Hugging Face's logo

Datasets:
Rosykunai
/
SGEMM_GPU_performance

Modalities:
Tabular
Formats:
parquet
Size:
100K - 1M
Libraries:
Datasets
pandas
Dataset card Viewer Files Community
1
Dataset Viewer
Auto-converted to Parquet
Full Screen
Search is not available for this dataset
MWG
int64
16
128
NWG
int64
16
128
KWG
int64
16
32
MDIMC
int64
8
32
NDIMC
int64
8
32
MDIMA
int64
8
32
NDIMB
int64
8
32
KWI
int64
2
8
VWM
int64
1
8
VWN
int64
1
8
STRM
int64
0
1
STRN
int64
0
1
SA
int64
0
1
SB
int64
0
1
Run_time
float64
13.3
510
__index_level_0__
int64
0
242k
32
128
32
32
8
32
32
8
1
4
1
0
0
0
60.91
65,656
32
64
32
16
16
16
16
8
2
2
0
0
0
1
37.4875
44,641
32
64
32
8
32
16
32
8
1
1
1
0
1
1
43.375
42,091
128
32
32
16
16
16
32
8
2
1
1
1
1
0
56.65
165,758
16
128
16
16
8
16
16
2
1
1
0
0
1
0
86.79
13,922
128
32
32
8
16
16
16
2
8
2
0
0
1
1
63.08
160,595
64
128
32
32
16
32
32
2
1
4
0
1
0
1
40.5625
141,445
128
128
32
8
32
16
8
8
2
1
1
1
1
0
200.7575
225,950
64
32
16
16
8
16
16
8
2
1
0
0
1
1
25.805
76,611
128
64
16
8
32
32
32
8
1
2
1
1
1
0
24.7525
175,934
32
64
16
8
32
32
32
2
1
1
0
0
1
0
36.9925
35,490
64
128
32
8
32
16
32
2
2
2
1
0
0
1
50.5675
131,209
32
64
16
16
8
16
8
8
2
2
0
1
0
0
39.9675
36,340
16
64
32
16
8
8
16
8
1
2
1
0
1
0
65.5825
10,122
64
128
16
8
16
32
8
8
2
2
0
0
1
0
233.7725
116,402
128
64
32
8
16
16
32
2
8
2
1
1
0
1
263.2975
186,717
32
128
32
8
32
32
16
8
1
2
0
0
0
1
76.43
60,385
32
64
32
8
16
16
32
2
1
1
0
0
1
1
38.6025
40,803
32
32
32
16
16
32
8
8
1
1
1
0
0
1
73.9325
30,793
128
64
32
32
16
16
32
8
1
2
0
1
0
0
129.145
199,604
128
32
32
8
16
16
16
2
4
1
1
1
0
1
37.38
160,557
64
128
32
16
8
32
8
2
1
1
0
1
1
0
121.9275
134,118
16
128
16
16
8
8
8
2
1
4
0
1
0
1
52.215
13,477
16
64
16
8
16
16
32
2
1
2
0
0
0
0
64.1225
6,800
64
16
16
8
16
8
16
8
4
1
1
0
0
0
67.6025
67,304
128
32
32
8
16
16
8
2
4
1
0
0
0
0
90.065
160,288
32
64
32
8
16
16
8
8
1
2
1
1
1
0
33.1325
40,542
64
128
32
16
16
32
16
8
1
2
1
0
1
1
35.67
137,339
64
64
32
8
8
8
32
2
8
1
1
0
1
1
215.3925
97,451
16
64
16
8
16
8
8
2
1
2
0
0
1
0
46.305
6,098
128
64
16
8
8
32
8
2
1
2
1
0
0
1
465.9275
171,449
128
64
32
8
8
16
16
2
8
4
1
0
0
0
498.085
183,480
32
128
32
8
8
8
32
8
1
2
1
1
0
0
265.205
55,500
32
128
32
8
16
8
16
2
1
8
0
1
1
1
47.8675
57,111
128
64
32
32
8
16
8
2
2
8
0
0
1
1
42.4075
197,587
64
128
32
16
16
8
16
2
1
8
1
1
1
0
25.615
135,262
128
128
16
8
32
32
16
2
2
1
0
0
0
1
319.9525
208,913
32
128
16
16
16
16
16
2
2
8
0
0
0
0
36.9525
53,360
64
128
32
8
16
32
16
2
2
8
1
0
0
0
281.625
129,080
128
64
32
8
16
8
16
8
1
4
0
0
1
0
211.6825
185,410
64
32
32
32
8
8
8
2
2
4
0
0
1
0
53.8325
85,394
128
32
32
16
32
32
32
8
4
1
1
1
0
0
129.075
166,716
32
128
32
8
8
32
8
8
1
1
0
1
1
1
156.935
56,391
128
128
16
32
8
16
16
8
2
2
0
1
1
1
97.6025
215,447
32
64
32
8
8
