Page 16 - Multipath MIPS
P. 16
GPUs GeForce 8800 Architecture
A quite revolution with
plenty of potential
Programming:
- Brook
- Cg
- CUDA
Bottleneck:GPU<->CPU
• 367 GFLOPS vs. 32 GFLOPS
• 86.4 GB/s vs. 8.4 GB/s
• Up to 10x is typical if kernels
have enough parallelism
• Up to 25x – 400x if a data,
control-flow suits the GPU*
*:http://courses.ece.uiuc.edu/ece498/al1/lectures/lecture1%20intro%20fall%202007.ppt
16