Module benchmarks/nvidia-geforce-gtx-1650/ssyr2

Benchmark results for ssyr2 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	0.0072	0.6236	0.0041	1.0938	57.0%
64	0.0072	2.3664	0.0041	4.1550	57.0%
128	0.0075	8.9957	0.0042	15.8788	56.7%
256	0.0082	32.3750	0.0051	52.4557	61.7%
512	0.0109	96.7988	0.0084	126.0421	76.8%
1024	0.0568	74.1015	0.0365	115.1608	64.3%
1280	0.0922	71.2778	0.0730	89.9759	79.2%
2048	0.2153	78.0288	0.1341	125.2818	62.3%
4096	0.8073	83.1855	0.5161	130.1270	63.9%

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

See also

• ssyr2.js — WebGPU benchmark script
• ssyr2.c — CUDA / cuBLAS reference script

Uplo sweep

Unless noted otherwise, every result above uses uplo = "lower". Real workgroups dispatch in increasing index order, so uplo = "upper" front-loads the heaviest rows first (worse — long-running heavy workgroups have nothing to overlap with) while lower back-loads them (better — light rows clear fast, the heavy tail gets full GPU to itself) — collapsed below by default, expand a uplo value to see its table and chart.

Nvidia Geforce Gtx 1650 — uplo = lower

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0068	0.6542	0.0041	1.0938	59.8%
64	0.0069	2.4758	0.0042	4.1231	60.0%
128	0.0073	9.1930	0.0042	15.8788	57.9%
256	0.0084	31.5133	0.0050	52.6222	59.9%
512	0.0121	87.0806	0.0084	125.8015	69.2%
1024	0.0369	114.1111	0.0366	114.8589	99.3%
1280	0.0543	121.0021	0.0731	89.8971	134.6%
2048	0.1210	138.8302	0.1341	125.3267	110.8%
4096	0.4444	151.1206	0.5161	130.1270	116.1%

Nvidia Geforce Gtx 1650 — uplo = upper

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0075	0.5996	0.0041	1.0980	54.6%
64	0.0076	2.2521	0.0041	4.1390	54.4%
128	0.0081	8.2846	0.0042	16.0000	51.8%
256	0.0085	31.0994	0.0049	53.8182	57.8%
512	0.0123	85.8333	0.0087	121.6236	70.6%
1024	0.0590	71.3078	0.0370	113.7163	62.7%
1280	0.0958	68.5752	0.0712	92.2192	74.4%
2048	0.2253	74.5765	0.1311	128.1562	58.2%
4096	0.8459	79.3965	0.5018	133.8236	59.3%

See also:

• uplo.ssyr2.js — WebGPU uplo-sweep benchmark script
• uplo.ssyr2.c — CUDA / cuBLAS uplo-sweep reference script

Lda sweep

Unless noted otherwise, every result above uses a tight lda (no padding). Padding the row stride changes throughput here — the exact mechanism and shape of that effect is routine-specific — collapsed below by default, expand a pad value to see its table and chart.

Nvidia Geforce Gtx 1650 — pad = 0

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0062	0.7179	0.0040	1.1200	64.1%
64	0.0064	2.6935	0.0041	4.1875	64.3%
128	0.0066	10.1256	0.0041	16.3750	61.8%
256	0.0079	33.4869	0.0049	53.9935	62.0%
512	0.0100	105.6410	0.0082	128.7500	82.1%
1024	0.0341	123.2015	0.0357	117.7921	104.6%
1280	0.0513	128.1398	0.0721	91.0736	140.7%
2048	0.1165	144.1669	0.1322	127.1322	113.4%
4096	0.4325	155.2801	0.5066	132.5641	117.1%

Nvidia Geforce Gtx 1650 — pad = 1

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0062	0.7179	0.0040	1.1290	63.6%
64	0.0063	2.7071	0.0041	4.1875	64.6%
128	0.0068	9.9336	0.0041	16.2481	61.1%
256	0.0081	32.6299	0.0050	52.9585	61.6%
512	0.0103	102.6791	0.0084	125.3232	81.9%
1024	0.0555	75.7453	0.0448	93.8301	80.7%
1280	0.0898	73.1184	0.0806	81.4765	89.7%
2048	0.2150	78.1333	0.1555	108.0473	72.3%
4096	0.8373	80.2036	0.7870	85.3333	94.0%

