To illustrate a wide range of possible workstation specifications, we used four cards from NVIDIA’s current Quadro range. These were the K620, K2200, K4200, and K5200. This wasn’t the entire range – we didn’t include the super-low-end K420, the mid-range K1200 or the ultra-high-end K6000 or M6000. The cards we chose represent the most frequently supplied graphics in workstations aimed at mainstream applications.
The K620 is a basic professional 3D accelerator. It still sports 384 CUDA cores, which is a decent quantity, and offers 2GB of frame buffer, but the latter is just DDR3 and the memory interface 128-bit, so only 29GB/sec bandwidth is available. The K2200 is much more potent as a 3D accelerator, with 640 CUDA cores and 4GB of GDDR5 memory, offering 80GB/sec bandwidth. With the K4200 we enter the high end. This card provides 1,344 CUDA cores, and although it has the same 4GB of GDDR5 memory, the 256-bit bus means the bandwidth is more than doubled compared to the K2200, to 173GB/sec. Finally, the K5200 is an ultra-high-end card with 2,304 CUDA cores and a whopping 8GB of GDDR5 memory. Like the K4200, it has a 256-bit memory bus, but faster memory ensures 192GB/sec of bandwidth.