If your PC is beginning to feel a bit tired, even though it's a comparatively recent model, do you really have to abandon it in favour of a new machine, or can clever upgrading restore cutting-edge performance?
There comes a time in the life of every PC when its perceived performance moves from inspiring to workaday, and you start to wonder how you can improve on the current limits for track count, plug-in count or polyphony. Depending on how 'future proof' your original purchase was, this generally happens sometime between six and 18 months after you purchased or built it, and once you hit this point you may increasingly find yourself wondering whether adding or upgrading to faster hardware components is in order. Should you consider radical measures, such as replacing the motherboard with one that supports a faster processor, or perhaps several processors, or installing a significantly faster hard drive, or even choosing RAID (Redundant Array of Independent Disks)? Just how much performance can you squeeze from a single PC? Let's find out.
By far the easiest upgrade to do is adding RAM, assuming you've still got some spare memory slots available to house it. Many motherboards support up to 4GB of RAM (and some even more), although, as I discussed in May's PC Notes, making use of more than 2GB with the familiar 32-bit versions of Windows XP can be tricky, and may not benefit that many applications at present. Adding more RAM, if you're short of it, can improve performance, so anyone with just 512MB should definitely upgrade to 1GB if at all possible, but beyond that the advantages are debatable. Many music PCs are supplied with 1GB of RAM as a matter of course, and I've never personally needed any more than this, but 2GB is starting to become the new standard. If you feel you'd personally benefit from more than 2GB of RAM, you ideally need to 'go 64-bit' (see later), when you'll be able to install and make use of 4GB or more.
Much higher on many musician's agendas when considering greater performance is adding another hard drive and creating a RAID array, especially since so many motherboards offer RAID as an option. After all, if you've already got a suitable motherboard with untapped RAID features, it will only cost you the price of an additional hard drive to get started. Even if your motherboard isn't so obliging there are some inexpensive PCI RAID controller cards available that should do the same job. So let's examine the various RAID options...
Most motherboards with RAID offer a choice of RAID 0, RAID 1, RAID 0+1, and JBOD (Just a Bunch Of Drives). JBOD (also known as spanning) takes all the drives in the array and turns them into one huge logical drive. If you want to record a large multitrack work of long duration this might be tempting, but you'll achieve no extra performance, and if any one of the drives fails you won't be able to fully access the data recorded on any of the other drives.
- RAID 0 is the option most musicians get excited about, since it offers 'data striping': reading and writing of data across two drives in parallel, doubling the sustained transfer rate, and hence the number of potential audio tracks or soft-sampler voices. The only disadvantage of RAID 0 is that if either of your drives develops a fault you may not be able to retrieve your data, so you have to make sure you back up regularly to another medium.
- RAID 1 provides 'data mirroring' for added security. The second drive holds a mirror image of the first (wonderful for long-term backup purposes), and also offers faster access times, but its writing speed is halved.
- RAID 0+1 offers combined 'data striping and data mirroring', so you would need four drives to put it into practice. While this scheme offers improved reading performance and added security, write performance is still slow, since every piece of data needs to be written twice, and for the musician, four drives mean more noise and more heat generation. I think we can assume overall that musicians will be most interested in RAID 0.
All motherboards that feature RAID options provide copious instructions on how to enable this feature in the BIOS, make your choice of array mode (0, 1, 0+1 or span), and (for the 0 and 0+1 options) choose a 'stripe' size that determines how the data is split between the drives. This figure can range from 2K to 64K or so, and audio playback performance generally benefits from higher values (see 'Technical Terms' box for details).
So should everyone with a RAID-equipped motherboard rush out and buy an extra drive so that they can split their audio and sample data and achieve double their current performance, while the rest of us replace our existing motherboard with one that can handle RAID or buy a PCI RAID controller card? Actually, in the majority of cases I've come across, the answer is probably no.
Many musicians are still unaware of just how capable the typical modern 7200rpm hard drive actually is. Way back in SOS June 2003, in 'The Right PC For The Job', I stated that "There's still some confusion about which drives are best for audio recording and playback, but it will help greatly if I start by saying that the vast majority of modern IDE drives will be perfectly fast enough for most musicians, without considering the added complications of SCSI or RAID." I also offered the advice to "stick with a single 7200rpm EIDE hard drive for audio purposes. It should be perfectly capable of running around 48 tracks of up to 24-bit/96kHz audio."