16
16
2
2
2
1
0
0
0
33.62
39,112
64
32
32
32
8
32
8
8
1
4
0
1
0
0
75.54
86,212
64
64
16
16
8
8
8
2
2
4
0
1
0
1
29.865
92,549
32
128
32
8
16
8
16
8
2
4
1
0
1
1
35.3825
57,355
128
32
32
16
16
8
16
2
8
1
0
1
0
1
128.785
164,869
32
128
16
8
16
8
16
8
1
8
0
0
0
0
28.915
49,424
32
128
32
16
8
8
16
8
2
8
1
0
1
1
40.43
61,019
64
64
32
8
16
16
16
2
4
1
0
1
1
0
34.585
100,454
128
128
32
32
16
16
32
2
2
4
1
1
0
1
29.035
240,445
128
128
16
32
8
32
16
2
1
1
0
0
0
1
266.9975
215,841
128
64
16
16
8
16
8
2
1
1
1
1
1
1
156.175
177,167
128
32
32
8
32
8
32
8
8
1
0
1
0
1
137.53
161,717
128
64
16
16
16
16
32
8
4
2
0
1
1
1
22.5575
179,767
128
128
16
8
32
32
32
2
2
2
1
1
1
0
260.02
209,230
16
32
16
8
8
8
8
2
2
1
1
0
0
1
68.295
1,561
128
128
32
16
16
32
16
2
1
4
1
0
1
1
227.3325
234,827
32
64
32
8
8
16
16
8
2
2
0
1
1
0
39.5925
39,206
16
64
16
8
8
8
8
2
1
2
0
1
0
1
49.065
5,237
64
128
32
32
16
32
16
2
1
1
1
1
0
1
41.97
141,165
64
64
32
32
8
16
8
8
2
2
1
0
1
1
33.14
108,859
16
128
32
8
32
8
16
2
1
1
1
1
0
0
66.58
16,684
32
32
32
8
8
16
8
8
2
2
0
1
0
1
41.2
27,205
128
128
32
32
8
16
32
2
2
4
1
0
0
0
320.1225
238,712
64
64
16
8
8
8
8
2
4
8
0
1
1
0
288.23
87,094
64
64
32
32
8
16
8
8
2
4
0
1
0
0
38.4075
108,868
128
128
32
8
16
8
32
8
1
1
0
0
0
0
489.63
221,888
32
64
16
8
16
8
8
8
1
4
0
0
0
0
39.0775
33,808
64
32
16
8
8
16
8
2
2
2
0
1
1
1
29.4675
72,967
128
128
32
16
16
32
32
2
2
1
1
1
0
0
353
235,228
128
32
16
8
32
16
16
8
4
1
1
1
1
0
29.6225
153,582
64
64
32
8
16
32
8
2
1
1
1
0
0
1
62.275
100,841
64
32
32
16
8
32
16
2
1
1
1
1
1
1
41.0275
83,535
64
128
16
8
16
8
32
8
1
2
0
0
1
1
131.0075
114,963
64
64
32
8
8
32
16
2
2
4
1
1
0
0
331.2025
98,812
32
32
32
8
16
32
32
8
1
1
1
1
1
0
41.0725
28,574
64
32
16
8
32
16
16
8
1
1
0
1
1
1
52.345
75,351
16
32
32
8
8
16
8
8
1
4
1
0
0
1
52.3825
3,577
128
32
32
32
8
32
16
8
4
2
0
0
1
1
27.77
168,339
16
64
32
8
32
8
8
8
2
2
0
0
0
0
62.0475
9,456
64
128
32
8
16
16
16
8
1
4
0
1
0
1
244.93
128,261
16
32
16
16
8
8
8
8
1
4
0
1
0
1
49.5225
2,709
32
32
32
8
32
16
16
2
1
1
0
1
0
1
134.005
28,933
32
64
32
8
8
8
32
8
4
2
0
0
0
1
49.4725
38,769
32
128
32
8
16
8
8
8
4
8
1
0
0
0
31.91
57,048
128
128
32
16
32
16
16
8
8
4
0
0
0
1
75.405
235,825
128
32
32
32
8
8
8
8
4
1
0
0
0
0
41.99
166,960
64
32
16
8
8
8
16
2
1
1
0
1
1
0
67.455
72,518
64
32
32
16
8
8
32
2
4
1
0
0
1
1
38.1675
82,691
32
16
32
32
8
8
16
8
1
1
1
0
0
0
122.9325
22,392
64
128
32
16
16
8
8
8
1
2
0
0
1
0
34.52
135,026
128
64
32
32
8
32
8
2
2
4
0
0
0
0
56.47
198,432
32
16
32
8
8
8
8
8
1
1
0
1
1
1
81.835
20,647
64
128
32
16
8
8
8
2
4
1
1
1
1
1
332.105
132,143
128
64
16
8
16
16
32
2
1
1
0
1
1
0
183.395
174,086
64
64
32
8
16
16
8
8
2
2
1
0
1
1
38.15
100,283
128
64
32
8
16
16
8
2
1
4
1
0
1
0
170.3375
185,866
End of preview. Expand in Dataset Viewer.

