J'ai donc retrouvé l'article de sound on sound :
If the proof of the pudding is in the eating, the proof of the laptop for musicians is surely in how much DSP power it provides for native-based applications, so I decided to run a few tests to see just how two T40 models stacked up against Apple's 17-inch Powerbook and a conventional Mobile Pentium 4-based laptop. In terms of audio hardware, I used the same Echo Indigo PCMCIA card on every system (see Laptop Audio box), although I also tried using the internal headphone output on the 17-inch Powerbook, which actually gave me exactly the same results in terms of CPU performance.
Of the two T40 models I used, one was a T40p with a 1.6GHz Pentium-M with 512MB RAM and ATI Mobility FireGL 9000 graphics with 64MB RAM, and the other was a standard T40 with a 1.3GHz Pentium-M and 768MB RAM, as described in the last section. The 17-inch Powerbook in question was the latest generation 1.33GHz G4 model with 1GB RAM and ATI Mobility Radeon 9600 graphics with 64MB RAM. And, finally, the Mobile Pentium 4 laptop used was a Compaq Evo N800w with a 2.2GHz Mobile Pentium 4, 1GB RAM, and also featured an ATI FireGL 9000 graphics chip set with 64MB RAM.
Conducting a fair performance test across multiple platforms is difficult, so I tried to come up with a situation that was relevant, simple, and as fair as possible. Using the latest version of Cubase SX 2, I created a test Project to see how many Reverb A plug-ins could be used, with a single A1 virtual synth rather than an audio file used as the audio source in order to keep the load on the computer focussed on the processing and avoid factoring in the varying overheads of hard disks and their associated controllers. The reverb plug-ins were placed as inserts on FX Channel Tracks rather than inserts on the A1 Channel itself, in order to keep the environment as consistent as possible when scaling the test beyond the eight insert effects allowed per Channel. Finally, all the gain controls (including the sends) were kept at 0dB, except the Master fader, which was set so there was no audible distortion.
I'll resist the urge to use some colourful metaphors to describe the results of the tests because, as you can see by the graph in Figure 1, overleaf, the outcome is easy to evaluate. Centrino-powered laptops are clearly as powerful as their Mobile Pentium 4 counterparts, with the 1.3GHz model offering the same level of performance as the 2.2GHz Mobile Pentium-4 system. And while the Centrino is admittedly a little more expensive than the Mobile Pentium 4, it offers the same performance with three or four times the battery life, and in a lighter, smaller form factor that doesn't get anywhere near as hot. After I ran the test, it would have actually been painful to touch the bottom of the N800w for more than a second, while the T40 remained casually warm.
Perhaps the biggest surprise was the relative performance of the 17-inch Powerbook. While it's fairly well-known that the G4's performance has been falling behind the competition, if you consider how much more you'll pay for this model over the Windows-based systems, and the fact that this is the highest-performing portable Mac money can buy, these facts would make me consider purchasing a Mac laptop only if I absolutely had to work on a Mac.
An interesting point about the Mac test was that even when the audio was breaking up, the actual responsiveness of the system, in terms of the user interface, wasn't impaired ? something that could definitely not be said of the Windows-based systems. This is fairly easy to explain: Mac OS X's Quartz Extreme technology offloads a large proportion of its Aqua user interface graphics processing to the graphics card, so its performance is constant no matter the load on the system's main CPU. In some ways this fact reflects poorly on the G4, since the Intel processors achieve more DSP while running the bulk of the user interface as well, but it does reflect well on the architecture of OS X itself for processor-intensive media applications. Intel and AMD users might expect to see these kinds of technologies in Microsoft's next-generation Windows operating system, Longhorn, although we may not see this until 2005 or 2006.
Figure 1: This graph shows the results from the test described in the main text. The solid colour shows the number of Reverb A plug-ins that could be run until the audio started failing, while the lighter region shows the number of plug-ins before the processor actually maxed out.
Mark Wherry
SOS February 2004
![[Vezøul]Numerø41](/data/avatars/s/29/29556.jpg?1125588453){kind=avatar}