Nothing has since changed my mind, especially now that modern SATA drives provide even faster performance than the IDE drives of 2003. To reinforce this point, I spent some time creating a Cubase SX song containing multiple mono tracks recorded at 24-bit/96kHz, and kept increasing their number until I began to experience drop-outs during playback. With my 7200rpm Seagate Barracuda SATA 80GB ST380013AS audio drive I managed a massive 76 tracks before this happened. The majority of musicians who are still running at 16-bit/44.1kHz should manage well over 100 tracks from a single drive without getting involved with RAID at all.
If you want to run more tracks than the figures above, you could simply try a faster single drive, such as one of Western Digital's 10,000rpm Raptor models. One of these could boost your track count significantly, as its sustained transfer rate exceeds 70MB/second on the outer parts of the drive, compared with about 60MB/second for the 7200rpm Seagate Barracuda drives more commonly used in musicians' PCs. Moreover, its random access time is significantly better, which should help those among you who are sample streaming. On the other hand, opinions seem divided about its acoustic noise levels, so bear this in mind if you want to build a quiet music PC. In addition, the largest model available was just 74GB when I looked, and that may not suit every musician.
If you're considering RAID because you're currently having problems running more than a couple of dozen audio tracks, you've got a problem elsewhere in your system. One thing to check is that your hard drives are benefiting from DMA, by looking in Device Manager in the Advanced Settings page for your various IDE ATA/ATAPI controllers. For each device on the chain, the Transfer Mode should read 'DMA if available', while the Current Transfer Mode should have an entry such as 'Ultra DMA Mode 5'. If you see any mention of PIO mode, this is the cause of your problem. Although Windows is supposed to make the optimum setting for you automatically, it's not unknown for it to make a mistake.
Don't forget that if your current PC is running out of steam, there are various ways to increase its processing power that don't involve physically upgrading the processor...
- Freeze Frame: Without spending any money, you can run many more plug-ins and soft synths by taking advantage of the 'freeze' functions offered by most major MIDI + Audio sequencers. When you're happy with the sound of a particular audio or soft-synth track that's using lots of plug-ins, freezing it will create a 'mixed down' audio track version with embedded effects, releasing the processing power they previously consumed for use elsewhere.
- Card Tricks: The most processor-intensive effect is still usually reverb, which is why dedicated DSP cards such as TC's Powercore and Universal Audio's UAD1 running high-quality effects are so popular for 'taking the strain' off your native processor. Moreover, you generally get better-quality results compared to most native plug-ins.
- The Hard Way: Yet another way to offload some of your effect processing is using a hardware effects unit, connected via a send/return loop to your sequencer. I discussed this approach back in SOS March 2004, but it's become even easier since for Cubase users, due to the new 'External FX' settings of Cubase SX 3.
So when does RAID move from desirable to necessary? Well, there seems to be a case for those running streaming samplers such as Gigastudio, particularly when creating virtual orchestras. This scenario gobbles polyphony like nothing else and can easily consume more than 200 voices. On well-tuned systems this seems to be about the limit for a single fast 7200rpm drive. Ironically, orchestral sample libraries also tend to be vast in size, so you may have to install more than one drive to be able to access all the instruments. If you need to run two drives anyway, it makes sense to turn them into a RAID array and get a performance benefit, and if you need three, your performance will go up again.
On the other hand, for sample streaming there's a school of thought that says that you may get better overall performance by letting those multiple drives act individually, with different sample libraries (for example, strings, brass, woodwind) on each one. This is because you're generally trying to access small sections from lots of files simultaneously, rather than reading or writing one huge file at enhanced speed. I don't have a definitive answer to this, but suspect the truth will depend on how you write your music, and how big an orchestra you're scoring for. Of course if your soft-synth polyphony requirements are beyond the capabilities of any one PC, installing each of these different instrumental groups on separate drives in independent PCs is another popular approach, and one that I'll be covering in more depth next month.
