VRAM from RAM). That's why you see the conversion speed compared to simple copy speed. 100% implies the conversion is already faster than the bus can keep up with (ie as fast as just copying the data) and hence as optimal as can be hoped for.

However, if you don't specify a source format, that operation is a copy (hence it should be close to the copy speeds) anyway.

It works as a benchmark generally in that it tries to measure how fast your main memory and video memory is for "best case" accesses (long aligned 32-bit transfers).

For guestimating your bus performance you should look at

Read VRAM
Write VRAM
RAM -> VRAM (and the move16 version if you have 040/060)
VRAM -> RAM (and the move16 version if you have 040/060)

If the move16 versions score significantly higher on 040/060, you have a good bus design ;-)

[Karlos]

@Red

Sure, I just wanted to measure everything :-)

I just thought I'd mention what the various figures reflected so that it would be easier to understand the output.

The most meaningful tests for the end use would be RAM->VRAM, since this measures how fast data can be pumped from memory across the bus and into the cards memory (which is important for lots of applications) and the pure read/write VRAM tests.

On the face of it, there aren't too many ways you can measure the bus other than by saturating it with traffic ;-)

[Karlos]

There's no PPC-WOS version of the C++ framework that this thing was designed to help test unfortunately. I didn't get time to work on that side of the tree much since the 680x0/OS3.x and x86/Win32 ones were already sucking up all my time.

Of course I can make a simple VBCC proggy that does the same job (aside from testing the conversion speed of course ;-)

I shall do that when I get a minute :-D