64 Core Processors - MM4M

LinuxDevices.com, Aug. 20, 2007
A startup founded by an MIT professor claims to have "solved the fundamental challenges
associated with multicore scalability." Tilera's first products include a 64-core Tile64 SoC
(system-on-chip), PCIe Express add-in board for networking and video-processing applications,
multicore-optimized Linux libraries, and an Eclipse-based multicore development environment
toolset.
Fabbed on 90nm process technology at TMSC, the Tile64 chip is "the first in
a family of chips that can scale to hundreds and even thousands of cores,"
Tilera said. The company plans to bake a 120-core version on 65nm
technology in the future, it added.
The Tile64 is based on a proprietary VLIW (very long instruction word) architecture, on which a
MIPS-like RISC architecture is implemented in microcode. A hypervisor enables each core to run
its own instance of Linux -- or other OSes, once they become available. Alternatively, the whole
chip can run Tilera's 64-way SMP (symmetrical multiprocessing) implementation.
Tilera Tile64 architecture
The crown jewel of the Tile64 architecture is a network-like "iMesh" switching interconnect said
to eliminate the centralized bus intersection that in previous multicore designs has limited
scalability, according to the company. Tilera's founder, MIT professor and serial entrepreneur Dr.
Anant Agarwal, has experimented with mesh-like chip interconnects since 1996, the company
said.
Each of the Tile64's cores clocks between 600MHz and 900MHz; each has its own L1 and L2
cache. L3 cache is handled in an interesting way, as Bob Doud, director of marketing, explains.
"We take all the L2 caches and consider them in aggregate to be the L3 cache," he said. "It's
highly effective, because if you reference your own cache, and don't find the data you're looking
for, a neighbor may have it, and that's faster than going off chip to external memory by a good

ways."
The Tile64 is implemented as a a system-on-chip (SoC) with no requirement for external
northbridge and/or southbridge chips. This saves power, at the expense of locking in a specific
peripheral mix, essentially tying the chip to specific verticals, according to the company. Doud
noted, "We did a lot of research, and believe we have the peripheral mix right for the markets we
are targeting -- networking and video. If we went into storage with a new processor, we'd add
fiber and disk drive interfaces."
Tilera claims that the Tile64 outperforms Intel's dual-core Xeon processor 10 times, while
offering 30 times better performance per Watt. Compared to TI's top-of-range TMS320DM648
DSP, performance per Watt is claimed to be 40 times better.
Another touted benefit is the ability to consolidate control- and data-plane functions on a single
device, with "solid-wall" processor boundaries reinforcing security and licensing containment
barrier. In this regard, the Tile64 chip resembles another heavily multicore MIPS64 chip,
Cavium's 16-way Octeon.
Software environment and tools
Tilera claims that existing, "unmodified" Linux apps will build for the Tile64 processor using the
company's toolchain. The toolchain includes a compiler licensed from SGI and based on
MIPSpro.
Alternatively, developers can port their applications to Tilera's iLib C library, aimed at exploiting
parallelism while still supporting standard system calls. The approach appears to resemble that
used in Intel's Threading Building Blocks, recently released under an open source license.
Finally, for users wishing to manually tune multi-core application performance, Tilera will offer a
full "MDE" (multicore development environment) toolsuite based on Eclipse. In addition to a full
IDE (integrated development environment), MDE includes a parallel debugger, along with an
application profiler aimed at helping developers figure out what parts of their code to optimize for
multicore.
Tilera is in discussions with "all major Linux support providers," Doud said, adding, "We'll have
ecosystem announcements coming out as we line them up."
Early markets, customers, reference implementations
The first Tile64 chips target network and video devices that require significant application
processing, such as surveillance systems, and firewalls with deep packet inspection. Early
customers in the networking space reportedly include 3Com and firewall vendor TopLayer, while
early video customers reportedly include U.K.-based high-definition videoconferencing
equipment provider Codian, and BackupTV, a vendor of network-based video recording and
other head-end services for cable TV network providers.
To hasten adoption, Tilera is offering processor daughterboards implemented as PCI Express
cards with six or 12 gigabit Ethernet ports. The cards can be used in production systems with
passive backplanes, or as targets in development hosts, Doud said. He declined to specify
pricing.

CEO Devesh Garg stated, "This is the first significant new development in chip architecture in a
decade. We developed this new architecture because existing multicore technologies simply
cannot scale. Moreover, customers have repeatedly indicated that the current multicore software
tools are very primitive because they are based on single-processor-core models. We're
introducing a revolutionary hardware and software platform that has solved the fundamental
challenges associated with multicore scalability."
Availability
The Tile64 is available now, in three variants differentiated by I/O mix and clock. Pricing starts at
$435 in 10,000 quantities, the company said. Tilera's iLib and MDE tools, and TilExpress-64
board are also available at undisclosed pricing.
Henry Kingman