PC DAW O.S Shootout - Part I : XP 32 v XP64 v Vista - Nuendo/Cubase :.

With the official launch of Windows Vista in January 2007 , the focus has definately been raised in relation to what benefit the new O.S will bring to the table for the Digital Audio community.

Lets brush away all of the hype and fanfare being generated about the slick new GUI, gadgets and improved security features, and lets get down to the crux of whether the O.S will bring any tangible improvement to the current XP32 / XP64 O.S, that the vast majority of us are using to drive our PC DAW solutions.

First off, I'll give you a rundown on the new Vista features that Microsoft have touted will be of most benefit to the Audio community , I'll then crank up the benchmarks and let the O.S's slug it out..

There are 3 main areas that have been touted as a direct benefit to the Professional Audio community

WaveRT Driver Model :

Whereas both WDM and ASIO operate at User Mode in the safety buffer of the HAL, WaveRT drivers operate directly at Kernel Mode, which will potentially allow lower latency. Traditionally drivers operating at Ring 0 - Kernel level have been avoided in audio environments , one exception being GSIF , due to the potential of a rougue driver bringing the whole system down, whereas drivers operating at User Mode will only effect the current process.

MMCSS -Multimedia Class Scheduler Service:

Quote from Microsoft : "The Multimedia Class Scheduler service (MMCSS) enables multimedia applications to ensure that their time-sensitive processing receives prioritized access to CPU resources. This service enables multimedia applications to utilize as much of the CPU as possible without denying CPU resources to lower-priority applications."

In short , applications that are optimised to use this feature will be less susceptible to glitches and dropouts, caused by background tasks interfering with the current process.

WDDM - Windows Driver Display Model:

Using the WDDM , the new Aero interface and any applications which use it are entirely rendered using the GPU.

Theoretically, this should speed up system responsiveness because it offloads a lot of work that the CPU was doing in drawing windows to the GPU on your video card

This leaves more CPU cycles available for processing your audio.

Well thats the theory.. :-)

While preparing for the shootout , there had been quite a bit of interest on the web in regards to substantial performance gains of Vista over XP that had been reported and posted by another party. I wont bother repeating or posting those results here, to be fair, the methodology and architectural configurations , although similar, were not identical, so the difference in results are difficult to guage.

For those interested, the discussion can be viewed at the SOS Forum. Here

One point that was raised, and that hadn't been covered previously was the performance of XP64 in comparison to XP32 and Vista , so I decided to include that O.S in the shootout to round out as many variables as possible.

The Lynx 2 was again chosen as the Referenace Audio Hardware , purely because it had proved both the most consistent and best performer in all previous testing, and also, Lynx had fully working drivers for both XP32/64 - Vista 32/64

The benchmarks used were Blofelds DSP40 , and also L-Factor II

The test applications chosen were Nuendo 3.2 and Cubase 4.02

Cubase 4 testing was added later in the proceedings after it was highlighted that N3 was not "officially" Vista compatible

Moving onto Vista, I firstly configured a highly tweaked Aero GUI, just to test the alleged performance gains attributed to the process being offloaded directly to the GPU.

The above results speak for themselves :-(

The behaviour of Nuendo running the test was also extremely erratic, constant pauses, some up to 20-25 seconds long, I kid you not, sluggish, more farping and parping than either XP32 or XP64.., the real time audio engine aspect of Nuendo under Vista took an absolute pounding..:-(

The install was tweaked to as high a level as I could, while still remaining Aero active.. , in short, the only thought I had was.. "What a piece of shite "... LOL !!

Next up I stripped the Vista install back to the Vista Basic GUI, definitely an improvement over the tweaked Aero configuration , especially on L-Factor II, still not up to XP64 tho.

What was really interesting to note was the difference in overall system behaviour, feeling a lot more fluid while running the tests, no long pauses as when using the stripped back Aero, so there is definately a resource hit when using Aero, even when it is stripped back. This could be also dependent on Video cards being employed. I am using smaller workstation based Nvidia Quadro cards on my systems, which wouldn't be high on the Vista rating..:-)

Either way,

I'd suggest avoiding Aero for Audio use.

My initial results with N3 certainly raised a few eyebrows , as the results were in strict contrast to some earlier reports.

It was suggested that the Vista results could have been seriously effected by N3 not being officially supported under Vista, and to rerun the tests using C4.

I ported the tests over to C4, and reran the tests across all 3 operating systems again.

The C4 results were at least consistent, but not in the way some had hoped.

C4 is consistently worse than N3 across all 3 O.S's, also the performance scaling on Vista is consistent to that of N3, proving that there is no Vista optimisation in C4, as I had suspected.

This raises a few serious questions in regards to the results and claims presented earlier where Vista had clearer trounced XP

1: " Vista outperforms XP without any trouble at all"

This is something that I could not quantify in any of my extensive testing , I found the exact opposite applies. XP clearly and easily out performs Vista.

2: Multithreading capability in XP is not optimised for Quad Cores and explains the issues of the system performing poorly under XP32 in those respective tests.

This is something that has absolutely no basis. I have always experienced a linear and progressive scalability using Single Quadcore and Dual Dualcore based systems using XP32. I experienced similar scalability in XP64 and Vista64 using both N3 and C4.

If there is any advantage with Vista in regards to better scalability per core, then it would only come into play above 4 cores.

Unfortunately with current state of the Steinberg MP capability, that answer will remain unconfirmed for now.

3. N3 is not Vista compatible, whereas C4 is officially supported.

My testing shows that C4 is no more Vista compatible than N3/SX3, if anything, and to top it off, N3 actually out performed C4 on Vista on these tests.

These tests are in no way conclusive in regards to DAW performance under Vista , I just wanted to dot the i's using the current applications and benchmarks.

Some companies are far more optimistic than others when it comes to Vista , Cakewalk in particular are leading an extensive push towards optimising their applications for Vista, which is not surprising considering their close affiliation with Microsoft.

They will be the first to utilise WavRT and the The Multimedia Class Scheduler service , their latest update , 6.2 actually has some MMCSS optimisation implemented

I am sure the coming 6-8 months or so will see more and more Vista optimised drivers and applications , Steinberg has indicated that version 4.1 of Cubase and Nuendo will be fully Vista optimised.