Altiverb vs Briscati?

[dannycurtean@yah]

Originally Posted by not_so_new

I am not disagreeing with any of that... and I am not making a stink about it... it sounds like a great piece of kit.

But my PC has multiple processors as well and I bet they are more powerful than what's in the M7. I am skeptical that "CPU power" is at the bottom of what makes the M7 sound so good.

You will also note that I didn't say anything at all about the time or effort that went into the design.

I pointed out above that the algo's are probably custom written for the hardware in the M7 but it's not like the M7 has magic technology inside. It's a computer, it has a CPU, mother board, memory and I/O like any other computer would have.

My feeling is that Casey and Co. have made a business decision to make a hardware reverb unit and more power to them for that but I don't believe there is magic in the hardware, it's still just a computer... a purpose build computer but a computer non the less.

This is EXACTLY what Ive been trying to say! This isnt just about the M7....AGAIN...its about all other reverbs. And Casey, I know I pissed you off, but I was kidding with the plugin thing man, relax man.....have a laugh...

[dannycurtean@yah]

Originally Posted by seancostello

Yes, but not CPU technology per se.

IIRC, the M7 uses 6 dual core Blackfin DSPs from Analog devices, with a 7th Blackfin used to control the other DSPs and run the interface.

I worked for Analog Devices for 7 years, programming audio tools and algorithms for the SHARC and Blackfin DSPs. I created many reverbs for these boxes, and programmed some specific functions for reverb generation (using VisualAudio, the soon to be sadly departed audio environment, if rumors are correct). In the past year, I have been primarily focused on development of reverb and other plugins for native processors. So I guess I can say that I work with 1's and 0's as well.

The Blackfin DSPs can perform signal processing tasks that far outstrip a PC of similar speed. With a single core Blackfin, I was able to program a reverb that used several hundred delay taps, feeding into a 16 delay line FDN reverb, with many cycles to spare. A reverb of such complexity could run on my MacBookPro, but it would take a fair amount of the CPU. Multiply my Blackfin results by 12, and you have some idea of the power available in the M7.

Could a PC run the algorithm? Maybe. Just by crunching the numbers, and adding in my experience with both Blackfins and native processors, and my guess is that the top of the line Mac Pros could run the algorithm...barely. This is assuming that the multiple cores could all be running in conjunction to process the algorithm, which is going to be quite a bit more difficult on a computer with a multitasking operating system than on a purpose built machine. The memory descriptions of the M7 algorithm make me think that the algorithm would not be able to run out of cache the whole time, which means that things would slow down a lot as data was transfered from the fast cache memory to the slower main memory.

Casey has said that the algorithm won't run on any current general purpose PCs, and he is the only person who knows what his algorithm is. My comparisons are based solely on number crunching capacities as they apply to reverbs.

Even if you could run the algorithm on a Mac Pro, I doubt that you would be able to run too much else on it. Let's be charitable, and presume that the algorithm is using 50% of the processor. You've essentially spent $1650 for the half of the computer that does the processing, plus whatever the plugin costs. You can't run multiple instances of the algorithm, as it uses a huge amount of resources by its very nature. At this point, the M7 looks like a better and better deal. Plus, will your Mac Pro boot up the plugin in 10 years? Talk to people that worked with the amazing audio programs that worked in OS9 about how long-term a software solution will be.

Having said that, I like plugins. I write them, and get paid for them. I have been pleasantly surprised by the performance I can get out of native code. People talk about the slow performance of computer RAM, but if you write your algorithm correctly, the delay lines for your reverb can all sit in the cache, which will speed things up enormously. The "classic" reverb algorithms of the 1980's should be able to run on PCs using only a few percent of the CPU, and maybe even less if some serious optimization work is done.

My question is, can algorithms be written that can sound as good as the Bricasti without using as many resources? Or, can reverb algorithms be created using a different paradigm than the recursive delay based algorithms, that can better take advantage of the increasingly parallel nature of modern CPUs and GPUs?

