Hardware
Twenty-First Century Bottomfeeders
It should become immediately clear to you that this material is rather old. The number of visits this area gets is astronomical.
VIA C7 and VIA Eden
The originals have upgraded their systems with the well-known,
hardware cryptographic and random number generation hardware and the
Mini-ITX and Nano-ITX platforms. I run a Via Esther processor, also
called VIA C7, for my home network DNS and DHCP services. I tried
mightily to get the OpenSSH and OpenSSL libraries to use the built-in
hardware encryption for my SSH gateway but it just doesn't seem to
"stick." I even tried to short-circuit the OpenSSL library to always
use it but my remote desktop session performance shows it just isn't
using the built-in hardware accelerator of the VIA. I blame both
OpenSSL for its immaturity (in spite of reaching 1.0) as well as the
VIA team for their unclear ABI and API documentation for the ASE
engine. I also notice that most of Wikipedia's articles do not
mention this engine anymore.
Here are the OpenSSL benchmarks without hardware acceleration for
Blowfish, an excellent all-around cipher.
The 'numbers' are in 1000s of bytes per second processed.
type 16 bytes 64 bytes 256 bytes 1024 bytes 8192 bytes
blowfish cbc 28418.58k 30366.93k 30874.79k 31028.22k 31072.26k
While the hardware accelerator is impressive, I can't get it to work
all the time, so I don't see the point in posting those numbers yet.
Since blowfish isn't in the hardware anyway (just AES) these numbers
offer a valid apples-to-apples comparison on a cipher that most
cryptographers would consider an efficient, all-around cipher. There
are now 2.0 GHz VIA C7 systems which I am considering but without
forcing the hardware acceleration I have been discouraged from using
the VIA platform for my encryption needs, which is disappointing since
I really like VIA and the architecture acquired with the purchase of
Centaur.
Intel Atom
Intel Israel took the Pentium-M platform and wrung the most they could
out of it, increasing the pipelines and cache, and doing lots of neat
tweaks to this aging platform giving it an entirely new life as a
low-power netbook and embedded product. While doing this they
de-emphasized the power consumption factor (watts) and instead
emphasize the "thermal design power," or "TDP" since they could not
compete with VIA on the power efficiency. This means the processor
produces less heat, but doesn't say much about its power consumption
which was the whole point of VIA's C3, C5, and C7 designs. This is
like the old RMS battles in the audio realm, where product designers
game the system for marketing reasons, trhing prove efficiency instead
of raw power consumption. It's a shame they did this, since VIA can't
compete on thermal dissipation, but they win hugely on power
consumption. Still, nobody knows this due to Intel's marketing
gimmick and people care about power dissipation when they should be
caring about power consumption. Furthermore, the chipsets that are
forced to be paired with the Atom are still power-hungry desktop
chipsets.
In spite of all this I give you OpenSSL benchmarks for Atom:
The 'numbers' are in 1000s of bytes per second processed.
type 16 bytes 64 bytes 256 bytes 1024 bytes 8192 bytes
blowfish cbc 47397.38k 51959.10k 53160.45k 53516.29k 53595.83k
Impressive, but not surprising, since Intel Atom inherits the
legendary math performance of the Pentium III.
AMD Neo
I don't have an AMD Neo but it is based on Athlon XP and Sempron which
always post impressive blowfish compute numbers. I have
decommissioned my Sempron 3000+ server but its blowfish numbers were
in the 80000.00k realm, beating out both Atom and VIA C7, but not in
power consumption.
|
|