Your browser is not supported. Please update it.

15 November 2012

Texas Instruments pushes Keystone towards supercomputers

When Texas Instruments unveiled its Keystone system-on-chip architecture, it was clear that the brand new platform was designed to expand TI’s reach beyond its traditional stronghold in base stations. How far it aims to stretch is more surprising though ‘ after predictably pushing further into network processors and small cells earlier this year, it is now targeting servers as well as a wide range of networking gear with its combination of multicore ARM processors, DSPs and accelerators.

Like Freescale, TI is looking to provide a single architecture and software base which run the whole RAN gamut from metrocell to macrocell, and then beyond the base station to the network server, which has traditionally been a separate segment, and now to the general purpose server. When it unveiled the second generation of its KeyStone architecture earlier this year, it boasted up to four ARM Cortex A15 cores and multiple C66x DSPs, but perhaps more significantly, hardware acceleration for switching, packet processing and media. This inclusion of packet processing and switching right on the chip makes KeyStone more suited to data-driven computing platforms, and will help further two aims close to TI’s heart ‘ bringing the DSP into the server, and integrating networking on-chip. In the latter case, integrating up to five 10-Gigabit Ethernet ports on the chip not only supports intensive real time information processing and movement, but could do away with the need for a top-of-rack switch.

Already this year, KeyStone SoCs have been unveiled targeting the Cloud-RAN, network server and embedded equipment. The latest members of the family add to the options in the networking and embedded spaces, and add models targeted squarely at servers, as TI seeks to extend its ARM/DSP/network architecture to every kind of equipment that is used in the cloud.

Marvell and start-up Calxeda have led the way into using ARM-based processors in servers, mainly addressing low power blade servers for cloud providers. TI has stated it is not primarily targeting the direct replacement of server blades, but instead is looking at the trend for large data centers to turn away from monolithic server designs in favor of ‘breaking jobs down and putting some things on a DSP or A15’, as Tom Flanagan, director of technical strategy at TI’s wireless unit, puts it. Power efficiency is also critical to these large centers, and TI says it can deliver as much as 10 times the MIPS per power as x86 in a heavy number crunching environment. Therefore, while the shift to cloud architectures is TI’s key driver, it is looking right up to high performance servers and supercomputers, where ‘ rather than seeking to oust the incumbent Intel x86 architecture – it will offer chips to offload heavy duty floating point tasks. In this role, it will take on Nvidia, which makes a similar play, though using its graphics processing units rather than a DSP-based solution like TI’s. Target functions would include media processing, video analytics and industrial imaging and control.

The latest Keystone chips, most of which are going into production now, consist of six 28nm SoCs based on ARM Cortex A15 CPUs, C66x DSPs plus security and networking blocks. As well as servers, they will target routers and switches, opening another front in TI’s war against Freescale, as well as challenging Cavium and LSI. The last of these has been the flagwaver for using ARM cores in the network infrastructure market, while Freescale also recently gave ARM a significant boost by including the Cortex platform in its PowerPC-based base station range ‘ though, for now, only at the low end.

Flanagan believes the open ecosystem of ARM will ensure its success in many markets, especially if chip vendors stick with a unified approach ‘ unlike some ARM users such as Qualcomm and Apple, TI uses the standard core design rather than adapting it, and relies on its accelerator and software efficiency to deliver a performance differential. The biggest of the new chips use up to four A15 cores and six DSPs to deliver up to 352GMACs and 19,600 Dhrystone MIPS.

Distinguishing features of its Keystone chips include a 256-bit interface from the cores to the SoC, clocked at the full 1.4GHz data rate of the cores, while some rivals use 128-bit interfaces clocked at only 33% to 50% of that rate. TI also squeezes up to 18Mbytes of aggregate memory on its top end SoCs and says it is unique in supporting Gigabit and 10Gbit Ethernet MACs.

The last of these features is critical to its positioning. Integrating networking onto the chip is a popular topic among chip designers as servers need to take on far greater levels of real time data analysis and other tasks where information must be processed in parallel and at ultra-high speeds. However, TI is rolling out 10Gbit capabilities imminently while Intel, for instance, is talking about 100Gbit networks-on-a-chip, but with no timelines. But adding integrated networking to a DSP/ARM architecture, in real world products, may allow TI to succeed in its aim of getting DSPs into servers.

While the high end server models will get the most attention, only two are devoted to that space. Another two come with just one DSP core and are aimed at routers and other wireless infrastructure plus cloud platforms. And two low end variants will focus on embedded and industrial products such as traffic or surveillance cameras, or VoIP, and come with no DSPs. These build on an earlier decision by TI, discussed back in March, to push KeyStone into the embedded world. Then, it release three chips for devices in sectors like industrial automation, imaging, video surveillance, medical, and audiovisual infrastructure. These are the dual-core TMS320C6657, which runs on two 1.25GHz DSP cores, delivering up to 40GFLOPs, and the C6655 and C6654, single-core parts providing up to 20GLOPS and 13.6GLOPS, respectively.

Among the new offerings, the lower end chips will ship in the second half of next year while their high end stablemates are sampling from next month. While lacking the glamor of the supercomputer chips, the networking infrastructure and the industrial sensor-based chips are aimed at markets where TI has strong roots, and so is trying to win more slots with well understood customers, rather than attacking a new space. But it is in servers where the biggest spoils are available, and where TI can finally make the concept of DSPs inside servers, which it has talked up for two or three years, a commercial reality ‘ not to mention supporting another key concept, fully integrated networking on a chip, well ahead of its noisiest exponent, Intel. As in mobile device processors, where TI is backing away from the tablet rat race just as so many players are entering it, in servers it is avoiding the most fashionable approach, the Calxeda-style cloud platform, and eyeing a somewhat different twist on the modern server. As such, for all its shift towards analog businesses, it may find a new lease of life for its DSP activities too.