Recently Acceleware celebrated its 5th Anniversary.  This was a significant milestone for our company, but it also caused me to reflect on the significant evolution that has occurred in the GPU/parallel computing space over the last five years.  An evolution for which we’ve had more than a front row seat, an evolution where we’ve been an active participant, and have evolved greatly ourselves in the process.  I’d like to highlight three main points in this evolution; I will call them the  three “A”s.  They are: Acceptance, Availability, and Awesomeness.  In summary, acceptance of GPU/parallel technology and the need for software enablers, increasing availability of hardware and software solutions that leverage GPU computing, and the sheer improvement of GPU and multi-core technology itself.

Acceptance.  When we started looking at hardware acceleration, it was a bewildering landscape of different technologies, multi-core CPU, FPGAs GPUs, and later Cell, none of which had gained significant penetration into the commercial market.  We eventually settled on GPUs and multi-core technology as they met the needs of our customers the best.  Cell and FPGAs have also settled into applications that require the unique benefits that fit their strengths.  As I’m sure was the case for every company doing hardware acceleration, when we approached software vendors (ISVs) with the idea of running their stuff on GPUs, they were highly suspicious of  a using a “gamer” technology.  They were doubtful that it would generate correct results, that it would be reliable, and that we could deliver a solution that an end-user would buy.  I wouldn’t be writing this blog had we not been successful, and for that I thank our awesome development team, and the vision and foresight of our ISV partners in adopting this technology. but the point I want to make is that the attitude of ISVs towards GPU computing  now is radically changed.  Both our existing partners and many of our end users are keenly aware of the latest advancements in GPU and multi-core technology and can’t wait for us to support each of the new hardware platforms as they are released.

We also have ISVs coming to us now, wanting to understand how we can enable their software to take advantage of parallel computing hardware.  Five years ago, most ISVs wouldn’t give us the time of day.  We don’t even have to explain to them what a GPU is anymore.  Thanks to NVIDIA and the work of others doing GPU computing, not to mention the marked slowdown in “free” performance gains from CPUs, GPUs are becoming more and more accepted as a technology.  It certainly makes my job easier.

Availability.  Beyond acceptance, the general availability of GPUs also makes our job easier.  In the days prior to the acceptance described above, high-end GPUs were somewhat difficult to procure given specific and significant requirements on the workstations that they plugged into.  This forced Acceleware into becoming a de-facto systems integrator.  Despite the fact that our core expertise was software and parallel programming, we were forced to test and validate hardware, carry inventory, and build workstations such that we could deliver our software libraries on a system that could use them.  The day that NVIDIA announced the Tesla Personal Supercomputer and it’s availability through a number of worldwide partners was a great day for Acceleware.  Some people here joked that NVIDIA had “stolen” our idea of a workstation solution, but believe me, that was one idea that we wanted stolen.  We weren’t experts at delivering hardware and it came at a great cost to our fledgling company.  NVIDIA’s Tesla Preferred Partners are experts at delivering these solutions and we have since been working with Colfax and Penguin Computing as our own preferred partners.  Even Dell (and HP I hope soon) are starting to offer high-end Tesla GPUs as standard orderable SKUs and systems.  This has allowed us to focus on what we do best delivering GPU/parallel enabled-software that runs on this fantastic hardware.

Awesomeness.  Fantastic hardware is an apt description for the latest Telsa GPUs.  We’ve seen performance more than quadruple in a little more than two years, as has the size of simulation and problem that our partners can run on a single GPU.  Amazing!  Furthermore, and this is where it’s great to be an engineer, it is cheaper and more robust.  The Tesla C1060 can outperform two of the previous generation cards for less than 1/6 the price.  $1350 vs. $7000+ MSRP.  This price performance advantage removes yet another roadblock to acceptance for GPU technology.   Now here’s the part I love the most, NVIDIA is spinning a new GPU about every 12-18 months and the performance to date have been significant 2X+.  Compare this to Intel’s more modest two year refresh cycle and performance gains.  This is where the investment in GPU / parallel technology will really pay dividends as once your code is written in parallel for GPUs, you get the gain for very little effort, even less if you use Acceleware libraries.  How awesome is that?  I just yelled out I “love my job.”

In summary, many burdens and roadblocks that Acceleware has lived with and jumped over in the past are evaporating.  The performance of GPUs makes people take notice.  The fact that others are already enjoying the benefits makes the technology easy to accept.  All at a cost and availability that is significantly lower than it was before.  The final key to the puzzle is the parallel code, and Acceleware is there to make that part just as easy for you.

Comments

its good to see Acceleware has a blog. As a long term investor in the company at times i've felt as though the company was not communicating w/ us or marketing its self properly. hopefully this will change. good luck

# Posted By austin | 6/27/09 10:47 AM

Maybe you can explain why your company still has a 3 million dollar market cap with all the opportunity currently . Do you expect and would you accept a takeover bid from NVIDIA?

# Posted By chris Parsons | 7/20/09 5:53 PM