I have been a strong proponent of accelerators for quite some time. Unfortunately, as indicated, it has been sadly, lacking success in trying to convince VCs and others to help fund the development we saw was needed. “The market will be there”, we said. “When” they asked. “Soon” we replied. That wasn’t good enough for them.
That was ~2 years ago. Now, free from much of the hype (though marketeers still inject a little every now and then), we see a rapidly developing accelerator marketplace. From leading the market to market bleeding edge, we are now seeing significant and sustained growth in accelerators.
HPC as a market has been, and will continue to evolve. We have been making this point for a while. It moves down market. It has to. Whats pushing it down-market are economic forces. Customers no longer want to spend $1M for their HPC. They don’t want to spend $100k for their HPC. $10k is a stretch. $1k would be great. The hardware has become far less expensive, as it is leveraging commodity parts in most cases. The applications have solidified behind a single OS instead of 10 disparate OSes. The 64 bit transition is largely over with, though quite a few non-HPC folks still run their 64 bit machines on 32 bit OSes … either due to a lack of understanding of what they get (literally for free) by switching to 64 bit OSes, or their natural conservatism, or … Whatever the reason, we are seeing strong upsurge in the use of HPC OSes on desktops, laptops, and clusters. Same OS everwhere, running windows apps either emulated, leveraging native versions, or via virtual machines.
The evolution of the HPC market is driving the cost of many other things down, and hopefully the ease of use, up. Our JackRabbit systems provide industry leading price performance, performance, and price, recent HPCwire article notwithstanding (which show a competitor winning price performance while ours is something like 50% better … go figure). Our DragonFly software provides a consistent web view of your HPC systems applications, making adding new applications to your HPC systems easy, and enabling you to work with your HPC systems from a web page. This software has many revolutionary ideas encapsulated within it, but the bottom line is to make it easy to use for everyone. Windows users who are GUI centric as well as command line users.
This lowering of barriers, price barriers, ease of use barriers, while raising performance targets, is exactly what end users ask for. The don’t want it to be complex. As simple as possible. As high performing as possible. And on a reasonable budget.
A PS3 has a 0.2TFLOP chip in it. Even if you can only get 10% of its power available to you, thats still a significant step above what you can do without it, at a $500 price point. A GPU accelerator card might be able to do 0.5TFLOP, of which, again, if you can get 10% of this power for your code, you are far ahead of what you can easily do on the desktop, or even in a number of cases, a small cluster.
Some of the FSB/HTX based FPGAs may have enough gates to do some interesting calculations. If they can be had for a reasonable price, and you have a good programming environment, end users would be able to use them.
The hard and fast growing segment of the market, the portion of the market which has been driving HPC for years, are growing more fond of the accelerators. This isn’t a fad, like some of the other things we have seen pop up and fade away. This has been growing and building for years. This is the direction of evolution of the market. If your low end HPC solution is above about $10k, you are probably not going to sell many of them. Which is a shame as the low end is becoming a volume market. Into which the accelerators will fit nicely.
I wish the VCs had understood. Its not too late, but they seem too busy finding the next great hype wave (SaaS, facebook/myspace/social-network-version-10000/web-85.0) that they may be missing where real market potential is. So what happens when HPC systems cost well under $10k, and you have apps for them? How many will you sell? Won’t be hundreds, thousands, or tens of thousands.
Its exhilarating seeing what we predicted come to pass, almost exactly as we predicted it. It is tremendously frustrating not having the capital to do much in this space.