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

Benchmark results for sger 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	0.0061	1.3750	0.0035	2.3891	57.6%
64	0.0061	5.4167	0.0038	8.6667	62.5%
128	0.0065	20.4356	0.0038	34.8354	58.7%
256	0.0082	64.2500	0.0045	117.0676	54.9%
512	0.0157	133.7352	0.0125	167.9386	79.6%
1024	0.0563	149.2605	0.0557	150.6315	99.1%
1280	0.0898	146.1130	0.0874	150.0165	97.4%
2048	0.2165	155.0415	0.2137	157.1197	98.7%
4096	0.8478	158.3591	0.8479	158.3382	100.0%

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

See also

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

Stride sweep

Unless noted otherwise, every result above uses unit stride (incx = incy = 1) — the normal case, and the coalesced, best-case GPU access pattern. Real usage sometimes passes a non-unit stride (e.g. operating on a row or column of a larger matrix, where incx = lda), which breaks memory coalescing and costs measurably more. This section sweeps a few representative strides to characterize that cost separately, collapsed below by default — expand a stride to see its table and chart.

Nvidia Geforce Gtx 1650 — stride = 4

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0061	1.3750	0.0035	2.4000	57.3%
64	0.0063	5.3197	0.0039	8.5246	62.4%
128	0.0066	19.9420	0.0038	34.8354	57.2%
256	0.0082	64.2500	0.0046	113.8270	56.4%
512	0.0164	128.2500	0.0142	148.2257	86.5%
1024	0.0570	147.2503	0.0562	149.3030	98.6%
1280	0.0901	145.5682	0.0877	149.5513	97.3%
2048	0.2180	154.0285	0.2137	157.1197	98.0%
4096	0.8485	158.2247	0.8482	158.2784	100.0%

Nvidia Geforce Gtx 1650 — stride = 32

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0061	1.3750	0.0044	1.9412	70.8%
64	0.0064	5.2261	0.0039	8.5950	60.8%
128	0.0080	16.4790	0.0039	34.1157	48.3%
256	0.0131	40.0682	0.0048	110.3893	36.3%
512	0.0358	58.6810	0.0172	122.3933	47.9%
1024	0.1434	58.5714	0.0566	148.3324	39.5%
1280	0.2238	58.6228	0.0879	149.1703	39.3%
2048	0.5704	58.8549	0.2146	156.4169	37.6%
4096	2.2586	59.4395	0.9700	138.4071	42.9%

Nvidia Geforce Gtx 1650 — stride = 256

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0061	1.3750	0.0035	2.4220	56.8%
64	0.0065	5.0980	0.0034	9.8113	52.0%
128	0.0080	16.4135	0.0036	36.3700	45.1%
256	0.0136	38.7012	0.0047	111.1351	34.8%
512	0.0362	58.0327	0.0173	121.6000	47.7%
1024	0.1434	58.5584	0.0567	148.1649	39.5%
1280	0.2248	58.3433	0.0881	148.9535	39.2%
2048	0.5704	58.8582	0.2149	156.1840	37.7%
4096	2.2733	59.0559	1.0057	133.4944	44.2%

See also:

• stride.sger.js — WebGPU stride-sweep benchmark script
• stride.sger.c — CUDA / cuBLAS stride-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	0.0061	1.3750	0.0042	2.0000	68.8%
64	0.0061	5.4167	0.0039	8.5950	63.0%
128	0.0065	20.2850	0.0038	34.6891	58.5%
256	0.0081	64.7559	0.0047	112.2730	57.7%
512	0.0158	132.9231	0.0171	122.8513	108.2%
1024	0.0563	149.2605	0.0558	150.3725	99.3%
1280	0.0898	146.0349	0.0875	149.9616	97.4%
2048	0.2162	155.2709	0.2133	157.3789	98.7%
4096	0.8457	158.7396	0.8480	158.3083	100.3%

Nvidia Geforce Gtx 1650 — pad = 1

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0061	1.3750	0.0038	2.2373	61.5%
64	0.0061	5.4167	0.0036	9.2444	58.6%
128	0.0067	19.8462	0.0037	35.7403	55.5%
256	0.0082	64.2500	0.0050	105.7749	60.7%
512	0.0187	112.4384	0.0147	143.0588	78.6%
1024	0.1008	83.2752	0.0744	112.8845	73.8%
1280	0.1652	79.4265	0.1167	112.4146	70.7%
2048	0.3994	84.0615	0.2931	114.5355	73.4%
4096	1.5094	88.9434	1.1857	113.2220	78.6%

