Update
As promised @Nuh - See below updated benchmark.
Block 962,000: 6,303 transactions, 8,006 inputs, split into 16 chunks of 501, proved on a single idle L40S. Guest mode 4 (chunk_prove) reads only the block and its prevouts (No accumulator, no bridge). This is why a block this close to the tip can be measured without having processed the chain up to it.
Per chunk:
- chunk 0 | 1314s | 2.62s per input
- chunk 1 | 1931s | 3.85s per input
- chunk 2 | 7660s | 15.29s per input
- chunk 3 | 6814s | 13.60s per input
- chunk 4 | 1963s | 3.92s per input
- chunk 5 | 293s | 0.58s per input
- chunk 6 | 292s | 0.58s per input
- chunk 7 | 591s | 1.18s per input
- chunk 8 | 1194s | 2.38s per input
- chunk 9 | 1195s | 2.39s per input
- chunk 10 | 1193s | 2.38s per input
- chunk 11 | 1195s | 2.39s per input
- chunk 12 | 1197s | 2.39s per input
- chunk 13 | 1190s | 2.38s per input
- chunk 14 | 1196s | 2.39s per input
- chunk 15 | 1167s | 2.38s per input
- total | 30385s | 3.80s per input
Aggregating those 16 chunk receipts into one block proof took a further 4,606s.
Compute:
- Proving - 30,385s / 8.44 hours / 86.8%
- Aggregation - 4,606s / 1.28 hours / 13.2%
- TOTAL - 34,991s / 9.72 hours / 4.37s per input