Nuendo / SX Benchmarks :.


MP Scalability : Reprise - Light at the End Of The Tunnel ? :.


Its been over 12 months since the original article that highlighted the scaling issue that were encountered specifically with the Steinberg line of DAW applications on Dual Quadcore systems. Steinberg at the time had officially acknowledged the issues being software related at their end, and had promised a fix in the new versions that were to be released in Q3 2007.

Both Nuendo and Cubase 4.1 were released as promised in Q3 2007 , however the x-scaling @ lower latencies was still identical to the previous versions. Steinberg had then promised the issue would be addressed with the next maintenance update - 4.1.2 , which was released early Jan 2008. Again the x-scaling issue remained , and this time no time frame was given for the issue to be addressed. While we were waiting for the fix from Steinberg, the tide was about to take a surprising turn.


Steinberg X-Scaling Bug Resolved ?    

While doing some routine testing and basic maintenance on the Dual Quad development system, I stumbled on a surprising fix that for the greater part resolved a lot of the scaling issues being experienced at lower latencies with the Steinberg app's on the Dual Quads for over 16 months.

The catalyst for the performance improvements ended up being a microcode update delivered within a major BIOS revision for the i5000x chipset, and from numerous reports have also been maintained in the current i5400 chipset. It isn't entirely clear what the specifics of the microcode updates where, suffice to say they had substantially improved arbitration and synchronization within various subsystems.

This meant that at least part of the solution to the x-scaling issues were in fact architectural, not purely software related as Steinberg had claimed.

This inevitably opened a can of worms, as Steinberg had officially acknowledged that the 2 issues that were highlighted and collectively dubbed " The X-Scaling Bug" i.e. accumulative resource loading with high I/O -caused by serialization of the ASIO driver, and the higher than normal diminishing return at lower latencies -caused by synchronization delays , where in fact software related at their end.


The accumulative resource loading with high I/O had already been resolved on the initial release of Nuendo 4.1 , without any official acknowledgment from Steinberg, which seemed a bit odd at the time.

When queried, it was explained by one of the moderators at Nuendo.com, that because the 2nd part of the bug was not resolved, Steinberg still considered the issue open.

The recent improvements with the scalability at the extreme lower latencies, at least in this series of tests, has gone a long way to resolving some of the remaining issues associated with the second part, which are the synchronization delays, well at least on the PC Platform. OSX is still not performing to the same extent, which suggests that there are even deeper software hooks to navigate there.

It is not by any means conclusive that the x-scaling bug has been quashed on XP/Vista , debate is still raging amongst end users and DAW builders whether the improvement within this testing methodology in fact constitute a fix. Steinberg in their usual manner, have totally ignored any and all efforts for communication or clarification on the subject.


Well that last line is not entirely correct, they did recently release an updated article in their Knowledge Base, stating that the performance issues on 8 Cores systems were still to be addressed - this was over 3 weeks after the performance improvements were disclosed.

The KB article seemed a bit odd in light of the quantum performance improvements , but without any official word from Steinberg in regards to the recent findings, we are again left to speculate on whether the x-scaling issues that they had acknowledged last year, are in fact resolved at their end.

This lack of communication is an ongoing and increasingly frustrating trend from Steinberg, there simply comes a point where you just need to throw your arms up, shake your head, and move on.

The above graphs highlight the x-scaling performance of not only Nuendo but also SONAR from 4-8 cores Pre and Post Fix using the earlier application specific benchmarks.

I have also included the results for the newer Blofelds DSP II benchmark which is using native VST3 plugins in Nuendo 4, again showing that the performance improvements have been maintained on the newer test session

Vin Curigliano
AAVIM Technology
April 9 2008

Part I



© AAVIMT 2008