Module benchmarks/nvidia-geforce-gtx-1650/ssyr2k

Benchmark results for ssyr2k on Nvidia Geforce Gtx 1650.

Nvidia Geforce Gtx 1650 — wgblas vs cuBLAS

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0164	0.7500	0.0161	0.7649	98.1%
64	0.0226	2.1772	0.0225	2.1818	99.8%
128	0.0424	4.6387	0.0326	6.0235	77.0%
256	0.1636	4.8080	0.0549	14.3259	33.6%
512	0.5920	5.3140	0.1823	17.2569	30.8%
1024	4.2568	2.9560	1.0230	12.3005	24.0%

Efficiency = wgblas GB/s ÷ cuBLAS GB/s × 100. 100% means parity with cuBLAS; values above 100% mean wgblas achieved greater throughput.

ssyr2k-default GB/s chart

ssyr2k-default ms chart

See also

ssyr2k.js — WebGPU benchmark script
ssyr2k.c — CUDA / cuBLAS reference script

Transpose sweep

Unless noted otherwise, every result above uses trans = "no-transpose". trans = "transpose" reads A with a cross-thread lda-strided mirror pattern instead of a coalesced one, and the gap grows with n — collapsed below by default, expand a trans value to see its table and chart.

Nvidia Geforce Gtx 1650 — trans = no-transpose

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
256	0.2584	3.0437	0.0631	12.4688	24.4%
512	0.8652	3.6360	0.2111	14.9036	24.4%
1024	4.9046	2.5655	1.1854	10.6145	24.2%
2048	37.0668	1.3579	7.6984	6.5379	20.8%

ssyr2k-transno-transpose GB/s chart

ssyr2k-transno-transpose ms chart

Nvidia Geforce Gtx 1650 — trans = transpose

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
256	0.1264	6.2233	0.0600	13.1037	47.5%
512	0.4708	6.6812	0.1867	16.8531	39.6%
1024	3.4580	3.6387	1.0238	12.2905	29.6%
2048	24.3519	2.0669	7.0502	7.1390	29.0%

ssyr2k-transtranspose GB/s chart

ssyr2k-transtranspose ms chart

See also:

trans.ssyr2k.js — WebGPU trans-sweep benchmark script
trans.ssyr2k.c — CUDA / cuBLAS trans-sweep reference script

Lda sweep

Unless noted otherwise, every result above uses a tight lda (no padding). Padding the row stride only matters for trans = "transpose" here (swept at both trans values below so that's visible in the data, not just claimed). Collapsed below by default — expand a trans value, then a pad, to see its table and chart.

Nvidia Geforce Gtx 1650 — trans = no-transpose (6 pads)

pad = 0

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
512	0.8646	3.6385	0.2096	15.0094	24.2%
1024	4.9411	2.5466	1.1857	10.6121	24.0%
2048	37.0573	1.3582	6.5787	7.6507	17.8%

ssyr2k-lda-no-transpose-pad0 GB/s chart

ssyr2k-lda-no-transpose-pad0 ms chart

pad = 1

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
512	0.5864	5.3646	0.1825	17.2388	31.1%
1024	3.7044	3.3967	0.9371	13.4279	25.3%
2048	35.1422	1.4322	6.6347	7.5862	18.9%

ssyr2k-lda-no-transpose-pad1 GB/s chart

ssyr2k-lda-no-transpose-pad1 ms chart

pad = 8

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
512	0.5527	5.6919	0.1813	17.3483	32.8%
1024	3.3584	3.7467	0.9333	13.4827	27.8%
2048	27.9512	1.8007	6.6158	7.6078	23.7%

ssyr2k-lda-no-transpose-pad8 GB/s chart

ssyr2k-lda-no-transpose-pad8 ms chart

pad = 16

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
512	0.6431	4.8917	0.1761	17.8605	27.4%
1024	3.7754	3.3329	0.9320	13.5010	24.7%
2048	30.7421	1.6372	6.6254	7.5968	21.6%

ssyr2k-lda-no-transpose-pad16 GB/s chart

ssyr2k-lda-no-transpose-pad16 ms chart

pad = 32

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
512	0.8645	3.6389	0.1823	17.2524	21.1%
1024	4.9235	2.5557	0.9318	13.5033	18.9%
2048	38.3435	1.3127	6.5718	7.6587	17.1%

ssyr2k-lda-no-transpose-pad32 GB/s chart

ssyr2k-lda-no-transpose-pad32 ms chart

pad = 64

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
512	0.8648	3.6375	0.1760	17.8718	20.4%
1024	4.9849	2.5242	0.9393	13.3959	18.8%
2048	38.8478	1.2956	6.5805	7.6486	16.9%

ssyr2k-lda-no-transpose-pad64 GB/s chart

ssyr2k-lda-no-transpose-pad64 ms chart

Nvidia Geforce Gtx 1650 — trans = transpose (6 pads)

pad = 0

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
512	0.4688	6.7095	0.1941	16.2084	41.4%
1024	3.4693	3.6269	1.0325	12.1873	29.8%
2048	24.4449	2.0590	7.0898	7.0991	29.0%

ssyr2k-lda-transpose-pad0 GB/s chart

ssyr2k-lda-transpose-pad0 ms chart

pad = 1

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
512	0.4643	6.7747	0.1960	16.0496	42.2%
1024	3.4437	3.6539	1.0780	11.6721	31.3%
2048	24.6881	2.0387	7.3393	6.8579	29.7%

ssyr2k-lda-transpose-pad1 GB/s chart

ssyr2k-lda-transpose-pad1 ms chart

pad = 8

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
512	0.4628	6.7965	0.1916	16.4196	41.4%
1024	3.3726	3.7309	1.0240	12.2886	30.4%
2048	24.0245	2.0950	7.1687	7.0211	29.8%

ssyr2k-lda-transpose-pad8 GB/s chart

ssyr2k-lda-transpose-pad8 ms chart

pad = 16

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
512	0.4688	6.7104	0.1907	16.4967	40.7%
1024	3.4218	3.6773	1.0206	12.3286	29.8%
2048	24.3062	2.0707	7.0943	7.0947	29.2%

ssyr2k-lda-transpose-pad16 GB/s chart

ssyr2k-lda-transpose-pad16 ms chart

pad = 32

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
512	0.4731	6.6494	0.1959	16.0614	41.4%
1024	3.4943	3.6010	1.0459	12.0303	29.9%
2048	24.7647	2.0324	7.2645	6.9284	29.3%

ssyr2k-lda-transpose-pad32 GB/s chart

ssyr2k-lda-transpose-pad32 ms chart

pad = 64

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
512	0.4733	6.6467	0.2018	15.5865	42.6%
1024	3.5001	3.5951	1.0926	11.5169	31.2%
2048	24.7978	2.0297	7.2059	6.9848	29.1%

ssyr2k-lda-transpose-pad64 GB/s chart

ssyr2k-lda-transpose-pad64 ms chart

See also:

lda.ssyr2k.js — WebGPU lda-sweep benchmark script
lda.ssyr2k.c — CUDA / cuBLAS lda-sweep reference script