|
Exploring Hyper-Threading Performance - PCMark, Cinebench, TMPGEnc When
I was sitting around deciding what Windows benchmarks I wanted to
run for Part 1 of this article, I decided that I wanted another synthetic
benchmark that simulated several different situations. There are several
popular suites available, and I ended up going with PCMark 2004. Reason
being it's relatively well-known and easy to acquire. It
incorporates things like file compression and encryption and even
divx encoding performance. It will even take advantage of multiple
processors. Woah,
big graph. Hyperthreading gives us an ~12.4% performance boost here.
We'll take it. We've
always been rather enamored with Cinebench here at 2CPU.com,
so its inclusion here is certainly not surprising. Obviously when
hyper-threading was disabled on my P4 test system, I was unable to
run the Multiple CPU portion of Cinebench's rendering benchmark. Xeons
have always ruled the roost in Cinebench and that doesn't look like
it's going to change anytime soon. Look at how they scale here from
1-2 physical processors when hyperthreading is disabled. That appears
to be a gain of 86%. Enabling HT on the Xeons gets us a whopping
110% improvement in performance as we scale from a single processor to multiple processors. The
uniprocessor P4 box doesn't see such a dramatic improvement, but it
does pocket a 18% increase in performance. Hooz
always loves putting on his magical video encoding cap and crawling
into his secret laboratory to brew up another benchmark for us. Since
DivX encoding seems so last year, we've replaced it
with a fancy DVD encode. We'll
be taking an episode of "Sex and the City" and using TMPGEnc
to encode the interlaced MPEG2 video file to a DVD compliant video
stream. (2-pass VBR with a minimum bitrate of 0, maximum of 8000 and
average of 4000). We'll frameserve it from AVISynth with no filtering,
cropping or resizing. This
really shows us what hyperthreading lets the P4C accomplish. Our Xeon
test system certainly won the race but hyper-threading didn't have
as much to do with it as we'd like. It accounted for a ~11% decrease
in overall encode time. The P4C test system shows a remarkable 30%
decrease in overall encode time. If HT will knock 17 minutes off a
relatively short video encode, what is it going to give you if you're
dealing with projects so big that they take days to encode? If you're
looking to upgrade your renderfarm, maybe several P4Cs housed in Shuttle
XPCs should be on your wishlist. Let's
move onto my conclusion. After spending so much time with these HT
graphs, I feel relatively qualified to provide some analysis on our
results and draw some conclusions.
|