DSP vs FPGA vs Sample rates...

[iomegaman]

So I'm in an interesting discussion on facebook regarding Antelope Audio plugins and 48kHz vs 96kHz...the author of the thread claims that at 48k the plugins from Antelope sound "harsh" on the high end, while at 96k they sound "smooth"...trigger words aside my first reaction was "We have no idea because Antelope has never released any info on if they upsample or what the filter cut-offs are that they are using...the best we have is subjective analysis...so the author ran the plugins through "plugin doctor" and there is some negligible difference maybe 1 db around and above 16kHz...well outside the realm of most people's hearing...


But it did get me thinking...I have a Discrete 8 Premium package, meaning I have all the plugins and its ALL FPGA>>>in other words I do NOT HAVE THE DSP PACKAGE that Antelope calls "synergy core"...so when I run the plugins it will all be FPGA...not DSP...and in my limited understanding I am assuming there are some coding difference between how you would code DSP vs how you would code FPGA...but Antelope loves to push the "synergy core" packages, not sure why, but the difference between my device and a Discrete 8 Synergy Core device is not all that and a bag of chips...I think there's only a handful of plugins they offer on the DSP side of things that I cannot run (Anteres and an LA2A, both I have in UAD Dsp versions...maybe one or two more but thats it)...the plugin count vs mine and the DSP version is only slightly + to the DP side...not enough to sweat green stamps over.


THE synergy core products they offer are basically the FPGA unit PLUS an additional DSP chip...


AS far as I recall Antelope is the only company Gyraf has allowed to make plugins of his devices (could be wrong dunno)...and i'm going to venture those are coded to FPGA only...


But it did get me thinking about how would you code a device that ran both DSP and FPGA plugins...would there need to be special cutoff filters between the code platforms that could introduce the suspected "harshness" the OP on face-nat-a-real-book was complaining about?

I'm going to run my own test using an analog Sine wave generator and run some plugin at both sample rates to see if I can get them to null...

I have no idea...but am open to bing educated here...I'm not far enough into the "Art of Digital Audio" to understand what the implication of FPGA +/vs DSP vs Sample rates vs filters are.


I do recall that McDSP makes a APB-16/8 which I believe is a FPGA VERSION device to run his plugins...and its pricey...but many people swear by it...and in his version the code sits between the AD and DA converters which he is converting at 32 bit...so it makes me wonder if the FPGA version of Antelope is doing bit-wise...

 

[john12ax7]

It's funny they are now promoting DSP, as they had previously touted the superiority of FPGA since it was "hardware". Their marketing is certainly a bit silly and misleading.

Now the 48k vs 96k thing is certainly real for plugins. A non-linear process will produce crud that is easier to deal with at 96k than 48k. Many plugins will upsample internally, process, then downsample back to the project rate to better minimize the crud.

 

[iomegaman]

It's funny they are now promoting DSP, as they had previously touted the superiority of FPGA since it was "hardware". Their marketing is certainly a bit silly and misleading.

Now the 48k vs 96k thing is certainly real for plugins. A non-linear process will produce crud that is easier to deal with at 96k than 48k. Many plugins will upsample internally, process, then downsample back to the project rate to better minimize the crud.

Yeah to be honest Antelope worries me quite a bit, just watched a youtube vid from a former employee (Marcel James) who recently left after 14 years with them and his youtube video was not very reassuring...they seem to be dipping into the "Aardvark" puddle of behavior again...which spells disaster for older devices like y Discrete 8 which became "legacy" less than 3 months after I bought it directly from them...


I've never had the horror stories with tech some people complain about...any issues I've had have been resolved within "Standard Audio Device Manufacturer Operating Rules of Customer Service" meaning I've never gone more than a week with no response...this is on par with Apogee, and Avid...once you get out of the "A"'s in audio world service improves for the most part....(Lynx and RME are Stellar, even Behringer seems to care these days)...
But they do seem to ignore some parts of the failing business model...for instance my Edge Solo modelling mic SOFTWARE from them is pretty much useless on my Apple M1 computer...the mic work inside the D8 software but they have yet to fix the Edge solo software at all, it fail in both Logic and PT>>>the DAWS won't load the software and its been this way for almost a year...(I don't really use it, the mic modeling phase here is over and we decided it's BS...regardless o what Slate and Acustica want to preach)...
THERE ARE Some bizzare tech advice workarounds that seem odd...for instance the AFX2DAW PLUGIN THAT supports the plugin I paid for FAILS IN LOGIC AU SCan...the work-around from their tech department is "use it anyway"...it crashes Logic plugin scan and then you go and check "use anyway"...which seem really not professional, but it does work in spite of Logic bitching.

