Advanced Micro Devices, the world’s second largest producer of x86 microprocessors, has announced that it encourages development of add-in co-processors or accelerators for servers running AMD Opteron chips. Just a little more than a decade ago both market leader Intel and AMD essentially killed the market of math1 co-processors by incorporating the technology into their central processing units (CPUs).

“Torrenza represents the industry’s first open, customer-centric x86 innovation platform, capitalizing on the Direct Connect Architecture and HyperTransport advantages of the AMD64 architecture to enable other processor and hardware providers to innovate within a common ecosystem. Torrenza” will enable a global innovation community to develop and deploy application-specific co-processors to work alongside AMD processors in multi-socket systems,” a statement by AMD reads.

AMD said that Torrenza platform will allow third-party developers of various co-processors and accelerators to fit them into standard AMD Opteron-based servers, either into a separate socket for processors, into Hyper-Transport HTX slot of PCI Express slot. Additionally, it is possible that AMD will built-in certain technologies into its Opteron processors or just put additional core onto a single piece of substrate with its server microprocessor, according to the company.

According to AMD’s senior vice president Marty Seyer. the first phase of Torrenza was nearly complete via an earlier HyperTransport investment that enabled partner silicon chipsets. The next phase would “entail licensing of coherent HyperTransport to members of a global ecosystem; developments already in progress include support for an HTX expansion slot”.

Application-specific co-processors will allow AMD to address markets of very specific machines. Like the interoperability between high-end graphics chips and physics accelerators allow AMD to address the market of high-performance computers for gaming, security accelerators or ultra-fast math1 co-processors will permit AMD to install Opteron chips into pretty lucrative scientific servers.