Если бы спрашивающий товарищ умел немного гулить, то сразу бы нашёл немного информации для размышления:
Axe FX Ultra: ADSP-TS201S @ 600 MHz (Dual core)
11R: 2 x TMS320C6727BZDH @ 275 to 350 MHz
Axe FX Ultra DSP has 3.6 GFlop (floating point operations per second) performance.
Each 11R DSP has 2.1 GFlop performance @ 350 MHz so 4.2 as 2 DSP present.
AXFX 3 runs a pair of Keystones at 1Ghz ea
AXFX 2 ran a pair of TigerSHARC's at 600Mhz ea
Axe-Fx's use the ADSP-TS201S
Helix runs a pair of SHARC's at 450Mhz ea
The AX8 has two ADSP-21469s and two microcontrollers.
https://wiki.fractalaudio.com/axefx2/index.php?title=AX8The TigerSHARC architecture is vastly superior to the SHARC, and it's a shame it has been discontinued. In our tests a TigerSHARC performs about 50% faster clock-for-clock. Then factor in the higher clock speed and it's about twice as fast. See here for independent benchmarks: http://www.bdti.com/MyBDTI/bdtimark/chip_float_scores.pdf. The newest SHARCs are the same as the 213xx except they have the FIR accelerator. As one can see a TigerSHARC (Axe-Fx's use the ADSP-TS201S) is about twice as fast. (...) The TigerSHARC has a more modern core and much better architecture (larger register file, wider buses, better DAG units, etc). It can do six floating point operations per clock. In practice this is two multiplies, two adds and two store/fetch instructions. This means a 4096 point IR only requires 2048 clock cycles. Secondly the TigerSHARC has a much wider bus than a SHARC. The TigerSHARC has a 512 bit bus whereas a SHARC has a 128 bit bus. The TS can move data around much faster and pipeline stalls due to bus contention are far less frequent. It can load or store eight words in a single clock cycle whereas a SHARC is hard pressed to do two. The TS also has a LOT more on-chip memory. A TS has 24 Mbits of on-chip memory compared to 5 Mbits for the best SHARC. This means more code/data in fast memory and less stalls waiting for data access. Also the TS has a superior cache unit which caches both instructions and data, as opposed to instructions only on a SHARC. There are numerous other improvements as well including better DMA engine, better interrupt handling, vastly superior 40-bit floating point support (which we use for amp modeling), etc. Having written hundreds of thousands of lines of code for both (including coding the SHARC's FIR accelerator) my experience is that the TS is a much better chip but it is much more expensive. We moved to SHARCs for our floor processors for a variety of reasons but performance was definitely not one of them. The FIR accelerator is nice and all but it only applies to cab modeling which is a small percentage of a typical preset. On an Axe-Fx II a stereo Hi-Res cab block only uses 11% of the DSP. On an AX-8 it's still a couple percent as the accelerator doesn't handle everything needed in the block. So the net savings is less than 10%. The rest of the effects then run about twice as fast which means almost twice as many effects per preset on an Axe-Fx compared to an AX-8. While "all about the code" has some merit, in reality it's "all about a lot of things" including the architecture and clock speed. Algorithm complexity is also very important and it becomes diminishing returns (i.e. it takes twice as much CPU to improve the sound quality 10%). If it were just about the code everyone would still be using the original 33 MHz SHARCs. The TigerSHARC enjoyed a long run as the best DSP on the market but, alas, all good things must come to an end. The good news is that Atomic, Fractal and Line6 are all using the same DSP family in their respective floor products so it's highly unlikely Analog Devices will discontinue them and there are now more choices for the consumer."
Размещение рекламы