Information

SGEMM GPU kernel performance Data Set available for download at https://archive.ics.uci.edu/ml/datasets/SGEMM+GPU+kernel+performance

We performed some filtering on this dataset.

Here is the original information of this dataset:

This data set measures the running time of a matrix-matrix product AB = C, where all matrices have size 2048 x 2048, using a parameterizable SGEMM GPU kernel with 241600 possible parameter combinations. For each tested combination, 4 runs were performed and their results are reported as the 4 last columns. All times are measured in milliseconds.

There are 14 parameter, the first 10 are ordinal and can only take up to 4 different powers of two values, and the 4 last variables are binary. Out of 1327104 total parameter combinations, only 241600 are feasible (due to various kernel constraints). This data set contains the results for all these feasible combinations.

The experiment was run on a desktop workstation running Ubuntu 16.04 Linux with an Intel Core i5 (3.5GHz), 16GB RAM, and a NVidia Geforce GTX 680 4GB GF580 GTX-1.5GB GPU. We use the 'gemm_fast' kernel from the automatic OpenCL kernel tuning library 'CLTune' (https://github.com/CNugteren/CLTune).

  • Note: for this kind of data sets it is usually better to work with the logarithm of the running times (see e.g. Falch and Elster, 'Machine learning-based auto-tuning for enhanced performance portability of OpenCL applications', 2015).

Download 

from datasets import load_dataset

dataset = load_dataset("Rosykunai/SGEMM_GPU_performance")

Reference

  • Rafael Ballester-Ripoll, Enrique G. Paredes, Renato Pajarola.
  • Sobol Tensor Trains for Global Sensitivity Analysis.
  • In arXiv Computer Science / Numerical Analysis e-prints, 2017
  • Cedric Nugteren and Valeriu Codreanu.
  • CLTune: A Generic Auto-Tuner for OpenCL Kernels.
  • In: MCSoC: 9th International Symposium on Embedded Multicore/Many-core Systems-on-Chip. IEEE, 2015
Downloads last month
29
Size of downloaded dataset files:
3.92 MB
Size of the auto-converted Parquet files:
3.92 MB
Number of rows:
193,349