I was more than a bit surprised to learn on Thursday morning that RapidMind had been acquired by Intel. Surprised not so much because it happened — 2009 has been a frenzy of acquisition activity — but because neither Intel nor RapidMind saw fit to issue a press release about the buyout. A statement on RapidMind’s Web site delivered the news as follows:
We are now part of Intel Corporation, the leader in software development tools for software parallelism and performance. This change is a great development for our customers and for easing the effort in parallel programming. With Intel, our shared vision for enabling parallelism will lead to strengthening our software developer tools for multicore and manycore processors and accelerators. Later this year, we will provide more details about the integration of the RapidMind platform and Intel software products and technologies, including the Intel Ct technology for data parallelism.
We continue to sell the RapidMind platform and we encourage you to contact us to explore your interest in solutions for software parallelism. We respect and strive to preserve and expand the investments our customers have made in our products. You are encouraged to visit the Intel Developer Forum in September 2009 to hear more about Intel software developer tools.
Please sign-up for the Intel data parallel newsletter at www.intel.com/software/data_parallel to get information direct from us.
After PeakStream was acquired by Google in 2007, RapidMind remained as the only software provider for a hardware-independent multicore development platform aimed at HPC workloads. The obvious question that now arises is if RapidMind’s platform will remain neutral as far as hardware (i.e., vendor) support. The last time I looked, the RapidMind platform included support for Intel CPUs, NVIDIA GPUs, AMD (both CPUs and GPUs), and the IBM Cell processor. Since Intel has little interest in adding value to other peoples’ silicon, it’s a little worrisome to see RapidMind’s software get swallowed like this.
The good news is that the RapidMind technology is getting funneled into Intel’s software group, which does maintain a degree of vendor neutrality. For example, its compiler and library products support both Intel and AMD x86 architectures. Also, because Intel’s Threading Building Blocks (TBB) is open source, support for PowerPC and SPARC architectures is now available for that technology.
According to a blog post by Intel multicore software evangelist James Reinders (hat tip to John West at insideHPC for the pointer), the company is “continuing to sell and support Rapidmind products.” The plan, apparently, is to mind-meld the RapidMind technology with Intel’s parallel software products and the upcoming Ct offering.
Ct (C/C++ for throughput computing) is Intel’s new computer language aimed at making data parallel programming accessible to average programmers. However, Ct is not intended to support standard GPU architectures (but is tailor-made to support Intel’s upcoming Larrabee architecture). And since none of Intel’s other software products support NVIDIA or ATI (AMD) technology, it wouldn’t be surprising to see RapidMind’s GPU support fall by the wayside. This may not be a huge tragedy in itself. As far as I can tell, the RapidMind platform has little market penetration today, and for GPU programming, most high-level implementations are being built on top of CUDA and OpenCL.
Because Ct is being backed by the world’s largest chipmaker, it stands to have a much brighter future than RapidMind probably could have attained on its own. In any case, we’ll have a better idea of what Intel has in mind fairly soon. Reinders says Ct is on track for a beta release before the end of 2009, with the integration of RapidMind technology set to be deployed in phases shortly thereafter.