Tascam report typical polyphony counts of between 250 and 300 voices with a twin-drive RAID setup, lots of RAM and a fast processor, which certainly ties in with the various user results I've been tabulating. As a single drive may sometimes manage 200 voices (although most users seem to achieve fewer), this equates to an improvement of between 25 and 50 percent. Comparing Carillon's published track-count figures for their RAID-equipped Core 4 system with their single-drive systems, the numbers jump from about 75 to between 98 and 108 mono 24-bit/96kHz tracks — so you can also expect an improvement of 3045 percent for audio track counts.
RAID could also benefit those who are using high sample rates, such as 192kHz. I still doubt that many people will really notice increased audio quality using this rate rather than 96kHz (most people seem to agree that it's largely a marketing 'feature' at the moment), but if you happen to be indulging in multitrack 24-bit/192kHz audio, RAID may be the only way to achieve more than around 40 simultaneous tracks.
The other obvious potential beneficiary of RAID is anyone involved in video recording and playback, as the bandwidth required for that is far greater than for audio. In the past, many musicians scoring for TV and film have locked their PCs to stand-alone video-playback machines, but it's generally far more convenient to have your MIDI + Audio sequencer running a video window. It seems to be the drive's seek time that limits the advantages of RAID over a single drive for sample and audio track streaming, because so many individual files are being accessed. However, if you're streaming a single huge video file alongside your audio tracks you'll probably find RAID benefits you rather more.
Extra processing power is the key to really pushing the boat out with PC performance, as it will let you run more plug-ins, soft synths, convolution reverbs and so on. In the past, increasing your processor power usually meant installing a faster version of the same model of processor in your motherboard, or buying a new motherboard to take advantage of a different and potentially faster range of processors. However, there are now so many processor ranges that making the best decision is rather more complex, even if you're absolutely determined to stick either with AMD or with Intel products.
As I explained in PC Notes June 2005, anyone considering a new, faster PC at the moment would be wise to make sure it's 64bit capable, to ensure they have the potential for improved performance without increasing the clock speed of the processor. Early indications are that having a fully 64-bit system may boost the performance of typical music applications by as much as 30 percent, which is equivalent to replacing one of today's fastest 3.8GHz processor models with one of 5GHz. Moreover, for anyone who needs more RAM, a fully 64-bit system removes the 4GB ceiling of 32-bit systems.
Intel introduced their first 64-bit Itanium processor back in 2000, but few people bought one, partly because the first Itanium chips weren't designed to run 32-bit applications at all. AMD's Opteron, introduced in 2003, could run 64-bit applications under a 64-bit operating system, but also had 64-bit 'extensions' in its core that allowed it to run 32-bit applications under a 32-bit or 64-bit operating system, and therefore proved rather more popular. Intel's subsequent Itanium 2 had a 32-bit emulator that translated 32-bit instructions into code that could be run on the Itanium chip (although this entailed a performance 'hit'). For the musician, the best news on the 64-bit front was AMD's follow-up to the Opteron, the Athlon 64, which was intended for low-end servers and personal use and was therefore significantly cheaper, but could run 32-bit applications with ease.
In July 2004, Intel came back with the Xeon EM64T (Extended Memory 64 Technology) processor, previously code-named Nocona, which also had 64-bit extensions compatible with the AMD ones. For the budget-conscious musician, Intel's latest Prescott Pentium 4 600 models also add 64-bit capability, courtesy of the same EM64T extensions. They also run at significantly lower temperatures when compared with the non-64-bit Pentium 4 500-series processors of the same clock speed, which would seem to remove the Prescott's previous disadvantage of requiring far more complex (and therefore expensive) cooling arrangements to remain quiet.
If you're after lots more processing power, I would personally recommend that you opt for a 64-bit-capable processor now, even if you intend to carry on running the 'traditional' 32-bit Windows XP for a while. A '64-bit-only' PC may not be a wise move at first, at least until any unforeseen hardware and software problems have been resolved, and because any hardware you own that doesn't have 64-bit drivers can't be used at all.
- Dual Core: A processor that combines two independent processors and their respective caches and controllers in a single die and package. Some advantages over running dual processors include the ability to clock some of the internal circuitry much faster than if the signals had to travel 'off chip', less space taken up on the motherboard, and slightly lower power consumption.
