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

Benchmark results for dscal 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	0.0079	0.0648	0.0033	0.1561	41.5%
64	0.0076	0.1339	0.0031	0.3333	40.2%
128	0.0076	0.2689	0.0033	0.6275	42.9%
512	0.0076	1.0734	0.0035	2.3704	45.3%
1024	0.0076	2.1649	0.0034	4.8075	45.0%
4096	0.0079	8.3083	0.0035	18.8756	44.0%
16384	0.0087	30.0073	0.0041	64.0000	46.9%
65536	0.0122	86.2316	0.0069	151.7037	56.8%
262144	0.0334	125.6683	0.0279	150.3119	83.6%
1048576	0.1059	158.4430	0.1002	167.3705	94.7%
4194304	0.4025	166.7384	0.3897	172.2153	96.8%
16777216	1.5551	172.6158	1.5483	173.3757	99.6%

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

See also

• dscal.js — WebGPU benchmark script
• dscal.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	0.0062	0.0821	0.0036	0.1441	56.9%
64	0.0061	0.1667	0.0034	0.3033	55.0%
128	0.0061	0.3333	0.0035	0.5872	56.8%
512	0.0061	1.3333	0.0038	2.1513	62.0%
1024	0.0061	2.6667	0.0038	4.3390	61.5%
4096	0.0067	9.7990	0.0042	15.7538	62.2%
16384	0.0095	27.5361	0.0062	42.0103	65.5%
65536	0.0288	36.3483	0.0270	38.7787	93.7%
262144	0.1022	41.0241	0.0991	42.3428	96.9%
1048576	0.3980	42.1555	0.3871	43.3439	97.3%
4194304	1.5762	42.5757	1.5388	43.6121	97.6%

Nvidia Geforce Gtx 1650 — stride = 5

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0063	0.0816	0.0028	0.1860	43.9%
64	0.0061	0.1667	0.0028	0.3721	44.8%
128	0.0062	0.3316	0.0028	0.7399	44.8%
512	0.0061	1.3333	0.0032	2.5990	51.3%
1024	0.0062	2.6256	0.0032	5.0693	51.8%
4096	0.0074	8.8276	0.0036	18.4505	47.8%
16384	0.0099	26.5974	0.0055	48.0469	55.4%
65536	0.0351	29.8706	0.0311	33.6773	88.7%
262144	0.1290	32.5079	0.1183	35.4680	91.7%
1048576	0.5041	33.2828	0.4649	36.0881	92.2%

Nvidia Geforce Gtx 1650 — stride = 32

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0063	0.0810	0.0028	0.1829	44.3%
64	0.0062	0.1658	0.0028	0.3678	45.1%
128	0.0063	0.3265	0.0028	0.7273	44.9%
512	0.0065	1.2549	0.0035	2.3273	53.9%
1024	0.0068	2.3925	0.0036	4.6126	51.9%
4096	0.0100	6.5223	0.0041	16.0000	40.8%
16384	0.0369	7.1111	0.0201	13.0550	54.5%
65536	0.1377	7.6125	0.0778	13.4737	56.5%
262144	0.5397	7.7716	0.3165	13.2510	58.6%

Nvidia Geforce Gtx 1650 — stride = 33

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
64	0.0062	0.1645	0.0028	0.3616	45.5%
128	0.0062	0.3282	0.0029	0.6995	46.9%
512	0.0066	1.2457	0.0033	2.4615	50.6%
1024	0.0068	2.4265	0.0035	4.7407	51.2%
4096	0.0092	7.1483	0.0041	16.0000	44.7%
16384	0.0392	6.6928	0.0227	11.5625	57.9%
65536	0.1541	6.8054	0.0900	11.6488	58.4%
262144	0.6237	6.7244	0.3858	10.8715	61.9%

Nvidia Geforce Gtx 1650 — stride = 255

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
512	0.0080	1.0261	0.0034	2.3814	43.1%
1024	0.0082	2.0000	0.0035	4.6335	43.2%
4096	0.0149	4.3854	0.0040	16.4498	26.7%
16384	0.0696	3.7647	0.0348	7.5294	50.0%
65536	0.3227	3.2497	0.1413	7.4203	43.8%

Nvidia Geforce Gtx 1650 — stride = 256

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
512	0.0092	0.8873	0.0041	2.0000	44.4%
1024	0.0099	1.6623	0.0036	4.4912	37.0%
4096	0.0156	4.2097	0.0041	16.0000	26.3%
16384	0.0655	4.0049	0.0331	7.9150	50.6%
65536	0.3256	3.2209	0.1352	7.7576	41.5%

See also:

• stride.dscal.js — WebGPU stride-sweep benchmark script
• stride.dscal.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	0.0062	2.6392	0.0028	5.8851	44.8%
65536	0.0098	107.0850	0.0058	182.0444	58.8%
1048576	0.1040	161.3690	0.1004	167.1304	96.6%
16777216	1.5770	170.2234	1.5653	171.4901	99.3%

Nvidia Geforce Gtx 1650 — alpha = 0

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
1024	0.0076	2.1695	0.0029	5.5652	39.0%
65536	0.0123	85.5561	0.0060	173.3757	49.3%
1048576	0.1063	157.8230	0.1004	167.0239	94.5%
16777216	1.5580	172.2914	1.5656	171.4568	100.5%

Nvidia Geforce Gtx 1650 — alpha = 1e-38

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
1024	0.0061	2.6667	0.0028	5.9535	44.8%
65536	0.0098	106.9103	0.0057	184.0899	58.1%
1048576	0.1038	161.5928	0.1004	167.1304	96.7%
16777216	1.5770	170.2234	1.5653	171.4866	99.3%

Nvidia Geforce Gtx 1650 — alpha = 1

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
1024	0.0061	2.6667	0.0027	6.0235	44.3%
65536	0.0099	106.3896	0.0057	184.6084	57.6%
1048576	0.1037	161.7424	0.1004	167.0771	96.8%
16777216	1.5768	170.2441	1.5654	171.4796	99.3%

Nvidia Geforce Gtx 1650 — alpha = 2.5

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
1024	0.0061	2.6667	0.0027	6.0235	44.3%
65536	0.0099	105.8740	0.0057	182.5515	58.0%
1048576	0.1037	161.8173	0.1004	167.1837	96.8%
16777216	1.5769	170.2337	1.5654	171.4778	99.3%

See also:

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