The hpcnotes HPC blog - supercomputing, HPC, high performance computing, cloud, e-infrastructure, scientific computing, exascale, parallel programming services, software, big data, multicore, manycore, Phi, GPU, HPC events, opinion, ...
Saturday 30 October 2010
Comparing HPC across China, USA and Europe
In my earlier blog post today on China announcing the world's faster supercomputer, I said I'd be back with more later on the comparisons with the USA, Europe and others. In this morning's blog, I made the point that the world's fastest supercomputer, in itself, is not world changing. But leading supercomputers, critically matched with appropriate expertise in programming and using them, togther with the vision to ensure use across basic research, industry and defence applications can indeed be strategically beneficial to a nation - including real economic impact.
There are plenty of reports and studies describing the strategic impact of HPC within a given organisation or at national levels (some are catalogued by IDC here), so let's take it as a premise for the following thoughts.
Friday 29 October 2010
Why does the China supercomputer matter to western governments?
There is a lot of fuss in the mainstream media (BBC, FT, CNET, even the Daily Mail!) the last few days about the world's fastest supercomputer being in China for the first time. And much ado on Twitter (me too - @hpcnotes).
But much of the mainstream reporting, twitter-fest, and blogging is missing the point I think. China deploying the world's fastest supercomputer is news (the fastest supercomputer has almost always been American for decades, with the occasional Japanese crown). But the machine alone is not the big news.
Thursday 23 September 2010
Is power-hungry supercomputing OK now?
We may be planning for a 1,000-fold increase in compute power in the next decade, but what about the extra power consumption ...
http://www.zdnet.co.uk/news/emerging-tech/2010/09/23/is-power-hungry-supercomputing-ok-now-40090137/
Monday 13 September 2010
Do you want ice with your supercomputer?
“Would you like ice with your drink?” It’s a common question of course. One that divides people – few will think “I don’t mind” – most have a firm preference one way or the other. There are people who hate ice with their drink and those who freak if there is none. National stereotypes have a role to play – in the USA the question is not always asked – it’s assumed you want ice with everything. In the UK, you often have to ask specifically to get ice.
Yet the role of ice in making our drinks chilled is misleading. I once had a discussion with a leading American member of the international HPC community about this. “No ice”, he was complaining as we headed out of a European country, “they had no ice for the drink”.
“I don’t get this obsession with ice”, I chipped in. “What?!” He looked at me as if I were mad. “Why do you like your coke warm?”
“Ah, but that’s just it”, I replied. “I hate warm drinks – I really like my coke chilled. But surely, in this modern world over a century after the invention of the refrigerator, it’s not unreasonable to expect the fluid to be chilled – without the need to drop lumps of solid water into it?”
“Ah, fair point”, he conceded.
What has this got to do with supercomputing? Perhaps the common thread is that usually we just accept the habitual choices of ways to do things – and don’t often step back to think – “are those the only choices?”
Maybe we should step back a little more often and ask ourselves what we are trying to achieve with HPC – and are the usual choices the only ways forward? Or are there different ways to approach the problem that will deliver simpler, better or cheaper performance?
Perhaps your business/research goals mean you need to conduct more complex modelling or you need faster performance. Maybe the drive of computing technology towards many-core processors rather than faster processors is limiting your ability to achieve this. (I have had several conversations recently, where companies are buying older technology because their software won’t run on multicore).
The “ice or no ice” question might be whether or not to upgrade your HPC with the latest multicore processors. But what about the “just chill the fluid” option? Well, how about upgrading the software instead, or as well?
NAG has plenty of case studies to show where enhancements to software have achieved huge gains in performance or capability (e.g., www.hector.ac.uk/cse/reports).
Sometimes buying more compute power is the right answer. Sometimes, extracting more efficient performance from what you have is the answer. Bringing them together - a balance of hardware upgrades and software innovations might well give you the best chance of optimising cost efficiency, performance and sustainability of performance.
Monday 30 August 2010
Me on HPC 2
Successful Deployment at Extreme Scale: More than Just the Iron
The Exascale Report
August 2010, by John West
[full article requires subscription, extracts here are not complete, and are modified slightly to support that]
"cost of science, not just the cost of supercomputer ownership"
"lead time, and funding, to get the user community ready"
"spend a year or more selecting a machine and then deploy it as quickly as possible, makes it very difficult to build a community and get codes ready ahead of time"
"software must be viewed as part of the scientific instrument, in this case a supercomputer, that needs its own investment. High performance computing is really about the software; whatever hardware you are using is just an accelerator system."
"a machine is deployed and then obsolete within three years. And the users often have no idea what architecture is coming next. There is no real chance for planning, or a return on software development investment."
