Hot and cold
Even with on-chip power/thermal management measures, cooling is still very important.
“The cooler the chips stay, the longer the life expectancy,” Eder said. “The mean time between failures (MTBF) numbers can double if you cool it down additionally 5 degrees, for example.”
But optimised cooling is not just about reliability. Intel’s Turbo Boost functionality, for instance, provides greater computer performance by dynamically overclocking the processor, but the feature limited by temperature constraints.
By providing better cooling within the constraints of the application, the processor can run at higher clock speeds for longer. To address this, congatec engineered a new COM Express-compliant heat-spreader and heat-piped-based cooling solution.
The new cooler improved cooling by 14 degrees compared to previous coolers. This yielded a calculated eight-fold increase in MTBF figures, and 30 to 40 percent more performance due to the ability to run on Turbo Boost mode for a longer period of time.
According to Eder, completed modules undergo intensive testing to ensure they can stand up to the rigours of industrial applications.
“Temperature is always an issue, even for extended CPUs where our module is specified for 0 to 60 degrees Celcius,” explained Eder.
“We do the tests down to -40 degrees, and up to 90-100 degrees, just to make sure that the design is rugged, even if we do not end up exceeding chip maker specifications.”
Where the module is specified as an extended temperature version, congatec will ensure all other components also perform under the extended temperature range, and test the systems to ensure they meet specifications.
In the short term at least, according to Eder, the trend towards increasing numbers of cores will continue, as will influence from the mobile and smartphone markets.
“Scaling will happen with additional processors, and with slight improvements from single processors. We will get many more processors in parallel onto the chips,” he said.
“I am also looking forward to the next ARM developments. Right now we see the Cortex A9 coming out. With ARM targeting the server market, and I am curious to see what emerges from there.”
On AMD’s part, increased integration is the way forward, with a strong push towards Heterogeneous System Architectures, beyond its current integration of CPUs and GPUs, or SoC approaches which also integrate DSP and FPGA functionality to boost performance and system flexibility.
Currently, in order to use the GPU to process data, a program running on the CPU queues work for the GPU using system calls through a device driver stack managed by a completely separate scheduler.
According to AMD, this introduces significant dispatch latency, with overhead that makes the process worthwhile only when the application requires a very large amount of parallel computation.
The aim for Heterogeneous System Architectures is to even more closely integrate different processing elements, unify their memory space, and reduce additional data handling and transfer – in short, to make the sum greater than its parts.
Applications will be able to create data structures in a single unified address space and initiate work items on the hardware most appropriate for a given task. Data sharing between the various compute elements will be simple.
Multiple compute tasks will be able to work on the same coherent memory regions, utilising barriers and atomic memory operations as needed to maintain data synchronisation.
“Heterogeneous architectures are still in their infancy, with new solutions coming down the line that will make them easily programmable you will start to see interesting combinations of processing solutions, that people may have not considered yet,” Swen told Electronics News.
For starters, heterogeneous architectures are expected to yield drastic improvements in the size, weight and power fundamentals for industrial and embedded applications.
Standards and compatibility
But even as chip makers continue looking into the future, legacy support is still of utmost importance.
“Industrial applications are generally slow to adopt new technologies,” Swen said. “This helps to ensure that the technology, and products offering it, are mature enough to deliver reliable operation.”
“For this reason it is important that solutions offer a migration path between existing and new technology.”
Compatibility and standards are thus a focus for both AMD and congatec. In pushing to improve the adoption of heterogeneous architectures, AMD is working with partners like ARM, Samsung and Qualcomm to unify approaches to the issue.
On its part, congatec sits on and advises a number of standards committees, with the aim of promoting competition and inter-vendor compatibility, which keeps prices down and encourages technological innovation.
“We are trying to stabilise and enhance standards, to make sure customers are not tied to one or another vendor. When we talk about standards, it must be exchangeable between vendors, which is very important,” Eder said.