Domain Expertise Vs. Compilers

Oded Computing Trends 2 Comments

Every so often people come up to us and ask, "Aren't compilers and compiler directives good enough for HPC applications?" or "Won't a compiler accomplish that for us?" While compilers have made massive progress in the last two decades, they are still nowhere near the point of putting us and many other HPC programmers out of business. Compilers are still a "one-size-fits-all" solution that needs to be able to deal with any and all input, whereas HPC programmers can be thought of as a designer-fitted solution. Application expertise brings a lot to the table that compilers cannot compete with: Our past experiences have helped us optimize applications that have irregular memory access patterns. While some applications such as matrix applications have regular and simple ...

How to Make GPU Hardware Decisions

Scott Computing Trends, CUDA, Hardware, OpenCL Leave a Comment

We get questions all the time about how to make GPU hardware decisions. We've seen just about every scenario you can imagine, and so we always jump at the chance to help others through this decision process. Here's a recent question from a customer. "I've just found your post on Analytic Bridge and have taken a look at your website ... I'm replacing my two Tesla M1060 cards (computing capability too low) and I'm considering used Tesla M2070s or the new GTX 760 cards. Could you offer any insight? I believe the GTX 760 cards may well outperform the older 2070s and are much cheaper." And here's our response. "The GTX 760 will probably outperform the M2070 for single precision ...

APU 2013 – Day 3 Recap

John Computing Trends, Events, OpenCL Leave a Comment

Big announcement here at #APU13! AMD CTO, Mark Papermaster, just announced 2 additions to the 2014 Mobile APU roadmap — AMD (@AMD) November 13, 2013 Today was the final day of AMD's APU 2013 conference. The theme of today was mostly focused on gaming topics, so it was not as relevant to technical computing as yesterday. However, the mobile product announcement from AMD in the tweet above was interesting. OpenCL is just as important in mobile computing as it is in HPC computing. Both ends of the spectrum have a need for speed and can achieve it through great data parallelism. AMD is looking to make better inroads into mobile computing with these APU announcements. Overall, APU 2013 was a fantastic ...

APU 2013 – Day 2 Recap

John Computing Trends, Events, OpenCL 1 Comment

Today was the first full day of AMD's APU 2013 conference. It was a whirlwind of heterogeneous computing. From the morning keynotes, three particular salient points stuck out to us: Mike Muller, CTO at ARM, talked about heterogeneous computing. He said it nicely with, "Heterogeneous computing is the future. It has also been our past, but we didn't notice because a few shiny companies overshadowed everything else." That is a great way to describe it. The future of heterogeneous computing involves the rise in importance of non-x86 processors. Throwing a few more MHz onto a CPU no longer is capable of satiating computational demands. Nandini Ramani, VP at Oracle, talked about the importance of Java for heterogeneous computing. She pointed ...

Application Time vs Solver Time

John ArrayFire, Computing Trends Leave a Comment

Last week, HPCwire ran an interesting article entitled, "Where has HPC's math gone?" The article analyzes the increasing importance of math solvers to successful HPC outcomes. As the number of cores grows, the percentage of time HPC codes spend in solvers increases significantly. The following chart illustrates this trend nicely:   ArrayFire is ideally suited for HPC applications that need to accelerate the toughest math problems. ArrayFire contains hundreds of math functions across numerous domains. In general, if the HPC community really wants to solve this problem, it will begin to invest more in libraries than in compilers that have no chance at optimizing these tough math problems automatically. Rather, it is only through expertly-tuned codes, such as those developed ...

ISC 2013 Keynote by Stephen Pawlowski of Intel

John Computing Trends, Events Leave a Comment

Stephen Pawlowski of Intel gave an interesting keynote today at ISC 2013. He continued the theme of yesterday's keynote to address challenges our market faces in getting to exascale computing. Here is a summary of the points he made during his talk: Getting to exascale by 2020 requires performance improvement of 2x every year Innovations anticipated include stacked chips and optical layers DRAM is not scaling with Moore's Law More power goes into transferring data than in computing Need to operate transistors near threshold New materials for DRAM needed. Resistive memory could replace DRAM. Need to explore both the big die and the small die paths as we approach 2020 Big die path leads to 10 billion transistors on a ...

ISC 2013 Keynote by Bill Dally of NVIDIA

John Computing Trends, Events Leave a Comment

Bill Dally of NVIDIA gave a wonderful keynote today at ISC 2013. He focused on addressing the challenges facing our market in getting to exascale computing. He talked about how Moore’s law is alive and well because transistors continue to double at an astonishing rate. However, the additional transistors are not translating into the same big performance gains as they did in the 1990’s. Whereas performance used to grow 50% per year, performance today is growing at a much slower pace. The biggest bottleneck to more performance is energy efficiency. Bill showed slides of chips and talked about the picojoules required to compute versus those required to move data and operands around the chip. The take home message was that ...

Are You Getting Left Behind?

John Computing Trends Leave a Comment

HPCwire posted a nice article today with trends from IDC on computer processing. These trends fall inline and corroborate things we've been saying here on this blog. Accelerators (including GPUs and co-processors) are taking off. Are you getting left behind? If you're reading this blog, you're probably at the bleeding edge, but nonetheless here are some interesting excerpts from HPCwire's market report (go read the whole thing): "While they expected to see a jump in coprocessor and accelerator uptake, they were wholly unprepared for the overwhelming positive response to GPUs and new entrants into the market, most notably Intel’s shiny new Phi." "Conway said that while accelerator and coprocessor adoption growth was anticipated, they had no idea that it would ...

History of the Modern GPU Series

John Computing Trends Leave a Comment

Graham Singer over at Techspot posted a series of articles a few weeks ago covering the history of the modern GPU. It is well-written and in-depth. For GPU affectionados, this is a nice read. There are 4 parts to the series: Part 1: (1976 - 1995) The Early Days of 3D Consumer Graphics Part 2: (1995 - 1999) 3Dfx Voodoo: The Game-changer Part 3: (2000 - 2006) The Nvidia vs. ATI Era Begins Part 4: (2006 - 2013) The Modern GPU: Stream processing units a.k.a. GPGPU Enjoy!

Parallel Software Development Trends for Dummies

John Computing Trends Leave a Comment

Last month, I posted two articles describing computing trends and why heterogeneous computing will be a significant force in computing for the next decade. Today, I continue that series with an article describing the biggest challenge to continued increases in computing performance - parallel software development. Biggest Challenge As I described previously, in order to use an accelerator, software changes must be made. Regular x86-based compilers cannot compile code to run on accelerators without these needed changes. The amount of software change required varies depending upon the availability of and reliance upon software tools that increase performance and productivity. There are four possible approaches to take advantage of accelerators in heterogeneous computing environments:  do-it-yourself, use compilers, use libraries, or use ...