Monday 19 July 2010
Time Machines and Supercomputers
I found a Linpack App for the iPhone last week. Nothing special, just a bit of five minute fun. It seems a 3G model achieves about 20 MFLOPS. [Note 1]
What's that got to do with time machines? Well it got me thinking "I wonder when 20 MFLOPS was the performance of a leading edge supercomputer?" Actually, it was before the start of the Top500 list (1993), so finding out was beyond the research I was prepared to do for this blog.
So I thought instead about the first supercomputer I used in anger. As soon as I name it, if anyone is still reading this waffle, you will immediately fall into two camps - those who think I'm too young to be nostalgic about old supercomputers yet - and those who think I'm too old to be talking about modern supercomoputers :-).
It was a Cray T3D.
You're still waiting for the time machine bit ... hang on in there.
My application on that T3D sustained about 25 GFLOPS. Which is about the same as a high end PC of recent years. What this means to me is that anyone who cares to apply the effort today with a high end PC, could get comparable results to that work of 15-20 years ago that needed the supercomputer.
Or, in other words, that supercomputer gave us a 15-20 years time advantage over everyone who didn't have supercomputers - or a few years over others with smaller supercomputers. [Note 2]
That is one of the key benefits of High Performance Computing - the ability to get a result before a competitor - you could say HPC is a time machine for simulation and modelling.
Now for the [Notes] - which actually contain the real story!
Note 1 : It's not really true to say the iPhone 3G can do 20 MFLOPs - all we can say is that particular App achieved 20 MFLOPs on that iPhone 3G. The result is a factor of both the software and the hardware. Better performance can come from optimising the application as much as from buying a more powerful phone.
Note 2 : If fact, even with the same supercomputer, it would be hard for most people to replicate the results - simply because there was as much value in the software (physics, algorithms, performance engineering, implementation, etc) and the associated validation and verification program as there was in the supercomputer.
The supercomputer offered us a time machine. But the attention to performance and scalability in the application enabled us to actually use that time machine to get results faster than others - even if those others used the same supercomputer. And the validation and verification effort meant that we could trust what our time machine was telling us.
Wednesday 30 June 2010
Me on HPC and multicore
What You Should Know about Power and Performance Efficiency
Scientific Computing, August 2010, Suzanne Tracy
"Components driving power consumption fall into two categories — those that, as consumers, we cannot control, and those we can. Power consumed by server hardware is increasing and is beyond our direct control as buyers (although manufacturers are working to optimize power efficiency). The biggest factors we can influence are design and deployment of HPC systems as a whole (datacenter included) and recognizing total cost of ownership (including power) when procuring."
"The primary strategy for optimizing power is to ensure proper total cost of ownership (including power) as the driver of procurement, not purely peak performance and initial capital cost. This enables the evolutions of datacenter optimization (e.g. run warm, “free-cooling,” hot aisles) and choices of power-efficient HPC system designs (e.g. more parallelism, lower power processors, etcetera) to be correctly attributed as delivering increased performance against cost."
"Optimizing software and algorithms is a key opportunity to dramatically improve the total cost of ownership of HPC solutions. By optimizing applications, fewer resources are required to deliver the results, thus reducing the power required. Equally, innovations in algorithms can deliver applications that are power-aware — that is, they recognize the energy consumed and the user can balance energy-cost against time-to-solution when selecting algorithms for a given simulation."
"The primary breakthrough will be the recognition of the role software (both implementation efficiency and algorithm design) has to play in delivering cost savings related to power efficiency. Beyond that, the key hardware technologies will be increased use of power switching across the system — while many modern processors will reduce power when not fully utilized, the ability to gate specific parts of the chip will improve, and the same capability will work into other parts of the system — memory, interconnect (maybe balancing power against bandwidth on a job-by-job basis), I/O, etcetera."
Multiple cores multiply programming
Scientific Computing World, June 2010, Paul Schreier
"When it comes to parallel programming, it’s easy to do something that looks right, but it’s difficult to be sure it is right and will do the same thing under all conditions," says Andrew Jones.
"We strongly urge people to use prepackaged routines such as these where other people have done the difficult work of dividing up the tasks in an optimal way," says Jones.
Personal Supercomputers?
Genomeweb, October 2009, By Matthew Dublin
"There is always going to be a class of computing power that is much bigger than anything that will physically fit on your desk because if you can buy something for $1,000 or $10,000 then there are going to be users that are prepared to buy hundreds of them for a million dollars," Jones says. "And there's always going to be something that is orders of magnitude bigger than what most people can afford but the cheap stuff gets more powerful."
