Now that software versions of synths, processors, and samplers are available for less than their hardware equivalents, many musicians are being tempted to dispense with their studio hardware and run everything inside their PCs. Martin Walker examines the feasibility of putting all your chips in one basket...
In August's SOS, I looked at some of the possibilities available to musicians using a wide variety of PC types, both old and new. The final section mentioned the possibility of running an entire software studio inside your PC, along with a fair few warnings about conflicts when attempting to run several simultaneous applications, such as a MIDI + Audio sequencer and a software synth and/or sampler. At the time I pointed out that you would need a very powerful processor to accomplish this feat, but since so many people seem determined to do it I've taken my research further!
There seems to be a significant number of PC‑based musicians who have extremely powerful PCs, often with Pentium III processors and 256Mb or more of RAM (although many of them do still use low‑cost consumer soundcards like the Soundblaster range). If you have invested a lot of money in buying the most powerful PC you can afford, it makes sense to get the most out of it, although running everything in software is not for the faint‑hearted, and a single unresolvable conflict between two music applications may completely sabotage your efforts. The problem is that there are now so many possible combinations of software that many musicians only discover after parting with their money that they can't use a new music application at the same time as the rest of their system.
However, this looks all set to change with the arrival of Steinberg's Rewire technology (watch out for an SOS feature on this shortly). One of the biggest stumbling blocks when creating software versions of both synths and samplers was that until Rewire arrived the only way you could add EQ and effects to individual sounds was to use a soundcard with multiple outputs feeding an external mixer, or (if available) to use whatever built‑in treatments were included with each software package. This has all changed now that Rewire‑capable applications can interface directly with Cubase VST v3.7 for Windows (and 4.1 on the Mac). Every single sound can be streamed into Cubase on a different channel and be treated separately. This is equivalent to having a synth or sampler with almost unlimited multiple outputs (up to 64 per Rewire device), and unlocks huge potential, as long as your PC has enough processing power to do what you need.
With this in mind, as well as the imminent launch of various VST Instruments (see the Cubase VST v3.7 for Windows review in SOS September '99 for more details), we are likely to see an ever‑increasing range of software‑generated music packages, but with less conflicts and the potential for easier integration. Let's see just how far you can take the concept of the completely PC‑based studio before it falls over.
There are various ways of creating a totally PC‑based studio, and for those committed to a software‑only solution, using multiple computers can make a lot of sense, especially if you still have an older PC that has since been superseded, and which can be pressed into service as a stand‑alone software synth or sampler. This approach can also reduce potential conflicts, since you never need to worry about the vagaries of running several applications in the same PC.
Even if you don't wish to run a software synth or sampler on a separate PC to avoid conflicts with a MIDI + Audio sequencer, it can often be easier to run them on a separate soundcard. Although some multi‑output soundcards do let you allocate individual stereo output pairs to several applications (multi‑client audio capability), you often don't find out until you try it. My Event Gina card, for instance, doesn't support this with its version 4.05 drivers, and even if you disable one pair of MME tracks inside Cubase VST for instance, the drivers still won't let you run this pair with a software synth.
If you do have several soundcards in your PC, it can therefore make sense to dedicate one to a software synth. Ironically, many consumer soundcards provide DirectX drivers with fairly low latencies, and if they also feature a digital output you can pair them up with a more sophisticated main card used for audio purposes. For instance, I often feed the S/PDIF output of my AWE64 Gold into the S/PDIF input of my Gina card. While this lets you achieve better audio quality for a software synth running on an older consumer card by transferring the audio using the digital path, the only worry is that you compromise the high‑end card by letting it slave to a lower‑quality clock signal with higher jitter.
Whether you run your software synths and samplers on a separate PC or soundcard, or all on the same one, you can never have too much processing power — even the fastest Pentium III processors are not a case of overkill. Those dedicated to the soft approach may already have noticed the new 'Soft Synth Station' being advertised by Turnkey for just such a purpose, with a Pentium III 500MHz processor, 256Mb of RAM, and a 10Gb hard drive. Don't forget that if you are happily using a 200MHz processor and 64Mb of RAM with your current MIDI + Audio sequencer, and are considering buying a software synth that recommends a 200MHz processor and 64Mb of RAM, you will really need a 400MHz processor and 128Mb of RAM to successfully run both together.
I discussed and tested the comparative performance of the Celeron and Pentium II ranges in SOS April '99, but in essence there is little to choose between them for music, and the Celeron is far cheaper. If you want to spend more money then the Pentium III is the one to go for, although this is still a serious investment, with the slowest 450MHz model just under £200 (processor only, obviously!), and the current fastest 550MHz version just over £500.
However, apart from the faster clock speed, simply plugging in a Pentium III processor won't automatically give you a great leap in performance terms (most tests show that a Pentium II 450MHz runs software at an identical speed to a Pentium III 450MHz). The 70 new SSE Instructions of the Pentium III (including the streaming Single Instruction Multiple Data ones) can certainly make a significant difference to audio software, but only if it has been specifically written to use them. Such software includes Steinberg's Cubase for Windows v3.7 (and the previous 3.65 version), and Seer Systems' Surreal software synth, which offer a seemingly typical 10 percent improvement in speed. This means that you'll get significant reductions in CPU overhead when using the built‑in Cubase EQ, but on all the other software and plug‑ins which haven't benefited from a PIII upgrade (and there hasn't exactly been a flurry of these) there will be no improvement at all. For those of us without bottomless coffers who are thinking of buying a new Intel processor, the Intel Speed graph (see right) should prove interesting.
If you currently have a slow Pentium II or Celeron processor, you will certainly be able to run every software synth I've reviewed to date, such as Seer Systems' Reality, Native Instruments' Generator and Reaktor, and the Nemesys Gigasampler. However, running software synthesis alongside a MIDI + Audio sequencer is another matter. Having plenty of RAM will help (my recent upgrade from 64Mb to 128Mb of RAM certainly helped in some respects), but you really need raw processing power. My recommendation for those with more sense than money is to upgrade to the Celeron 466MHz model. This currently costs about £116 (my Pentium II 300MHz cost more than double that when I bought it a year ago), and will outperform a Pentium III 450MHz with the majority of non‑tweaked software.
While it's comparatively easy to generate analogue waveforms such as sine, triangle, pulse, and sawtooth, calculating complex waveforms using various forms of synthesis, and then filtering and otherwise treating them, requires a great deal of processing power. There are two main designs of soft synth: modular and 'fixed'. Each has its advantages and disadvantages.
Most people are excited by the possibilities of modular types like those in the Native Instruments range (see this month's review of their Reaktor starting on page 108), since you can patch together any combination of modules to get the sound you desire, using 'virtual patch cords'. However, while this versatility is wonderful, it does have a penalty in slightly greater processor overhead. In addition, unless you really know what you're doing, your own designs may use more processor power than they really need. For instance, simply turning down an LFO depth control to zero doesn't necessarily lower CPU usage, since the LFO will still be connected and using processor power (even though they now have no effect on the sound).
On the other hand, 'fixed' soft synths like Reality offer a huge number of pre‑defined controls, but the order in which sections like oscillators, filters, amplifiers, envelope generators, LFOs, and so on are connected is permanently fixed. You can still coax a vast number of sounds from designs like these, and the absence of patch cords and all the extra interfacing code normally means that processor overhead can be slightly lower. If you intend to create entire songs using software generated sounds then you may well get more voices from a design like this, at the expense of some versatility.
When it comes to software samplers, the processor power available in the host PC is perhaps not quite so important, since the waveforms are not being created from scratch, they're simply digital recordings being played back, and only need filtering and further expression. However, once again, there are two distinct types. Most use some of your system RAM to store samples, so having plenty of this is important — if you wish to have 32Mb of samples available then a realistic minimum amount of system RAM is 96Mb, with 128Mb being preferable.
In the case of the Nemesys Gigasampler, samples are streamed directly from the hard drive, so the CPU and speed of the hard drive together determine the likely number of notes. However, since RAM is used to buffer the different sounds, the amount available will also determine how many different instruments can be played at once. Users report 64 voices with a Pentium 233MHz, although once again this is without the additional overhead demanded by a MIDI + Audio sequencer.
However much processor power you have, you are still likely at some stage to run out. Since so many people are interested in buying software synthesizers, I should point out the consequences of processor overload, since they can be extremely frustrating, and generally result in one of two things. Either you get a nasty pop, click, or glitch in the audio output, or it may stop altogether.
Other warning signs that your processor is flagging are slow and juddery screen updates, and long waits before a keypress registers. This is largely because developers sensibly give much greater priority to generating audio than they do to graphics. Don't underestimate this aspect — at one stage during my research for this feature it took my PC a full 30 seconds to redraw the screen when switching between Reaktor and Cubase, and there is little you can do in the meantime. The safest thing is probably to wait until all screen activity has finished, and then de‑activate whatever it is that caused the processor overload.
