Actually there is one reason : you've got an external upsampler that you want to use instead of the MC one (a hardware one for instance).
And that's fine, because MC allows you to do this. However, one common misconception about computing in general that annoys me has to do with hardware vs software.... I'm not suggesting that you are making this mistake, TontonGuigui, I just figured this was a good opportunity to rant a little.
[nerdrant]
When you take an algorithm and reproduce it in silicon (make a "hardware" version), that process does not magically make it "better" or "higher quality" than an identical software implementation of the same algorithm. In fact, because of transistor budgets and error correcting capabilities (coupled with the fact that integrated circuit design is hard), many hardware implementations are of LOWER theoretical quality than their software counterparts.
All making a hardware version accomplishes is making the algorithm run FASTER (and making the system cost more).Sometimes, "faster" does equal "better" when software isn't fast enough to do the job in real time. A perfect example of this is in 3D graphics rendering. While a full software rendering of a 3D scene will usually be more technically perfect, it is so much faster to render on custom GPU hardware, that it allows you to achieve high framerates for real-time playback (or it lets you get your cinematic masterpiece to theaters more quickly, and time is money). However, GPUs cheat. They are generally of measurably LOWER quality than a pure software renderer in certain ways (AA and AF for example) because the hardware vendors take shortcuts in hardware to save on their transistor budgets, and end up with smaller (cheaper) pieces of silicon when they order them in bulk from TSMC. Usually these shortcuts don't impact visible quality, but there are certainly times where they do.
Implementing things in hardware can also be useful for security applications, because it makes tampering with the algorithm more difficult. Note that it makes it more difficult, but certainly not impossible. Many "hardware" implementations are actually done using
FPGA chips which are essentially somewhere between hardware and software, and even if it is custom silicon, hardware is just as prone to bugs as software. It also can serve to make logic more opaque to reverse engineering, though this sometimes backfires (you can, after all, look at any integrated circuit with an electron microscope and trace the actual electron paths). Security concerns are the main reasons that things like
TPM and other "DRM-like" security schemes were implemented in hardware. Of course, that didn't help prevent people from cracking AACS immediately, so this is certainly of dubious value in many cases.
However, for audio processing, the general purpose CPU in your computer is
plenty powerful enough to process TONS of audio streams in real time on the fly. There is no technical reason to make the algorithms
faster. Now, you may believe that your hardware maker implemented a proprietary (and apparently secret from the entire mathematical world) algorithm that is
better than what MC uses. That's fine if that's what you believe, I suppose, but you should understand that THAT is what you are "believing". I would personally tend to believe that a good, high quality algorithm is a good high-quality algorithm regardless of how it is implemented. And, frankly, those high-end component makers do suffer from some serious cost constraints. Even though you may feel like you pay a premium for the product, getting custom bulk silicon runs from TSMC or Global Foundries is EXTREMELY expensive and time consuming and requires a huge team of dedicated hardware engineers to do properly. To manufacture using new process tech, we're talking investments of BILLIONS of dollars a year in engineering time and just raw silicon cost. I'd bet that the vast majority of those "high end" chip designs are probably machine designed and implemented using FPGA chips due to cost concerns. Those few that are using actual custom silicon, if any actually exist, probably are built using process tech that is YEARS behind. Plus, if you discover a bug, it is pretty expensive to fix it (to order a whole new respin from TSMC), so they probably just ignore it.
Being "hardware" doesn't automatically make it "better". In fact, in many real-world cases, the
opposite is true.
[/nerdrant]
