Vista Performance Tests
January 2008
It's difficult to know exactly what a computer is capable of. What is the difference, in terms of audio and plug-in performance, between certain CPU's? How does Vista really stack up against XP on different machines and in different circumstances? Does the software version I'm using make any real difference? Part of our mission at Rain is to demystify the technology surrounding computer music technology. We believe in giving the best information we can to help you make informed choices.
In our previous performance testing we took a single, cutting edge machine and ran tests to determine whether Windows Vista would make a viable DAW platform. The results were very positive and caused a lot of discussion and even some controversy amongst the computer music community. What we were trying to show, and what we certainly demonstrated, was that Vista is capable of being a successful platform for DAW applications provided you have compatible software and hardware - this remains the case.
In this testing we used four different systems in order to make our results a great deal more comprehensive and to give us a better chance of catching any errors or anomalies. Our focus is on comparing systems rather than comparing operating systems although both are dealt with. No test environment can be perfect but we offer up our results as reliable and hopefully useful information.
Each system was tested using Windows XP Professional 32bit, Vista Ultimate 32 and Vista Ultimate 64, fully updated and tuned for audio production by Rain technicians.
Music Software and Hardware
- The test software was Steinberg's Cubase version 4.1.
- The audio interface was the excellent Focusrite Saffire Pro 26 using software driver set 2.1rc1. Interface buffer set to 4ms giving an Output Latency in Cubase of 9.5ms.
- For each desktop system we used a firewire card with the Via VT6306 VIA Fire II chipset as specified by Focusrite. The notebook used its own on-board Firewire controller.
The Tests
Most benchmarking software used to test computers seems to produce results that don't bare any relation to anything you are actually likely to do. Knowing how many bits of data a system can process in any given second is largely unhelpful in an audio context. What we want to know is how many plug-ins or virtual instruments a system can run and then play music through them without glitching. So that's what we did.
Plug-ins
- We start with a 32 track project of 24bit 96kHz audio.
- On the master bus we add a multiband compressor and a Roomworks reverb.
- Next we add 4 bands of EQ onto each track followed by a compressor.
- Then we add another plug-in (as specified below), one at a time, one track at a time until the audio begins to glitch.
- If we can load 32 of the new plug-in then we start adding another to each track and so on in the same way.
- Once glitching occurs we remove plug-ins until stable playback is maintained.
- We add up all the plug-ins and that gives us our number.
Here are the plug-ins we used:
- Channel EQ
- Dynamics Compressor - Ballad Piano
- Stereo Chorus - Background Vocal
- Mod Machine - Dance Vocal
- Rotary - Rockrotary
- Maximizer - Remastered
- Pingpong Delay - Shuffle
- Roomworks SE - Bright Plate
Polyphony - Spectra
Rarely would you find a piece of music that has every instrument playing at the same time, however for the purposes of measuring polyphony that's what we had to do. Rather than measuring from a single instrument we restricted each instance of the Spectra VSTi to 8 notes and then added another instance. A separate MIDI track was created for each instance and a sequence of 8 notes were played and held for a number of bars to ensure stable playback over time. Once glitching occurs a few notes were muted until stable playback was achieved giving our instance and polyphony count.
Each instance was loaded with "Ana Strings" and set to 8 voices.
Polyphony - HalionOne
A slightly different polyphony test this time using a sample based instrument. The test was the same but the polyphony of each instance varied depending on the loaded instrument.
Here's what we loaded and its corresponding number of voices that sounded in response to the 8 notes:
- Cello 1 - 8 voices
- Viola 1 - 8 voices
- Violin 1 - 8 voices
- Stereo Strings - 16 voices
- Oboe - 8 voices
- Clarinet - 8 voices
- Fr horn - 16 voices
- Trumpet - 8 voices
- Piano - 8 voices
- Accordian - 8 voices
- Suitcase - 8 voices
- Aah Choir - 16 voices
- Ooh Choir - 24 voices
- Sweep Pad - 16 voices
- Warm Pad - 16 voices
- Medium Brass - 24 voices
- Brass & Strings - 32 voices
- Aah Voice Attack - 32 voices
- Woodwind Ensemble - 16 voices
- Mogul - 32 voices
- Silk Chorus Guitar - 16 voices
- TX Digital Piano - 16 voices
- Lovely Ensemble - 32 voices
- Legendary - 16 voices
- Slow Movie Strings - 32 voices
- Digital Organic - 32 voices
- Church Organ - 24 voices
- Sitar - 8 voices
- Two Synths - 16 voices
Track Count
Bit simpler this one - keep adding audio tracks until the system chokes. Again we started with the 32 tracks of 24bit 96kHz audio and then made physical copies of the audio files and imported them into Cubase. As these were music tracks rather than test tones you could see how or if the performance varied with the dynamics of the piece and it was also easier to hear the glitches.
