Pipe is -very- slow

[Mem5]

OK, new try with :
- wordlist : 38M words (more than the 10M advised in the FAQ)
- [amplifier] rules list of 2000 rules applied on GPU (more than the 1000 advised in the FAQ)

Give :

- Straigh use of cudahashcat :
"ATTENTION! The wordlist or mask you are using is too small....."
Duration : 1 minute 10 seconds

- piping hashcat :
hashcat-cli64.exe --stdout *.dic | cudaHashcat64.exe -w 3 -r 2k.rule hash.txt
Duration : ~3 minutes

For fast hashes such as MD5, it is crucial to expand the candidates on the GPU with rules or masks in order to achieve full acceleration

It seems that I did what the FAQ advise, why can't I get full speed ?

For fast hashes, I would rather not using pipe, right ?