Also the releasing new product while old product has issues sends a really concerning message about priorities and financial stability...but Avid has shown that you can still symp around doing this crap and get away with it.
I will adit the FPGA versions are very good sounding in spite of the pita the device is...

UAD STILL Leads in ease of use and quality product a far as I' concerned.




"Their marketing is certainly a bit silly and misleading."
That is a huge understatement...when I bought the D8 it was promoted WITH D8 Synergy Core sort of adverts, little did I know they were trying to move out old stock it was several months before I realized I had NOT bought a SyneryCore model, the difference in the ads were so minute and subtle as to be a complete sleight of hand act...thats when I discovered they had made my 3 month old unit "legacy"...I was almost ready to return it because between feeling misled and them breaking promises on M1 support I wa pissed, but thy got M1 drivers pretty much b4 a lot of other vendors (Apogee told us our Symphony MKI would never be supported AFTER SELLING US A THUNDERBRIDGE TO FUTURE PROOF OUR DEVICE...then changed their minds when they got push back ...but it took a year).

Bottom line it's a decent device with lots of options, great clocking and sounds really good but the company make e nervous.

 

[sahib]

.............and in my limited understanding I am assuming there are some coding difference between how you would code DSP vs how you would code FPGA...

I am quite lost with what you are trying to get to. So I'll just respond to this bit.

FPGA is a hardware, a programmable chip, which you can use for DSP. On the other hand DSP is done both using dedicated hardware or a computer (software). They say that FPGA will eventually kill dedicated DSP chip.

FPGAs are programmed using hardware description languages (HDL) Verilog and VHDL. On the other hand dedicated DSP chips are programmed using programming languages like C. They are completely different things, though there are some similarities.

 

[abbey road d enfer]

I'm going to run my own test using an analog Sine wave generator and run some plugin at both sample rates to see if I can get them to null...

Good chance they won't null. Any phase difference will result in non-nulling. And then you won't know which one is correct.
Now the difference between 48k and 96k is perennial, and can be traced to the anti-alias and reconstruction filters, which have very different slope and hence phase response.

 

[gyraf]

..agree, that company makes me nervous too. I'm not entirely sure it was a good idea to let them model some of my units, their customer service kinda messes up the positive aspects I'd expected (mainly in the form of legitimization)

/Jakob E.

 

[mhelin]

You can build a softcore DSP on FPGA hardware, too. Usually there are either HW microcontroller or softcore processor (NIOS and MicroBlaze most common but I think RISC-V softcore does exist) that controls the lets say the audio streaming and algorithms used on the FPGA. These days SoC (System On a Chip) FPGA's are very common, they have something like ARM Cortex-A9 MPU, external memory controllers and some memory, bus, peripherals etc. on the hard processor system (HPS) side and the FPGA on the other side of the SoC chip.

 

[iomegaman]

You can build a softcore DSP on FPGA hardware, too. Usually there are either HW microcontroller or softcore processor (NIOS and MicroBlaze most common but I think RISC-V softcore does exist) that controls the lets say the audio streaming and algorithms used on the FPGA. These days SoC (System On a Chip) FPGA's are very common, they have something like ARM Cortex-A9 MPU, external memory controllers and some memory, bus, peripherals etc. on the hard processor system (HPS) side and the FPGA on the other side of the SoC chip.

Yea I think Antelope initially thought the FPGA plugin path would be easy once they got it off the ground but have since discovered that in the audio industry market of ideas its harder to move the "new" needle above the noise floor...they have lately pushed the DSP side of things and frankly have not developed any newer plugins since inception that would land on the FPGA side of things as far as I can tell...(Their DSP versions has a few more plugins but not many)

I got the Discrete 8 and the premium plugin package which include ALL of the FPGA versions (since my device has no DSP chip)...for $1k... one of the first things I did was carefully lift the security foil on the bottom and check inside to see what was in there, its a Xilinx FPGA and an AKM dac...


Honestly the plugins sound really good, frankly they are on par with UAD stuff and in a few places a bit better (they are the only company with Gyraf stuff and its probably 90% as good as the analog stuff I have built)...I can run a ton more instances on an FPGA than my Sharc UAD Apollo device...like probably 5x as many...which is not much of a problem as far as FPGA code goes from what I understand.

But like I said I think they are finding the FPGA development side not as fruitful as their original impressions...hence the pushing of DSP.


I only post this next link as a "Caveat Emptor"...I have no horse in this race (other than a $1k interface)...could be a disgruntled employee but 14 year is nothing to ignore...


Marcel James on the rise and fall of Antelope...