I understand that, and I tried to address that in my post. I did learn a thing or two from your post though and I understood the main problem with running the M7 for example on a native CPU. But I also mentioned the other technologies out there, like GPU parallel computation, or just running a sort of eco system for that algorithm to have virtual resources dedicated to it but processed on the native back end. That could be counter intuitive but you could do it. For example, you could run several virtual "sections" of the plugin stages in different sequences. So for example if your reverb has 100 steps or stages, you run 10 different segments of 10 processes in sequence. That way you dont need th massive parallel needs of having 10-12 core machines to run this thing properly. This is what I mean about optimizations. Cause at the end of the day, and in its most basic stages, a reverb is just a sequential time delay.....

But I could be horribly wrong. And in that case, something like a parallel processing unit (like a GPU) or some sort of UAD type thing dedicated to the M7 with multiple instances and recall would be great. Until then, we have the hardware units.

[Barish]

Originally Posted by dannycurtean@yah

But I could be horribly wrong. And in that case, something like a parallel processing unit (like a GPU) or some sort of UAD type thing dedicated to the M7 with multiple instances and recall would be great. Until then, we have the hardware units.

Man...

The guy is saying "stop it, there will be no M7 plug-in", and still you keep going on with this. Even though you acknowledge that you piss him off by doing this, you keep on going and you expect him to have a laugh and allow you to keep doing it.

Are you disturbed or what?

Now that I've noticed you've also quoted yourself in your own sig line, I'm starting to think that you really are self-obsessed of some sort. You just don't seem to read lips.

Give up dude. You have gotten pretty annoying with this already and it's clear that your constant abusive expression of your wishes will not get you an M7 plug-in from Casey & Co in the foreseeable future. Just see it and move on.

B.

[The MPCist]

Originally Posted by Barish

Now that I've noticed you've also quoted yourself in your own sig line, I'm starting to think that you really are self-obsessed of some sort. You just don't seem to read lips.

And I thought I was the only one who thought that was strange...Hmmmm... Anyways, now that someone's mentioned it, I've got to second it.

It's weird dude.... really weird....

Also, Danny, just stop with the plugin talk. Thanks.

[colinmiller]

Originally Posted by Foot In Mouth

Altiverb colors impulses a bit, due to all it's processing. The same impulse played through Logic's space designer sounds a bit clearer and more spacious. The impulses available for Altiverb are generally good, but not great. I have done some impulses that are on par with the Bricasti or better. Convolution as a technology is great. It just hasn't reached it's full potential yet.

I have always felt the exact opposite. I was so excited when Logic came out with space designer, but everything sounded so 2 dimensional to me. Altiverb on the other hand gave me the depth I couldn't get with space designer. I don't know about any of the built in stuff though as I only use my own home made IRs.

[seancostello]

Originally Posted by dannycurtean@yah

I understand that, and I tried to address that in my post. I did learn a thing or two from your post though and I understood the main problem with running the M7 for example on a native CPU. But I also mentioned the other technologies out there, like GPU parallel computation, or just running a sort of eco system for that algorithm to have virtual resources dedicated to it but processed on the native back end.

GPUs will not run delay-based algorithmic reverbs efficiently at all. A GPU is designed to work with large blocks of data in a feedforward system. Reverbs work with arbitrary numbers of small blocks of data in a feedback system, and the blocks all need to be able to talk to some shared memory.

The massively parallel architectures such as a GPU will work for convolution reverbs, and possibly for architectures that have lots of parallel things that can work independently from each other. Good reverbs tend to not work exclusively in parallel - there are some parallel parts, but they need to be coupled to each other.

Modern SIMD engines, such as Altivec and SSE3, have some instructions that are useful for computing parallel filters and the like. The parallel architecture doesn't help out when working with delay lines that can "wrap around" at arbitrary locations, which all algorithmic reverbs rely upon heavily.

That could be counter intuitive but you could do it. For example, you could run several virtual "sections" of the plugin stages in different sequences. So for example if your reverb has 100 steps or stages, you run 10 different segments of 10 processes in sequence.

This works with real time convolution, but again, this isn't how algorithmic reverbs work. Working with large blocks of samples only gets you so far, and running them in sequence will destroy the specific timings that are crucial to the reverb. The programming you describe works for feedforward systems, but reverbs are feedback.

