I was testing BeamSync last night on two Macs with near identical hardware; DLSD 15” 1.67 vs SLSD 17” 1.67 both use the 7447A CPU and the Mobility Radeon 9700 128MB GPU (however the DLSD’s GPU runs approx 20% faster MHz according to ATIcellerator II).
BeamSync disabling increases Quartz Performance on both machines with much higher FPS results, but only slight visual change. However only the DLSD’s display caused visual window tearing. The older 17” display was smooth and tear-free with BeamSync disabled. Also an interesting find between these two machines, the 17” SLSD resulted in slightly higher FPS in my Quartz Debug tests even though the GPU runs faster in the 15” DLSD. The resolution isn’t identical (1440x900 on SLSD vs 1440x960 on DLSD) but is so minor it’s strange to result in such varied frame rates - I assumed the faster GPU in the DLSD would result in higher FPS.
Off-topic, but the DDR2 Memory bandwidth in the DLSD scored lower than the DDR memory in the SLSD in all Geekbench tests. This got me thinking about the “memory tax” ideas and how a DLSD could be possibly “unlocked” at a software level.
Also totally off topic, but I picked up my new (German QWERTZ) DLSD PB G4 15” during the week and found an interesting sticker on the RAM modules...
Seems the reseller was prepared for the imminent release of the PowerBook G5 and labeled their DDR2 SO-DIMMs accordingly
...
The “memory tax” tweaks are interesting and for some part make a degree of logic. Like disabling a high bandwidth I/O to retain system bus /memory controller bandwidth. This kind of makes sense, especially for old school hardware. In the past, I specifically recall witnessing LocalTalk, Ethernet and (PCI) USB access causing system slow downs on pre-G3 Macs.
I also have a recollection that Firewire had lower CPU overhead than USB on file transfers on G3 and G4 hardware. Another observation I posted was of a PCMCIA-CF adapter activity causing 100% overhead on my G4 PowerBooks due to what I assumed was non-DMA I/O.
However, in the case for the “Classic” and “Mac OS ROM” memory taxes... I find these hard to believe.
On my two test machines mentioned earlier, the 17” SLSD which had higher FPS and faster memory bandwidth, had Tiger with Classic (9.2.2) installed on a separate partition, whereas the 15” DLSD only had Leopard installed.
So, did I uncover another inverse case of the poster’s theories or did I misunderstand the process of “introducing” Classic / Mac OS ROM which would supposedly cause memory bandwidth to be taxed for the sake of some kind of Classic bandwidth/DMA requirement...
EDIT: When I get home tomorrow I’ll try installing Tiger and Classic on the DLSD and re-run the tests to confirm if it is actually the non-existence of these files causing a memory tax /bandwidth reduction, but I have my doubts...
BeamSync and QE notes: try toggling BS and QE via Quartz Debug (Xcode Tools) to see instant results.
From memory, disabling BeamSync in the Secrets sys prefs doesn’t seem to do anything.
The best example of QE at work is to run Terminal with the “Pro” theme. This will cause the Terminal window to be semi-transparent. On QE capable Hardware, moving the non-opaque window should be buttery smooth. Disable Quartz Extreme in Quartz Debug and then move the transparent window again to see software (CPU/AltiVec) alpha blending at work. It will stutter. Go back into Terminal Prefs and set the window color to be Opaque (Alpha 100%) and movement should be smooth again. The gains of QE being specifically ideal for transparent window compositing.
QuartzGL / Quartz 2D Extreme attempts to offload view based CG/Quartz rendering to OpenGL surfaces and in some cases can improve some rendering performance, but in many OpenGL enabled apps, will cause glitches and rendering issues. It doesn’t work properly, so best to leave it off.
Last edited:
