An explanation of the rationale behind the use of an FPGA-based system for our solution's implementation. A comparison is made between FPGAs, GPUs and ASICs.
7. 7
Power requirements
GPUs
High speedup
Hundreds of watts
(Tesla K40 by Nvidia max 235W)
ASICs
Highest possible
Lowest possible
FPGAs
High speedup
Tens of watts
(Virtex7-VC707 by Xilinx abt 15W)
8. 8
Flexibility
GPUs
High speedup
Hundreds of watts
Programmable
ASICs
Highest possible
Lowest possible
Application-specific
FPGAs
High speedup
Tens of watts
Programmable
9. 9
Costs
GPUs
High speedup
Hundreds of watts
Programmable
About 3000$ each
(Tesla K40 by Nvidia)
ASICs
Highest possible
Lowest possible
Application-specific
Too high if few
circuits required
FPGAs
High speedup
Tens of watts
Programmable
About 3500$ each
(Virtex7 VC707 by Xilinx)
10. 10
The choice
GPUs
High speedup
Hundreds of watts
Programmable
About 3000$ each
ASICs
Highest possible
Lowest possible
Application-specific
Too high if few
circuits required
FPGAs
High speedup
Tens of watts
Programmable
About 3500$ each
11. 11
The choice
GPUs
High speedup
Hundreds of watts
Programmable
About 3000$ each
ASICs
Highest possible
Lowest possible
Application-specific
Too high if few
circuits required
FPGAs
High speedup
Tens of watts
Programmable
About 3500$ each
12. 12
The choice
GPUs
High speedup
Hundreds of watts
Programmable
About 3000$ each
ASICs
Highest possible
Lowest possible
Application-specific
Too high if few
circuits required
FPGAs
High speedup
Tens of watts
Programmable
About 3500$ each
13. 13
The choice
GPUs
High speedup
Hundreds of watts
Programmable
About 3000$ each
FPGAs
High speedup
Tens of watts
Programmable
About 3500$ each
14. 14
The choice
GPUs
High speedup
Hundreds of watts
Programmable
About 3000$ each
FPGAs
High speedup
Tens of watts
Programmable
About 3500$ each
16. MATRIX INVERSION
16
First case study
• Intense computation
• Useful in different applications
(e.g. fluid dynamics lumped parameters model)
reusability
17. MATRIX INVERSION
17
First case study
So… FPGAs!
- High speedup vs GPPs
- Low power requirements
- Programmable
- Affordable costs
18. ANY QUESTIONS?
HAMSprojectContact us!
You can find us…
hams.necst@gmail.com
chiara1.gatti@mail.polimi.it
guido2.lanfranchi@mail.polimi.it
www.facebook.com/hams.project
https://twitter.com/HAMS_project
http://www.slideshare.net/HAMSproject
https://vimeo.com/necstxohw16
https://www.youtube.com/channel/UCaovqRpUc7D_Uf2WJHL0rvA