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

Benchmark results for dger 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	0.0065	2.5819	0.0038	4.4746	57.7%
64	0.0067	10.0000	0.0037	17.7778	56.2%
128	0.0078	33.8361	0.0039	67.6721	50.0%
256	0.0118	89.2700	0.0058	181.2452	49.3%
512	0.0348	120.7059	0.0283	148.7293	81.2%
1024	0.1132	148.3743	0.1014	165.5520	89.6%
1280	0.1759	149.1432	0.1564	167.7594	88.9%
2048	0.4340	154.6928	0.3949	170.0029	91.0%
4096	1.7271	155.4646	1.5932	168.5294	92.2%

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

See also

• dger.js — WebGPU benchmark script
• dger.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	0.0082	2.0625	0.0038	4.4370	46.5%
64	0.0082	8.1092	0.0038	17.4790	46.4%
128	0.0097	27.2026	0.0040	66.8502	40.7%
256	0.0158	66.7939	0.0060	175.9144	38.0%
512	0.0404	103.9810	0.0285	147.3939	70.5%
1024	0.1311	128.1250	0.1016	165.3174	77.5%
1280	0.1979	132.5851	0.1568	167.3143	79.2%
2048	0.4892	137.2568	0.3951	169.9478	80.8%
4096	1.8132	148.0813	1.6602	161.7304	91.6%

Nvidia Geforce Gtx 1650 — stride = 5

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0068	2.5024	0.0037	4.5913	54.5%
64	0.0068	9.7196	0.0037	17.7778	54.7%
128	0.0080	33.0240	0.0040	66.5807	49.6%
256	0.0118	89.3913	0.0058	182.2493	49.0%
512	0.0354	118.6341	0.0284	148.2257	80.0%
1024	0.1174	143.0362	0.1019	164.7982	86.8%
1280	0.1843	142.3333	0.1568	167.3314	85.1%
2048	0.4484	149.7414	0.3951	169.9478	88.1%
4096	1.7035	157.6169	1.6828	159.5606	98.8%

Nvidia Geforce Gtx 1650 — stride = 32

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0067	2.5143	0.0037	4.5322	55.5%
64	0.0075	8.8511	0.0039	16.9106	52.3%
128	0.0102	25.8000	0.0041	64.5000	40.0%
256	0.0230	45.7206	0.0059	177.8162	25.7%
512	0.0717	58.6286	0.0287	146.5714	40.0%
1024	0.2696	62.2982	0.1018	164.9018	37.8%
1280	0.4173	62.8736	0.1572	166.9055	37.7%
2048	1.0518	63.8345	0.3973	168.9897	37.8%
4096	4.1800	64.2352	1.7498	153.4460	41.9%

Nvidia Geforce Gtx 1650 — stride = 33

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0067	2.5385	0.0036	4.6316	54.8%
64	0.0069	9.6520	0.0038	17.7021	54.5%
128	0.0082	32.2500	0.0041	65.0079	49.6%
256	0.0141	74.4253	0.0058	181.7459	41.0%
512	0.0376	111.8161	0.0283	148.3932	75.4%
1024	0.1616	103.9311	0.1017	165.0834	63.0%
1280	0.2537	103.4172	0.1573	166.7358	62.0%
2048	0.7482	89.7327	0.4035	166.4162	53.9%
4096	3.6365	73.8354	1.7513	153.3142	48.2%

Nvidia Geforce Gtx 1650 — stride = 255

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0068	2.4906	0.0038	4.4000	56.6%
64	0.0069	9.5853	0.0038	17.6271	54.4%
128	0.0082	32.2500	0.0041	64.5000	50.0%
256	0.0137	76.5914	0.0058	181.7459	42.1%
512	0.0382	110.0821	0.0284	147.8087	74.5%
1024	0.1434	117.1429	0.1019	164.8500	71.1%
1280	0.2171	120.8491	0.1574	166.6680	72.5%
2048	0.7429	90.3724	0.4015	167.2253	54.0%
4096	5.1912	51.7223	1.7627	152.3261	34.0%

Nvidia Geforce Gtx 1650 — stride = 256

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0069	2.4558	0.0038	4.4746	54.9%
64	0.0077	8.6307	0.0037	17.7778	48.5%
128	0.0103	25.6000	0.0039	67.6721	37.8%
256	0.0231	45.5939	0.0058	180.7473	25.2%
512	0.0717	58.6286	0.0283	148.3932	39.5%
1024	0.2698	62.2354	0.1017	165.0834	37.7%
1280	0.4174	62.8591	0.1574	166.6680	37.7%
2048	1.0558	63.5936	0.3974	168.9489	37.6%
4096	4.9544	54.1941	1.7582	152.7142	35.5%