Nvidia Geforce Gtx 1650 — pad = 8

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0061	1.3750	0.0037	2.2957	59.9%
64	0.0061	5.4167	0.0039	8.6307	62.8%
128	0.0068	19.5640	0.0036	36.5310	53.6%
256	0.0082	64.2500	0.0053	99.3837	64.6%
512	0.0198	106.2524	0.0165	127.0097	83.7%
1024	0.1044	80.3922	0.0735	114.2857	70.3%
1280	0.1679	78.1098	0.1152	113.8192	68.6%
2048	0.3901	86.0543	0.2970	113.0422	76.1%
4096	1.4906	90.0652	1.1752	114.2394	78.8%

Nvidia Geforce Gtx 1650 — pad = 16

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0061	1.3750	0.0034	2.4558	56.0%
64	0.0061	5.4167	0.0033	10.0000	54.2%
128	0.0066	19.9420	0.0036	37.0224	53.9%
256	0.0082	64.2500	0.0045	116.6525	55.1%
512	0.0207	101.4900	0.0181	115.8095	87.6%
1024	0.1024	82.0000	0.0734	114.3604	71.7%
1280	0.1624	80.7485	0.1161	112.9879	71.5%
2048	0.4002	83.8901	0.2931	114.5417	73.2%
4096	1.4832	90.5131	1.1884	112.9644	80.1%

Nvidia Geforce Gtx 1650 — pad = 32

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0061	1.3750	0.0034	2.4673	55.7%
64	0.0061	5.4167	0.0034	9.8113	55.2%
128	0.0067	19.7512	0.0036	37.1892	53.1%
256	0.0082	64.2500	0.0045	116.2403	55.3%
512	0.0167	126.0345	0.0150	139.7106	90.2%
1024	0.0578	145.2935	0.0573	146.5922	99.1%
1280	0.0885	148.1460	0.0881	148.9535	99.5%
2048	0.2171	154.6529	0.2191	153.1963	101.0%
4096	0.8724	153.8779	0.8621	155.7315	98.8%

Nvidia Geforce Gtx 1650 — pad = 48

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0061	1.3750	0.0034	2.5143	54.7%
64	0.0061	5.4167	0.0035	9.4545	57.3%
128	0.0066	20.1366	0.0036	36.6933	54.9%
256	0.0082	64.2500	0.0045	116.6525	55.1%
512	0.0225	93.2727	0.0188	111.4839	83.7%
1024	0.1040	80.7260	0.0749	112.1607	72.0%
1280	0.1614	81.2608	0.1187	110.4906	73.5%
2048	0.3961	84.7542	0.2949	113.8333	74.5%
4096	1.5014	89.4164	1.1886	112.9461	79.2%

Nvidia Geforce Gtx 1650 — pad = 64

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0062	1.3608	0.0035	2.4332	55.9%
64	0.0061	5.4167	0.0034	9.9048	54.7%
128	0.0068	19.5177	0.0036	36.8571	53.0%
256	0.0082	64.2500	0.0045	115.8310	55.5%
512	0.0164	128.2500	0.0150	139.8594	91.7%
1024	0.0594	141.3793	0.0568	147.8726	95.6%
1280	0.0887	147.9055	0.0885	148.1996	99.8%
2048	0.2258	148.6697	0.2229	150.5904	98.7%
4096	0.8724	153.8779	0.8831	152.0184	101.2%

Nvidia Geforce Gtx 1650 — pad = 128

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0061	1.3750	0.0039	2.1909	62.8%
64	0.0063	5.3061	0.0036	9.2857	57.1%
128	0.0068	19.4717	0.0036	36.2105	53.8%
256	0.0082	64.2500	0.0046	114.2222	56.3%
512	0.0164	127.8754	0.0150	139.7106	91.5%
1024	0.0581	144.5730	0.0572	146.8383	98.5%
1280	0.0912	143.8821	0.0880	148.9806	96.6%
2048	0.2273	147.6757	0.2280	147.2404	100.3%
4096	0.9423	142.4729	0.9073	147.9729	96.3%

See also:

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

alpha sweep

alpha is a plain multiplier here: the kernel applies it unconditionally, with no branch for any particular value. A flat sweep is therefore the expected result and is recorded as a measured null. Levels include 0, 1 and a denormal-producing 1e-38 because those are the values a shader could special-case if it ever grew a branch — and strsm is the routine where one does.

