Archicad versions 9 + 10:
Since version 10 has been released and benchmark results have been recieved, I have compared (4) identical systems with both version 9 & 10. It does appear that version 10 is between 9% - 34% faster than version 9 in rendering. The mean average across these 4 results = 25% faster. Some of this increase is likely to be just the new library format - see not below.
Amounts of RAM:
It is clear that test 3 benefits from more than 1Gb of ram. Tests 1 & 2 benefit only marginally. The useful limit appears to be 2Gb of Ram using 32bit windows. Apparently windows can be tweaked to use up to 3Gb for the one program. 64 bit windows can use far greater amounts of Ram & does appear to be slightly faster, see below.
64 bit windows:
64 bit processors and 64 bit windows have been around for over a year now, albiet with relatively poor driver and application support. Archicad 10 is supported by 64bit windows but from the single test completed using identical systems, the difference appears to be only about a 10% improvement in rendering speed. Other machines, in particular the AMD machines seem to benefit much more when used in 64 bit.
Multiple processor machines:
Recently traditional dual processor systems have been replaced by dual core systems. Although Apple already have dual, dual core & quad machines (4 & 8 processors). Lightworks in Archicad is supposed to benefit from multiple processors but only when the exact anti-aliasing setting is used - see note below).
The new core2duo processors do offer a far greater boost in performance than other duo processor systems. It appears the apple dual Xeon machines aren't as fast for the rendering even with 8 cores - I think it might be because of their lower memory bandwidth. Amounts and speed of Ram seems to make a big difference.
Intel Mac's:
Not surprisingly Mac's are now considerably faster with in the new Intel processors & will figure a lot more in the top results. They can also be booted as PC's.
Video cards:
Graphisoft recommend only a standard full open GL video card. In my testing & experience the speed of the video card adds little if anything to rendering performance. I am finding it difficult however to find a video card that performs without screen and rendering errors.
Loaded libraries
I also discovered a significant improvement in speed of the render when the old 9 library was replaced by the version 10 library. I suspect it may be because of the ‘container file' format of storing the library items – but I haven't tested this theory.
Multi-core processing & Anti-aliasing:
It does appear LW is able to use multiple cores but only if the Antialiasing is set to a specific value.
On the 4 machines I have in my office two have the antialiasing setting on a 3 point slider bar, the other 2 have it on a 4 point slider bar. On the 3 point bar below only the ‘Best' position allows LW to render with multiple cores. On the 4 point bar – don't use the ‘Best' setting as it will take interminably long and is not multicore friendly (only the 3 rd setting is).
