Yes something not ok, try setting by hand for example 120/16 2 threads
I really hurried to push the v8 version, so of course bugs are to be expected, and we still have time to fix them until the fork
Hey Doc,
Vega 64 - Cryptonight v8 analysis:
Using binary kernals : 583 h/s
Not Using Binary Kernals, Kernals compiled on each run and deleted
Driver: 18.5.1 GPU: 1600 Mem: 1100 Volt : 840mV
Intensity : 119 Worksize : 16 : 2019 H/s (94.338% on cn7)
Intensity : 119 Worksize : 32 : 1963 H/s (91.745% on cn7)
Intensity : 120 Worksize : 16 : 2025 H/s (94.626% on cn7)
Intensity : 120 Worksize : 32 : 1983 H/s (92.663% on cn7)
Intensity : 121 Worksize : 16 : 2032 H/s (94.953% on cn7)
Intensity : 121 Worksize : 32 : 1975 H/s (92.289% on cn7)
Intensity : 122 Worksize : 16 : 2038 H/s (95.234% on cn7)
Intensity : 122 Worksize : 32 : 1992 H/s (93.099% on cn7)
Intensity : 123 Worksize : 16 : 2043 H/s (95.467% on cn7)
Intensity : 123 Worksize : 32 : 1984 H/s (92.710% on cn7)
Intensity : 124 Worksize : 16 : 2049 H/s (95.748% on cn7)
Intensity : 124 Worksize : 32 : 2000 H/s (93.458% on cn7)
Intensity : 125 Worksize : 16 : Failed
Intensity : 125 Worksize : 32 : Failed
The above are avg HR from a run of 6 Vega 64s and the highest HR setting were tries multiple times for consistency.
Great work doc, so thats a 4.25% drop from v7 !! not bad considering devs said 7% drop
That's my man, i really love it when someone takes time and tries himself something
So it looks like the binary kernel is faulty ? I will check that out.
Thank you for your efforts
edit:
And YES, the binary IS faulty. Great that i pushed the 1.6.8 version 10 days before the fork so we can smooth these kind of things out