Follow us on Facebook
Follow us on Twitter
Signalogic on LinkedIn

Home > Products > HPC > CIM Arrays > TI Multicore CPUs

TI Multicore CPUs


TI C66xx series multicore CPUs--
up to 8 cores at 1.25 GHz per core



TI OMAP 3630 used inside Motorola Droid II phones.
The 3630 is under the Elpida SDRAM (MCM stacking)
(picture Copyright Grandmax 2010)



TI Davinci DM6467 inside a panoramic camera
(picture copyright Ephel Labs, 2010)



Texas Instruments multicore CPUs are suitable for CIM for a number of reasons. First and foremost, they process standard C code, without forcing Linux developers into a series of function calls, or "APIs" (as is the case with CUDA and OpenCL), or forcing developers into programming at the "assembly language" level using a cryptic, narrowly defined instruction set. This is not to say that API based acceleration is not good -- in many cases it can be quite effective and yield excellent performance. But for Linux development, it's easier and less resource intensive to a) avoid rewriting existing, working code so that it instead uses a series of APIs, and b) not have to think in terms "how to write code to fit my accelerator" when creating new code.

Second, TI multicore CPUs offer high performance per mm2 and per mW; i.e. they have high performance per square area and per mW of energy (power). In a word, they have high performance density. Examples visible in the market include OMAP devices found in smart phones and DaVinci devices found in digital cameras (see pictures at right).

TI Advantages

Due to TI's long and involved experience in embedded products, TI multicore chips excel at both SIMD processing (like Nvidia) and general-purpose processing (like Intel). TI chips have for many years been found in small, embedded products such as mobile phones and digital cameras, which require very low energy consumption and small package size, while still maintaining high performance. The pictures at right show two examples of consumer products where compute-intensive performance is crucial, even with package size and power consumption constraints. The smart phone runs compute-intensive voice and video codecs, and the digital camera does H.264 video compression.

Here are some specific TI multicore CPU advantages:

In part, TI obtains these advantages due to its chip design and fab process, developed through painstaking research and practical experience over more than 30 years. While Intel has focused during this time mostly on performance and input/output required for desktop applications (e.g. programs that run under Microsoft Windows), TI has focused on both performance and reduced energy consumption. As one of the world's top 5 semiconductor manufacturers, it's safe to say that TI's performance-to-energy ratio is the best available, making it the optimized solution for increasing server efficiency.

Lab Measurements

Here is a data point for a TI multicore CPU (2008 chip) vs. a quad-core Penryn x86 server (this one from a convolution with filter and data lengths about 37,000 and 196,000):

Penryn quad x86 Server w/ OpenMP 1 C6472 w/ CIM OpenMP
1 sec 0.65 sec

With the SigC6678 CIM array card currently under development, 500x acceleration vs 8-core x86 will be obtainable for certain compute-intensive applications.

CPU Types Used with CIM

Here are some of the TI CPU types currently supported by CIM technology:

1 With dynamic scheduling (workload adjustment) and O3 optimization level