We request a performance test to a big DAW developer, We hope have some result into 4 week. Maybe be a Dante PCIe LAN NIC will be a solution for us. I already Netgear network product hardware beta test, depending the results I request a specifict product for us that will compatible with Waves NetGrid and Vienna Library.
Enrique Silveti. Acustica Audio customer and technical support.
Some subjective test feedback. My two machines are identical. Not cutting edge anynore but very powerful all the same.
All my work is at 96 khz so I am pushing the limits. Systems Windows 7 64 bit versions 3 Gigabit lan ( 3 ports teamed by using Link Aggregation through an HP procurve 1810G-8) 3 Ghz intel Xeon workstations (Cloverton CPU) 16 GBs RAM each Server version has FX teleport as well.
Loaded up a song in Reaper and began adding instances.
I struggled with the nebula server version a little to be honest. It was eating a lot of local cpu and had less efficiency than FX teleport.It protested at 3 R2Rs. I think there may be a core loading balancing issue actually. One Core takes up too much of a load until it reaches near 100% and then the computer crackles and slows down and gives up. If one instance is over 50 % then if I add another it runs on a second core but if I add a third it goes back to the first core and overloads it. However It is cool that the controls are available on the local host machine but the efficiency was not as good as FX teleport sadly because of that core bug.
With FX teleport I was able to run 9 R2R tapes with no audible dropouts. Server CPU usage was about 90%. When I added one more it peaked at 100% and crackles appeared. Let me say I had NO crackles at all until then. There were however samples lost along the way as reported by the program but nothing audible. Local cpu usage was about 4 percent fully loaded.
I think the remote CPU usage is the same except for the bug.
So far I am pleased with being able to do all of this remotely and will be using both the server version and FX telport depending on my needs but there is a small core loading bug ( It may just be my system) that makes the server less usable and efficient. I will update when VSL ensemble pro 5 is released.
The next step!!!!! Thinking out loud here. . If money were not an issue ( and of course it is) Have you guys thought of comissioning someone to make a DSP card just for running Nebula? I mean working at 96 khz means mandatory bounce downs at the moment and no matter how powerful cpus become, the 11 kernel programs will eat them for breakfast. Workflow killer right there. Not the end of the world but not great. Cuda seems iffy at this stage unless you have got it running smoothly. But a card built from the ground up just to run Nebula and nothing else.........hmmmmm. Interesting.
This idea may not be popular with others who hate hardware as a dongle but it is an area worth investigating I think. I know you are a small company and I know this sounds silly in this day and age with really fast computers but even they have their limits. Imagine being able to have a DSP card that has oodles of memory, fast ASIC processors that can crunch the Nebula maths more efficiently than a CPU.
I still love my 4 powercore express cards. They have lots of power and great plugins but sadly very few instances. Only 16 CL1Bs for example.But they place only a fraction of a few percent CPU load, almost zero. But they are pci express X1. There are more powerful solutions out there of course like the soon to come Protools HDX. Now that is expensive but also AWESOME performance that no CPU can give. As an aside, multiple instances of Nebula interferes with my midi transmission under heavy load. Too much latency I guess.
Don't know if you know much about Digital Cinema but for example The RED Rocket Card ( Red Digital Cinema ) can plough through 5K footage in realtime that a modern CPU simply chokes on. Dedicated DSP solutions are still viable because they take the load off the CPU and because they have only one task to perform they are much more efficient as far as I know.
With PCI express cards V2 giving up to 80 Gb/s throughput the sky is the limit for dedicated hardware. I did some casual maths and to run say on every channel a pre-amp, tape, eq , distortion thingy and some console ins and outs on my system I would be able to run 3 or four channels befor the whole thing collapsed. This is a Dual CPU Xeon workstation. Score of 7.5 in the CPU benchmark (top i7 is just above 10)
No-one can realistically afford the top performer anyway ( 33.5 ) and it has three cpus and won't work with modern Windows versions. ( maybe windows server ). But I bet it costs about the price of a good car!!!!
So the top i7 is about thirty percent faster than my system.On my identical server I can run 9 consoles before it dies. That's it!!! So maybe 13 or 14 instances on the fastest affordble systems at 96 khz with the 11 kernel programs? Seems a long way from being able to run a whole console in real time on multiple channels. With my main system AND my server about 18 to 20 instances of the heavy crunchers. That's it!!!
But let's say some hardware guru were able to get cards with say a few gigs of RAM on each and really fast DSP solutions I would pay a few hundred a card to get this thing efficient. I would spend that on servers anyway LOL. Imagine using this mother board with say 6 DSP cards
Lets say each card could process 15 heavy Nebula instances, maybe more of the light ones. Then all of a sudden it becomes a possibilty to do a whole mix with minimal CPU hit. You may even get lower latency and other benefits. That would be the revolution. Even if each card were priced at say 7-800 Euros ( if that will be worth anything soon, LOL) I would buy 6. That would rival buying a console worth many tens of thousands.It could also be an alternative revenue stream for you.
h-man wrote:If money were not an issue ( and of course it is) Have you guys thought of comissioning someone to make a DSP card just for running Nebula?
That's the same thing I asked to Enrique a couple of years ago. Acustica Audio don't like this idea, but I think it would be kick-ass.
At the beginning, I was thinking that having a powerful computer will solve any problem regarding Nebula, but I think i won't be possible for a big while. My hot dream is to have R2R, TB+, console Input, three EQ bands and console output on every channel, just for beginning a new project. I know it's not very realistic to think this way (ok, I admit it: I'm crazy), but sometimes I'm overmixing a little because I'm not having the possibility of summing my tracks in this way.
Some guy told me that in the future would be nice to have a bunch of extremely powerful computers in rack mode, all of them with Nebula Server. Maybe then we will see this. But it's not nearby, that's for sure
I think your "dream" of up to 5-6 nebula instances per channel is more realistic then you think. I mean too, it must be possible to open 6 nebula´s per (maybe 24/32) channel. Freezing is not realy a solution for the future. Mixing is intuitiv. Nebula is a top product! Top sounding! What is to do for a project (for example) with 6 nebula´s in 24 channels? Yes I know, 144 or more nebula´s are a little unrealistic but in idea most normal. The idea of h-man with a own dsp-card is a good idea. Maybe a great cuda usage with a pure-cuda-nebula-version and we have a "dsp"-card for nebula. Cuda cards are buyable in some sizes and the next CPU´s will bring up an own GPU support. Yes, maybe we are all dreamer but for me is nebula a main product. Please Giancarlo push it forward. Whats the situation with a pure cuda version at the moment?