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

Benchmark results for cscal 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	0.0836	0.0035	0.1475	56.6%
64	0.0061	0.1675	0.0034	0.3019	55.5%
128	0.0060	0.3413	0.0036	0.5740	59.5%
512	0.0061	1.3368	0.0040	2.0317	65.8%
1024	0.0061	2.6947	0.0034	4.8302	55.8%
4096	0.0061	10.6945	0.0038	17.3559	61.6%
16384	0.0065	40.5545	0.0042	62.2966	65.1%
65536	0.0082	128.0000	0.0068	153.4801	83.4%
262144	0.0282	148.6077	0.0272	154.0211	96.5%
1048576	0.1018	164.7409	0.0992	169.1252	97.4%
4194304	0.3895	172.3073	0.3877	173.1038	99.5%
16777216	1.5458	173.6538	1.5646	171.5655	101.2%

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

See also

• cscal.js — WebGPU benchmark script
• cscal.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	0.0060	0.0851	0.0034	0.1509	56.4%
64	0.0060	0.1698	0.0032	0.3200	53.1%
128	0.0060	0.3404	0.0033	0.6124	55.6%
512	0.0061	1.3403	0.0034	2.3814	56.3%
1024	0.0061	2.6667	0.0036	4.5714	58.3%
4096	0.0065	10.0887	0.0039	16.7869	60.1%
16384	0.0096	27.2159	0.0061	42.6667	63.8%
65536	0.0275	38.1911	0.0268	39.1962	97.4%
262144	0.1011	41.4785	0.0987	42.5006	97.6%
1048576	0.3901	43.0062	0.3869	43.3600	99.2%
4194304	1.5438	43.4697	1.5401	43.5754	99.8%

Nvidia Geforce Gtx 1650 — stride = 5

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0061	0.0833	0.0027	0.1905	43.7%
64	0.0061	0.1675	0.0029	0.3516	47.7%
128	0.0061	0.3351	0.0027	0.7619	44.0%
512	0.0061	1.3333	0.0031	2.6256	50.8%
1024	0.0061	2.6667	0.0029	5.6889	46.9%
4096	0.0066	9.9902	0.0033	19.8835	50.2%
16384	0.0101	25.9241	0.0060	43.8075	59.2%
65536	0.0343	30.5672	0.0312	33.6082	91.0%
262144	0.1208	34.7119	0.1183	35.4632	97.9%
1048576	0.6553	25.6019	0.4653	36.0558	71.0%

Nvidia Geforce Gtx 1650 — stride = 32

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0118	0.0435	0.0027	0.1905	22.8%
64	0.0124	0.0829	0.0026	0.3902	21.2%
128	0.0125	0.1641	0.0028	0.7232	22.7%
512	0.0142	0.5779	0.0035	2.3379	24.7%
1024	0.0149	1.0964	0.0035	4.7407	23.1%
4096	0.0211	3.1101	0.0039	16.7869	18.5%
16384	0.0503	5.2145	0.0197	13.2771	39.3%
65536	0.4833	2.1695	0.0777	13.4903	16.1%
262144	0.3174	13.2136	0.3174	13.2142	100.0%

Nvidia Geforce Gtx 1650 — stride = 33

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
64	0.0072	0.1429	0.0033	0.3137	45.5%
128	0.0071	0.2870	0.0028	0.7399	38.8%
512	0.0078	1.0449	0.0032	2.5473	41.0%
1024	0.0080	2.0562	0.0032	5.0443	40.8%
4096	0.0100	6.5327	0.0037	17.5794	37.2%
16384	0.0256	10.2400	0.0227	11.5625	88.6%
65536	0.1059	9.9057	0.0901	11.6322	85.2%
262144	0.4256	9.8543	0.3851	10.8904	90.5%

Nvidia Geforce Gtx 1650 — stride = 255

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
512	0.0078	1.0557	0.0032	2.5600	41.2%
1024	0.0080	2.0480	0.0036	4.6126	44.4%
4096	0.0100	6.5327	0.0039	16.9256	38.6%
16384	0.0367	7.1452	0.0350	7.4881	95.4%
65536	1.6196	0.6474	0.1412	7.4287	8.7%

Nvidia Geforce Gtx 1650 — stride = 256

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
512	0.0118	0.6966	0.0041	2.0157	34.6%
1024	0.0121	1.3527	0.0038	4.3390	31.2%
4096	0.0160	4.0878	0.0039	16.6504	24.6%
16384	0.3381	0.7754	0.0326	8.0511	9.6%
65536	0.1372	7.6418	0.1339	7.8289	97.6%

See also:

• stride.cscal.js — WebGPU stride-sweep benchmark script
• stride.cscal.c — CUDA / cuBLAS stride-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 = 0+0i

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
1024	0.0071	2.2960	0.0027	6.0952	37.7%
65536	0.0102	102.4000	0.0051	205.4420	49.8%
1048576	0.1021	164.3278	0.1004	167.1837	98.3%
16777216	1.5462	173.6053	1.5499	173.1949	100.2%

Nvidia Geforce Gtx 1650 — alpha = 1+0i

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
1024	0.0061	2.6667	0.0027	6.1687	43.2%
65536	0.0082	128.0000	0.0053	198.5939	64.5%
1048576	0.1012	165.7828	0.1004	167.1837	99.2%
16777216	1.5462	173.6107	1.5658	171.4392	101.3%

Nvidia Geforce Gtx 1650 — alpha = 2.5-3.75i

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
1024	0.0061	2.6667	0.0026	6.2439	42.7%
65536	0.0082	127.5019	0.0056	186.7122	68.3%
1048576	0.1011	166.0190	0.1004	167.1837	99.3%
16777216	1.5462	173.6125	1.5658	171.4375	101.3%

Nvidia Geforce Gtx 1650 — alpha = 0+1i

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
1024	0.0061	2.6877	0.0026	6.2439	43.0%
65536	0.0082	127.2544	0.0051	204.1620	62.3%
1048576	0.1013	165.5995	0.1004	167.1837	99.1%
16777216	1.5459	173.6412	1.5658	171.4375	101.3%

Nvidia Geforce Gtx 1650 — alpha = 1e-38+0i

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
1024	0.0061	2.6667	0.0026	6.2439	42.7%
65536	0.0082	128.0000	0.0058	180.0439	71.1%
1048576	0.1014	165.4949	0.1004	167.1837	99.0%
16777216	1.5462	173.6053	1.5664	171.3727	101.3%

See also:

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