Deep Learning Benchmarks (RTX 2080 TI vs. RTX 2070)
- Lambdalabs Benchmark, see repository here
- Yusaku Sako's benchmark, see repository here
- TensorFlow 1.12 for CUDA 10 + cudnn 7.3.1, from here
-
nvidia-docker
with nvidia/cuda:10.0-cudnn7-devel
image, from here
- AMD Ryzen 7 2700X CPU
- G.Skill Trident Z, 32 GB, CL14 14-14-34
- ASRock X470 Taichi Ultimate
- Zotac RTX 2080 TI AMP
- Zotac RTX 2070 AMP Extreme
Results (Lambdalabs benchmark)
syn-replicated-fp32-1gpus
Config |
2700X-GeForce_RTX_2080_Ti |
2700X-GeForce_RTX_2070 |
Difference |
resnet50 |
307.54 |
209.46 |
98.08 (32 %) |
resnet152 |
117.2 |
81.59 |
35.61 (30 %) |
inception3 |
205.31 |
136.94 |
68.37 (33 %) |
inception4 |
85.66 |
57.98 |
27.68 (32 %) |
vgg16 |
181.17 |
120.9 |
60.27 (33 %) |
alexnet |
3835.94 |
2510.35 |
1325.59 (35 %) |
ssd300 |
156.91 |
108.53 |
48.38 (31 %) |
syn-parameter_server-fp32-1gpus
Config |
2700X-GeForce_RTX_2080_Ti |
2700X-GeForce_RTX_2070 |
Difference |
resnet50 |
306.02 |
208.58 |
97.44 (32 %) |
resnet152 |
116.98 |
81.62 |
35.36 (30 %) |
inception3 |
205.14 |
137.0 |
68.14 (33 %) |
inception4 |
85.63 |
58.01 |
27.62 (32 %) |
vgg16 |
181.14 |
120.97 |
60.17 (33 %) |
alexnet |
3834.12 |
2512.17 |
1321.95 (34 %) |
ssd300 |
156.78 |
108.57 |
48.21 (31 %) |
syn-replicated-fp16-1gpus
Config |
2700X-GeForce_RTX_2080_Ti |
2700X-GeForce_RTX_2070 |
Difference |
resnet50 |
496.29 |
344.88 |
151.41 (31 %) |
resnet152 |
187.12 |
131.68 |
55.44 (30 %) |
inception3 |
305.53 |
200.72 |
104.81 (34 %) |
inception4 |
118.36 |
83.98 |
34.38 (29 %) |
vgg16 |
264.88 |
170.18 |
94.7 (36 %) |
alexnet |
5274.9 |
3297.37 |
1977.53 (37 %) |
ssd300 |
207.26 |
145.86 |
61.4 (30 %) |
syn-parameter_server-fp16-1gpus
Config |
2700X-GeForce_RTX_2080_Ti |
2700X-GeForce_RTX_2070 |
Difference |
resnet50 |
501.49 |
347.51 |
153.98 (31 %) |
resnet152 |
190.38 |
133.54 |
56.84 (30 %) |
inception3 |
307.93 |
202.11 |
105.82 (34 %) |
inception4 |
120.55 |
85.35 |
35.2 (29 %) |
vgg16 |
269.87 |
173.37 |
96.5 (36 %) |
alexnet |
5371.81 |
3349.07 |
2022.74 (38 %) |
ssd300 |
207.19 |
146.78 |
60.41 (29 %) |
Results (Yusaku Sako's benchmark)
Precision |
vgg16 eval |
vgg16 train |
resnet152 eval |
resnet152 train |
32-bit |
29.0ms |
91.4ms |
43.6ms |
191.2ms |
16-bit |
18.7ms |
60.2ms |
25.5ms |
135.0ms |
Precision |
vgg16 eval |
vgg16 train |
resnet152 eval |
resnet152 train |
32-bit |
42.6ms |
130.6ms |
65.1ms |
264.2ms |
16-bit |
29.0ms |
94.2ms |
39.5ms |
183.9ms |
Network |
RTX 2080 TI |
RTX 2070 |
Difference |
vgg16 eval |
29.0ms |
42.6ms |
13.6 ( 32 %) |
vgg16 train |
91.4ms |
130.6ms |
39.2 ( 30 %) |
resnet152 eval |
43.6ms |
65.1ms |
21.5 ( 33 %) |
resnet152 train |
191.2ms |
264.2ms |
73.0 ( 28 %) |
Network |
RTX 2080 TI |
RTX 2070 |
Difference |
vgg16 eval |
18.7ms |
29.0ms |
10.3 ( 36 %) |
vgg16 train |
60.2ms |
94.2ms |
34.0 ( 36 %) |
resnet152 eval |
25.5ms |
39.5ms |
14.0 ( 35 %) |
resnet152 train |
135.0ms |
183.9ms |
48.9 ( 27 %) |
Coming here soon!
Contact (Bugs, Feedback, Contribution and more)
For questions about dl-benchmarks
, please open an issue
here.