Hello,
GFLOPS numbers of various DSPs are given in this directory www.ti.com/.../products.page:1230:0] Could you tell me, this GFLOPS are Single Precision or Double Precision ?
This thread has been locked.
If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.
Hello,
GFLOPS numbers of various DSPs are given in this directory www.ti.com/.../products.page:1230:0] Could you tell me, this GFLOPS are Single Precision or Double Precision ?
Hello Ran,
The page is "C66x Multicore Digital Signal Processor Selection". Some GFLOPS number are given in that page. URL is given below;
Lets see how they got to this number
If you look at the DSP core you see the following -
The internal connection to the M unit is 128 bit wide, so you can get 4 32-bit values in a cycle. For single precision floating point it is 4 values
The S and L units each has 64 bit internal buses, and each can do 2 32-bit single precision additions
All together there are 8 single precision operations in a single cycle (look at the http://www.ti.com/lit/ug/sprugh7/sprugh7.pdf
to see what instructions provide the operations that I specified above)
So 8 floating point instruction per cycle per side in the DSP core
There are two sides - so 16 instructions in a cycle
DSP clock is up to 1.25GHZ so there are 20G instructions per second per core
There are 8 cores and thus we have 160 G floating point instructions in a second or 160 GFLOPS
Ran