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

Benchmark results for daxpy 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	0.0080	0.0954	0.0028	0.2791	34.2%
64	0.0078	0.1979	0.0028	0.5455	36.3%
128	0.0075	0.4120	0.0029	1.0726	38.4%
512	0.0077	1.5934	0.0028	4.4138	36.1%
1024	0.0071	3.4830	0.0029	8.5333	40.8%
4096	0.0066	14.9854	0.0031	31.6701	47.3%
16384	0.0082	48.0000	0.0039	99.9024	48.0%
65536	0.0164	96.0000	0.0072	217.0066	44.2%
262144	0.0451	139.6364	0.0386	162.8898	85.7%
1048576	0.1564	160.8739	0.1455	172.9372	93.0%
4194304	0.6058	166.1681	0.5739	175.4058	94.7%
16777216	2.5907	155.4204	2.2885	175.9442	88.3%

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

See also

• daxpy.js — WebGPU benchmark script
• daxpy.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	0.0066	0.1171	0.0028	0.2727	42.9%
64	0.0080	0.1916	0.0031	0.4948	38.7%
128	0.0065	0.4706	0.0031	1.0000	47.1%
512	0.0065	1.9010	0.0033	3.7282	51.0%
1024	0.0065	3.7647	0.0033	7.3846	51.0%
4096	0.0077	12.8000	0.0037	26.8297	47.7%
16384	0.0118	33.4367	0.0066	59.9415	55.8%
65536	0.0425	36.9981	0.0405	38.7940	95.4%
262144	0.1502	41.8983	0.1549	40.6257	103.1%
1048576	0.5796	43.4205	0.6127	41.0734	105.7%
4194304	2.6846	37.4965	2.4372	41.3026	90.8%

Nvidia Geforce Gtx 1650 — stride = 5

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0081	0.0943	0.0028	0.2697	35.0%
64	0.0071	0.2157	0.0028	0.5486	39.3%
128	0.0073	0.4192	0.0028	1.0787	38.9%
512	0.0065	1.8916	0.0033	3.7647	50.2%
1024	0.0068	3.6226	0.0034	7.2113	50.2%
4096	0.0100	9.8147	0.0036	27.4286	35.8%
16384	0.0154	25.5202	0.0101	38.7634	65.8%
65536	0.0513	30.6338	0.0498	31.6090	96.9%
262144	0.1878	33.4937	0.1928	32.6375	102.6%
1048576	0.7358	34.2038	0.7653	32.8831	104.0%

Nvidia Geforce Gtx 1650 — stride = 32

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0082	0.0938	0.0029	0.2609	35.9%
64	0.0065	0.2353	0.0028	0.5455	43.1%
128	0.0066	0.4649	0.0030	1.0323	45.0%
512	0.0071	1.7376	0.0036	3.3982	51.1%
1024	0.0072	3.4362	0.0038	6.4810	53.0%
4096	0.0143	6.8571	0.0048	20.4800	33.5%
16384	0.0492	8.0000	0.0264	14.8675	53.8%
65536	0.1767	8.9035	0.1026	15.3265	58.1%
262144	0.6918	9.0942	0.4085	15.4015	59.0%

Nvidia Geforce Gtx 1650 — stride = 33

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
64	0.0064	0.2388	0.0029	0.5333	44.8%
128	0.0066	0.4660	0.0030	1.0267	45.4%
512	0.0069	1.7696	0.0036	3.4439	51.4%
1024	0.0073	3.3833	0.0037	6.6494	50.9%
4096	0.0143	6.8571	0.0049	20.1443	34.0%
16384	0.0519	7.5805	0.0305	12.8940	58.8%
65536	0.1884	8.3478	0.1195	13.1616	63.4%
262144	0.7738	8.1310	0.4854	12.9620	62.7%

Nvidia Geforce Gtx 1650 — stride = 255

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
512	0.0084	1.4713	0.0036	3.3982	43.3%
1024	0.0086	2.8444	0.0037	6.6207	43.0%
4096	0.0219	4.4912	0.0061	16.0000	28.1%
16384	0.0992	3.9652	0.0508	7.7356	51.3%
65536	0.4045	3.8886	0.1991	7.8991	49.2%

Nvidia Geforce Gtx 1650 — stride = 256

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
512	0.0084	1.4545	0.0038	3.2542	44.7%
1024	0.0089	2.7626	0.0038	6.4000	43.2%
4096	0.0187	5.2603	0.0046	21.4825	24.5%
16384	0.0840	4.6829	0.0486	8.0949	57.9%
65536	0.4272	3.6819	0.1935	8.1277	45.3%

See also:

• stride.daxpy.js — WebGPU stride-sweep benchmark script
• stride.daxpy.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 = -3.75

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
1024	0.0081	3.0296	0.0028	8.7273	34.7%
65536	0.0164	96.0000	0.0074	213.7043	44.9%
1048576	0.1550	162.3518	0.1456	172.8612	93.9%
16777216	2.6100	154.2742	2.2885	175.9442	87.7%

Nvidia Geforce Gtx 1650 — alpha = 0

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
1024	0.0062	3.9385	0.0030	8.3027	47.4%
65536	0.0143	109.7143	0.0050	313.0701	35.0%
1048576	0.1556	161.6842	0.0352	715.5887	22.6%
16777216	2.3295	172.8493	0.5197	774.7144	22.3%

Nvidia Geforce Gtx 1650 — alpha = 1e-38

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
1024	0.0072	3.3907	0.0029	8.5333	39.7%
65536	0.0143	109.7143	0.0071	222.9116	49.2%
1048576	0.1562	161.1046	0.1456	172.8612	93.2%
16777216	3.8175	105.4768	2.2885	175.9491	59.9%

Nvidia Geforce Gtx 1650 — alpha = 1

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
1024	0.0064	3.8400	0.0029	8.4862	45.2%
65536	0.0164	96.0000	0.0072	217.0066	44.2%
1048576	0.1559	161.4353	0.1455	172.9182	93.4%
16777216	2.5743	156.4105	2.2884	175.9553	88.9%

Nvidia Geforce Gtx 1650 — alpha = 2.5

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
1024	0.0079	3.1156	0.0029	8.4396	36.9%
65536	0.0164	96.0000	0.0072	219.4286	43.7%
1048576	0.1564	160.8739	0.1456	172.8802	93.1%
16777216	2.3238	173.2754	2.2883	175.9639	98.5%

See also:

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