See also:

• stride.dger.js — WebGPU stride-sweep benchmark script
• stride.dger.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	0.0066	2.5631	0.0038	4.4746	57.3%
64	0.0067	9.9522	0.0038	17.6271	56.5%
128	0.0077	34.4000	0.0040	66.5807	51.7%
256	0.0115	91.8883	0.0057	183.2646	50.1%
512	0.0347	121.0954	0.0283	148.7293	81.4%
1024	0.1130	148.6265	0.1014	165.6043	89.7%
1280	0.1755	149.4831	0.1568	167.3655	89.3%
2048	0.4327	155.1618	0.3949	170.0374	91.3%
4096	1.6589	161.8568	1.5932	168.5277	96.0%

Nvidia Geforce Gtx 1650 — pad = 1

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0066	2.5507	0.0037	4.5128	56.5%
64	0.0069	9.6296	0.0039	17.0492	56.5%
128	0.0080	32.8270	0.0039	67.6721	48.5%
256	0.0118	89.3913	0.0060	174.0529	51.4%
512	0.0348	120.7059	0.0307	136.8000	88.2%
1024	0.1156	145.2533	0.1102	152.3588	95.3%
1280	0.1961	133.7532	0.1720	152.5709	87.7%
2048	0.4523	148.4488	0.4387	153.0566	97.0%

Nvidia Geforce Gtx 1650 — pad = 8

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0066	2.5631	0.0038	4.4370	57.8%
64	0.0067	10.0000	0.0038	17.3333	57.7%
128	0.0077	34.1863	0.0039	67.6721	50.5%
256	0.0116	90.8729	0.0058	182.7556	49.7%
512	0.0358	117.2571	0.0301	139.4881	84.1%
1024	0.1171	143.3684	0.1085	154.7170	92.7%
1280	0.1957	134.0374	0.1696	154.6868	86.7%
2048	0.5006	134.1330	0.4328	155.1331	86.5%

Nvidia Geforce Gtx 1650 — pad = 16

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0066	2.5631	0.0040	4.2581	60.2%
64	0.0067	10.0000	0.0037	17.7778	56.2%
128	0.0077	34.1157	0.0039	67.6721	50.4%
256	0.0116	90.6226	0.0057	185.8531	48.8%
512	0.0348	120.7059	0.0284	148.2257	81.4%
1024	0.1204	139.4447	0.1019	164.7724	84.6%
1280	0.1880	139.5829	0.1569	167.1778	83.5%
2048	0.4728	142.0037	0.3966	169.3033	83.9%

Nvidia Geforce Gtx 1650 — pad = 32

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0082	2.0625	0.0038	4.4000	46.9%
64	0.0082	8.1250	0.0037	17.7778	45.7%
128	0.0097	27.3377	0.0039	67.1219	40.7%
256	0.0149	70.8202	0.0058	181.7459	39.0%
512	0.0395	106.3815	0.0282	148.8979	71.4%
1024	0.1290	130.1587	0.1019	164.8759	78.9%
1280	0.1924	136.3559	0.1567	167.4339	81.4%
2048	0.5325	126.0961	0.3956	169.7210	74.3%

Nvidia Geforce Gtx 1650 — pad = 48

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0102	1.6500	0.0037	4.5322	36.4%
64	0.0102	6.5000	0.0037	17.8541	36.4%
128	0.0116	22.8698	0.0039	67.6721	33.8%
256	0.0184	57.0616	0.0057	185.8531	30.7%
512	0.0494	85.0570	0.0286	146.8993	57.9%
1024	0.1556	107.8947	0.1023	164.2053	65.7%
1280	0.2388	109.8687	0.1569	167.1949	65.7%
2048	0.6028	111.3888	0.3990	168.2714	66.2%

Nvidia Geforce Gtx 1650 — pad = 64

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0116	1.4505	0.0041	4.1089	35.3%
64	0.0123	5.4167	0.0041	16.2500	33.3%
128	0.0142	18.5946	0.0041	64.5000	28.8%
256	0.0208	50.6092	0.0058	181.2452	27.9%
512	0.0536	78.4516	0.0287	146.5714	53.5%
1024	0.1645	102.1012	0.1023	164.1282	62.2%
1280	0.2623	100.0354	0.1572	166.9225	59.9%
2048	0.6587	101.9324	0.3964	169.3648	60.2%

