Date
1 - 5 of 5
Benchmark Platform
|
Jim Lill
I'd like to compare/benchmark various WD platforms. Seems like wsprd decode is the heavy load. Would timing the decode of a single 2 minute wav file by wsprd be a good way to compare.
-Jim |
|
|
|
Rob Robinett
Yes. On Mon, Apr 11, 2022 at 11:04 PM Jim Lill <jim@...> wrote:
--
|
|
|
|
Jim Lill
Here are some results, timed to the nearest second: XU4 5 secs 10W needs fan for 26 channels, max reliable channel
count Atomic Pi 3 secs 15W no fan channel count TBD Intel© Core™ i5-9600K CPU @ 3.70GHz × 6 core, Ubuntu 1 second many Watts channel count TBD -Jim On 4/12/22 11:25, Rob Robinett wrote:
|
|
|
|
Phil Karn
I wrote the Fano decoder in wsprd. The Fano decoder is one of a family of 'sequential' decoding algorithms where decoding time increases as SNR decreases. More specifically, it is a Pareto-distributed random variable with a "tail" that becomes unbounded as the SNR decreases towards something called the "computational cutoff limit". Many blocks decode quickly but a few can take a really long time. That's why the decoder has a configurable limit on decoding time.
toggle quoted message
Show quoted text
Sequential decoding algorithms do a lot of data-dependent branching that can really impair the branch prediction algorithms in modern, deeply pipelined CPUs. All this led to the development of more efficient codes for low SNRs. So at the very least make sure you use the same parameters and test input file(s) for all runs to make the results comparable. On 4/12/22 02:04, Jim Lill wrote:
|
|
|
|
Jim Lill
Relative to Phil's comments... same commands and wav file used
for all. SNR was modest for those sigs decoded On 4/12/22 12:23, Jim Lill wrote:
|
|
|