目前,我正在研究GPU编程,使用CUDA和OpenCL进行编程,所用的显卡分别为NVIDIA Tesla K40C和AMD Firepro W8100。
通过查询相关的资料,我整理了一下这两款显卡的性能参数,并进行了对比。表格如下:
显卡名称 | AMD Firepro W8100 | NVIDIA Tesla K40C | |
显卡制造商 | AMD | NVIDIA | |
Graphics Processor | GPU Name: | Hawaii GL40 | GK110B |
Architecture: | GCN 2.0 | Kepler | |
Process Size: | 28 nm | 28 nm | |
Transistors: | 6,200 million | 7,080 million | |
Die Size: | 438 mm² | 561 mm² | |
Graphics Card | Released: | Jun 23rd, 2014 | Oct 8th, 2013 |
Production Status: | End-of-life | End-of-life | |
Bus Interface: | PCIe 3.0 x16 | PCIe 3.0 x16 | |
Launch Price: | # | 7,699 USD | |
Memory | Memory Size: | 8192 MB | 12288 MB |
Memory Type: | GDDR5 | GDDR5 | |
Memory Bus: | 512 bit | 384 bit | |
Bandwidth: | 320.0 GB/s | 288.4 GB/s | |
Clock Speeds | GPU Clock: | 824 MHz | 745 MHz |
Memory Clock: | 1250 MHz | 1502 MHz | |
Render Config | Shading Units: | 2560 | 2880 |
TMUs: | 160 | 240 | |
ROPs: | 64 | 48 | |
Compute Units: | 40 | 15 | |
Pixel Rate: | 52.74 GPixel/s | 44.70 GPixel/s | |
Texture Rate: | 131.8 GTexel/s | 178.8 GTexel/s | |
Floating-point performance: | 4,219 GFLOPS | 4,291 GFLOPS | |
Board Design | Slot Width: | Dual-slot | Dual-slot |
Length: | 11 inches | 10.5 inches | |
TDP: | 220 W | 245 W | |
Outputs: | 4x DisplayPort | No outputs | |
Power Connectors: | 2x 6-pin | 1x 6-pin + 1x 8-pin | |
Graphics Features | DirectX: | 12.0 | 12.0 |
OpenGL: | 4.6 | 4.6 | |
OpenCL: | 2.0 | 1.2 | |
Shader Model: | 5.0 | 5.0 | |
CUDA: | × | 3.5 |
另外,这两款显卡均支持多GPU编程技术。