| |
|
|
 |
|
Review of our dual Xeon PC
(and
a delve into the technology behind it)
Written originally for the Poweroid
9220, some parts also apply to other dual
Xeon PCs we sell.
First Looks:
Looking at the specs of the Poweroid 9220, the very first thing that jumps
out at you is the dual Xeon "Nocona" CPUs (Central Processing Units,
or "processors"). These are Intel's latest version of the Intel
Pentium Xeon Dual Processor professional CPUs. Intel has done a lot of
tweaking to these chips, moving from the previous Gallatin core. What
kind of tweaking? For starters, the FSB (Front Side Bus) has been boosted
all the way from 533 MHz to 800 MHz. The FSB matters
because, when you have a dual processor setup, and each one of those has a virtual "twin"
thanks to HyperThreading (HT), together they create an awful lot of traffic congestion
between the CPUs and the rest of the PC.
Implications
of the FSB boost & EM64T
Going from 533MHz to 800MHz on the FSB allows for
more headroom, which makes HT even more effective. The other
gem in the Nocona revision, is EMT64 support. For now, this feature
doesn't perform any active function. But, once Microsoft gets around to bringing us a 64
bit enabled version of Windows XP, the ability to make use of more than 4GB of
memory without resorting to a kludge solution will make a sizeable difference to tasks such as
video rendering.
Other
new features: SSE2/SIMD/Hyperthreading
The Xeon processors also continue with their other standard
features such as SSE2 support, and the ability to slow themselves down should
they reach an unacceptable internal temperature (for example, should one of the
fans on the heatsinks die). Both of these are fairly important with the
recent boost in clock speed. SIMD (Single Instruction Multiple Data)
instructions found in the form of SSE/2/3 benefits directly from clock speed.
The more, the better. And the new Xeons have that in spades.
Unfortunately, that also means added heat. So the precautions of having
each chip proactively monitor it's own temperature and respond to excess heat will ensure you don't
have a very expensive piece of charcoal instead of functioning silicon.
A bit more information on both SSE/2/3 and HyperThreading is
relevant here.
The main software the Poweroid is going to be used with is likely video editing
software like Adobe's products (Premier, After Effects
etc). Software like this is aware
of the new enhancements, and can take advantage of them. Let's look
at the duality issue first, adding HT makes your task manager in Windows show
four CPUs (2 physical CPUs and two logical ones). Your
video editing application and Windows XP will treat your
machine as a quad processor machine and you will see
appropriately increased performance.
While two of those
CPUs may not be
"real" they do allow the two actual processors to be more efficient with
carrying out the tasks they are assigned by the O/S. Programs like the
Adobe family also take advantage of SIMD support. Many operations in video
editing involve doing the same calculation over and over again, just on
different pixels. SIMD allows for such repetitive work to be done in
parallel on larger data sets. Again, this is an optimisation that you'll
not notice yourself, except in the time taken to complete your workload.
The
new, improved flavour motherboard chipset
Connecting these two (or four) processors
together is a motherboard with Intel's new "Tumwater" chipset. The Tumwater
development brings all the
latest technologies - which have just been brought to desktop machines - to the
Xeon workstation platform. They include PCIe, DDR2, SATA RAID
and, of course, that massive amount of memory support. Each of these
additions on its own wouldn't be much to get excited about, but all of them together in one chipset is quite impressive. It's quite a bit of future-proofing all
bundled together.
PCI Express
PCIe, of course, is
the new standard for internal connectivity replacing the aged and slow PCI bus
as well as the erstwhile AGP slot. As far as graphics are concerned, at
the moment, PCIe doesn't provide any increase in performance over AGP. Even
today's high performance AGP graphics cards aren't fully utilising the bandwidth available to them
under AGP
8X. From now on though, most of the new introductions will be in the PCIe
format. This makes it easier for you to add
graphics performance via future upgrades and/or to
modify the PC for use as a professional 3D graphic
workstation.
DDR2
DDR2 is another of the "it doesn't really help now, but it sure will later"
improvements. Apart from it's lower power requirements over DDR it has
much, much more headroom for clock speed increases. Another big gain is in how DDR2 is made. With it's FBGA (fine ball grid array) packaging, and
on-die termination, there is less chance for errors to occur (as the signalling
is much better). This results in the ECC (error correcting code) part of the
memory having less work to do, allowing for somewhat reduced latency. Standard,
older DDR modules, with their TSOP (thin small outline
package) memory have "legs" hanging out which pick up Electro Magnetic
Inteference (EMI), which can cause bit
errors. By moving to FBGA, all of those communication lines have been
taken internally and underneath, like on the CPU itself. As in many things,
shorter route for data travel is invariably better.
Other features of the
motherboard
SuperMicro, the maker of this particular motherboard,
didn't leave all the connectivity to Intel. In
addition to the already mentioned PCIe and PCI slots, there are also two 100MHz
PCI-X slots, and a 133MHz PCI-X. These 64 bit slots are typical of servers and
workstations and not to be confused with PCIe (lower
case "e). What are these here for?
Well, currently there are hardly any cards for PCIe. And PCI is really
too slow to handle much more than the bandwidth required by products like the Matrox RT.X100
card.
So should you need more connectivity - beyond the onboard SCSI - you'd want to use one of those higher bandwidth slots. Further,
a lot of server grade products come designed for use on
the 64 bit PCI-X slots and don't fit into standard PCI
slots.
Speaking of SCSI, why on earth would you need that? The first and foremost
reason, is speed. Occupying that SCSI connector in the Poweroid 9220 is a 15k
rpm, 36GB drive with 16MB of cache. By setting up your Windows
partition on this drive disk performance is about as good as it can possibly get.
And, since 36GB really isn't going to cut it for any kind of digital video
editing, two massive 300GB IDE drives each carrying 16MB of cache as well are
hooked up to the board. When it comes to GB/$, SCSI isn't really
competitive with PATA or SATA. Matrox also has very specific
recommendations for how the disks should be set up to allow for the real time
effects the card is capable of. The "system" disk shouldn't be on the same
channel as the "A/V drive", and neither should the "export" disk. With how
Poweroid sets up the 9220, this isn't an issue. The Windows system disk is
all by itself on the SCSI channel, one 300GB disk has a SATA channel all to
itself, and so too does the other 300GB disk. Disk performance won't be an
issue for this box.
Last,
but not least
Speaking of the Matrox RT.X100 card,
this is the real star of the system.
It's intelligent, with the card not only taking advantage of
hardware-accelerated effects, but also harnessing the power of the CPU to assist
in creation of lots of real-time filters and effects. Because of that
combination of both hardware and software playing together, it's important for
everything else in the system to be up to snuff and capable of catering to all
the RT.X100 demands. Considering the dual Xeons and other carefully chosen
components you can
have multiple layers of video playing back in real time, with hardware effects
on top of that. The RT.X100 is able to scale with whatever CPU it is paired
with; the more processing power you hook up to it, the more work you can do
in real time. And since the Poweroid 9220 has about as much horsepower as you can shove into an ATX case,
this is an
outstanding machine.
|
|
Article
on dual core
|
