ACE : Aural Computing Engine™
Serve as catalyst for next-generation of music and sound experiences
Analogue tools, synthesizers, acoustic spaces and real instruments consist of complex frequency-dependent nonlinearities.
To implement those features in a discrete-time system it is necessary to deploy advanced nonlinear differential equations and mathematical transfer functions. For accurate simulation of nonlinearities in a virtual environment, the output of those equations may have significantly more bandwidth than the input signal. Until today, the fidelity of those techniques was limited by computational resources. It is time to open the doors of unlimited creativity.
Aural Computing Engine give us the necessary means to work with thousands of tracks, effect, virtual instruments with unlimited polyphony in real-time and leading low-latency performance. Software developers can now implement highly advanced algorithms that computing bottlenecks would not permit. ACE is not just about bringing analog qualities back inside the digital domain but more importantly is about creating new tools that neither analog nor the underpowered Digital 1.0 can ever enable. HPCmusic technologies achieved this by reinventing the compute platform and the design methodologies of today.
ACE is a new platform that ignites Digital 2.0 era for music creation and the audience experience. We have overcome all barriers of underpowered Digital 1.0 discrete-time systems. We compute without fighting with conventional CPU or DSP constraints. There is no way you can overload an HPC music production system when we work with 96kHz, 192kHz, 384kHz or 1.5MHz by using any amount of tracks, effects and virtual instruments at any buffer size. Moreover, ACE allows us to incorporate different sound qualities in the same project so we can push hard when we want to simulate analog synths or luscious reverbs. We are now ready for accurate solid-state and thermionic circuitry simulations.
ACE is not only about music production. Even more importantly is a reproduction engine capable of delivering the audience experiences of the future. Intel and HPCmusic demonstrated Digital 2.0 and few of those experiences at SC15. At this critical juncture of entertainment evolution, with 3D Cinema, 6K, 4K TV with HDMI 2 and In-Vehicle Infotainment, the industry creates a roadmap for the next music revolution. A true quality upgrade of the overall music experience is on-going. ACE is here to keep music production on par with those innovations and it will provide the necessary tools, specifications and revolutionary technologies so that music professionals will be able to produce and deliver the next generation of content to meet the demands and exceed expectations of the audience.
High-Performance Computing Audio Networking Protocol
HPC384 Spec. is a multi-layer non-blocking fat tree proprietary audio network topology based on Infiniband and Omni-Path interconnects. It was developed between 2012 and 2015 by HPCmusic in collaboration with Mellanox Technologies and Intel Corporation.
It supports an unlimited amount of channels and at ultra-low latency and tested at bandwidth up to 6.1MHz. The spec. was designed to support the deployment of the Aural Computing Engine.
1 Gigabit - 96KHz Stereo
100 Gigabit - 768KHz Multi Channel
Lecture about transient response
High Performance Computing at 384kHz
By using real time sample-rate as a performance index, we measure how BIG is the amount of DATA we are able to process in a very low latency (microseconds). From our point of view, extreme sample-rates have nothing to do with frequency responce that we inherently get in advanced computing platforms. Aural Computing Engine allow us fully-blown component based simulations of analog gear, true physical modeling of real instruments with unlimited polyphony and true room acoustics simulations.
Tests and Measurements - The sound differences to date
On another test, we render the first ever reverb at 1.5MHz using U-He Zebra 2 (HS Eternal Chimes) clocked at 3.1MHz as our sound generator. This sound is quite likely the most mathematically complex and harmonically rich single sound ever created in the digital domain. More importantly, it was created using Native VST instrument and effect technologies.
We carefully measure latency and network overload. As of December 2016, 7.1 x 1536kHz is fully operational on HEXE Desktop Supercomputer . Currently, we conduct tests of digital sound production up to 6.1MHz/24bit PCM.
Software developers will gradually take advantage of the massive computational resources of HPC for music systems therefore the sound quality gain of the architecture will increase exponentially.
384kHz downsampled to 44.1kHz
Spectral Analysis of 384kHz at 192kHz inside Adobe Audition CS6
What do high-performance computing for music production actually mean for music professionals?
1. Composers, songwriters and sound engineers gain access to HPC4D sound and a new palette of possibilities to work with.
1a. When creating music in the analog, mechanical and early digital world, you’re limited by the primitive rules, the bottlenecks and system architectures of those worlds. We’re at the beginning of something that’s not bound by those rules.
1b. An unlimited amount of computational resources translates into an infinite amount of tracks/instruments for professionals to shape music, as well as an infinite amount of effects to shape sound. We’re looking at a common set of tools to enable people to create new music technologies and new user experiences.
1c. Offline Rendering speed of complex algorithms is now as fast as real-time.
1d. Advanced physics modeling, analog circuit simulators, detailed simulating of non-linear / non-stationary characteristics.
