kaivalnp commented on PR #15742:
URL: https://github.com/apache/lucene/pull/15742#issuecomment-3940097582
Latest JMH benchmarks on AWS Graviton3:
```
Benchmark (size)
Mode Cnt Score Error Units
VectorUtilBenchmark.binaryHalfByteDotProductBothPackedVector 1024
thrpt 15 18.930 ± 0.037 ops/us
VectorUtilBenchmark.binaryHalfByteDotProductSinglePackedVector 1024
thrpt 15 21.819 ± 0.085 ops/us
VectorUtilBenchmark.binaryHalfByteDotProductVector 1024
thrpt 15 26.335 ± 0.196 ops/us
VectorUtilBenchmark.binaryHalfByteSquareBothPackedVector 1024
thrpt 15 16.055 ± 0.030 ops/us
VectorUtilBenchmark.binaryHalfByteSquareSinglePackedVector 1024
thrpt 15 16.465 ± 0.063 ops/us
VectorUtilBenchmark.binaryHalfByteSquareVector 1024
thrpt 15 18.935 ± 0.146 ops/us
```
..which is a slight improvement from earlier
`knnPerfTest.py` results on Cohere v3 vectors, 1024d, `dot_product`
similarity:
Baseline with `-reindex`
```
recall latency(ms) netCPU avgCpuCount nDoc topK fanout maxConn
beamWidth quantized visited index(s) index_docs/s force_merge(s)
num_segments index_size(MB) filterStrategy filterSelectivity overSample
vec_disk(MB) vec_RAM(MB) bp-reorder indexType
0.918 2.806 2.803 0.999 500000 100 100 64
250 4 bits 8223 93.42 5351.94 141.31 1
2276.79 null N/A 1.000 2204.895
251.770 false HNSW
```
Candidate with `-reindex`
```
recall latency(ms) netCPU avgCpuCount nDoc topK fanout maxConn
beamWidth quantized visited index(s) index_docs/s force_merge(s)
num_segments index_size(MB) filterStrategy filterSelectivity overSample
vec_disk(MB) vec_RAM(MB) bp-reorder indexType
0.918 2.655 2.654 1.000 500000 100 100 64
250 4 bits 8220 96.46 5183.39 131.73 1
2276.76 null N/A 1.000 2204.895
251.770 false HNSW
```
Baseline search-only, use existing index
```
recall latency(ms) netCPU avgCpuCount nDoc topK fanout maxConn
beamWidth quantized visited index(s) index_docs/s force_merge(s)
num_segments index_size(MB) filterStrategy filterSelectivity overSample
vec_disk(MB) vec_RAM(MB) bp-reorder indexType
0.918 2.666 2.665 0.999 500000 100 100 64
250 4 bits 8220 96.46 5183.39 131.73 1
2276.76 null N/A 1.000 2204.895
251.770 false HNSW
```
Candidate search-only, use existing index
```
recall latency(ms) netCPU avgCpuCount nDoc topK fanout maxConn
beamWidth quantized visited index(s) index_docs/s force_merge(s)
num_segments index_size(MB) filterStrategy filterSelectivity overSample
vec_disk(MB) vec_RAM(MB) bp-reorder indexType
0.918 2.540 2.539 0.999 500000 100 100 64
250 4 bits 8220 96.46 5183.39 131.73 1
2276.76 null N/A 1.000 2204.895
251.770 false HNSW
```
Seems like there's a small improvement with this PR
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