Nvidia Geforce Gtx 1650 — pad = 8

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0062	0.7179	0.0041	1.0938	65.6%
64	0.0063	2.7071	0.0041	4.1712	64.9%
128	0.0072	9.2949	0.0042	16.1231	57.6%
256	0.0082	32.3750	0.0052	50.8466	63.7%
512	0.0102	103.0000	0.0084	126.2835	81.6%
1024	0.0594	70.8276	0.0475	88.4928	80.0%
1280	0.0963	68.2447	0.0763	86.1255	79.2%
2048	0.2145	78.3198	0.1797	93.4929	83.8%
4096	0.8446	79.5138	0.7629	88.0304	90.3%

Nvidia Geforce Gtx 1650 — pad = 16

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0061	0.7292	0.0035	1.2785	57.0%
64	0.0061	2.7917	0.0038	4.5042	62.0%
128	0.0066	10.1256	0.0038	17.4667	58.0%
256	0.0077	34.2479	0.0041	64.7500	52.9%
512	0.0096	109.8667	0.0067	156.9524	70.0%
1024	0.0560	75.0534	0.0448	93.7967	80.0%
1280	0.0886	74.1083	0.0695	94.5120	78.4%
2048	0.2171	77.3848	0.1444	116.3172	66.5%
4096	0.7887	85.1498	0.7446	90.1963	94.4%

Nvidia Geforce Gtx 1650 — pad = 32

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0061	0.7292	0.0034	1.3084	55.7%
64	0.0061	2.7917	0.0036	4.7857	58.3%
128	0.0065	10.3251	0.0036	18.4670	55.9%
256	0.0077	34.5333	0.0041	64.2481	53.7%
512	0.0098	108.0656	0.0066	160.0000	67.5%
1024	0.0343	122.5126	0.0392	107.3110	114.2%
1280	0.0512	128.3000	0.0598	109.9224	116.7%
2048	0.1154	145.6466	0.1414	118.8447	122.6%
4096	0.4485	149.7458	0.6225	107.8878	138.8%

Nvidia Geforce Gtx 1650 — pad = 48

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0061	0.7292	0.0035	1.2903	56.5%
64	0.0061	2.7917	0.0036	4.7434	58.9%
128	0.0065	10.3251	0.0036	18.7982	54.9%
256	0.0077	34.5333	0.0041	64.7500	53.3%
512	0.0097	108.9587	0.0067	158.4615	68.8%
1024	0.0587	71.6772	0.0455	92.4120	77.6%
1280	0.0874	75.1941	0.0696	94.3382	79.7%
2048	0.2160	77.7918	0.1658	101.3426	76.8%
4096	0.8060	83.3226	0.8364	80.2987	103.8%

Nvidia Geforce Gtx 1650 — pad = 64

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0061	0.7292	0.0035	1.2727	57.3%
64	0.0061	2.7917	0.0036	4.7644	58.6%
128	0.0065	10.2745	0.0038	17.7627	57.8%
256	0.0077	34.6054	0.0041	64.7500	53.4%
512	0.0096	109.5017	0.0067	158.4615	69.1%
1024	0.0348	120.7126	0.0412	102.1414	118.2%
1280	0.0500	131.2532	0.0592	110.9622	118.3%
2048	0.1198	140.2955	0.1498	112.1435	125.1%
4096	0.4436	151.3877	0.5980	112.3014	134.8%

Nvidia Geforce Gtx 1650 — pad = 128

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0061	0.7292	0.0035	1.2670	57.6%
64	0.0063	2.7347	0.0037	4.6207	59.2%
128	0.0065	10.3251	0.0037	18.0690	57.1%
256	0.0075	35.2681	0.0041	64.7500	54.5%
512	0.0099	106.3226	0.0071	147.8027	71.9%
1024	0.0347	121.3813	0.0420	100.2716	121.1%
1280	0.0512	128.3000	0.0631	104.1238	123.2%
2048	0.1206	139.3091	0.1837	91.4572	152.3%
4096	0.4859	138.2080	0.6361	105.5731	130.9%

See also:

• lda.ssyr2.js — WebGPU lda-sweep benchmark script
• lda.ssyr2.c — CUDA / cuBLAS lda-sweep reference script