VEGASVErsatile GBT Astronomical Spectrometer ROACHROACH 2 Xilinx Virtex
Ø天文终端混合架构是如何规划的? Ø一个应用实例 üVEGAS:VErsatile GBT Astronomical Spectrometer
硬件配置 Ø ROACH(ROACH 2) üXilinx Virtex 5( Virtex 6) ü 4 x. CX 4: 4 x 10 Gb. E( 8 x. CX 4 or SFP+ ) üA/D:ADC 2 x 1 G-8( ADC 1 x 10 G-4 ) Ø GPU üNvidia GTX TITAN üCUDA Cores: 2688 üBase Clock: 837 MHz üMemory Clock : 6 Gbps üMemory Interface :GDDR 5 üMemory Bandwidth : 288. 4 GB/s=6. 0 Gbps*384 -bit/8 üPCIE 3. 0 x 16,32 GB/s(PCIE 2. 1 x 16,16 GB/s)
三、混合架构相关器系统软件 Ø FPGA开发 üMATLAB/Simulink+Xilinx System Generator üPython Ø CPU+GPU混合编程 üCUDA 5. 5 üNsight集成开发环境
ROACH相关机架构 ROACH FPGA+GPU混合实时相关器架 构 ROACH GPU
CUDA程序 Øcorr_main. cu : üTop-level file üN_FFT、N_Sub. Bands、PFBOn、Num. Acc ücuda. Malloc、cuda. Memcpy Øcorr_kernels. cu : üCUDA kernels üDo. PFB、Do. FFT、Correlation、Accumulate Øcorr_dataread. cu:data read from 10 Gb. E module Øcorr_plot. cu : plot、save ØMakefile : nvcc make Øcorr_gencoeff. py : generate PFB coefficients
A. A* Phase(A. B*) Amp(A. B*) B. B* 70 MHz, -10 d. Bm, N_FFT=32768, N_Sub. Bands=1 C_mix=1/8, 400 Mbps
Phase(A. B*) Amp(A. B*) B. B* A. A* N_FFT=1024
Phase(A. B*) Amp(A. B*) B. B* A. A* N_FFT=1024, Num. Acc=100
N_FFT=1024, Sub. Band=8
A. A* Phase(A. B*) Amp(A. B*) B. B* 383 MHz, -10 d. Bm, N_FFT=32768, N_Sub. Band=1 C_mix=3/8, 800 Mbps
Phase(A. B*) Amp(A. B*) B. B* A. A* N_FFT=1024
N_FFT=32768, N_Sub. Band=8
Thank you ! 23
PFB technique 1024 samples 分成 4个子块, 每块 256个点,并行处理
- Slides: 27