REGISTER | LOGIN
Breaking News
News & Analysis

Baidu Releases AI Benchmark

Intel, Nvidia post results training neural nets
9/26/2016 04:30 PM EDT
2 comments
More Related Links
View Comments: Newest First | Oldest First | Threaded View
WaveJin
User Rank
Rookie
Need more than benchmarks
WaveJin   9/27/2016 5:04:54 PM
Great article Rick. At Wave we welcome the goals of DeepBench. Although in its current form, it is useful mainly for comparing GPUs and CPUs, we are looking forward to contributing to this effort to enable fair comparison of non-GPU architectures. More fundamental to Baidu's goal of achieving 100x improvement in compute power, we believe this can only be met through discontinuous versus continuous innovation. If we only focus on incremental improvements of low-level operations such as matrix multiplications or convolutions, we will miss out on opportunities for key innovations.

Wave's CTO, Dr. Chris Nicol, recently wrote an interesting blog about the need for discontinuous innovation in computing architectures to achieve real and game-changing improvements in computing systems for deep learning rather than being constrained by certain types of benchmarks and associated designs.

http://wavecomp.com/blog/benchmarks-can-either-encourage-or-constrain-innovation

While the initial work on DeepBench is very useful, it needs to be complemented with the willingness to think out-of-the-box and innovate at all levels to meet the 100x challenge posed by Baidu. Thank you again for a really insightful article and we are looking forward to continuing discussions on this topic.

joe.raffa
User Rank
Author
On the topic of 100x faster processors...
joe.raffa   9/27/2016 11:42:38 AM
NO RATINGS
http://www.meetup.com/ACM-SIGAI-Bay-Area-Chapter/events/233406926/

 

Like Us on Facebook
EE Times on Twitter
EE Times Twitter Feed