Absolutely nothing, although I suppose if you insisted on non-native int or FP formats you would take a big performance hit. Bruno uses SHARCs instead of FPGAs. Some people here don't seem to understand that you can do whatever you want in a DSP and constantly complain about "sound quality" anytime they see one inside a device.
Last edited:
Syn08 knows what an FPGA is. Yes, it is programmable logic, but the actual structure is more complicated.
It is just a DSP front-end for a rudimentary DAC. That's all these FPGA DACs are. It is truly nothing novel.
Maybe you can explain how an "FPGA DAC" differs from the ESS Sabre implementation...
Last edited:
Short of a DSP, the true question is FPGA vs. CPLD. I always disliked FPGAs for the very complicated timing scenarios. CPLD allow a deterministic timing analysis, which is kind of WYSIWYG of digital design. That's one of the prices for the tremendous flexibility of FPGAs. OTOH, CPLDs are too small to implement a complex design.
I don't believe a FPGA implemented DAC (from the digital filter(s) to the Delta Sigma Modulator) can have a competitive price/performance ratio comparable to a custom chip (which, make no mistake, is not designed from scratch, at the device level, but various already optimized IPs are used, anyway); FPGAs could be used in prototypes (for example to check a new digital filter properties) but otherwise I don't think there's much in any audio grade DAC that needs to be prototyped before sending a DAC chip custom design to a silicon foundry. If there is any required analog part on the DAC chip, then that cannot be implemented on a FPGA, anyway.
Shortly, using a FPGA is (at least) economically a dumb idea to implement a DAC.
I don't believe a FPGA implemented DAC (from the digital filter(s) to the Delta Sigma Modulator) can have a competitive price/performance ratio comparable to a custom chip (which, make no mistake, is not designed from scratch, at the device level, but various already optimized IPs are used, anyway); FPGAs could be used in prototypes (for example to check a new digital filter properties) but otherwise I don't think there's much in any audio grade DAC that needs to be prototyped before sending a DAC chip custom design to a silicon foundry. If there is any required analog part on the DAC chip, then that cannot be implemented on a FPGA, anyway.
Shortly, using a FPGA is (at least) economically a dumb idea to implement a DAC.
this is pretty easy:
"Maybe you can explain how an "FPGA DAC" differs from the ESS Sabre implementation..."
first let's accept that there is no such thing as an "FPGA DAC", DACs like the PS Audio, and Chord use the FPGA to handle various processing chores like re-clocking, volume control, PLL(s), and oversampling using custom filters and rates, to a very high rate. What outputs from the fPGA is a very high rate digital signal, which is then converted to analog through a discrete filter circuit of varying type. So the actual DAC is a discrete converter.
The actual functional difference between this approach and an ESS chip, is that the ESS chip is resource limited: it has less processing power onboard, and therefore cannot apply as sophisticated filters, modulators, and oversampling as can be done if a powerful FPGA or a SHARC processor (or two SHARCs as Bruno's Mola Mola designs use). Now engineers can debate all day long whether that additional processing power, and more sophisticated algorithms which it allows actually result in better sound quality, or not, but the fact remains that there is a significant difference.
The extension of this approach is the even greater processing power offered by running the oversampling and any other needed DSP functions in a full fledged computers and then sending the high rate a data to a discrete DAC which does no additional processing onboard outside of the digital receiver and re-clocking functions. For example, my preferred settings in HQPlayer software require at a minimum, an I7-9700K processor to output DSD 256 to a discrete DAC for direct conversion. My new server is built around an i9-9900K chip and 32 GB of RAM to be sure I have enough processing power. IME, the modulator in HQPlayer sounds fantastic, and is worth this effort.
How is this relevant to the AKM 4499? Well the AKM 4499 has a wonderful operation mode: direct DSD, which bypasses all of the chips processing stages, and sends incoming DSD data right to the switched resistor conversion stage, so DACs which enable the use of this mode of operation with this chip are excellent candidates for use with HQPlayer's more advanced oversampling/filtering approach.
"Maybe you can explain how an "FPGA DAC" differs from the ESS Sabre implementation..."
first let's accept that there is no such thing as an "FPGA DAC", DACs like the PS Audio, and Chord use the FPGA to handle various processing chores like re-clocking, volume control, PLL(s), and oversampling using custom filters and rates, to a very high rate. What outputs from the fPGA is a very high rate digital signal, which is then converted to analog through a discrete filter circuit of varying type. So the actual DAC is a discrete converter.
