Table update with bitcrack key/rate per sec:
1. RTX 2080ti 1,200-1,300 Mkey/sec (price $1,200-1,500)
2. RTX 2080 810 Mkey/s | 300 Mkey/s (2M addresses)
3. GTX 2070ti 805 Mkey/sec
4. RTX 2060 620 Mkey/sec
5. GTX 1080ti 345 Mkey/sec (price $700-1,000)
6. Tesla K80 250 Mkey/sec | 100 Mkes/s (2M addresses)
7. GTX 1080 130 Mkey/sec (price $500)
8. GTX 1070ti 100 Mkey/sec
9. GTX 1070 80-90 Mkey/sec (price $450-550)
10. GTX 1060 69 Mkey/sec
11. GTX 1050ti 64 Mkey/sec (price $150-200)
12. GTX 980 70-80 Mkey/sec
13. GTX 770(2Gb) 120 MKey/sec
14. GTX 680 109 Mkey/sec (price $150)
15. GT 640 9 Mkey/sec
16. RX 480 107 Mkey/sec (price $130-150)
17. RX 470 105 Mkey/sec (price $150-300)
18. RX 580 89 Mkey/sec (price $200-250)
19. RX 560 50 Mkey/sec (price $100-150)
20. R9 280/290x 20 Mkey/sec (price $50)
For Nvidia GTX 1050Ti (4gb) on Ubuntu 16.04 and Cuda 10.1, with the parameters that you can find on this topic (-b 500 -t 128 -p 128) you can get 69 MKey/sec with 15 targets.
Why 15 ? Because it's the limit in bitcrack between classic comparison and bloom filter to find adresses.
With the parameters : -b 300 -t 128 -p 512 -> 83 mKey/sec.
Blocks must be a multiple of Streaming Multiprocessors (SM) count.
Threads must be a multiple of 32. The true count in each SM is better...
GTX 1050Ti have 6 SM with 4 units of 32 ROPs -> 6*4*32 = 768 cuda core !