A brief technical overview about GPU power consumption and performance, with references to the latest architecture developed by Nvidia: Maxwell and Tegra X1.
Co-Author: Pietro Piscione (https://www.linkedin.com/pub/pietro-piscione/84/b37/926)
1. Power Consumption and
Performance trends on GPUs
Computer Architecture
Authors:
Piscione Pietro
Villardita Alessio
A.Y. 2014/2015
Degree: Computer Engineering
3. Why GPU ?
● Multimedia and
general purpose
applications
and recently, also
● High Performance
Computing applications,
"autonomous machines"
and automotive
CAD and
3D apps
Office
and PDF
readers
Multimedia
Audio/Video
Browsers
4. Power Consumption Overview
Domain
GPU power
consumption
Desktop and Workstation
(no mobility) 300W (50%)
Notebook
(mobility)
50W (71%)
And the smartphones ?
6. GPU power consumption on mobile
How much power
does a mobile
application require
in order to execute?
Why is the GPU so
“energy-hungry”?
The answer is in the
GPU Architecture.
14. GPU vs CPU
GPU
● hundreds of simpler cores
CPU
● few very complex cores
● thousand of concurrent
hardware threads
● single-thread performance
optimization
● maximize floating-point
throughput
● transistor space dedicated
to complex ILP
● most die surface for
integer and fp units
● few die surface for integer
and fp units
24. Tegra X1 vs K1: power consumption
Architectural solutions to
improve energy efficiency:
● big.LITTLE processing
design
● Using of cluster migration
● Cache coherence solution
DRAM Energy efficiency
25. Conclusions
Power consumption reduction: current and future trends
involve both Hardware AND Software design
● Reduce data movement
● Lots of local processing in parallel
● Efficient caching and memory usage
● Where data lives
● Where computation happens, how it is scheduled
Energy efficiency must now be key metric.