VIA C7 Microprocessor
The C7 in the latest EPIAs is an evolution of the old C3 Nehemiah core that showed up in the EPIA-based Tranquil PC we reviewed a couple of Christmas's ago, and its followup not long after. The C7, codenamed Esther, doubles the L2 cache size compared to Nehemiah, to 128KiB (exclusive, 16-way associative), and you get the Pentium-M's quad-pumped bus interface for connection to the outside world.
Built on IBM's 90nm SOI (silicon-on-insulator) process, Esther scales to around 2GHz and while not advancing things in terms of the execution core, does add features over Nehemiah, mostly in terms of security.
VIA recently released their Cryptographic Service Provider (CSP) tools for Windows XP and its APIs, as well as code to use the hardware security features under Linux. You see, C7 has hardware-level encryption features. They include hardware SHA-1 and SHA-256 hashing with a 20Gb/sec throughput (at 2GHz, scaling back for slower Esthers), hardware AES and RSA (again with a 20Gb/sec throughput) encryption (the big deal, RSA being new in Esther) and 2 enhanced (over Nehemiah) hardware random number generators for feeding the SHA hardware.
None of these features are available (that we know of anyway) on any other consumer x86 or x86-64 processor available on the market today, providing an inroad for VIA and the C7 for application designers looking to provide hardware-accelerated secure programs (provided they map to Esther's abilities).
Esther's main pipeline is 16-stage and optimised at the branch stage over Nehemiah, giving further speedups and it seems that the hardware can switch between full- and low-power states in a single cycle of execution. SSE3 is supported, too, bringing VIA in line with AMD when providing support for Intel's latest SIMD instruction set revision. We've no idea how many SSE ops it can push at peak, though.
The CPU itself is teeny (although we don't show it to scale), VIA leaving the die bare this time compared to C3 which only ever came in a complete package replete with heatspreader. Clock-for-clock performance should be up on the C3, then, and it scales to much higher clocks too. Let's take a look at some synthetic performance tests before some real-world stuff.
