As more musicians and audio professionals begin to embrace G5 Power Macs, we take a look at quantifying just how much host-based DSP power you can expect. Plus, news of the 20-inch iMac and an explanation of 'Journaling'.
While the issue of how many reverbs or sampler voices someone can run on their computer is often perceived as being somewhat dull, these benchmarks do at least provide some way of judging how useful a new system is going to be for a musician or audio engineer, in the way that comparing the performance of a Photoshop filter (the kind of test one would normally see in the 'general' Mac magazines) wouldn't. So when I finally got my hands on a G5 this month, it wasn't because of any desire to wear a hooded raincoat that I was interested in seeing exactly what this machine was capable of. After reading so many comments from G5 users in other industries, coupled with Apple's claims and the need for a major leap forward in professional Mac desktop performance, I couldn't resist finding out for myself whether the G5 really is the ultimate platform for a host-based studio (for more on the G5, see last month's review).
The G5 in question was a dual-2GHz machine that came from the Apple Store with 1GB of memory installed, although I also ordered an additional 1GB chip from Crucial. I'd read many reports that recommended installing 2GB of memory to get the full performance benefits from the system, so I decided to try out the same tests with both 1GB and 2GB installed. In the tests described throughout the rest of this article, there was no discernible difference between performance with 1GB and with 2GB of memory — which is interesting, since the performance difference has been fairly significant with applications such as Photoshop. I can only speculate that this is because graphics and video applications buffer larger blocks of data when compared to music and audio software, where the buffers are kept fairly small.
The music software I used was Emagic's Logic Platinum 6.3.2 (running on Mac OS 10.3.1), since it also offers effects and instrument algorithms and is claimed to be fully optimised for the G5 processor. Being a curious sort, I asked Emagic co-founder and programming genius Gerhard Lengeling about why the 64-bit nature of the G5 makes such a big difference, and he explained: "The 64-bit buss and architecture means a lot. In one cycle of the system clock, 64 bits get moved instead of 32, and a great number of things benefit from that because moving data is involved in practically everything."
While the above comment might seem self-explanatory, it's worth mentioning that this improvement doesn't happen to 32-bit code by default. Emagic have to specifically tell Logic to group two 32-bit words into a single 64-bit word where appropriate, meaning that a great deal of thought was put into how to make the most of the G5 processor. This is a different issue to the fact that existing 32-bit code simply runs faster because the G5 can execute more instructions in the same amount of time as the G4.
In judging the performance of the system during the tests, I looked at both Logic's Audio System Performance window and Panther's newly improved Activity Monitor. You can find Activity Monitor in the Applications / Utilities folder on your computer, and it's really handy for monitoring the CPU, System Memory, Disk Activity, Disk Usage and Network activity of your system. On the CPU page, the percentage of processor resources soaked up by user-run applications such as Logic (as opposed to the OS) is indicated by the 'User' parameter, which you'll notice referenced throughout the test results.
It seems that hardly a month goes by without a new Mac appearing, and this month Apple released a new high-end iMac with a 20-inch screen that can only be described as, well, rather big. Technically speaking, the screen offers a resolution of 1680 x 1050 pixels, using the same graphics chipset as the previous flagship 17-inch iMac (the Nvidia GeForce 5200 Ultra with 64MB DDR video memory). Indeed, aside from the larger screen the 20-inch iMac is functionally identical to the 17-inch model, featuring a 1.25GHz G4 processor with 167MHz system buss, 256MB PC2700 RAM, an 80GB 7200RPM Ultra ATA/100 hard drive, and a 4x Super Drive. In terms of connectivity, there are two Firewire 400 ports, three USB 2.0 ports (with two USB 1.1 ports on the keyboard), a video connector for VGA, S-Video and composite output, a headphone jack and a line-level input.
