at the moment my problem is not RAID but rather speed in blackmagic disk test that didn't showed speed increase when changing card from delock to sonnet (it should due to better switch on the card). and yes, my idea was to buy two additional blades and make RAID 0 out of 3 blades (4th is system and it, of course, stays separate)
I'm not sure why you would expect a speed increase without RAID.
The delock is x8 which is enough for a PCIe 2.0 slot to do PCIe 3.0 x4.
but I did what you told me to and here are results:
1. pcitree.sh gave me empty list (nothing) just #===================================. Ok, I don't have DirectHW.kext installed and I tried sudo nvram boot-args="debug=0x144" but again nothing
If you are using Open Core, then the boot-args need to go in the config.plist. A restart is required.
You also need to install pciutils.
Type
sudo lspci -A detect
to get a list of methods. The one we want is AppleACPIPlatformExpert, which doesn't require DirectHW.kext
but does require
debug=0x144
2. Amorphous gave similar results for both blades (first one is system disk) but kind of strange fall in later results
Those results look good. They indicate that the blades are running at PCIe 3.0 speed (I don't think there was much doubt about that since you have a MacPro5,1). You can just click the SEQ1M QD8 button to do the benchmark faster.
3. ATTO testing both blades in the same time also gave strange results. System disk is much slower in writing in higher MiB with really significant drop. I got the same result while testing only system disk in ATTO
You got good result at the 8MB block size. Maybe the rest are from throttling? Maybe need to wait awhile before trying again. You can change the settings of the ATTO benchmark to only include 8MB to 16MB block sizes to make the benchmark go faster.
You don't show the ATTO result for the total. The total should be near 6000 MB/s for x16 card maybe (guessing) 3500 MB/s for x8 card.