使用pinned memory 进行 cudaMemcpyAsync 时,PIC-E最高能到达6GB/s的速率, 但是把cudaMemcpyAsync最后的参数用cudaStreamCreate创建的streamId替换0时
PCI-E的传输效率只能到达5GB/s,只有用streamId=0时的80%。 下面是有改写bandwidthTest程序测得的结果。
谁能帮忙解释一下这个问题。
还有一点,当cudaMemcpyAsync使用0作为streamId的时候,CUDA支持的overlap就没有作用了。
Running on…
device 0:GeForce GTX 295
Shmoo Mode
Host to Device Bandwidth for Pinned memory
…
Transfer Size (Bytes) Bandwidth(MB/s)
1024 164.1
2048 363.7
3072 529.2
4096 687.7
5120 834.7
6144 976.0
7168 1091.6
8192 1225.6
9216 1321.7
10240 1439.5
11264 1559.9
12288 1648.9
13312 1746.9
14336 1833.7
15360 1916.1
16384 2010.2
17408 2081.0
18432 2120.9
19456 2247.4
20480 2302.3
22528 2431.7
24576 2532.6
26624 2644.0
28672 2799.8
30720 2855.7
32768 2966.5
34816 3053.6
36864 3123.8
38912 3205.3
40960 3274.4
43008 3359.7
45056 3415.9
47104 3492.1
49152 3529.7
51200 3609.8
61440 3806.8
71680 4014.0
81920 4179.8
92160 4308.4
102400 4426.0
204800 4993.5
307200 5012.2
409600 5182.6
512000 5274.6
614400 5357.7
716800 5413.0
819200 5451.5
921600 5479.4
1024000 5506.7
1126400 5527.0
2174976 5632.8
3223552 5670.4
4272128 5690.0
5320704 5700.6
6369280 5709.9
7417856 5707.5
8466432 5723.6
9515008 5751.6
10563584 5749.7
11612160 5755.9
12660736 5745.7
13709312 5761.6
14757888 5760.2
15806464 5762.9
16855040 5760.1
18952192 5766.4
21049344 5767.4
23146496 5768.8
25243648 5748.8
27340800 5770.4
29437952 5771.2
31535104 5771.8
33632256 5769.8
37826560 5773.6
42020864 5774.0
46215168 5774.3
50409472 5775.1
54603776 5774.7
58798080 5775.6
62992384 5774.7
67186688 5773.9
Running on…
device 0:GeForce GTX 295
Shmoo Mode
Host to Device Bandwidth for Pinned memory
…
Transfer Size (Bytes) Bandwidth(MB/s)
1024 21.3
2048 43.7
3072 56.2
4096 89.1
5120 111.1
6144 132.7
7168 154.7
8192 175.9
9216 196.7
10240 209.1
11264 241.9
12288 260.5
13312 281.4
14336 297.0
15360 319.9
16384 338.7
17408 358.7
18432 380.4
19456 397.6
20480 420.4
22528 456.5
24576 494.6
26624 531.4
28672 565.7
30720 603.4
32768 642.5
34816 679.5
36864 716.7
38912 751.0
40960 784.1
43008 814.9
45056 838.6
47104 884.5
49152 904.4
51200 947.5
61440 1094.4
71680 1243.9
81920 1376.3
92160 1492.7
102400 1607.4
204800 2529.6
307200 3099.0
409600 3479.8
512000 3783.5
614400 4006.2
716800 4160.8
819200 4248.1
921600 4316.8
1024000 4366.0
1126400 4378.0
2174976 4545.4
3223552 4585.9
4272128 4597.3
5320704 4618.1
6369280 4614.7
7417856 4484.7
8466432 4450.9
9515008 4455.1
10563584 4461.6
11612160 4462.4
12660736 4468.0
13709312 4467.3
14757888 4469.0
15806464 4470.0
16855040 4472.5
18952192 4472.5
21049344 4478.0
23146496 4477.1
25243648 4481.2
27340800 4482.9
29437952 4480.4
31535104 4486.3
33632256 4486.7
37826560 4475.9
42020864 4468.3
46215168 4460.5
50409472 4455.6
54603776 4455.4
58798080 4457.3
62992384 4457.3
67186688 4458.1