zkVM framework benchmark

[hero78119]

This issue is for tracking benchmark result in rolling update fasion

[hero78119]

Benchmark create_proof
Commits: 7b5ce9f

RAYON_NUM_THREADS=16 cargo bench --bench riscv_add --package ceno_zkvm on ¹:

• 2^20 add instances take [737.24 ms 740.09 ms 742.92 ms], ~1.41Mhz
• 2^21 add instances take [1.8537 s 1.8630 s 1.8716 s], ~768khz

RAYON_NUM_THREADS=8 cargo bench --bench riscv_add --package ceno_zkvm on ²:

• 2^20 add instances take [1.0532 s 1.0681 s 1.0801 s], ~981khz
• 2^21 add instances take [2.6562 s 2.6826 s 2.7102 s], ~782khz

Footnotes

1. AMD EPYC 9R14 (16 phy cores) + 64GB memory ↩

2. AMD R7 5800x (8 phy cores) + 32GB memory ↩

[hero78119]

Benchmark create_proof
Commits: a35d642

RAYON_NUM_THREADS=32 cargo bench --bench riscv_add --package ceno_zkvm on ¹:

• 2^20 add instances take 0.3870s, ~2.7Mhz
• 2^21 add instances take 0.9035s, ~2.3Mhz

Footnotes

1. AMD EPYC 9R14 (16 phy cores) + 64GB memory ↩

[hero78119]

Benchmark create_proof on master branch after #91 merge
Command: RAYON_NUM_THREADS=32 cargo bench --bench riscv_add --package ceno_zkvm
Commits: 0a40721

RAYON_NUM_THREADS=32 cargo bench --bench riscv_add --package ceno_zkvm on ¹:

• 2^20 add instances take 0.3870s, ~2.7Mhz
• 2^21 add instances take 0.9035s, ~2.3Mhz

more or less the same as prev result

Footnotes

1. AMD EPYC 9R14 (16 phy cores) + 64GB memory ↩

[hero78119]

command

RAYON_NUM_THREADS=32 cargo bench --bench riscv_add --package ceno_zkvm -- --save-baseline baseline

A throught put drop (+43.176%) on command between 2 commits on master, see comparison 5a68423...71b1aa2

Need investigate reason