While our computers today are often extremely fast, most applications aren’t optimized for the hardware platform they are running on. Libraries are often compiled for a generic platform and can’t make use of the specific features of a given CPU, for example.
Bitfusion, which debuted today at TechCrunch Disrupt NY, wants to automate all of this for developers and promises that this first iteration of its solution can result in speed improvements of up to 4x. That’s only by building and selecting pre-optimized versions of popular open-source libraries that many popular applications rely on. Some of the markets the company is targeting include pharmaceutical and bioinformatics companies, as well as data analysis software.
The company, which was founded by three former Intel employees in Austin, also today disclosed that it has raised a $1.45 million seed funding round from Data Collective, Resonant VC, and Geekdom — the investment vehicle founded by a number of RackSpace founders and investors.
The founders have worked on everything from chip design to public cloud infrastructures and supercomputers. CEO Subbu Rama previously also founded SocialStock, a 2012 Disrupt NY Battlefield finalist.
“One of the main things we bring to the table is that we try to bring hardware and software together,” said Rama. “We are trying to focus on the area that nobody is focusing on: utilizing hardware better.”
Bitfusion re -compiles, curates and collects libraries that it can optimize for different platforms. But that’s only the first step in the company’s roadmap and is mainly meant to help it build a customer base. In our interview, Rama called this “the easy stuff.”
The team plans to go beyond the CPU and start using the GPU and FPGAs to enable performance boosts of up to 100x. The company says that hardware manufacturers today can’t agree on a standard for accessing many of the more powerful features of their chips.
A number of groups have lobbied for OpenCL to be the standard for interfacing between hardware accelerators and scientific computing. But the reality today is that developers can’t rely on OpenCL support and instead are faced with a plethora of competing APIs like CUDA, HAS, AVX2 and others. The Bitfusion team believes it can help bridge these worlds by doing the work of optimizing applications and libraries for these developers and their (sometimes rather exotic) chips.
One area the team seemed especially excited about is FPGAs, which are essentially reprogrammable chips. Those FPGAs could be reprogrammed to solve a certain problem and, because doing anything in hardware is always going to be faster than in software, developers should see huge speedups from this. While FPGAs aren’t all that standard today, some of the large semiconductor manufacturers are thinking about them, including this technology on their CPUs in order to bring more room for speedups and flexibility to their chips. Binary translation is another area that the Bitfusion team is targeting to speed up applications.
The team plans to work with enterprise customers to build solutions for their specific needs and architectures. For them, Bitfusion will offer an appliance with built-in hardware accelerators that they can use in their own data centers. In addition, the team is working with RackSpace to build an accelerated cloud that its users can boost up to speed up their applications.[gallery ids="1154494,1154493,1154492,1154491,1154490,1154489,1154488,1154487,1154486,1154485,1154483,1154482,1154481,1154480,1154479,1154478,1154477,1154476"]
Q: How big is the market for this?
A: We are going after the large number of small and medium businesses that don’t have the in-house expertise to do this. Also media and virtual reality companies. .
Q: What is your business model?
A: We have three different business models. Software, our appliance (with hardware accelerators), and the accelerated RackSpace Cloud.
Q: Is this plug-and-play?
A: The technology we are developing is the largest set of software-accelerated libraries out there. And then we deploy these libraries to specific applications in the machine learning and bioinformatics space, for example. All of this is done automatically.
Q: Why aren’t you going after large-scale enterprises?
A: Large enterprises can build their own hardware and have the skills to do this for their specialized applications. We can work with the enterprises, too, though and we are already working with a large data center customer.