Nvidia Geforce Gtx 1650 — pad = 128

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0123	1.3750	0.0037	4.5517	30.2%
64	0.0128	5.2000	0.0038	17.3333	30.0%
128	0.0145	18.2453	0.0038	68.8000	26.5%
256	0.0227	46.4633	0.0058	181.7459	25.6%
512	0.0573	73.2857	0.0285	147.3939	49.7%
1024	0.1797	93.4307	0.1069	157.1022	59.5%
1280	0.2683	97.7863	0.1575	166.5664	58.7%
2048	0.6842	98.1373	0.3971	169.0986	58.0%

See also:

• lda.dger.js — WebGPU lda-sweep benchmark script
• lda.dger.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	0.0066	2.5507	0.0038	4.4184	57.7%
64	0.0068	9.8578	0.0037	17.7778	55.5%
128	0.0078	33.9753	0.0039	67.1219	50.6%
256	0.0118	89.3913	0.0058	180.7473	49.5%
512	0.0348	120.7059	0.0282	149.2364	80.9%
1024	0.1131	148.5213	0.1013	165.7350	89.6%
1280	0.1752	149.7288	0.1566	167.4852	89.4%
2048	0.4334	154.9098	0.3949	170.0167	91.1%
4096	1.6592	161.8225	1.5931	168.5395	96.0%

Nvidia Geforce Gtx 1650 — alpha = 0

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0082	2.0625	0.0016	10.5600	19.5%
64	0.0082	8.1250	0.0016	42.0202	19.3%
128	0.0096	27.6583	0.0016	165.1200	16.8%
256	0.0150	70.3658	0.0016	671.3470	10.5%
512	0.0383	109.7143	0.0016	2680.1633	4.1%
1024	0.1228	136.8092	0.0016	10820.6182	1.3%
1280	0.1823	143.9326	0.0016	16731.4297	0.9%
2048	0.4529	148.2443	0.0016	40741.2812	0.4%
4096	1.7308	155.1355	0.0015	174805.3281	0.1%

Nvidia Geforce Gtx 1650 — alpha = 1e-38

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0066	2.5507	0.0037	4.5128	56.5%
64	0.0067	9.9522	0.0037	17.7778	56.0%
128	0.0079	33.6293	0.0039	67.6721	49.7%
256	0.0116	90.4979	0.0057	184.2913	49.1%
512	0.0348	120.9282	0.0282	149.2364	81.0%
1024	0.1132	148.3953	0.1013	165.8398	89.5%
1280	0.1753	149.6605	0.1565	167.5879	89.3%
2048	0.4327	155.1733	0.3950	169.9754	91.3%
4096	1.6609	161.6572	1.5932	168.5294	95.9%

Nvidia Geforce Gtx 1650 — alpha = 1

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0066	2.5631	0.0042	4.0000	64.1%
64	0.0067	9.9522	0.0043	15.6391	63.6%
128	0.0079	33.6293	0.0045	58.3463	57.6%
256	0.0115	91.5049	0.0069	151.9446	60.2%
512	0.0348	120.7059	0.0287	146.5714	82.4%
1024	0.1130	148.6686	0.1004	167.2136	88.9%
1280	0.1752	149.7288	0.1549	169.3185	88.4%
2048	0.4331	155.0185	0.3892	172.5332	89.8%
4096	1.6589	161.8552	1.5683	171.2082	94.5%

Nvidia Geforce Gtx 1650 — alpha = 2.5

n	wgblas ms	wgblas GB/s	cuBLAS ms	cuBLAS GB/s	efficiency
32	0.0066	2.5507	0.0040	4.1905	60.9%
64	0.0067	9.9048	0.0038	17.3333	57.1%
128	0.0078	33.9753	0.0039	67.1219	50.6%
256	0.0116	90.6226	0.0058	181.7459	49.9%
512	0.0348	120.7059	0.0283	148.5611	81.2%
1024	0.1130	148.5844	0.1014	165.5520	89.8%
1280	0.1755	149.4967	0.1564	167.7250	89.1%
2048	0.4331	155.0185	0.3949	170.0305	91.2%
4096	1.6593	161.8131	1.5933	168.5141	96.0%

See also:

• alpha.dger.js — WebGPU alpha-sweep benchmark script
• alpha.dger.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	0.0066	2.5507
64	0.0068	9.8578
128	0.0078	34.0454
256	0.0116	90.9986
512	0.0348	120.7059
1024	0.1128	148.8583
1280	0.1748	150.0439
2048	0.4339	154.7327
4096	1.6555	162.1853

Nvidia Geforce Gtx 1650 — layout = row-major

n	compute ms	GB/s
32	0.0066	2.5507
64	0.0067	9.9522
128	0.0076	34.5439
256	0.0115	91.6323
512	0.0348	120.7059
1024	0.1128	148.8372
1280	0.1756	149.3878
2048	0.4326	155.2135
4096	1.6609	161.6603

See also:

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