- Dual Processor: Two independent processors, each mounted on its own die and in its own package. To run more than one physical processor you need a motherboard and associated chip set that supports this, plus an operating system that does the same, such as Windows XP Professional.
- Hyperthreading: An Intel technology, available on some Intel processors such as the Xeon, Pentium 4 Northwood and Prescott ranges, which lets processors appear to Windows XP and Linux 2.4x as two 'virtual' processors instead of one physical one. They each share the various internal 'sub-units', including the all-important floating-point unit, but can run two separate processing 'threads' simultaneously. According to my measurements, this can result in a typical 10 percent reduction in CPU overhead at low audio interface latencies with applications that specifically support it, but doesn't provide anywhere near the same performance improvement as true dual processors.
- Multi-threading: Separating the software application into multiple fragments of code is the key to multi-processing: individual threads can run on different processors simultaneously. Even with a single processor, switching rapidly between the various threads ensures that multiple processes appear to run simultaneously. However, on a computer with multiple processors different threads can run on different processors in true parallel fashion. Operating systems that support multi-threading (and, indeed, use it themselves) include Windows 2000 and XP.
- RAID (Redundant Array of Independent Disks): A technique that turns two or more hard drives into a single, higher performance drive, either by duplicating data for greater security, or splitting it among the drives for faster performance.
- Stripe Size: Not to be confused with stripe width (which refers to the number of simultaneous stripes that are read from or written to a RAID array, and always equals the number of drives in the array), stripe size is also referred to as block or chunk size. When data is passed to the RAID controller, it is divided by the stripe size to create one or more blocks. These blocks are then distributed among drives in the array, leaving different pieces on different drives. In most cases, the relatively large sizes of audio and sample files tend to benefit from a large stripe size such as 64KB (or more).
Both the Athlon 64 and Pentium 600 series are ideal for anyone about to buy a 64-bit capable PC that has a reasonable longevity, but for those who want the fastest performance available, sticking with one processor is no longer the ultimate solution, especially now that (as I reported in PC Notes January 2005) clock speeds seem to have reached a ceiling — for the time being, anyway. The answer is to run multiple processors in parallel, with each handling part of the total load, and by choosing a motherboard and associated chip set that supports two physical processors rather than one, (along with a compatible operating system like Windows XP Professional), you can typically boost your performance by between 50 and 70 percent compared with a single processor of the same clock speed.
For example, by far the best processor performance I've measured to date (running plug-ins and soft synths) came from the pre-EM64T Dual Xeon PC supplied by Red Submarine and reviewed in SOS June 2004. At the time, its twin Xeon 3.06GHz processors provided roughly equivalent performance to a Pentium 4 Northwood of a theoretical 5.6GHz! Other musicians have created systems running dual AMD Opteron processors with similarly impressive results — and with this CPU range, systems using up to eight processors are possible, although I've never heard of any musician running more than two, because of current audio software limitations (see Hyperthreading in the 'Technical Terms Explained' box, above, for more details).
Of course, the next logical step from a dual-core system must be one that also has dual physical processors. To the operating system this will appear as four processors, but it's currently difficult to predict how audio software will benefit — or, indeed, cope. Steinberg currently recommend disabling the Hyperthreading feature of Intel's Xeon processors if you're running a dual Xeon PC, as otherwise the dual processors will appear to the operating system as a total of four processors (two physical and two virtual). Many plug-ins and soft synths may have severe problems with this arrangement — at the moment, anyway. However, as more and more musicians start trying out similarly-equipped computers, it seems inevitable that 'quad' processing will eventually become the next goal for musicians aiming for the ultimate PC, so I suspect that software developers will have to sort out any current teething troubles.
While PC systems running dual Xeon or Opteron processors have already proved their worth for anyone seeking a significant hike in performance, there's now another way to have multiple processors in your PC. Both AMD and Intel have abandoned their pursuit of ever-higher clock speeds in favour of a dual-core approach, where two processors sit side-by-side on the same chip to provide faster performance.
As I write this, AMD have just officially launched the world's first dual-core 64-bit processors. The 800 series Opteron is available now and designed for high-end four- to eight-way servers, while the Opteron 200-series is aimed at two-way servers and more general-purpose workstations, and should also be available by the time you read this. However, for the musician probably the best news is the unveiling of the dual-core Athlon 64 X2, due to be launched in June.
These new ranges have two cores on a single die, plus memory, I/O and dedicated caches, but use the same infrastructure as their predecessors, including the same physical package as the single-core versions. The dual-core Opterons use the familiar Socket 940 format, while the dual-core Athlon 64s will use Socket 939. So for many AMD motherboard owners, moving to dual-core simply involves buying a new processor and updating the BIOS.
Performance benefits for the musician are uncertain at the moment, but it seems that a dual-core Athlon X2 could provide more than 60 percent more power than a similarly clocked Athlon 64. Some of the Opteron benchmark tests I've seen also suggest that a dual-core model outperforms twin processors of the same clock speed, probably because having both cores on the same chip neatly bypasses some of the potential bottlenecks of dealing with two individual cores on physically separate chips, interfaced via the motherboard. The dual-core model also generates less heat, which should make it better for use in a quiet music PC. Overall, it looks as if dual core may be the preferred route of the future if you're considering any sort of PC containing multiple processors.
Meanwhile, Intel have produced a dual-core Xeon, due to ship in early 2006, and are also developing dual-core versions of their Itanium and Pentium 4. The new Pentium 4 (codenamed Smithfield) will be released as the Pentium D range, and is expected to out-perform the existing P4 Prescott by something like 40 percent when running at the same clock speed, depending on the application. This is probably the most interesting new dual-core processor from Intel for most musicians, since it will be the cheapest, and because it's expected much earlier than the dual-core Xeon, in the latter half of 2005. It won't feature Hyperthreading, like its predecessors, although this will feature in a new and more expensive dual-core Pentium Extreme Edition.
Sadly, all these new dual-core processor ranges from Intel will require specific motherboards with new chip sets, which not only means a more expensive upgrade, but also a longer wait before you can go ahead, especially since new chip sets may result in teething troubles. For anyone wanting to buy a new PC at the moment, the AMD option is more enticing: you can buy a machine with a single-core processor right now that should be compatible with the forthcoming dual-core Athlon 64s or Opterons, and then buy a dual-core processor for it later on.
It used to be possible to buy a computer that would last you unchanged for at least 18 months, but nowadays this is more difficult. If your current PC is running out of juice, fitting more RAM may improve its performance in some cases, and at the very least let you load more simultaneous soft synths. Installing a faster hard drive or RAID array may also let you run more audio tracks, although, as I've explained, many musicians shouldn't find this necessary. However, for the vast majority of musicians the only real way past a genuine bottleneck is more processing power, and this generally means a more radical upgrade.
If you already have a 64-bit-capable computer based on an Athlon 64 or Opteron you may be able to achieve this by moving across to Windows XP Professional x64, once things have settled down. All the signs are that you'll also be able to pop a dual-core equivalent of these processors into the same motherboard later on, for even better performance — assuming that the motherboard manufacturer releases a suitable BIOS update to permit this. I can't guarantee that this will happen in all cases; sometimes it makes more sense for motherboard manufacturers to introduce a new model to support a new processor, even if the old one could be forced to do it with a little development effort. Nevertheless, many existing AMD users have some cause for celebration.
Unfortunately, for the rest of us keen to achieve a serious improvement in processor performance, going 64-bit will probably mean buying a new motherboard as well as a new processor, or even buying a completely new PC. If you really must have lots more CPU power now, upgrading to a motherboard supporting Dual Xeon or Dual Opteron processors will provide lots more guaranteed performance, and if you want to buy a completely new PC there are certainly some extremely capable new systems already available from specialist music retailers in these formats.
However, so many major technological changes have either just arrived, are imminent or are on the horizon that, unless you really must have that much faster new computer immediately, I'd advise you to sit tight for a month or two until the dust settles. Then the various new processors will actually be available to the public, and systems built using them will have started to appear for testing with 64-bit versions of existing audio software. Overall, I don't envy anyone about to make a major upgrade to their current PC (or buy a completely new one for music purposes), since there are so many changes afoot. All I can say is that it's a very exciting time!