That way you dont need th massive parallel needs of having 10-12 core machines to run this thing properly. This is what I mean about optimizations. Cause at the end of the day, and in its most basic stages, a reverb is just a sequential time delay.....

A reverb is not a sequential time delay. It is a whole crapload of delays, not necessarily sequentially related to each other, that may or may not communicate with each other.

But I could be horribly wrong. And in that case, something like a parallel processing unit (like a GPU) or some sort of UAD type thing dedicated to the M7 with multiple instances and recall would be great. Until then, we have the hardware units.

A UAD type thing that can run multiple M7 instances would need to have several M7 units on board. The Quad-core UAD2 is probably not as powerful as a single M7, just from a signal processing perspective. Back when I worked at Analog Devices, the 300 MHz ADSP-21369 SHARC ended up being about as powerful as a single core 600 MHz Blackfin. The UAD-2 Quad Core uses 4 21369s, while the M7 uses 6 dual-core Blackfins. I know that Universal Audio has a couple of my old co-workers that work there now, and they are skilled at creating optimized assembly code building blocks for the SHARC. However, it is safe to assume that similar things are happening at Bricasti. Plus, there are some useful intrinsics for the Blackfin that allow assembly code to be called from C language (same thing for the SHARC), so the code can end up being very efficient on both platforms.

Since you would need the DSP hardware of the M7 on the board, why not have it in a hardware box, that will work 10 years from now? The idea of having a hardware accelerator card for reverbs, that requires a PC or Mac to run it, has been tried before by Lexicon. These cards are great bargains now, assuming you can find an OS9 Mac with the right slot to run things.

The M7 is not a quick profit product. From all accounts, and from Casey's postings, it is clearly intended to be used for a few decades. The digital algorithm inside of it is not the entire unit. It also has the custom hardware required to run such an algorithm (without choking) in 2009, high quality A/D and D/A convertors to work with any studio environment, a low noise power supply, and a durable case to protect the hardware from harm. It is a future desirable artifact, just like the hardware processors of old.

[not_so_new]

Originally Posted by seancostello

GPUs will not run delay-based algorithmic reverbs efficiently at all. A GPU is designed to work with large blocks of data in a feedforward system. Reverbs work with arbitrary numbers of small blocks of data in a feedback system, and the blocks all need to be able to talk to some shared memory.

The massively parallel architectures such as a GPU will work for convolution reverbs, and possibly for architectures that have lots of parallel things that can work independently from each other. Good reverbs tend to not work exclusively in parallel - there are some parallel parts, but they need to be coupled to each other.

Modern SIMD engines, such as Altivec and SSE3, have some instructions that are useful for computing parallel filters and the like. The parallel architecture doesn't help out when working with delay lines that can "wrap around" at arbitrary locations, which all algorithmic reverbs rely upon heavily.

This works with real time convolution, but again, this isn't how algorithmic reverbs work. Working with large blocks of samples only gets you so far, and running them in sequence will destroy the specific timings that are crucial to the reverb. The programming you describe works for feedforward systems, but reverbs are feedback.

A reverb is not a sequential time delay. It is a whole crapload of delays, not necessarily sequentially related to each other, that may or may not communicate with each other.

A UAD type thing that can run multiple M7 instances would need to have several M7 units on board. The Quad-core UAD2 is probably not as powerful as a single M7, just from a signal processing perspective. Back when I worked at Analog Devices, the 300 MHz ADSP-21369 SHARC ended up being about as powerful as a single core 600 MHz Blackfin. The UAD-2 Quad Core uses 4 21369s, while the M7 uses 6 dual-core Blackfins. I know that Universal Audio has a couple of my old co-workers that work there now, and they are skilled at creating optimized assembly code building blocks for the SHARC. However, it is safe to assume that similar things are happening at Bricasti. Plus, there are some useful intrinsics for the Blackfin that allow assembly code to be called from C language (same thing for the SHARC), so the code can end up being very efficient on both platforms.

