5 ms·
What's the purpose of a "dedicated" soundcard? I'm an audio professional - from consumer to prosumer to pro, the best soundcards these days use USB (well, exce
by riprowan 10y ago
What's the purpose of a "dedicated" soundcard?
I'm an audio professional - from consumer to prosumer to pro, the best soundcards these days use USB (well, except for the Avid stuff).
I don't think this has anything to do with DRM. I think it has to do with the logic of building one device that works on everyone's computer and doesn't need drivers.
Historically, PCI / PCIe was required because it was the only way to move enough audio data at low latency back and forth to the computer. Historically what has plagued the manufacturers of dedicated devices is the difficulty of finding good driver developers and then getting drivers certified. I watched companies actually fail because they build a great audio device and couldn't get working drivers written for it once it was built (I'm thinking of Gadget Labs but I'm positive there was another one too).
Once USB became fast enough to do the job - and once USB Class drivers became a thing - the decision to build devices for USB (instead of for PCI which not all computers even have) became a no-brainer.
- speeder 10y agoDedicated soundcard, and drivers with direct access to hardware have nothing to do with audio professionals. It is about processors on the card (that back then had about half of the amount of transistors that the PC main CPU had, so they could pack quite a punch in processing power) that could do realtime audio calculations. I am not talking just about "effects" from the audio professional sense, I am talking about realtime creation of audio according to the scene geometry and stuff happening, for example A3D could calculate actual reflections, not just simulated reverb/echo. If you want to see an example of this in real life: go near a window that has a barrier in front of it of some kind, in a relatively enclosed space. Then wait for some noise to happen outside, in the direction opposite to the window. Suppose you was facing "forward", the window (and the wall beyond it) is to your left. Then an airplane passes by your right. The naive way to do it (that is what most games today do without a soundcard to do it for them), is play sound from the right speaker (be it in a 5.1 or in the right speaker of a headphone), or to simulate it is "outside", play it muffled. The correct way to do it, is play from the left, because it reflected on the wall, and came inside through the window. Vortex soundcards for example, made by Aureal, could calculate that (back then, only one reflection, so sequences of reflections that could go around corners didn't worked, but back then was 1996, this was already amazing, the Aureal+QSound sound card demo still impresses people to this day: https://www.youtube.com/watch?v=IUDTlvagjJA https://www.youtube.com/watch?v=IUDTlvagjJA ) Mind you, these can't be done with USB, the protocol has too high latency (even with USB 3.1, due to the handshakes needed), in fact evne PCIe has too high latency compared to PCI, and when the switch from PCI to PCIe happened, soundcard makers had to put a ARM chip on the PCIe cards to do some calculations that originally were supposed to be done by the PC main CPU, because the latency of PCIe was too high.
- riprowan 10y agoThanks for the lesson in acoustics but I'm not sure what any of this has to do with OP. He wanted a Soundblaster. Also this doesn't add up: > in fact evne PCIe has too high latency compared to PCI, and when the switch from PCI to PCIe happened, soundcard makers had to put a ARM chip on the PCIe cards to do some calculations that originally were supposed to be done by the PC main CPU, because the latency of PCIe was too high. You say that the convolution was supposed to be performed by the host CPU, but had to be moved to the PCIe card due to latency. That doesn't add up. If the convolution can be calculated in realtime by the CPU, then the only data that has to travel to the soundcard is the 16/44 (or 2496) audio which is easily handled by PCI, PCIe, USB, etc. with negligible latency. If a processor had to be introduced, it was to perform convolutions because the host CPU couldn't keep up.