Parallella: A Supercomputer For Everyone on Kickstarter

http://www.kickstarter.com/projects/adapteva/parallella-a-supercomputer-for-everyone
If the Kickstarter video causes your mouse to disappear, click off screen and it should come back.

Note the video and project is not directly focused on synthesis, however the application (like the iPad as a synthesis platform) should be obvious. The following is a write-up on that potential in via reader cheater:

"I'm in no way affiliated with those guys, but this thing opens so many possibilities for the synthesis aficionado that I have to let you know what's in store. Why is this cool? Simply put, it's a Kickstarter for an open platform which could enable synthesis methods that are currently not possible, unless you lug around an expensive workstation, and put them in a desktop-module style case. For those intents and purposes it's a self-contained system with a 45 GHz CPU, the size of a DVD box.

Below are some examples of synthesis methods that could work out thanks to the radically different way this thing works. Each of those is slow on current PCs and because of the parallel nature of the chip in question are tons faster to execute on it:

1. granular synthesis. Ever try to run one of those in VST? They sound better the more parts you can have. Current PCs can only run so many parts, and you need thousands per voice. This thing could easily run a polyphonic synthesizer.

Here are some demos of what you can do with a granular synthesizer:
https://www.youtube.com/watch?v=jgcVr6lTzDs
http://www2.camelaudio.com/music/AlchemyAnalysisDemo02.mp3
http://www2.camelaudio.com/music/Alchemy-PresetDemo-JohnLehmkuhl.mp3
https://www.youtube.com/watch?v=hTODhn-UROc

See how Alchemy only allows 10 grains per voice? You could have an order of magnitude more, with each grain bank (Alchemy has 10) modulated separately.

More:

Steinberg Padshop: https://www.youtube.com/watch?v=jgcVr6lTzDs
First steps in Granular Synthesis with synthesis pioneer Curtis Roads: http://www.youtube.com/watch?v=ajdRGF5NHIs

If anyone of you knows the harmonizer, chorus and resynthesis sounds produced by the Eventide hardware audio processors, this is the way they did a lot of the cool things people buy those boxes for. I don't know the exact specs of the H8000, but based on when it came out an educated guess says the Parallella is several hundreds of H8000's in a box. Some demos: http://www.sweetwater.com/store/detail/H8000FW


2. Additive synthesis.

Current additive synthesizers allow a fair amount of partials. For example, Alchemy allows 600 sine waves per voice. You could have an order of magnitude more, and each sine wave could be separately modulated. What does this mean? You start off with a life-like sound and get controls that do crazy stuff to it, such as harmonic stretching.

Sure, you can do "additive synthesis" already. But what we have right now is child's play compared to what it can be. This project could take it to the next level.


3. Ambiance synthesis.

This isn't even a synthesis method yet because it's so computationally inefficient on current hardware. It's sort of an upgrade from granular or additive synthesis. Basically, you take your sound's partials (be it decomposed to sine waves via additive resynthesis or grains via granular synthesis) and you place each in 3D space. Then, you render it via convolution (modulate loudness and add reverb / eq / filtering) and finally via a good HRTF model so that the stereo field is represented well. This creates pads like you've never heard before. Explained simply via examples? Imagine you're walking down a rainy street, and each time a raindrop hits the ground it sends out a note from a simple monosynth like the Minimoog. Thousands upon thousands of raindrops. Or take a violin sound, and place each partial somewhere else in space. Kind of like candles around a dark room. Then, have objects move around this 3D space and occlude the partials, kind of like stopping the light of each candle separately from reaching your eyes. And then have them move around, complete with the doppler effect and raytraced reverb.

The well known THX sound is an example which could be realized using this method, and barely uses any of the features I mention. For example they didn't have HRTF in 1982: https://www.youtube.com/watch?v=FWkJ86JqlPA

4. Volterra kernel synthesis.

Everyone knows convolution: reverb, EQs, filters, etc. Convolution is linear, this means it cannot model what gives sounds grit and character: the nonlinearities. Volterra kernels are a direct upgrade from convolution and they rely completely on great amounts of matrix multiplications.

Here's how a prototype Parallella handles matrix multiplication compared to a desktop PC:

<iframe width="640" height="480" src="http://www.youtube.com/embed/4sMWbaV1sRQ" frameborder="0" allowfullscreen></iframe>

It basically kicks the PC's ass, a chip which could cost a couple bucks. It's going to get an order of magnitude faster, as they will go from 16 cores spread across several dies to one chip with 64 cores. All synthesis methods I describe rely heavily on matrix multiplication, but especially this, and for the price of a DAW you could buy a mass-produced system that has literally hundreds of those Parallella chips on it.

Volterra kernels are behind the notorious Acoustica Audio Nebula 1/2/3 sound processor. It's notorious for eating CPUs.

So how does this sound? Here's a guy using it for mastering:
http://www.youtube.com/watch?v=ir97ETQCslc

However I assure you it can do a lot more stuff. Check for yourself here: http://www.acustica-audio.com/index.php?option=com_content&view=article&id=23&Itemid=228 - you will be able to hear how cool it sounds and see your CPU go up in flames. The Parallella could eat it for breakfast.

Lots of demos from CDSoundMaster, who makes presets for Nebula below. You can really hear the difference, bear in mind none of the hardware below has been component-modeled, it's all impulse responses from actual hardware:

Pultec EQ demo (check out the sound coloring..):
http://monumentalaudio.com/mac-pc/COOLTEC-EQP1A3S/EQP1A3S-AUDIO-DEMO.zip

Tape recorder demos (nice soft saturation.. something convolution can't do):
http://cdsoundmaster.com/r2r_mp3s/SharonVocal1.mp3
http://cdsoundmaster.com/r2r_mp3s/string1a.mp3
http://cdsoundmaster.com/r2r_mp3s/Drum1.mp3
http://cdsoundmaster.com/r2r_mp3s/windchimes.mp3
http://cdsoundmaster.com/r2r_mp3s/hidekey.mp3

Tube boost demos (listen to the fuzz.. That's what Volterra kernels can and convolution can't):
http://monumentalaudio.com/ACQUA/VTCTB/BEFORE-VTC-TB.mp3
http://monumentalaudio.com/ACQUA/VTCTB/AFTER-VTC-TB.mp3
http://monumentalaudio.com/VTC-TUBE-BOOSTER/VALVES-OF-VICTORY/VALVES-OF-VICTORY-DRY.mp3
http://monumentalaudio.com/VTC-TUBE-BOOSTER/VALVES-OF-VICTORY/VALVES-OF-VICTORY-MULLARD.mp3
http://monumentalaudio.com/VTC-TUBE-BOOSTER/VALVES-OF-VICTORY/VALVES-OF-VICTORY-TUBE-BOOSTER.mp3
http://monumentalaudio.com/VTC-TUBE-BOOSTER/TUBE-BOOSTER-AUDIO-SAMPLES/BASS1-BEFORE-TUBEINSTANCE1-2-3-4-5.mp3
http://monumentalaudio.com/VTC-TUBE-BOOSTER/TUBE-BOOSTER-AUDIO-SAMPLES/DRUM1-BEFORE-TUBEINSTANCE1-2-3.mp3
http://monumentalaudio.com/VTC-TUBE-BOOSTER/TUBE-BOOSTER-AUDIO-SAMPLES/DRUM-AND-BASS-BEFORE-TUBEINSTANCE1-2.mp3
http://monumentalaudio.com/VTC-TUBE-BOOSTER/TUBE-BOOSTER-AUDIO-SAMPLES/GUITAR1-BEFORE-TUBE1INSTANCE-2-3.mp3
http://monumentalaudio.com/VTC-TUBE-BOOSTER/TUBE-BOOSTER-AUDIO-SAMPLES/SCORE1-BEFORE-TUBE1INSTANCE-TUBE2INSTANCE.mp3

Compressor demos (that's right, Volterra kernels can emulate compressors):
http://monumentalaudio.com/TDC/TDC-DEMOS-1/TDC-Demo-1.zip

5. Each of those CPUs can easily run several VA voices. You could have hundreds of voices on one of those. Need more hundreds of voices? Buy another $99 box and hook it up to your PC digitally or use it standalone.

So how is this different from all other hardware synth platforms you've seen before, all of which were supposed to be open-ended but ultimately failed to deliver and died off? Let's make a quick comparison between the Parallella and those other things. Here we have the Korg OASYS, the Soundart Chameleon, Nord Modular (haven't seen updates to that platform in a while), Creamware, Protools, Eventide, many different homebrew FPGA synths, etc:

A. First I'll mention points of failure common to all projects:

A1. They all focused on synths only. Parallella is a multipurpose computer with tons of applications in many industries. It's certain to find popular use. Synthesis will be a small portion of it. Let's face it, synthesizers *are* a fringe market. The VST ecosystem boomed because it uses mainstream hardware that has lots of uses. The Parallella is similar: open platform (just like the Intel x86 platform which VST runs on), open documentation, cheap (a chip could cost a couple bucks in bulk). Trying to leverage synthesizers as your main market inevitably fails because there just aren't so many people interested in that, and every synth manufacturer will tell you that. If moog modulars were as common as laptops they'd cost $500 a pop but they don't.

A2. Not enough openness on the software side. All the software was tied to the vendor. The software was supplied by the vendor or the SDK was and it cost loads. This is a computer which runs a Ubuntu Linux desktop. The compiler is free, you get a full toolchain. You can write your own programs for it. Could you write your own programs for the OASYS? No.