"I don't think there's anything wrong with the term 'personal supercomputing' if it successfully gets a whole lot more people making use of the compute power that's available," Jones says. "It's marketing, but it's perfectly valid marketing, aimed at an audience that would normally not go anywhere near large-scale supercomputers. ... HPC can do so much for people trying to do simulations and modeling that whatever we call it to get more people to using it, the better."
With virtualization, high-performance computing becomes more mainstream
SearchServerVirtualization.com, November 2008, By Jo Maitland
"Scheduling jobs, queuing jobs, shoring up resources, determining policies such as rejecting a job that doesn't have an estimate of how long the job is going to take … these are typical HPC skills but start to overlap when you're managing a virtualized compute environment," said Andrew Jones.
Jones said he does not believe mainstream computing will ever catch up with HPC. "By definition, HPC will always be more powerful than mainstream computing," he says.
Tuesday 22 June 2010
Technical computing futures part 2: GPU and manycore success
In my previous blog, I suggested that the HPC revolution towards GPUs (or similar many-core technologies) as the primary processor has a lot in common with the move from RISC to commodity x86 processors a few years ago. A new technology appears to offer cheaper (or better) performance than the incumbent, for some porting and tuning pain. Of course, I’m not the first HPC blogger to have made this observation, but I hope to follow it a little further.
In particular, my previous blog suggested the outcome might be: “at first the uptake is tentative ... but in a few years time, we might well look back with nostalgia to when GPU’s were not the dominant processor for HPC systems” – in other words, hard going initially, but GPU/many-core will “win” eventually. I even ended up with an ambitious promise for my next blog (i.e. this one): “an idea of what/who will emerge as the dominant solution ...”
Continuing the basis of using the past to guess the future, my prediction is that the next steady state of HPC processors will be GPU-like/manycore technologies (for most of the FLOPS at least) and, just like the current steady state (x86), those few companies with the strongest financial muscle will eventually own the dominant market share. However, other companies will have pioneered many of the technologies that make that dominant market share possible, enjoying good market share surges in the process.
I can even have a go at predicting some of the path that might get us to the next steady state of HPC architecture. NVIDIA has already shown us that GPUs for HPC are sometimes a good solution – and importantly, that a good programming ecosystem (CUDA) really helps adoption. Over the last year or so, I’d say the HPC community has moved from “if GPUs can work in this case ...” to “how do I make GPUs work across my workload?”
As Intel’s Knights processors bring us many-core but with a familiar x86 instruction set, we might learn that getting good performance across a broad range of applications is possible, but critically dependent on software tools and hard work by skilled parallel programmers. AMD’s Fusion with tighter links between CPU & GPU, could show that the nature of the integration between the many-core/GPU unit and the rest of the system (be it CPU, network, main memory etc) will affect not only maximum performance on specific applications, but maybe more importantly the ease of getting “good enough” performance across a range of applications.
I don't know of any GPU/many-core/accelerator announcements from IBM, but it’s always possible IBM will throw in another useful contribution before the dust settles. They were one of the first into many-core processors for HPC acceleration with Cell and they cannot be easily counted out of top end HPC solutions - e.g. the forthcoming Blue Waters (POWER7) and Sequoia (BG/Q) chart-toppers.
But back to my “winner” prediction. When the revolution settles into a new steady state of mostly GPU/many-core for HPC processors, there won’t be (can’t be) critical distinctions between the various products anymore for most applications. Whichever product we consider (whether GPU or x86-based or whatever), many-core is sufficiently different from few-core (e.g. 1-8 cores) to mean that the early winners have been those users who are easily able to move their key applications across to get step changes in cost and performance.
The big winners in the next stages of the GPU/manycore emergence will be those users who can move the bulk of their high-value-generating HPC usage to many-core processors with the most attractive transition (economy and speed) compared to their competitors.
So what about the dominant solution I promised? For the technology to be pervasive, first there must be greater commonality between offerings (I stop short of standardization) so that programmers have at least a hope of portability. Secondly, users need to be able to extract the available performance. Ideally these would mean a software method that makes many-core programming “good enough easily enough” is discovered – and if so, that software method will be the dominant solution, across all hardware.
Or, if the magic bullet is still not market ready, skilled parallel programmers will be the dominant solution for achieving competitive performance and cost benefits - just like it is for HPC using commodity x86 processors today.
Wednesday 16 June 2010
Supercomputing's future: Is it CPU or GPU?
