Benchmarks

Every number on this page is reproducible. No simulated hardware, no fabricated results.

Reproducibility guarantee: All Python results can be verified by running python -m atomik_core benchmark on your own machine. FPGA results are from Gowin/Vivado synthesis reports and hardware test suites (80/80 parallel bank tests, 9/9 ATOMiK core tests, 6/6 display pipeline tests). Synthesis reports and test logs are in the source repository.

Verified Results

AMD Ryzen 7 5700U (8 GB)

Verified

Python 3.12 / Kubuntu 24.04

Throughput (ACCUM)~5 M ops/s
Rollback (10k undo)~140x faster
Change detection~330,000x less memory
Bandwidth (1 MB state)131,072x reduction

Methodology: python -m atomik_core benchmark

Developer's machine. Run the same command on your hardware to compare.

Tang Nano 9K (GW1NR-9K, $13.50)

Verified

Single bank, N=1, 81 MHz core clock

Throughput94.5 Mops/s
LOAD latency64 cycles
ACCUM latency70 cycles
READ latency99 cycles
LUT usage477 (5.5%)

Methodology: Synthesis reports + 80/80 hardware tests

Production SoC: PicoRV32 @ 25.2 MHz + ATOMiK @ 81 MHz. Deterministic latency (stdev < 0.5 cycles).

Tang Nano 9K — 16 banks

Verified

N=16 parallel banks, 66 MHz

Throughput1,056 Mops/s
LUT usage1,779 (20.6%)
LUT scaling3.7x for 16x throughput

Methodology: Synthesis sweep + 80/80 hardware tests

Sub-linear LUT scaling: ~65 LUT + 64 FF per additional bank.

Xilinx Zynq XC7Z020

Projected

N=512 parallel banks, 135.6 MHz

Throughput69.7 Gops/s
LUT usage23,542 (44.3%)
Fmax (N=1)444.4 MHz
Marginal LUT/bank~34

Methodology: Vivado synthesis ceiling characterization (6 configs, 4 strategies each)

Synthesis-validated. Board integration pending (AX7020 in hand). N=512 is the XC7Z020 placement limit.

Hardware acceleration tiers

Same algebra at every tier. Code once, accelerate anywhere.

~5M ops/s
Python SDK
pip install, zero deps
~500M ops/s
C Library
Single-header, any compiler
69.7G ops/s
FPGA (Zynq N=512)
512 parallel banks, synthesis-validated

Run it yourself

1
Install
pip install atomik-core
2
Benchmark
python -m atomik_core benchmark
3
JSON output
python -m atomik_core benchmark --json

The benchmark suite runs five tests: rollback, change detection, multi-node convergence, bandwidth, and raw throughput. All operations are O(1) -- constant time regardless of state history.

Run the benchmark on your machine and share your results

pip install atomik-core && python -m atomik_core benchmark