A3. Not enough openness on the hardware side. Could you build your own Nord Modular? The chips were available but the firmware wasn't. Or your own Korg OASYS? You couldn't do that either. You'll be able to buy Parallella chips from Aapteva for several dollars a pop and build your own system. And in case Adapteva blows up, the specification is open so anyone would be able to come in and build a Parallella-compatible chip if they wanted to. Kind of like the Tron 2 quote (amazing analog synth score)... "remove yourself from the equation". Well, Adapteva did just that. Here's what many people think of Korg: http://www.youtube.com/watch?v=eGundKbInZI A4. Too expensive. Because you're only selling to a select public (audio engineers or synth musicians) you only sell so many units. Economy of scale dictates that something like the OASYS has to cost $5000, while a comparably complex Parallella can go for $99 and offer much better things.

A5. Not hackable. Too difficult to write software for. With Parallella you get a fully working computer which you can hook up to a monitor and start messing about with programming it right away. You get Python support

A6. Always leveraged closed-platform solutions from external vendors. Why did the Nord Lead 3 stop production? The chip vendor stopped making the chip. Why are there no updates to Creamware, Protools, Eventide? Chip vendors stopped making faster and faster versions of the chip family and instead introduced new, faster, but incompatible chips. This platform is open from the very beginning, and it's decreed so by the very people who came up with the chips and are building them in a plant.

B. Now some specific things:

B1. Korg OASYS - killed by the whim of a hardware giant who wasn't really that interested in making it in the first place.

B2. Soundart Chameleon - fringe audience, cost too much for what it gave us, difficult to program, quickly outdated hardware, fell into obscurity quickly

B3. Nord Modular - tied to the low-level platform it was on. Motorola DSP chips galore. You program the Parallella in portable high-level programming languages, which means programs can be fairly easily recompiled for newer iterations of the platform. Same with Creamware, Protools, Eventide.

B4. FPGA synths: Usually based off evaluation boards that cost a lot (several hundred dollars) and use toolkits that are very complex. Plus, programming an FPGA is like nothing you've done before, and I say that as a computer programmer. FPGAs have their spotlight because, like the Parallella, they allow highly parallel computation. With platforms like CUDA and now the Parallella you will see a steady decrease of those kinds of projects.

So you can see the Adapteva Parallella isn't a dead end like those other synth platforms.

But why is it suited for audio? Some points:
1. Linux, unlike Windows, supports realtime operation. This means as a stand-alone box the Parallella could have tiny delays of microseconds, not milliseconds. It has audio integrated, so you can use that as a desktop output. Or get out a soldering iron and put some 1/4" jacks on the thing.
2. However, because it's an open OS, you get driver support. So you can plug in a usb midi adapter and it works right away. So do many audio cards and many other peripherals.
3. And, since the thing uses only a couple Watt energy, you can rack it and it won't have any annoying coolers.
4. It's small so it's good on-the-go. Use it with your iPhone or iPad.
5. Ever wished you could hook up your desktop synth to a big LCD display? This has HDMI output. Or use DisplayLink for even more HDMI outputs via USB.
6. It can be hooked up to Ethernet. Have your DAW send audio to it for processing via a special VST plugin. If you want to process more tracks shell out another $99. Doesn't use up PCI or PCI-e slots. Ethernet has been compatible with standards since the late 60s, so you're not buying into a dead-end.
7. Once it gets mainstream adoption, it's likely it'll get more and more new versions. As a side-effect, old versions will be even cheaper second-hand. Example: you can get a fairly recent nVidia GeForce 8800 for $5 on eBay. By comparison, old digital hardware for the sound market can double in value as opposed to becoming 25x cheaper than when it launched. So, as you go on and your needs grow, you'll be able to expand more and more cheaply.
8. Uses so little power that it can be battery-operated. Hook it up to an audio interface or use the existing audio and use it as a modern-day Nagra.
9. Wait and see how CSound, Puredata, and Processing get ported to the platform to leverage the new processing capabilities.

What other uses do you have for this? Once you buy this thing there's already a lot of Linux audio apps (some work on Arm, some don't, but it's getting better). You know things like Ardour, CSound and Pure Data are great fun. Plus, it's a good general purpose PC for browsing the 'net, and you get it for $99.. that's nothing... And it needs no fans, so it's super-quiet (I know that I am constantly annoyed by the fans whirring in my laptop).

The possibilities really are endless, and if the hardware platform lifts off I can easily see loads of synth projects springing up for it. Instead of buying yet another Rompler, pledge $99 here (you only actually pay if the project lifts off), and you could help make things really happen: http://www.kickstarter.com/projects/adapteva/parallella-a-supercomputer-for-everyone

I hope I got some of you guys interested.

Cheers,
D."