Intel is once again adding a new computing form factor to the mix. At CES Intel announced its new Intel Compute Card. It combines CPU, GPU, DRAM, storage, WiFi, and communications inside a small modular housing slightly larger than a credit card and about 5mm thick. Intel already offers its Compute Stick, but it is limited in its interface options. The compute Stick only supports HDMI along with USB and WiFi, making it a bit limited. Unlike the Compute Stick which seemed to be promoted as a highly portable computer that can turn and HDMI monitor into a useful PC, the Compute Card is intended to provide the brains for a number of applications, such as smart TV’s, appliances, IoT devices, etc.
The question is, what does it offer in these applications that ‘hardwired’ processing does not allow?
The Compute Card has a proprietary connector set on its end that allows it to plug into its host. Intel describes the interface as a modified USB-C. This enables it to connect to a wide range of devices, such as hdmi, storage, PCIe, potential future interfaces, etc. Because standard interfaces are not brought out to connectors, it will not operate as a standalone device.
The Compute Stick was said to have low performance, although it was improved in the second generation. The Compute Card is planned to offer a wide range of CPU’s not just the low-end Atom cores. The upper end of the power dissipation is said to be around 6W. One nice difference from its predecessor is that the Compute Card has no cooling fan, which could have been a potential reliability issue. Apparently the dock (or socket) provides some cooling in addition to a ‘locking mechanism’ to prevent removal where security is a factor.
So, is the Compute Card an embedded processor or is it a portable compute device? In the embedded processor space there is a wide range of options, from both Intel and through ARM based processor providers. It seems that major appliances that need would be built around a specific processor and chipset. While the Compute Card touts upgradability and future proofing for its hosts, it’s not clear that upgrading the processor, if it proves practical, will extend the life of appliances. Indeed, smart appliances may not actually outlive their processing units.
Nevertheless, it is conceivable that repairs could be made easier by a plug in compute unit. But this could be offset by connector or thermal issues with the Compute Card packaging/dock.
To be fair, I was skeptical of virtual machines, but as technology improved they became practical. This took a few decades in actuality. And, they never be as efficient as bare metal, rather they offer flexibility and convenience without a severe penalty. In fact I know of several websites that solved their server bandwidth issues by reverting to bare metal. But that is another story.
Intel has signed up partners to help develop Compute Card enabled products. These partners include Dell, Lenovo, HP, Sharp, as well as InFocus, Seneca and others. The rest of us will need to wait until June of 2017 to get pricing and detailed specifications. Likewise, they will be available for purchase in the middle of the year.
A lot of the utility depends on the actual specifications, price/performance ratio, and the details of the necessary ancillary hardware, such as the dock. Presumably the Compute Card will run Windows and likely Linux too. The OS will also play heavily into potential applications and market acceptance.
It is too early to tell if this will be part of a significant shift in the development of smart products or for the Internet of Things. I plan on watching it with guarded expectations. Much of what we have seen recently points to the extremely high utility of products based on custom SOC’s and advanced packaging such as TSMC’s CoWoS and InFO technology. For instance, ARM cores are available for silicon integration through a large number of SOC and Virtual ASIC vendors. However, just as with virtual machines, only time will tell if there will be significant market acceptance for the Intel Compute Card.