The actual functional difference between this approach and an ESS chip, is that the ESS chip is resource limited: it has less processing power onboard, and therefore cannot apply as sophisticated filters, modulators, and oversampling as can be done if a powerful FPGA or a SHARC processor (or two SHARCs as Bruno's Mola Mola designs use). Now engineers can debate all day long whether that additional processing power, and more sophisticated algorithms which it allows actually result in better sound quality, or not, but the fact remains that there is a significant difference.
The extension of this approach is the even greater processing power offered by running the oversampling and any other needed DSP functions in a full fledged computers and then sending the high rate a data to a discrete DAC which does no additional processing onboard outside of the digital receiver and re-clocking functions. For example, my preferred settings in HQPlayer software require at a minimum, an I7-9700K processor to output DSD 256 to a discrete DAC for direct conversion. My new server is built around an i9-9900K chip and 32 GB of RAM to be sure I have enough processing power. IME, the modulator in HQPlayer sounds fantastic, and is worth this effort.
How is this relevant to the AKM 4499? Well the AKM 4499 has a wonderful operation mode: direct DSD, which bypasses all of the chips processing stages, and sends incoming DSD data right to the switched resistor conversion stage, so DACs which enable the use of this mode of operation with this chip are excellent candidates for use with HQPlayer's more advanced oversampling/filtering approach.
Last edited:
I know the answer, I was asking ggetzoff.
There is no need for that much processing power if it's implemented more elegantly. The idea that you need 75W of CPU to do a modulator for audio rates is insane. Nevermind the fact that a Core i9-9900k isn't actually all that fast for DSP. It lacks AVX-512 and isn't going to touch a GPU either in many cases.
There are also significant advantages afforded with an IC that can be made on a custom process that cannot be realized with these "FPGA DACs".
Last edited:
I've often thought it seems ridiculous that such a huge bandwidth is supposedly necessary in the digital domain to reconstruct the audio bandwidth.
Hehe, probably I should have said "relatively", I keep forgetting the precise qualifying language required. 😉
If you design it yourself it must be better than those pre-packaged ICs that are already extreme overkill, surely. 😉
You are implying that people like Rob Watts or Bruno are either charlatans or deluded.
TCD
Rob Watts probably is. Bruno isn't, he doesn't make extraordinary claims or say things in YouTube videos that are nonsense that I'm aware of. There's nothing wrong in selling a super-expensive piece of overkill, it's just the dishonest marketing that is the issue.
FWIW, HQplayer actually utilizes GPU for additional processing.
Again, you are implying that Jussi Laako is either a charalatan or suffering delusion despite the hundreds if not thousands of people who are using the
latest processing intensive modulators in HQplayer.
TCD
To clarify, I don't think Jussi is either one, since I haven't seen him make audibility claims about his software. If he wants to provide crazy overkill and people want to buy it, then good for him.
I will not claim to know exactly how his software is implemented, however, I am deeply suspicious that this is useful.
At work I am part of a team doing advanced processing, including machine learning, on 4k video with 12-bpp color depth at high framerates. We settled on huge FPGAs that cost around $10-20k each in low quantities, but we've benchmarked many solutions, including GPUs and x86 processors. You should see what we can achieve in real-time with a Quadro RTX 6000 and a 45W embedded x86 CPU. I highly doubt this kind of power is required to do anything useful in audio except maybe hundreds of channels of processing.
When a functional block can be implemented in an ASIC using < 1W and be good enough to yield industry leading performance (ESS, AKM DACs), I am going to say that any implementation that requires a 5 GHz x86 CPU consuming 50-150W is either insane or badly implemented. Given that it at least appears Jussi knows what he's doing in terms of implementation, I'm going to say the whole thing is insane and everyone is delusional.
I don't need to be an expert in designing SDMs to know this doesn't pass the smell test. It's even crazier than people thinking you need (or would want) a 10 trillion tap interpolation filter to "do it right".
BTW, the fact that the fastest CPU for his software is a 9900k or 10900k indicates it is almost completely driven by single-threaded performance...
Last edited: