2CAudio Kaleidoscope | It's A Trip | Latest Update 1.1

[Andrew Souter]

I must be doing something wrong, setting max res to 512 didn't appreciably change the performance. I'm still getting 40+ tracks, although it's a bit spikier than at 64.

yes, your initial test is invalid if you have max resonators to 64, you are only processing 64 resonators not 512 as intended, so you are cheating by a factor of 8!

also to change the result properly, delete ALL tracks except one. Then set Max resonators to 512, then go to the main page of the GUI and move the slider up to 512, or reload the preset. Now you will see the number 512 in white on the right side of the GUI, yes?

Now, duplicate that track, and watch your CPU cry...

If you can get 6 I will be very surprised. Prob 4 to 5 I guess...

Simon's result of 16 (on a lucky day when the stars align) is about the best I have seen with the build you guys have and he and I have the same hardware.

[SJ_Digriz]

Just curious what CPU you have? I've been holding off updating for a long time because my setup is damn stable. However, it looks like performance has finally jumped a bit and I'm seriously considering one of the Skylake processors, like a 6700.

[Sampleconstruct]

Just curious what CPU you have? I've been holding off updating for a long time because my setup is damn stable. However, it looks like performance has finally jumped a bit and I'm seriously considering one of the Skylake processors, like a 6700.

MacPro (latest ashtray), 2.7 Ghz, 12 cores, 32 GB RAM - Andrew has the same.

[SJ_Digriz]

are those the E5 v2 2697s?

Edit: Sorry .. shouldn't derail the thread. Carry on with killing our CPUs

[Sampleconstruct]

are those the E5 v2 2697s?

Edit: Sorry .. shouldn't derail the thread. Carry on with killing our CPUs

2,7 GHz 12-Core Intel Xeon E5 with 30 MB L3-Cache.

[Andrew Souter]

Yes, this is the max Ivy-Bridge generation Xeon E5 2600 series, single socket. "2697 v2":

http://ark.intel.com/products/75283/Int ... e-2_70-GHz

There are actually 18 Core Haswell Xeon E5s out there in the wild already also for the last 12-18 months already...

http://ark.intel.com/products/81061/Int ... e-2_30-GHz

22-core Broadwell Xeons E5s are just about ready to ship also. A theoretical 2016 Mac Pro would likely use such a CPU if released...

On windows it is possible to build a 2 socket machine with 2 or these CPUs... You are talking a VERY expensive computer for such things, but if you work on scoring hollywood movies etc for example, and this is your PRIMARY work tool that you spend 12 hours a day with every day, you can afford a $15k + computer. You are in fact rather silly NOT to IMHO if this is your level of professionalism... PT systems still cost more than that for example... IMHO you should always buy the fastest, most powerful computer you can. This is the central tool of the modern studio. Nothing else will give a better ROI to your productivity etc. Period.

Denis just built a "normal" desktop using Skylake 6700K (4 core) and has pretty good results also. Almost 75% of my 12 core... so price/performance ratio is very good for this CPU if you don't want to get into the crazy cost of Xeons... and/or don't make a living from you music/audio/computer work...

[Andrew Souter]

are those the E5 v2 2697s?

Edit: Sorry .. shouldn't derail the thread. Carry on with killing our CPUs

yes.

more than fine to talk about CPUs/hardware. I like the topic and it is very relevant to products like KS that push the edge of what is possible with the the latest hardware... ask/say whatever you want on this topic...

[SJ_Digriz]

are those the E5 v2 2697s?

Edit: Sorry .. shouldn't derail the thread. Carry on with killing our CPUs

yes.

more than fine to talk about CPUs/hardware. I like the topic and it is very relevant to products like KS that push the edge of what is possible with the the latest hardware... ask/say whatever you want on this topic...

hehe, I have built my own main/work PC about every 2 to 3ish years for a long long time. This one has lasted a little longer for whatever reason. I am an ERP/logistics/finance developer by trade. I record, edit, mix local orchestras, school bands, quartets etc.. for fun. And of course closet guitar/keyboard god is in there somewhere (legend in my own studio). I also just can't help but get all nerded out with distributed processing and such. So, I have a pretty extensive VEP setup with orchestra libraries on multiple servers. Spending 5 or 6k on a build would not be out of the question. I think 15k might take some splainin' to the missus.

By the way I dig the concepts of your programs. Cool twist on normal.

Anyhow, I'm a Windows guy. My next build keeps getting procrastinated but is building internal momentum. I'm truly torn between going silly dual Xeon, or just too expensive regular i7 line. Kaleidoscope and B2 are part of the reason .. so I'm telling my wife it's your fault:D

[lxegs]

with core i7 950 , i can have 6 instances with preset 1

[Andrew Souter]

hehe, I have built my own main/work PC about every 2 to 3ish years for a long long time. This one has lasted a little longer for whatever reason. I am an ERP/logistics/finance developer by trade. I record, edit, mix local orchestras, school bands, quartets etc.. for fun. And of course closet guitar/keyboard god is in there somewhere (legend in my own studio). I also just can't help but get all nerded out with distributed processing and such. So, I have a pretty extensive VEP setup with orchestra libraries on multiple servers. Spending 5 or 6k on a build would not be out of the question. I think 15k might take some splainin' to the missus.

By the way I dig the concepts of your programs. Cool twist on normal.

Anyhow, I'm a Windows guy. My next build keeps getting procrastinated but is building internal momentum. I'm truly torn between going silly dual Xeon, or just too expensive regular i7 line. Kaleidoscope and B2 are part of the reason .. so I'm telling my wife it's your fault:D

The middle ground solution between Xeon and "normal" i7 is of course the "E" CPUs. Like Haswell-E, and soon-to-be Broadwell E. These are about $1k per CPU typically and only allow one per motherboard, and do not require extreme server/workstation motherboards etc. For the $3-5k or or so system range, this is prob the best choice IMHO... At this point, I'd wait for Broadwell E as it should be here soon-ish...

If I were building a new super workstation/server on the win platform, I'd wait for the Skylake Xeons, which are likely 12-18 months away... There is rumored to be a completely new platform for this generation called "Purely" which offers major advancements over the current platform. So I would NOT personally build a dual Broadwell Xeon system for win right now. I'd wait for this.

But if Apple releases as single Broadwell-based 22-core Mac Pro, it'd be a nice jump from the current Mac Pro. For Mac OSX, this would be quite interesting...

[Andrew Souter]

So, I have a pretty extensive VEP setup with orchestra libraries on multiple servers.

a VEP farm with 6700Ks might be the smartest solution at the moment in terms of price/performance... Not entirely sure, but most likely true... Intel definitely inflates Xeon prices since no one else can complete with their raw performance at the moment...

there is that certain magic/satisfaction of just being able to load some massive host project file that has 8 billion heavy VIs an FX plugs all live, all local though, I agree... that is the ideal in the "money is no object" category....

Maybe/hopefully AMD Zen will be competitive again and put some pressure on Xeon pricing?

Or some 64-bit ARM CPU (Qualcom?) Although that would be a nightmare for most of us developers here who have wasted gazillions of hours trying to optimize our code for x86 etc. by writing our stuff in assembly etc...

[krankyone]

Andrew, I guess I'm surprised by this. To be fair, it's now several years since I developed software, so I have to assume a lot has changed. But on some of the hairier computational programs I worked on (image processing, video processing) I didn't do very much assembly work. I tried. When I started as a programmer, this was always a guaranteed way of improving performance. But later in my career compilers had become so efficient that I could rarely get significant improvements by dropping down to the machine level. Believe me, I tried.
Most of the performance improvements eventually came from using better algorithms and writing more efficient higher level code.

[Andrew Souter]

Andrew, I guess I'm surprised by this. To be fair, it's now several years since I developed software, so I have to assume a lot has changed. But on some of the hairier computational programs I worked on (image processing, video processing) I didn't do very much assembly work. I tried. When I started as a programmer, this was always a guaranteed way of improving performance. But later in my career compilers had become so efficient that I could rarely get significant improvements by dropping down to the machine level. Believe me, I tried.
Most of the performance improvements eventually came from using better algorithms and writing more efficient higher level code.

It's actually an interesting topic in real-time DSP. Certainly to get the most out of an intel CPU one has to use SSE/AVX etc. The question then becomes does manual inline assembly code offer any benefit over using various compiler intrinsics? One would hope that the compiler is smart enough to handle it and convert what you want to do to fast vectorized code that is also efficient in its memory usage. This is the ideal. Unfortunately, for a very long time this was not true, and many people found the manual "hand written" assembly in SSE or AVX or whatever ended up faster. Others would use intrinsics and then look as disassembly and try to learn what the compiler did and try to fool it into doing the right thing by editing the code at the C or intrinsics level. This is another approach. As we move forward into the future, compilers become better and better, but sometimes the hand-written approach still gives better results in our experience. This is an active topic of research for us. We do not have all the answers on this, and in fact it would be interesting to make a post in the DSP-Development thread to see what other developers think on these topics in 2016.

Personally I do not code at the assembly level either via use of intrinsics or via manual in-line assembly. I write the ideas/algs/math in pure C and the other half of 2Caudio, Denis, handles assembly and general optimization, i.e. "making things fast". I am not personally an expert on these topics. He is. My job is to know, discover, and design what sounds the best and what is most interesting creatively/aesthetically as a musician/composer/producer/engineer/sound-designer etc. and to solve the math problems involved in achieving this. First we need the right answers. Second we have to figure how to calculate them quickly.

In general however, it is safe to say that in most things we deal with, memory speeds, cache, RAM, etc are still the main limiting factor to performance. Optimizing this stuff for something like KS on a multithreaded basis, very quickly becomes an extremely hairy problem to solve...

We (mostly Denis) have/has spent the majority of the last year reevaluating all this stuff and trying to squeeze the absolute most we can out of x86 (Intel) CPUs. It's an extremely deep, and often maddeningly frustrating topic.

If you were to look at the credits of any major DAW and see a list of 50 or more developers, there are probably a few who study this topic and NOTHING else.... it's incredibly subtle and complicated...

It would be really, really, really, rad if we could simply write C code and be done... that would be like heaven to us!

[krankyone]

This has me nostalgically remembering disassemblers and profiling tools and Knuth and cpu architectures. The only way to really do that kind of work is to be very obsessive. It was a 24/7 kind of thing for me. I'm grateful for that period of my life (made a great living) and also pretty grateful that this period of my life is over. But thank you and Denis for KS and for what you are doing to make it more powerful.

[deastman]

This has me nostalgically remembering disassemblers and profiling tools and Knuth and cpu architectures. The only way to really do that kind of work is to be very obsessive. It was a 24/7 kind of thing for me. I'm grateful for that period of my life (made a great living) and also pretty grateful that this period of my life is over. But thank you and Denis for KS and for what you are doing to make it more powerful.

Yes, a huge +1 on the thanks for the great work you guys are doing. My only assembly experience dates back to the 6502 days, and most of my development work had been in lazy managed languages like Java. I'm in awe of what you and all the other realtime DSP developers do!