先看结论: 基本上只和PCIE的带宽有关
1. i5-8250U 1.6GHz
2. PCIE
3 4K 105fps = 10.45Gps 4K 一帧 = 99.5Mbit
4K YUV444 + mask = 3840*2160*4*8 = 265.42Mbit
4K YUV444 = 3840*2160*3*8 = 199.07Mbit
i9-9900K + RTX2080Ti :
H->D : 199.07Mbit * (1000/3.47) /1000 = 57.37Gbit /s = 7.17GB/s
D->H : 265.42Mbit * (1000/4.8)/1000 = 55.30Gbit/s = 6.91GB/s
MX150 :7.17GB/s * 3.47 / 11.73 = 2.12GB/s
H->D : 6.91GB/s * 4.8 / 13.87 = 2.39GB/s
GTX1060 : 7.17GB/s * 3.47 / 6.42 = 3.88 GB/s
H->D : 6.91GB/s * 4.8 / 8.96 = 3.70 GB/s
4. 4K 60fps I420 = 60*100Mbps = 5Gbps 如果YUV444的话 则需要10Gbps
5. 实测
硬件 | CPU/Hz | GPU/Hz | PCIE | DDR | GDDR | CPU-->GPU/ms | Handle | GPU--->CPU |
i5-8250U + MX150 | 1.8G | 1468 - 1532M | ?? | DDR4 |
DGGR5 6.008 Gbps |
11.73 | 6.97 | 13.87 |
i5-8标压+GTX1060+DDR4 | 2.80G | 1759M | X16-3.0 | DDR4 2400MHz 2.4*16/8=4.8GB/s | GDDR5 8GHz 8Gbps | 6.42 3.88GB/s | 2.13 | 8.96 3.70GB/s |
i9-9900K + RTX2080Ti | 3.60G | 1545M | X16-3.0 15.8GB/s | DDR4 4266?? 4.266*16/8 = 8.53GB/s | GDDR6 14Gbps | 3.47 7.17GB/s | 0.01 | 4.80 6.91GB/s |
推测的公式: 因为只有16根线连接了显存:
对于1060 台式机
CPU---->GPU CPU频率*16 = 2.80G*16 = 44.8Gbps 即 2.22 ms / 帧I420 4.44ms/帧YUV444
GPU---->CPU GPU频率*16 = 1.759G*16 7.0ms/帧 YUV444 + 2.33 ms / 帧 Mask = 9.33ms /帧
共计 4.44 + 7.0 +2.33 = 13.77ms /帧 即 72.6fps YUV444
对于MX150 笔记本:
CPU---->GPU 6.91 ms 一帧YUV444 如果是8位PCIE 那就是 13.82ms
GPU----->CPU 10.94 ms 一帧YUV444 如果是8位PCIE 那就是21.88 ms
fps 如果是8位PCIE 那就是28.01fps
PCIE 与传输速率关系 感觉图有问题??