Fahlens Corner

Users' Stress Test Reports

The following is a compilation of audio stress test results with the first breed of single cpu nForce4 chipset platforms (NF4) featuring PCI Express and/or SLI video card slots. The reports originate from DAW users who have run several audio stress test files deliberately designed to bring the systems to the limits and find out if there is any overhead left at high I/O bandwidth/cpu/memory/low latency loads. This is viewed as a relevant test approach to meet pro DAW users' demand, where just one click/pop during a recording session may be detrimental.

The observed performance issue - increased cpu resource usage and/or necessity to use significantly higher latency settings to eliminate pops/clicks with NF4/PCI-E/SLI, and to a lesser degree, Intel 925/915/PCI-E - is detected by users with several DAW softwares, such as Cubase SX, Nuendo 2/3, Samplitude 8, Sound Forge, and Sonar 4. The tests have been done with specific audio stress test files which strive to image the maximum possible glitch-free resource load following various DAW workflows (restricted to, or a combination of, number of streaming audio channels, sample playback, DSP effects and VSTi). The specific audio stress test files used and cpu load/latency setting results, etc.. can be found at on-going threads such as:

Nuendo-Forum

A limited number of audio stress test results with the Intel 915 PCI-E chipset , Layla 24 soundcard and Intel Prescott 3.2 GHz/1 MB @ L2 cache/2 MB @ L2 cache can be found here:

Nuendo-Forum

With the Thonex audio stress test file (downloadable from http://www.adkproaudio.com/downloads.cfm), including memory-intensive data communication via samples/VSTis, and used with Cubase SX/Nuendo 2/3, soundcard latency has to be increased to approx. 2048 samples buffer latency setting with the NF4 to receive glitch-free audio recordings whereas with the NF3, and equivalent software/hardware/soundware equipment, the minimum latency can be significantly reduced (some 128-256 samples). Likewise, cpu load values are significantly higher with the NF4 than with the NF3 ditto - based on exactly the same audio stress test files and equivalent hardware peripherals.

The following NF4 mainboards, PCI/PCI-E/SLI video cards, and audio sound cards have been tested:

Mainboards: Gigabyte GA-K8NF-9, Gigabyte K8NXP-SLI, Asus K8N-SLI

Video cards: Matrox RTX100, ATI X100, ATI X300, ATI X800, ATI 6800

Soundcards: Tascam 1884 (Firewire), RME MADI (PCI), RME Fireface800 (Firewire)

Based on DAW stress test results all combinations have been found to perform significantly poorer than the NF3/Intel 875 AGP ditto - with similar soundcards/drivers.

A comparative reading of audio stress test results from the Intel 915/925 and the NF4 platforms also shows that the performance penalty is less in combination with low-performing PCI-E video cards but still significantly reduced than compared with the NF3/Intel 875-AGP ditto. Interestingly, equivalent low-performing PCI-E video cards tested on the Intel 915 platform have resulted in less performance penalty and in some cases are almost on par with audio performance on the well-proven Intel 875-AGP platform (the NF3-Socket 939 has an edge over the Intel 875 in terms of number of VSTi/samples simultaneously possibly engaged and minimum latency settings). It has so far not been of any help to try latency setting tools (Powerstrip, WPCREDIT) to rectify the problem on the NF4 platform (PCI-E devices show up as 'PCI' with 'bus mastering' features).

It thus so far seems as if the current breed of PCI-E/SLI bus/driver implementation creates artifacts which hog resources off the PCI-E bus as both the Firewire and PCI soundcards communicate and stream audio data via the PCI bus on the PCI-E/SLI equipped mainboards) - seemingly the more demanding PCI-E video cards cause more performance penalty on audio streamed via the PCI bus even when only used in 2D video modes.

It also appears as if the memory-cpu performance is presently reduced on the NF4 platform, causing significantly higher cpu load values on memory-intensive audio stress test files than with the NF3 ditto.

It is interesting that the Intel PCI-E platforms with - on paper - slower interconnectivity between the chipset subsystems currently seem to have less problems in cases with demanding PCI/Firewire audio plus PCI-Express I/O data transfer. Based on compiled user reports the NF4 currently appears to get more bogged down and performance reduced - including memory/chipset subsystems - in cases where there is a time critical competition between PCI-E and PCI bus resources.

According to this article (http://www.theinquirer.net/?article=19313) the Nvidia nForce4 A02 chipset revision is possibly limited to 4x Hypertransport capacity (800 MHz HT), whereas the nForce3 250 chipset performs at full HT speed, and the NF4 A02 revision is identified to have some issues with PCI Express. The article states that the identified issues will be fixed with the A03 chipset revision. This link suggests that the NF4 A03 silicone started to ship in mid-November (2004) and that the A02 revision was aimed towards the AMD 64 @ socket 754 platform (http://www.nforcershq.com/modules.php?name=News&new_topic=45). It is not confirmed whether or not this is related to present audio stress test results.

Verdict

At present the NF4 single cpu chipset cannot be recommended for demanding pro DAW work. Multimedia users, recording hobbyists or semi-pro DAW users who use a limited combination of streamed audio tracks, samples and/or VSTis may not face the aforementioned limitations and performance may be as good as on any comparative platform.

The single cpu nForce3 @ socket 939 (AMD 64 cpu) and Intel 875 (Pentium IV cpu) chipsets are well-proven under different DAW workflows and match well with RMEs soundcards for pro audio usage.

DAW users who want to invest in single cpu/PCI-Express/SLI-equipped motherboards are currently on their own territory and glitch-free DAW performance under heavy load cannot be guaranteed. If used - it is currently recommended to try with low-performing video cards (such as ATI Radeon X300 or Nvidia 6200), in combination with PCI-E equipped mainboards as to minimize the risks of PCI-PCI Express performance penalty in time-critical DAW-PCI data transfer events.

Users interested in DAW performance development of NF4/PCI-E equipped mainboards are encouraged to download DAW stress test files, such as Thonex (Cubase SX/Nuendo; http://www.adkproaudio.com/downloads.cfm) or the Orangine (Cubase SX/Nuendo; http://www.citriq.com/sonomatica/sistemas/Orangine-Stress%20Test%20SX.rar), follow the adjoined instructions, and report their detailed hardware configuration and test results. Readers are further encouraged to follow future and updated reports via subject matter-related threads at major DAW forums (including the RME forum).

Acknowledgement

A big thank you goes to ADK (http://www.adkproaudio.com) and individual users who generously have shared audio stress test results with the community via various public DAW forum.

Disclaimer

The author/RME cannot be held responsible for any eventual damages as a consequence of using the information found in this document. The author has done his best to cross-check and verify information presented in this document but there is no liability to guarantee against any eventual damages caused by the user. The user and reader are fully responsible for any eventual damages caused by implementing or misusing information in this document. This information is provided 'as is' and may be subject to change if audio stress test results with the NF4 single cpu chipset show verifiable signs of improvements.

Any imperfections left in the document are solely the responsibility of the author. Readers who may want to get in dialogue with the author (e.g. to suggest improvements, corrections, etc). should make contact at www.agronova.se/music.html.

March 9, 2005

Anders Fahlén, Agronova, Sweden