Since you would need the DSP hardware of the M7 on the board, why not have it in a hardware box, that will work 10 years from now? The idea of having a hardware accelerator card for reverbs, that requires a PC or Mac to run it, has been tried before by Lexicon. These cards are great bargains now, assuming you can find an OS9 Mac with the right slot to run things.

The M7 is not a quick profit product. From all accounts, and from Casey's postings, it is clearly intended to be used for a few decades. The digital algorithm inside of it is not the entire unit. It also has the custom hardware required to run such an algorithm (without choking) in 2009, high quality A/D and D/A convertors to work with any studio environment, a low noise power supply, and a durable case to protect the hardware from harm. It is a future desirable artifact, just like the hardware processors of old.

Thanks Sean, that is a very informative answer.

[StoneinaPond]

Originally Posted by seancostello

The digital algorithm inside of it is not the entire unit. It also has .... a low noise power supply.....

Actually there two, with a separate power supply for the digital section.

And yes, thanks for the very informative and detailed post.

[KevWind]

Wow wee after the last few posts ,,, I have just one question... What in the hell is a "moreso" ?

[dannycurtean@yah]

Originally Posted by seancostello

GPUs will not run delay-based algorithmic reverbs efficiently at all. A GPU is designed to work with large blocks of data in a feedforward system. Reverbs work with arbitrary numbers of small blocks of data in a feedback system, and the blocks all need to be able to talk to some shared memory.

The massively parallel architectures such as a GPU will work for convolution reverbs, and possibly for architectures that have lots of parallel things that can work independently from each other. Good reverbs tend to not work exclusively in parallel - there are some parallel parts, but they need to be coupled to each other.

Modern SIMD engines, such as Altivec and SSE3, have some instructions that are useful for computing parallel filters and the like. The parallel architecture doesn't help out when working with delay lines that can "wrap around" at arbitrary locations, which all algorithmic reverbs rely upon heavily.

This works with real time convolution, but again, this isn't how algorithmic reverbs work. Working with large blocks of samples only gets you so far, and running them in sequence will destroy the specific timings that are crucial to the reverb. The programming you describe works for feedforward systems, but reverbs are feedback.

A reverb is not a sequential time delay. It is a whole crapload of delays, not necessarily sequentially related to each other, that may or may not communicate with each other.



A UAD type thing that can run multiple M7 instances would need to have several M7 units on board. The Quad-core UAD2 is probably not as powerful as a single M7, just from a signal processing perspective. Back when I worked at Analog Devices, the 300 MHz ADSP-21369 SHARC ended up being about as powerful as a single core 600 MHz Blackfin. The UAD-2 Quad Core uses 4 21369s, while the M7 uses 6 dual-core Blackfins. I know that Universal Audio has a couple of my old co-workers that work there now, and they are skilled at creating optimized assembly code building blocks for the SHARC. However, it is safe to assume that similar things are happening at Bricasti. Plus, there are some useful intrinsics for the Blackfin that allow assembly code to be called from C language (same thing for the SHARC), so the code can end up being very efficient on both platforms.

Since you would need the DSP hardware of the M7 on the board, why not have it in a hardware box, that will work 10 years from now? The idea of having a hardware accelerator card for reverbs, that requires a PC or Mac to run it, has been tried before by Lexicon. These cards are great bargains now, assuming you can find an OS9 Mac with the right slot to run things.

The M7 is not a quick profit product. From all accounts, and from Casey's postings, it is clearly intended to be used for a few decades. The digital algorithm inside of it is not the entire unit. It also has the custom hardware required to run such an algorithm (without choking) in 2009, high quality A/D and D/A convertors to work with any studio environment, a low noise power supply, and a durable case to protect the hardware from harm. It is a future desirable artifact, just like the hardware processors of old.

Ahhh...I understand now. I have absolutely no problem admitting I was wrong, and Im glad I learned what I learned. And its kind of funny, I never though of even thinking of getting a hardware unit.....now im looking at all the possibilities

Im glad that there are people like you that have great knowlege, know how to explain it and be calm and professional as opposed to every other dickhead (including myself) that shouts like a moron pointing useless shit out.

I honestly appreciate your input and your time to correct me, seriously.

