ABCDEFGHIJKLMNOPQRSTUVWXYZAAAB
1
Useful Life2
2
Electricty Rate$0.09
3
4
Example Sealing RigAmortized CapexPower/Unit (w)
5
ComponentSpecCostUnit CostHourly Cost/UnitPower (watts)
6
CPU32-core e.g. EPYC 7543$4,500$141$0.00807.0
7
Memory16 x 64GB$5,120$5$0.00030.2
8
NVMe5 x 3.4TB$2,600$153$0.0087
9
GPU2 Nvidia 2080Ti$2,000$1,000$0.0571250.0
10
Misc.Chassis, HDD, Networking, Etc.$5,500$5,500$0.3139
11
Total$19,720
12
13
Sealing Capacity15 Sectors x 4/day x 32GB1920
14
Capex/TB/Day$10,271
15
16
CapexOpEx
17
Sealing ProcessCore operationComputeStorageMemory (GB)I/OTime (h)
Compute Cost
Memory Cost
Everything ElsePowerTotal
18
PC1SHA Hashing for DRG2 CPU Cores384GB64GB2 GB/s4$0.0642$0.0731$0.0860$0.0105$0.2338
19
PC2Poseidon Hashing for Merkle tree1 GPU0.25$0.0143$0.0807$0.0056$0.1006
20
C1NA\
21
C2MSM and NTT for SNARK creation1 GPU0.25$0.0143$0.0807$0.0056$0.1006
22
23
Total Cost/32 GB$0.43
24
Total Cost/TB$13.92
25
26
27
Optimization #Opportunities for Cost ReductionBenchWho?
28
1Replace DDR with NVMe for PC1PC1NVMe Mfg.
29
2Improve SHA ThroughputPC1CPU + FPGA Mfg.
30
3Accelerate Merkle treePC2GPU/FPGA Mfg.
31
4Accelerate SNARKC2GPU/FPGA Mfg.
32
5Cost Optimization e.g. GPU vs. FPGA, different system configurations, etc.NAAll + OEMs
33
34
Benchmark Instructions
35
RUST_LOG=trace FIL_PROOFS_USE_MULTICORE_SDR=1 FIL_PROOFS_USE_GPU_COLUMN_BUILDER=1 FIL_PROOFS_USE_GPU_TREE_BUILDER=1 FIL_PROOFS_VERIFY_CACHE=0 FIL_PROOFS_MAXIMIZE_CACHING=1 RUSTFLAGS="-C target-cpu=native" cargo run --release --features cuda --bin benchy -- window-post --size 32GiB --cache ./cache_benchy_run_32g 2>&1|tee large_32G_4.log
36
37
38
39
40
41
42
43
44
45
46
47
48
49
174,980,149
50
51
4,576,000,000
52
52962.96296
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100