NVIDIA has launched its fastest GPU accelerator for computing servers. The Tesla K40 delivers up to 25% more performance compared to the Tesla K20X. AMD’s FirePro S10000 is still faster but with more power consumption…
NVIDIA Tesla K40 in numbers:
– 12GB of ultra-fast GDDR5 memory, 384-bit
– full GK110 GPU (Kepler) with 2880 CUDA cores
– core clock: 745MHz, boost clock: 875MHz
– TDP: 235W
– Price: around USD $5500
– Peak FP64 performance: 1.43 TFLOPS
– Peak FP32 performance: 4.29 TFLOPS
The Tesla K20X for the sake of comparison:
– 6GB of GDDR5 memory, 384-bit
– GK110 (Kepler) with 2668 CUDA cores
– core clock: 732MHz
– TDP: 235W
– Price: around USD $3800
– Peak FP64 performance: 1.31 TFLOPS
– Peak FP32 performance: 3.95 TFLOPS
And the AMD FirePro S10000:
– 12GB / 6GB of GDDR5 memory, 384-bit
– dual-Tahiti GPU with 2×1792 stream processors
– core clock: 825MHz
– TDP: 375W
– Price: around USD $3600
– Peak FP64 performance: 1.48 TFLOPS
– Peak FP32 performance: 5.91 TFLOPS
FP32 performance per Watt:
– Tesla K40: 18.25 GFLOPS/Watt
– Tesla K20: 16.80 GFLOPS/Watt
– FirePro S10000: 15.76 GFLOPS/Watt
Links:
http://nvidianews.nvidia.com/News/NVIDIA-Unveils-World-s-Fastest-Accelerator-for-Data-Analytics-and-Scientific-Computing-c15.aspx
http://www.anandtech.com/show/8729/nvidia-launches-tesla-k80-gk210-gpu
Tesla K80, GK210 GPU with Compute Capability 3.7 officially announced and released.