2. This is the end of the well-known loudness war, which degrades the sound quality of modern music.
For more than a decade, mastering engineers have been 'forced' to compete in the 'average loudness' of music due to lack of other sound ways of differentiating their products and deliver a better music experience. As a result, every new hit record has less dynamic range than the previous one. By pushing the sound capabilities beyond analog, engineers and producers will care less about loudness and more about spatial sound information, texture, timbre and other sound quality that deeply contribute the music experience.
3. Advanced Digital 2.0 sound processing enables new sound and music possibilities. This is akin to photo processing, in which the processing of the large raw file looks better than processing the jpg. Low-quality processing creates digital artifacts like aliasing.
4. High quality algorithmic multi channel audio allow mixing engineers and producers to utilize true surround mixing and deliver a revolutionary audience experience (see HPC4D).
5. In a movie world that innovates with 3D, 8K and HDR picture technologies, sound and music will eventually find its place on par with those quality developments and help on shaping new kind of magic that will usher the cinematic experience into the future.
Low Latency Supercomputing for Music
Intel SSF coupled with HPCmusic ACE delivers a real-time platform for virtual instruments and effects up to 1.5MHz. Omni-Path interconnect enables us to implement adaptive-intelligence load balancing so that we can use of advanced DSP algorithms in real-time music production.
Cost per GFLOPS
HPC for music cost per GFLOPS is roughly 35x better than AVID (NASDAQ:AVID) Protools HDX DSP-accelerated solution.
Steinberg GmbH underlines that "When choosing a processor, please be aware that floating point operations are crusial for audio performance". Here is updated look of FLOPS (FLoating-point Operations Per Second) on processors and systems.
The degree to which the processing can add capacity without disruption and without incurring excessive overhead (nonproductive processing) is largely determined by the scalability of the particular computing platform. There are many solutions that try to deliver scalable computational power by using dedicated DSP. The two most popular ones come from Universal Audio and Avid Technologies. One Universal Audio UAD2 SHARC DSP (Analog Devices ADSP-2136) delivers 2.4 GFLOPS and their flagship PCIe DSP card puts together 8 of them for $1500 (19 GFLOPS). An Avid Protools PCIe HDX system delivers 38 GFLOPS and the entry level HDX DSP PCIe card (18 Ti C672x Chips running at 350MHz ) costs approximately $7500.
A Universal Audio DSP system that operates up to 192kHz has a cost of $78 per GFLOPS
An Avid Protools HDX System that operates up to 192kHz has a cost of $197 per GFLOPS
An entry level HPCmusic system based on Aural Computing Engine, can operate up 1536kHz real-time and costs $1.9 per GFLOPS
Even by rough calculations and without taking into account the economies of scale that HPC systems can deliver, value for money of HPC for Music™ is at least 100x than Avid Protools HDX card.
Dynamic resource allocation and adaptive intelligence load balancing allow more than 90% CPU utilization on our HPC systems.
HPC384 systems for music production can scale up to a huge supercomputer but they do not necessarily have to be huge systems inside machine rooms. More form factors soon... ;)
A brief Bibliography
Technion - Israel Institute of Technology
Performance analysis of dual source transfer-function generalized sidelobe canceller, Jan 2006
A. Askenfelt, ed., Five Lectures on the Acoustics of the Piano, Stockholm: Royal Swedish
Academy of Music, lectures by H. A. Conklin, Anders Askenfelt and E. Jansson, D.
E. Hall, G.Weinreich, and K.Wogram, Oct 1990
S. Bilbao, Wave and Scattering Methods for the Numerical Integration of Partial Differential Equations,
PhD thesis, Stanford University, June 2001
Technion - Israel Institute of Technology
In-kernel Integration of Operating Systems and InfiniBand Primitives for High-Performance Computing Clusters, Sep 2005
S. Bilbao, “Time-varying generalizations of all-pass filters,” IEEE Signal Processing Letters, May 2005
AES - Audio Engineering Society
O. J. Bonello, “Modular parametric equalizer-filter, a new way to synthesize the frequency
response,” Audio Engineering Society Convention, Preprint 1170, Oct. 1976
Electronics Letters Vol27
P. Bowron, M. R. J. Motlagh, and A. A. Muhieddine, “Harmonic characterisation of feedback
systems incorporating saturation nonlinearities, Oct 1991
Tokyo Institute of Technology
PHoM — a polyhedral homotopy continuation method for polynomial systems, Dec 2002
IEEE Cluster 2013 Conference
Influence of InfiniBand FDR on the Performance of Remote GPU Virtualization, Sep 2013
IEEE 18th International Conference on Parallel and Distributed Systems
Comparing the Performance of Blue Gene/Q with Leading Cray XE6 and InfiniBand Systems, Dec 2012
IEEE Cluster 2013 Conference
Distributed Resource Exchange: Virtualized Resource Management for SR-IOV InfiniBand Clusters, Sep 2013