X-Fi DSP
All X-Fi cards are based on the same DSP and share the following characteristics:
With the X-Fi, Creative Labs wants to slightly modify our approach to sound cards and the technical characteristics to which we usually refer to in the audio world. Noise signal report, standard resolution or even converter quality are still applicable, but the manufacturer also adds data coming from the processor and 3D world. This is calculation power, number of transistors, core frequency, fabrication process, and voltage.
To be honest we don’t really think this brings anything to the user. From a marketing point of view, however, such figures are better for sales, especially in the subjective domain of sound. We do give you these figures, though, without any hesitation, because they help to understand the technical choices made by Creative Labs for the X-Fi.
The X-Fi, labelled CA-20K1 PAG, includes 51.1 million transistors, twice more than a… GeForce 2 Ultra. The difference with an Audigy chip is even greater as it only has 4.61 million. If you are thinking why do we need 11 times more transistors with the X-Fi, Creative Labs’ answer is clear: “to have a higher calculation power”. So, X-Fi delivers 10340 MIPs at the time when an Audigy 2 delivered a mere 424! If you are still unconvinced and ask why such calculation power is necessary for the X-Fi, you have to know that the chip’s architecture is fixed and that each function has a defined calculation power:
71% of the calculation power is devoted to a Sample Rate Converter. In short, SRC operations are necessary when the DSP receives a signal, which isn’t in the standard sampling rate. For Creative Labs’ and e-Mu DSP, this is 48 kHz. If the input signal is 44.1 kHz, it has to be processed to arrive to the DSP with the right resolution. With the Audigy 2 this operation was carried out by a Cirrus Logic CS8420 chip (with 350 000 transistors). With the X-Fi, it’s processed internally. The main interest here is to have a perfect synchronisation with the DSP and processing part.
It also helps Creative to adapt SRC operations to its needs. The most recurrent problem in terms of SRC is the addition of harmonics, a cause of distortion. With the CS8420, the THD+N at 1 kHz is -117 dB, with the X-Fi, the THD+N at 997 Hz is -135 dB (for 44.1 KHz to 48 kHz operations, the most delicate possible). As DACs in the Sound Blaster Elite Pro (Cirrus Logic CS4398) provide -107 dB (at 1 kHz) THD+N, SRC operation will be the most transparent as distortion and noise level produced will be inferior to what the converter will add in the end. With a DSP normally set at 48 kHz, the interest of a good SRC is obvious, but the calculation power of X-Fi puts in perspective a power 24 times higher than the previous line. Indeed with the Audigy 2, Sample Rate Converting was processed with an external chip and with the X-Fi it represents 71% of the calculation power added by this operation.
If we compare the Audigy 2 and X-Fi without taking into account the SRC, the calculation power is 3000 MIPs, approximately 7 times more than the Audigy 2. This is already quite a good result.