When you have a MIDI + Audio sequencer running as well as a soft synth, the calamitous possibilities multiply. You may get glitches in the sequencer audio tracks (and if you happen to be recording audio at the time this can be even more frustrating) or your sequencer may stop prematurely. Worst of all, your PC may crash altogether. If you have a utility such as Norton's CrashGuard running behind the scenes, this may catch the crash and let you continue or terminate the offending application gracefully. If not (and many musicians prefer not to run any background utility unless absolutely necessary) your machine may exhibit the dreaded blue screen warning, or even worse, lock up completely, requiring a complete reset. No‑one ever wants this to happen, but when creative inspiration strikes, it's absolutely the last thing you want.
Even with a powerful PC and Rewire, which greatly reduces the possibilities of clashes by running all audio applications using the same software engine, running several music applications simultaneously can still result in a clash of requirements. Your software synth will vary greatly in its processor requirements depending on how many notes are currently being played, and your MIDI + Audio sequencer will vary in its requirements depending on how many audio tracks are currently active, and what effects are being used.
There are four possible answers to the problem of running out of processor power as a result of attempting to play too many simultaneous software‑generated notes. The first (and easiest) is simply to lower the sample rate of your sounds (most soft synths have suitable options). This will result in a significant drop in processor resources, without necessarily ruining the sounds (in fact, in many cases using a lower sample rate may give your sounds more desirable bite and grunge). However, while some software applications let you choose sample rate on a patch‑by‑patch basis to save processor power, others only provide a global setting.
The second answer is to carefully monitor your sequences to ensure that your note requirements never exceed the number that you know your system can cope with. Although this sounds tedious, I'm not necessarily talking about totting up by hand — many modern analogue sounds (as well as solo and bass voices) tend to be monophonic, so making sure that this option is selected in your soft synth will ensure that only one voice is ever used for these sounds.
Most (but not all) soft synths only use processor power for the voices they are currently playing. The first voice will take a significant amount of overhead, while additional voices using the same 'engine' will takesignificantly less (see the screenshot on page 160). The Native Instruments range offers another way of working; here it is up to the user to specify the maximum number of voices used by each Instrument. Using one voice again takes the biggest initial chunk of processor power, and additional ones each take less, but you will be wasting processor power if you enable six voices and then only use four of them — so it pays to be careful.
Running everything in software is not for the faint‑hearted, and a single unresolvable conflict between two music applications may completely sabotage your efforts.
Thankfully, the third alternative is far simpler than note‑counting. Most software synths and samplers have options to cap the number of simultaneous notes, and as soon as an additional one is required the oldest one currently sounding is abruptly terminated to make way for the new one. This can be distinctly audible in some situations, but those of us who started making music years ago with MIDI will already know all the tricks. You can try placing obvious and important sounds like drums, bass and lead lines in the topmost tracks of your sequencer, and let the chordal ones move down a bit. Cubase is supposed to give a higher priority to the topmost eight tracks in its Arrange page — these tracks will then be dealt with first, and any note‑stealing should occur elsewhere, where it is less noticeable. You can avoid overlapping notes by taking advantage of any Legato functions in your sequencer — these adjust note lengths to reduce the total number of notes playing at any time, although if there is any decay after the Note Off you will still be using more than one note until the end of the decay phase.
For those applications that let you specify a maximum number of simultaneous notes, run some simple tests to see how many your PC can manage comfortably, but make sure you are running your sequencer with a typical number of hard disk audio tracks and effects at the same time to give it a more realistic scenario.
The main problem with capping CPU usage power solely by the maximum number of notes is that the processor power taken by a synth may vary hugely depending on the type of synthesis, as well as what other features such as filters and effects are currently being used. A fourth approach is therefore to cap the maximum amount of processor power that can be used by the software synth, and thankfully nearly all of the available soft synths can do this. Usually, this method works by note stealing; when the allotted processor power is used up, and a further note is requested, the 'oldest' note is stopped to make way for it. As long as your PC is powerful enough to run eight or more software notes, this is unlikely to be very noticeable, but as with older analogue synths, the effects will be more obvious if you can only manage four or less notes.
Sharing processor resources between a sequencer and soft synth can be done in various ways. You can set up your soft synth to be capped at a certain maximum processor percentage, at a certain maximum number of voices, or a combination of both. The safest way is to set a sensible 'Maximum CPU percentage' value of something like 50 percent, leaving the remaining 50 percent for your sequencer. Then, if you find yourself running out of processor power, try reducing the maximum number of notes a little, and see if you can hear the effects.
Many musicians will run out of processor power long before they are satisfied with the number of software voices, and here it is vital to have some additional hardware support, such as a soundcard‑based synth, soundcard sampler, or external MIDI hardware (see the 'Halfway House' box). As long as your basic sounds are catered for in hardware (external MIDI modules or built‑in soundcards) then you may be able to splurge 50 percent of your processor power to add an otherwise unattainable sound using physical modelling, FM synthesis and several off‑the‑wall software effects. It's up to you.