I was lucky enough to be able to get my hands on a 20-inch iMac soon after they were released. As I unpacked it, schoolboy humour left me helpless against the way Apple pack the spherical speakers against the neck of the computer, not to mention the lavatory-inspired styling of the polystyrene once you actually remove the packaging! Joking aside, the screen does have quite an impressive presence once you can see the iMac sitting on your desk; and if you're wondering whether Apple made the base bigger or not, despite being a little fatter the base of the 20-inch model has the same 10.6-inch diameter as its predecessors. While the diameter has stayed the same, though, the same cannot be said about the weight; at 40 pounds (18kg), the 20-inch iMac is around twice the weight of the 17-inch and 15-inch models, which weigh in at 22.8 pounds (10.4kg) and 21.2 pounds (9.7kg) respectively, and is even a pound heavier than a G5 tower. In other words: you wont believe how heavy this iMac is!
On a slightly unrelated note, I couldn't resist ordering the 20-inch iMac with Apple's matching white Bluetooth keyboard and mouse. And here it's worth remembering that Bluetooth is a build-to-order option on the iMac (just as it is on every other Mac except the Powerbook range). The reason this is worth mentioning is that unless you choose the Bluetooth option at the time of ordering, you'll be unable to add the upgrade internally at a later date and you'll have to get a USB Bluetooth adaptor if you want to use Apple's wireless keyboard and mouse.
If you're doing Mac-based performance tests with Logic, what better place to start than with the trusty Platinum Verb — come to think of it, though, does anyone actually use this effect for anything other benchmarking? For those who can't stand the suspense, I managed to get the G5 to handle 120 stereo Platinum Verb instances simultaneously, based on playing back either two audio tracks or two Audio Instrument tracks as a source signal. Compare this to the 20 stereo instances coaxed from the 1.33GHz 17-inch Powerbook in last December's Apple Notes, and the G5 starts to look quite impressive. However, the really interesting thing was to look at the way in which Logic made use of the two processors, by observing the Audio Performance window, where the CPU bar is in two vertical halves to illustrate the independent performance of the processors.
To begin with, I set up eight busses in Logic, each with 15 instances of Platinum Verb, which is the maximum number of insert effects you have available to each buss. Logic is actually quite clever at only activating audio processes as they're required, so keeping dormant Platinum Verbs in a song uses only memory, rather than taking up cycles of the CPU's processing time.
I started with two Audio Instruments, using Emagic's simple polyphonic software synth as my audio source, and started sending signals from the Audio Instruments to the reverb busses. Activating the first send on the first track (enabling the first set of 15 Platinum Verb s) took the first CPU meter to 30 percent and left the second CPU meter at zero, with the User display in the Activity Monitor registering 19 percent. Enabling a further three sends with the next 45 Platinum Verbs (now 60 in total) took the first processor to 95 percent, with the User processing showing 49 percent. However, even with the second processor not in use, our old friend the Core Audio System Overload message appeared when I enabled the fifth send on the first track, and the song politely stopped.
Normally, you might think that you'd reached the processing capacity of your system at this point, but the interesting thing is that after the song stopped and I selected a different track on the Arrange window and pressed play again, Logic rebalanced the processing tasks to utilise the second processor. At this point, both CPU meters in Logic settled to around 50 percent and the User value also showed 50 percent, which suggests that Logic seems to assign processing tasks to the first processor during song playback, whereas it will be able to balance its tasks more usefully once the song is stopped. So if you get overload messages during playback and you have a dual-processor Mac (G5 or G4 based), it's probably because you've overloaded the first processor rather than the whole system.
You might have picked up on the fact that I mentioned selecting a different track on the Arrange window in the last paragraph, and this highlights another issue that's worth knowing about. As you may already know, Logic uses the currently selected Arrange window track as the MIDI Thru output. When an Audio Instrument track has been selected, Logic needs to make sure the Audio Instrument is constantly active to receive and play back notes. What this means is that if you have an Audio Instrument track that doesn't play constantly throughout a song and you have this track selected on the Arrange window, more processing power will be required. So in the above example Logic would have been unable to rebalance the processing because the track containing all of the processing tasks was being kept active, since it was selected for MIDI Thru. Suddenly, Logic's 'No Output' object becomes quite handy, since you can assign it to a track and select it as the MIDI Thru track to avoid this issue.