Graphics processing units are a hot topic, but that does not assure them a place in supercomputing's future ...
http://www.zdnet.co.uk/news/it-strategy/2010/06/16/supercomputings-future-is-it-cpu-or-gpu-40089202/
Tuesday 8 June 2010
Revealing the future of technical computing: part 1
I recall some years ago porting an application code I worked with, which was developed and used almost exclusively on a high end supercomputer, to my PC. Naively (I was young), I was shocked to find that, per-processor, the code ran (much) faster on my PC than on the supercomputer. With very little optimization effort.
How could this be – this desktop machine costing only a few hundred pounds was matching the performance of a four processor HPC node costing many times that? Since I was also starting to get involved in HPC procurements, I naturally asked why we spend millions on special supercomputers, when for a twentieth of the price, we’d get the same throughput from a bunch of high-spec PCs?
The answer then (and now) was that I was extrapolating from only one application, and that application could be run as lots of separate test cases with no reduction in capability (i.e. we didn’t need large memory etc, just lots of parameter space). However, the other major workload (which I also ported and also ran fast on the PC) would not have been able to do the size of problem we wanted on a PC – we needed the larger memory and extra grunt from parallel processing. (We did look at the newfangled Network Of Workstations emerging at the time but decided it might be a wolf in sheep’s clothing. Sorry.)
In the end, we had to find a balance between (a) speed at lowest cost for the one application; (b) the best capability for the other application (i.e. fastest solution time for the largest problems); (c) ease of programming – to get a good enough (fast-enough) code developed with the limited developer effort and funding we had; and (d) whole life affordability.
Why do I foist this reminiscence on you? Because the current GPU crisis (maybe “crisis” is a bit strong – "PR storm" perhaps?) looks very much the same to me. The desktop HPC surprise of my youth has evolved into the dominant HPC processor and so for some years now, we have been developing and running our applications on clusters of general purpose processors – and a new upstart is trying to muscle in with the same tactic – “look how fast and how cheap” – the GPU (or similar technologies – e.g. Larrabee, sorry Knights-thingy).
The issues are the same: (a) for some applications, GPUs offer substantial performance improvements for considerably less cost than a “normal” HPC processor; (b) for other applications, the limits such as off-card bandwidth etc mean that GPU’s cannot deliver the required capability; (c) the underlying concern is ease of programming for GPUs; (d) affordability – sure GPU’s are cheap to buy, but what about power costs when in bulk, or code porting costs, etc?
Maybe the result will be the same as when commodity processors and clusters eventually exploded to leave custom supercomputer hardware as the minority solution. At first the uptake (now) is tentative - and painful. Some will have great success stories, many will get burnt. But in a few years time, we might well look back with nostalgia to when GPU’s were not the dominant processor for HPC systems.
I’ll continue on the future of HPC in my next blog in a few days, including an idea of what/who will emerge as the dominant solution ...
Tuesday 23 March 2010
What’s the next revolution in technical computing?
It’s a question that absorbs the attention of the technical computing community, especially those working at the leading edge of technology and performance (high performance computing, HPC). What is the next disruptive technology? In other words, what is the next technology that will replace a currently dominant technology? Usually a disruptive technology presents a step-change in performance, cost or ease-of use (or a combination of these) compared to the established technology. The new technology may or may not be disruptive in the sense of discontinuous change in user experience.
Why is identifying disruptive technology so important? First, those who spot the right change early enough and deploy it effectively can attain a significant advantage over competitors as a result of a substantial improvement in technical computing capability or reduction in cost. Second, identifying the right technology change in time can help ensure that future investments (whether software engineering, procurement planning, or HPC product development) are optimally spent.
However, in a field as fast moving as technical computing, spotting the next disruptive technologies of specific relevance to your individual needs can easily become a full time activity (which is why NAG helps to do this for others).
One very credible candidate for disruptive change in HPC right now is GPU computing (or related products that might be in development). However, at the Newport conference recently, the discussion turned to what the next disruptive technology to hit HPC would be (after the possible GPU disruption). One suggestion, made by John West (of InsideHPC fame), was that the next disruptive technology could be in software, especially programming tools and interfaces. This builds on the fact that parallel computing is no longer a specialist activity unique to the HPC crowd – parallel processors are becoming pervasive across all areas of computing from embedded to personal to workgroup technical computing. Parallel programming is thus heading towards a mass market activity – and the mass market is unlikely to view what we have in HPC currently (Fortran plus MPI and/or OpenMP, or limited tools, etc) with much favour. I’m not knocking any of these, but they are not mass-market interfaces to parallel computing. So perhaps the mass market, through volume of people in need – and companies driven by economics will come up with a “better” solution for interfacing with supercomputers.
As a HPC community we lost control of much of our hardware to the commodity market some years ago. Maybe we now face losing control of our software to the commodity community too.