The trick is to spend some time running your music applications while keeping an eye on any processor usage meters provided, to get a feel for how much different features take. Once again, Rewire technology will make this far easier, since you will only have a single CPU meter to watch, rather than trying to judge which application pushed your processor 'over the edge' when you get a glitch.
Individual software applications may provide other performance optimisation choices, depending on your particular combination of processor power andavailable RAM. For instance, Bitheadz' Unity DS1 has an 'Expand Samples When Loaded' option, and choosing to use this requires more RAM, but reduces CPU requirements. Although you will have to read the manuals carefully, tweaking settings like these may help your PC to achieve more simultaneous notes.
While it's an exciting prospect to those without rackmounting samplers and synths to buy a couple of pieces of software and achieve the same results using a single PC and soundcard, only those with particularly powerful PCs will even be able to attempt this. Many modern hardware synths and samplers have 64‑note polyphony, and replacing such a hardware studio with one solely created in software is a tall order indeed.
Currently the main limitation is available processing power, but thankfully clock speeds seem to be rising more rapidly than any other aspect of PC design. There will be a Pentium III 600MHz device from October '99, and the final Celeron 533MHz model in the current range is expected by the end of this year. Another interesting possibility comes from AMD, whose new Athlon processor seems to have excellent floating‑point capabilities. A new Celeron range with 100MHz FSB and the same SIMD extensions as the Pentium III is expected next year, starting with a 550MHz model, and sometime next year we should see the appearance of 1GHz clock speeds. With this amount of power, many more musicians will be tempted to take the plunge and upgrade to a more powerful processor.
However, some cautionary words are in order. Firstly, because there are so many end‑of‑line bargains available on MIDI synths these days, switching to software synthesis and sampling doesn't always make for as big a saving as you might think. In fact, many enthusiastic users of software‑only systems don't see them as a cheap alternative at all — rather a different approach that brings them various advantages, such as being able to work in a large‑screen environment and having more flexibility to specify exactly what sort of sounds they use.
In short, I would suggest that there is no point in selling your existing MIDI synths and samplers and changing to a software‑only setup unless it gives you a significant advantage, such as being able to access an exotic form of synthesis, or hundreds of megabytes of samples in a single song. Attempting to create every sound in software is really tempting providence, and unless you have an extremely powerful PC, it is at present far better to use sounds derived from a mixture of software and hardware.
For those who want to move on from a hardware‑only setup but aren't prepared to rely on running everything in the PC, there is another hybrid solution that many people find appealing — combining musical software and hardware, but within the PC. Of course nearly all musicians are likely to need at least one external MIDI keyboard instrument to input data (unless you use another sort of input device such as a MIDI guitar or Wind Controller), but modern soundcards can now provide most of the other facilities previously only available in rackmounting boxes.
Many modern soundcards include excellent MIDI synths and samplers, and these can provide many of the basic sounds needed for your music, leaving a software synth to add more unusual synthesized or evolving sounds that cannot be easily sampled, or a specialist software sampler like the Nemesys Gigasampler to add really long samples that are not restricted by available RAM size. You could also use a DSP‑based product like Creamware's Pulsar to take much of the strain from your processor.
Not only does this mixed approach often result in a more stable PC system, since you end up running less simultaneous software applications, but it also greatly reduces the total processor power required.
Many musicians who started out with hardware MIDI keyboards and modules still believe that this hardware offers the best way to work, due to its general reliability, hands‑on interface, and the fact that it tends to have a good second‑hand resale value if you decide to update your gear in the future. For live work, a rack of MIDI boxes will also stand far more abuse and dodgy mains supplies, whereas taking a PC on stage and keeping it running reliably during a gig could prove a nightmare.
There is also a lot to be said for being able to switch on a synth and start playing within a second or two, and you don't suffer latency problems or conflicts with other applications. Some musicians have tried software synthesis or sampling, had a few problems, declared it impractical and unreliable, and then returned to their relatively crash‑free hardware racks.
In addition, musicians who use software synth and sampler packages point out the advantages of buying into an evolving system — in a year's time you will probably have a more powerful computer capable of many more notes and effects, even when running the same synthesis software, and by that time there will probably be further upgrades with more enticing features at little or no additional cost as well. Users can even suggest new features to the developers that stand a much better chance of appearing in future versions than with the hardware equivalents.
Mind you, those of the hardware persuasion will counter this by saying that their boxes work straightaway, and don't need the number of bug fixes that the software camp do. It's an on‑going and complex argument!