(And BTW, to the first DICK that wrote about my sig, I believe in my own mindset.....and second of all, go fuuck urself ).

But since the MPCist said its weird...I changed it.....happy?


And in the right order of things, this thread was horribly hijacked...by me.....so if im sorry. Altiverb is great, Bricasti is great-er.

BTW, speaking of plugins, will there be any sort of plugin control for the unit that can be saved with the session...something like USB or something? Wouldn't that be a great idea, or is it there just no point to that? Casey?

[dannycurtean@yah]

Originally Posted by Barish

Man...

The guy is saying "stop it, there will be no M7 plug-in", and still you keep going on with this. Even though you acknowledge that you piss him off by doing this, you keep on going and you expect him to have a laugh and allow you to keep doing it.

Are you disturbed or what?

Now that I've noticed you've also quoted yourself in your own sig line, I'm starting to think that you really are self-obsessed of some sort. You just don't seem to read lips.

Give up dude. You have gotten pretty annoying with this already and it's clear that your constant abusive expression of your wishes will not get you an M7 plug-in from Casey & Co in the foreseeable future. Just see it and move on.

B.

Are you seriously THAT stupid? I said "Until then....we have the hardware"...DUMBASS

Read idiot....man....

[Warp69]

Originally Posted by dannycurtean@yah

Im glad that there are people like you that have great knowlege, know how to explain it and be calm and professional

I couldn't agree more - Sean is very knowledgeable and it's always a pleasure to read his posts.

[Barish]

Originally Posted by dannycurtean@yah

Are you seriously THAT stupid? I said "Until then....we have the hardware"...DUMBASS

Read idiot....man....

You read what you are written to.

There is no "until then". Don't you get it?

Casey said they WILL NOT make a plugin version.

That means there is no "until then".

What part of "There will be no plugin" don't you understand?

Twit.

B.

[BorrelNote]

Originally Posted by Tube World

In a mix, I doubt anyone would care which one you used.

[StoneinaPond]

In a mix, I doubt anyone would care which one you used.

I would.

[jamwerks]

Originally Posted by dannycurtean@yah

BTW, speaking of plugins, will there be any sort of plugin control for the unit that can be saved with the session...something like USB or something?

I've noticed that M7 has midi I/O. Maybe a system exclusive midi message could be used for recall.

[OurDarkness]

This whole thread makes me wonder if I am missing a big something reverb-wise with the Bricasti M7, when I have an Eventide H-8000FW. I also wonder what's inside the H-8000 FW (I know there are two Motorola DSP 56303 and lots of custom made chips plus I know that it uses FGAP for conversion), so I can only imagine what the folks at Eventide have done with their algorithms.

Casey, do you think you make an application that will interface with the M7 so that users can actually design reverbs? Eventide has VSig, why not something for the M7. I know that reverb design (in a higher programming level - say a visual language) is not that difficult, plus the processing power is there, so do you think it's possible? Maybe for ver. 3.0?

[colinmiller]

Originally Posted by jamwerks

I've noticed that M7 has midi I/O. Maybe a system exclusive midi message could be used for recall.

Or just a simple program change if you save your own presets. Obviously not if you tweak parameters for every song and have too many to make presets for each.

[dannycurtean@yah]

Originally Posted by Barish

You read what you are written to.

There is no "until then". Don't you get it?

Casey said they WILL NOT make a plugin version.

That means there is no "until then".

What part of "There will be no plugin" don't you understand?

Twit.

B.

Ah, you want to go grammatical on me? OK!

If you read my poats I CLEARLY stated SEVERAL TIMES that I am NOT talking JUST about the Bricast, but about ALL reverb hardware units.

"Until then, we have the hardware units."

Now if you would be kind enough to go back, read what the hell I wrote CAREFULLY, then you would also understand that the conclusion I came up with ends with a "S" or as others would put it, its PLURAL. "UnitS"

And Im sorry for calling you what I called you, nobody should have to go through that. I got caught up.

[thethrillfactor]

Originally Posted by jamwerks

I've noticed that M7 has midi I/O. Maybe a system exclusive midi message could be used for recall.

Casey has already said in the past the V2 will address this.