Nvidia Geforce Gtx 1650 — alpha = -3.75

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0061	1.3750	0.0035	2.3891	57.6%
64	0.0062	5.3886	0.0035	9.4977	56.7%
128	0.0065	20.3350	0.0037	35.8956	56.7%
256	0.0081	64.7559	0.0045	117.4857	55.1%
512	0.0159	131.9879	0.0127	165.8182	79.6%
1024	0.0563	149.0486	0.0556	151.0650	98.7%
1280	0.0898	146.0870	0.0997	131.5322	111.1%
2048	0.2159	155.5126	0.2142	156.6972	99.2%
4096	0.8458	158.7215	0.8479	158.3382	100.2%

Nvidia Geforce Gtx 1650 — alpha = 0

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0075	1.1234	0.0014	6.0000	18.7%
64	0.0077	4.2975	0.0014	23.6364	18.2%
128	0.0082	16.1250	0.0014	91.7333	17.6%
256	0.0100	52.8875	0.0014	378.1149	14.0%
512	0.0184	114.0000	0.0015	1367.9999	8.3%
1024	0.0604	138.9463	0.0015	5704.3477	2.4%
1280	0.0930	140.9871	0.0017	7883.0767	1.8%
2048	0.2213	151.6681	0.0014	24397.3945	0.6%
4096	0.8561	156.8142	0.0016	85618.9453	0.2%

Nvidia Geforce Gtx 1650 — alpha = 1e-38

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0061	1.3750	0.0035	2.4332	56.5%
64	0.0061	5.4167	0.0036	9.1228	59.4%
128	0.0065	20.4356	0.0035	37.8716	54.0%
256	0.0082	64.2500	0.0046	115.0210	55.9%
512	0.0158	133.3279	0.0126	167.0840	79.8%
1024	0.0565	148.7107	0.0557	150.6315	98.7%
1280	0.0895	146.5046	0.0875	149.9890	97.7%
2048	0.2161	155.3514	0.2131	157.5325	98.6%
4096	0.8458	158.7276	0.8479	158.3382	100.2%

Nvidia Geforce Gtx 1650 — alpha = 1

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0067	1.2662	0.0040	2.1036	60.2%
64	0.0068	4.9289	0.0040	8.2540	59.7%
128	0.0071	18.6366	0.0041	32.2500	57.8%
256	0.0087	60.2491	0.0054	96.7530	62.3%
512	0.0165	127.5029	0.0134	156.3428	81.6%
1024	0.0573	146.4286	0.0563	149.0486	98.2%
1280	0.0901	145.5165	0.0879	149.2789	97.5%
2048	0.2173	154.4594	0.2138	157.0374	98.4%
4096	0.8455	158.7906	0.8490	158.1353	100.4%

Nvidia Geforce Gtx 1650 — alpha = 2.5

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0067	1.2692	0.0034	2.4673	51.4%
64	0.0069	4.8485	0.0037	9.0435	53.6%
128	0.0077	17.1642	0.0036	36.6933	46.8%
256	0.0087	60.2491	0.0046	115.0210	52.4%
512	0.0165	127.1326	0.0128	164.3654	77.3%
1024	0.0573	146.4286	0.0565	148.5003	98.6%
1280	0.0903	145.2587	0.0877	149.6332	97.1%
2048	0.2176	154.2890	0.2134	157.3317	98.1%
4096	0.8438	159.1068	0.8476	158.3860	100.5%

See also:

• alpha.sger.js — WebGPU alpha-sweep benchmark script
• alpha.sger.c — CUDA / cuBLAS alpha-sweep reference script

layout sweep

Column-major swaps the effective m/n and flips the transpose flag internally, changing which axis is contiguous and therefore how the matrix reads coalesce. wgblas-only: cuBLAS is column-major and has no layout argument, so there is no reference curve to compare against.

Nvidia Geforce Gtx 1650 — layout = column-major

n	compute ms	GB/s
32	0.0072	1.1707
64	0.0066	5.0363
128	0.0068	19.3803
256	0.0084	62.5399
512	0.0164	128.2500
1024	0.0572	146.8794
1280	0.0901	145.5682
2048	0.2170	154.6871
4096	0.8454	158.7936

Nvidia Geforce Gtx 1650 — layout = row-major

n	compute ms	GB/s
32	0.0072	1.1681
64	0.0073	4.5815
128	0.0078	16.8834
256	0.0100	52.8026
512	0.0184	114.0000
1024	0.0606	138.5794
1280	0.0925	141.7427
2048	0.2216	151.5147
4096	0.8581	156.4516

See also:

• layout.sger.js — WebGPU layout-sweep benchmark script