SP-mod private version 2 is soon ready. Here are som stats:
(faster than version private 1)
qubit:
gtx 970: +2.2% 12070 / 11810 (+260khash) +4.7% 12370 / 11810
gtx 960: +2.6% 7900 / 7730 (+170khash)
gtx 750ti:+2.5% 4860 / 4740 (+120khash)
x11
750ti: 3090/3045 +1.42% (+55KHASH)
970: 8130/7950 +2.22% (+180KHASH)
960: 5170/5119 +1% (+51KHASH)
x13
750ti: 2426/2400 +1.08% (+26KHASH)
970: 6550/6400 +2.3% (+150KHASH)
960: 4096/4080 +0.4% (+16KHASH)
x15
750ti: 2084/2060 +1.17% (+24KHASH)
970: 5645/5520 +2.22% (+75KHASH)
960: 3540 /3525 +0.42% (+15KHASH)
lyra2v2
750ti: 4630/4610 +0.43% (+20khash)
960,970 slower than private #1
reporting 0.5% increases is rather meaningless (It is within error measurement...)
I have optimized SIMD. one of the 15 algos in x15.
In lyra2v2 I have merged two kernals. Faster on compute 5.0 but slower on 5.2
The X series is not like neoscrypt or lyra2v2 where 90% of the work is done in the memory kernal.
You need to add the improvements in private #1 to get the total increase over my opensource.
release 1 sp-mod private (bether than release 78):
X11 is up 30khash (750ti) (+1%)
x13 is up 100khash(750ti) (+4.3%)
x15 is up 66Khash(750ti) (+3.3%)
quark is up 200khash(750ti) (+3.2%)
lyra2v2 is up 250khash(750ti)(+6%)
release 2 sp-mod private (bether than sp-mod private 1):
x11: +1.42% (+55KHASH(750ti))
x13: +1.08% (+26KHASH(750ti))
x15: +1.17% (+24KHASH(750ti))
lyra2v2: +0.43% (+20khash(750ti))
qubit:+2.5% (+120khash(750ti))