One topic that's caused some debate amongst Mac musicians is whether a feature known as Journaling in Panther should be enabled or disabled. Although it was first introduced as an option in Mac OS 10.2 Server, Journaling is now standard and enabled by default in both the Client and Server versions of Mac OS 10.3. Put simply, this is a technique whereby any changes made to files on a disk are logged in a Journal, so that, in the event of an error while data is being written (such as a power cut), the disk can be restored to a healthy state by the system going back through the Journal and performing the necessary reconstruction. Journaling helps to improve the integrity of a filing system. Indeed, in OS X Journaling is implemented as a backwardly compatible extension to the standard Mac HFS+ filing system.
So why do we need to wonder whether Journaling should be left active? Well, as you can probably guess, the process of the system creating a continuous log of all the changes on a disk decreases the overall performance of that disk, which isn't a good thing if you're depending on getting the ultimate data throughput to your storage system. The situation isn't actually as bad in practice as you might think, and it's only really a problem when files are being written to a disk, so if you use your Mac's hard drive mainly for playing back samples with software such as EXS24 or Kontakt, this will be significantly less of an issue than if you're recording a large number of audio files simultaneously.
It's also worth mentioning that, in theory, Journaling slows down the system far more (relatively speaking) when you're writing a large number of small files to a disk than a small number of large files, which should make it less of an issue for audio and video users. As a general recommendation for the ultimate performance, though, it's best to disable Journaling. However, if you don't notice any significant loss in the number of simultaneous tracks you can record and play back with Journaling active, it may not be an issue to worry about, and you might even benefit from the increased integrity if you suffer a power failure (since Macs never crash...).
To disable Journaling in Panther (it will be active regardless of whether you upgraded from a previous version of OS X or carried out a clean install), load Applications / Utilities / Disk Utility. In Disk Utility, select the volume on which you want to disable Journaling and choose File / Disable Journaling.
Getting back to the tests, I activated the first send on the second Audio Instrument track (making five sends in total and 75 Platinum Verbs), and this took the processor usage to 75 and 60 percent in Logic, with the User measure at 65 percent overall. Enabling busses 2-4 on the same track, so that 120 Platinum Verb s were indeed running, maxed out the CPU meters in Logic, while the User level was at 92 percent. It's worth pointing out that the User parameter doesn't allow for a certain amount of overhead required by the operating system itself. Also, when 120 Platinum Verb s were running, having either of the Audio Instrument tracks selected on the Arrange window produced Core Audio Overload errors.
Now, you could say that it was unfair to use a software instrument as the signal source, since this might require more processing resources than simply playing back an audio file — a potentially fair point. However, I got the same results using two audio tracks playing an audio file: adding more Platinum Verb s during playback caused the first processor to eventually overload, while stopping and pressing play again rebalanced the load.
While Platinum Verb isn't the last word in tools for gauging a system's performance, it does at least provide a familiar and relative guide, since so many users have done similar tests on different systems. Even with my own experiments, you could quantify that a dual-2GHz Power Mac was six times more powerful than a 17-inch Powerbook, for example. But in terms of G5-specific optimisations, reverb is one effect that does particularly benefit from the 64-bit nature of the G4, although more from an addressing perspective than simply one of processing, since reverb algorithms deal with complex calculations of large numbers, which can now be carried out in fewer operations with data being retrieved and manipulated as longer words.
The G5 is obviously a competent system but, from what I've seen, software that's written well for the platform can make it an amazing system. The issue of how many Platinum Verbs can be run isn't really the point, of course: the tests are really just a way of demonstrating relative DSP performance, meaning that you could expect proportionally the same amount of improvement from a range of effects and instrument algorithms, rather than just a reverb. It's also interesting to note that having two processors makes a huge difference, as I'll explain in a little more detail next month. This means that while both the dual 1.8GHz and 2GHz Macs will offer the kind of improvement described here, the performance of the single-processor 1.6GHz G5 Power Mac will be far less impressive.
After the Platinum Verb experiments, I moved on to looking at the number of voices Emagic's EXS24 sampler was capable of playing back on the G5 — and I have to say that the results were astounding. Unfortunately, they'll have to wait until next month because we've almost reached the end of the page. But — in the style of the time-honoured 'guess the weight of the jar' game at at the summer fête — if anyone feels like guessing the number of EXS24 voices I managed, the closest guess could win a scale model of Steve Jobs whittled from Wensleydale!