1 buckets per second much better whats more you can

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: Intel, AMD, NVIDIA) is no longer an advantage •  Processor chips become commodity –  Get manufactured wherever it is cheap •  Innovation shifts elsewhere •  Or more starkly stated: –  Innovation in computing stops 8/27/13 cs420: intro 32 Alterna3ve: Parallelism •  If we Pind killer apps that need all the parallel power we can bring to bear –  With 50B transistors, at least 100+ processor cores on each chip •  There is a tremendous competitive advantage to building such a killer app –  So, given our history, we will Pind it •  What are the enabling factors: –  Finding the application areas. –  Parallel programming skills 8/27/13 cs420: intro 33 A few candidate areas •  Simple parallelism: –  Search images, scan Piles, .. •  Speech recognition: –  almost perfect already –  But speaker dependent, minor training, and needs non- noisy environment –  Frontier: speaker independent recognition with non- controlled environment •  Broadly: ArtiPicial intelligence •  And, of course, HPC (high performance computing) 8/27/13 cs420: intro 34 Parallel programming Skills •  So, all machines will be (are?) parallel •  So, almost all programs will be parallel –  True? •  There are 10 million programmers in the world –  Approximate estimate •  All programmers must become parallel programmers –  Right? What do you think? 8/27/13 cs420: intro 35 Programming Models innova3ons •  Expect a lot of novel programming models –  There is scope for new languages, unlike now –  Only Java broke thru after C/C++ •  This is good news: –  If you are a computer scientist wanting to develop new languages •  Bad news: –  If you are a application developer •  DO NOT WAIT FOR “AutoMagic” Parallelizing compiler! 8/27/13 cs420: intro 36 Digging Deeper •  First law of holes: –  If you in a hole, stop digging! •  But we are in a hole, and we cannot help but dig! •  Let me explain 8/27/13 cs420: intro 37 Latency to Memory •  To process data, we must bring it to the processor and take the resulting data back to memory •  DRAM, the main inexpensive memory chip –  Once you supply an address to it, it gets you the data after 50- ish nanoseconds •  Doesn’t improve that much over time: 80 - > 30 ns –  A single core clock is 2 GHz: it beats twice in a nanosecond! –  So, you are working with someone who is 100 times slower.. –  Not just that: a single core can do 4+ additions/cycle –  We are talking about putting hundreds of cores on a chip?! 8/27/13 cs420: intro 38 Latency vs bandwidth •  Imagine you are putting a Pire...
View Full Document

Ask a homework question - tutors are online