The Results
Plug-ins
All the systems performed really well, particularly pleasing was the intel based notebook results giving 115 plug-ins running on a 32 track project all from a 2.2GHz processor. The scaling from one system another is quite consistent although it's interesting that moving from a dual-core E6850 to a quad-core QX6850 didn't result in a doubling but rather about a 45% increase. This demonstrates that there's more than pure CPU number crunching going on.
With the operating system comparison the XP and Vista 32 results are much more consistent with other results that we've seen, with XP being slightly ahead. The Vista 64 results are a good improvement and largely down to the Cubase 4.1 64-bit patch. In our previous testing using Cubase 4.01 Vista 64 showed no advantage over Vista 32 whereas here it does and we can see that it even pips XP on most of the systems. The test projects were all built using 32 bit plug-ins and no special changes have been made to make the best out of a 64-bit OS so this shows only the beginning of Vista 64's potential.
Polyphony Spectra
The impressive notebook managed to run 12 instances of the Spectra, not quite managing 8 notes on the last one for a score of 94 voices. It's interesting to note that this is the only system where XP won out in this test. With the others Vista 64 gains the highest marks. Comparing the Dual to the Quad again we see it's much closer to a doubling of polyphony showing that for software synths CPU power is the most important factor. The results on the AMD Athlon system were really hard to tell apart with just one or two voices between them.
Polyphony HalionOne
The 64-bit patch for Cubase 4.1 also contained a 64-bit version of HalionOne. However, the results for Vista 64 are disappointing considering the 64-bit environment. Although certainly no slouch, we were hoping for a marked increase in performance for Vista 64 in this test. Perhaps working up a test project with more 64-bit specific instruments would yield better results.
Again, CPU is the main contributor to the results with the QX6850 demonstrating that it has twice the cores of the E6850. XP Pro takes the crown as far as the OS is concerned.
Track Count
This was completely OS independent as there was no difference between XP, Vista 32 or Vista 64. The technology, however, did make a difference. The notebook managed a whole 70 tracks of 24bit 96kHz audio, which is extremely impressive for a mobile system and a credit to the 7200RPM Seagate Momentus hard drives. Both the AMD based desktop and the Intel dual-core desktop ran the test project from the second partition on the OS drive and shows how the Intel drive controller has a slight edge over the AMD chipset. The RAID0 drive on the quad-core Intel desktop demonstrates a solid 16% performance increase over a single drive, which, although not as impressive as one would expect, shows its suitability for multitrack recording.
Conclusions
Overall all the systems performed brilliantly with the Focusrite Saffire Pro and Cubase 4.1. The 4.1 update came right in the middle of the performance testing so we were able to see if it brought any improvements in terms of power. We saw a 10-15% improvement in the plug-in and Spectra tests and a whopping 40% improvement to the HalionOne tests - that's from one update.
In terms of the systems the dual-core notebook is a feisty little machine. The T7500 processor on the Santa Rosa 965 chipset far outpaces desktop machines of just a year ago - aided of course by the Cubase update and the excellent Saffire Pro drivers. The first AMD based system we've tested performed admirably as an entry level machine giving good value for money, even with onboard graphics which is usually seen as a no-no in computer music. The dual-core Intel desktop certainly whizzes along as expected but the real winner here is the QX6850 - its performance was breathtaking, head and shoulders above the rest and at the time of writing it's the fastest desktop processor available. The QX6850 is available as an upgrade on our Element and Nimbus systems. It's not cheap, but it is awesome.
The comparison between operating systems is very interesting. In the year that Vista has been available to the general public, the support from music software and hardware manufacturers has been slow but steady. Issues remain, like those of working at very low (<2ms) latencies, but on the whole it is not far off XP in terms of DAW performance. Vista 64 certainly looks very encouraging winning half of the tests by a small margin. Vista 64 should be the future, it holds the promise of far greater performance than we've seen here, but support by the music software and hardware manufacturers has been even slower than Vista 32. Performance in XP has no where to go, it's reached its peak, we've got everything out of it that we're going to, whereas Vista 64 hasn't even begun to realise its potential and hopefully, over time, we'll all begin to benefit.
So, what OS should you choose for your system? The simplest answer is to stick with XP - it's stable, fast and is tried and tested with absolutely everything. Vista 32 brings lots of advantages in terms of ease of use, future compatibility and security but has no advantage in DAW performance terms. However, Vista 32 is certainly no slouch and if the attraction of the other non-musical features tempts you to Vista then you won't be losing too much DAW performance on a properly configured machine. Vista 64 is the way to go in terms of performance and future proofing but only provided that the hardware and software you want to run is compatible - that is most important. But with 64-bit, Vista 64 compatible versions of GigaStudio and other sample based instruments from EastWest and Vienna Symphonic just around the corner the future of Vista 64 is looking quite rosey.
Robin Vincent is CTO of Rain Computers, Inc, President of Rain Recording UK and author of music technology books PC Music - The Easy Guide and The Guitarists Guide to Computer Music.
