testcuda/seaborn_plot.py at main · weimin023/testcuda · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
import seaborn as sns
import matplotlib.pyplot as plt

sns.set_theme(style="whitegrid")
sns.set_theme(style="ticks")

limi = 100

plt.figure(figsize=(10, 6))
x_sticks = ["16*16*16", "64*64*64", "128*128*128", "256*256*256", "512*512*512", "1024*1024*1024"]
data1 = [0.0124, 0.038, 0.07, 0.31, 2.40, 18.91]
data1 = [x * 1e3 for x in data1]

data2 = [0.0128, 0.0137, 0.02, 0.04, 0.25, 1.87]
data2 = [x * 1e3 for x in data2]

data3 = [0.0106, 0.0137, 0.02, 0.02, 0.1, 0.62]
data3 = [x * 1e3 for x in data3]

data4 = [0.0114, 0.011, 0.01, 0.02, 0.06, 0.32]
data4 = [x * 1e3 for x in data4]

sns.lineplot(x=x_sticks, y=data1, label="(1024) CUDA Core (Naive)", marker="o")
sns.lineplot(x=x_sticks, y=data2, label="(1024) CUDA Core (Shared Mem)", marker="s")
sns.lineplot(x=x_sticks, y=data3, label="(256) CUDA Core (Shared Mem + Threads Reuse)", marker="^")
sns.lineplot(x=x_sticks, y=data4, label="(32) Tensor Core (Naive)", marker="D")

plt.xlabel("Data Size (M*K*N)", fontsize=20)
plt.ylabel("Runtime (us)", fontsize=20)
plt.suptitle("Performance Comparison of Matrix Multiplication (RTX4060)", fontsize=20, y=0.95)
plt.title("Using CUDA Cores and Tensor Cores", fontsize=15)
plt.ylim(0, limi)

plt.xticks(rotation=45, fontsize=15)
plt.tight_layout()
plt.show()


plt.figure(figsize=(10, 6))
x_sticks = ["16*16*16", "64*64*64", "128*128*128", "256*256*256", "512*512*512", "1024*1024*1024"]
data1 = [0.0164, 0.0634, 0.118, 0.858, 6.767, 81.956]
data1 = [x * 1e3 for x in data1]

data2 = [0.0152, 0.0149, 0.0178, 0.0638, 0.4427, 6.812]
data2 = [x * 1e3 for x in data2]

data3 = [0.0142, 0.0136, 0.0167, 0.0371, 0.1996, 2.711]
data3 = [x * 1e3 for x in data3]

data4 = [0.0138, 0.0126, 0.0137, 0.0256, 0.122, 1.6]
data4 = [x * 1e3 for x in data4]

sns.lineplot(x=x_sticks, y=data1, label="CUDA Core (Naive)", marker="o")
sns.lineplot(x=x_sticks, y=data2, label="CUDA Core (Shared Mem)", marker="s")
sns.lineplot(x=x_sticks, y=data3, label="CUDA Core (Shared Mem + Threads Reuse)", marker="^")
sns.lineplot(x=x_sticks, y=data4, label="Tensor Core (Naive)", marker="D")

plt.xlabel("Data Size (M*K*N)", fontsize=20)
plt.ylabel("Runtime (us)", fontsize=20)
plt.suptitle("Performance Comparison of Matrix Multiplication (AGX Orin)", fontsize=20, y=0.95)
plt.title("Using CUDA Cores and Tensor Cores", fontsize=15)
plt.ylim(0, limi)

plt.xticks(rotation=45, fontsize=15)
plt.tight_layout()
plt.show()