and Analysis of the Factors Influencing the Performance of these Chipsets
We have recently reviewed in detail the new generation of Intel chipsets, now it's high time to pay our attention to off-the-shelf mainboards based on these chipsets. Besides, taking advantage of a specific character of some of the mainboards under review, we shall correctly compare DDR2 and DDR and analyze startup failures of the integrated i915G video.
Short comparative characteristics of all mainboards under review
are provided in a summary table below:
This time we have an opportunity to estimate correctly the effect
of using different memory types. The fact is, when you try to compare
a mainboard from the X1 manufacturer based on the Y1 chipset with
the Z1 memory type and a mainboard from the X2 manufacturer on the
Y2 chipset with the Z2 memory type, it's very difficult to determine
the contribution of Factor Z to the total difference. Fortunately,
Gigabyte launched a model supporting both DDR and DDR2, plus we got
two practically identical mainboards from Foxconn based on the same
chipset, but with memory slots for different generations of RAM. Besides,
in case of Foxconn mainboards we can measure the degree of performance
slow down due to the integrated i915G video – again for different
memory types. In order not to overcrowd the diagrams, we provide only
the distinctive features of each system – we hope it won't be
difficult to correlate them with specific models.
We'll start with low level memory tests in RightMark
Memory Analyzer. We are not interested in maximal real read bandwidth,
this value is approximately the same for all contenders, as it is
limited by the CPU bus bandwidth. For read without prefetch operations
the i925X mainboard is approximately on the level with DDR mainboards,
new chipsets with DDR2 are lagging behind by 4%. The use of integrated
video proportionally decreases the results approximately by 4-5% in
comparison with the use of an external video card (mainboards with
DDR2 experience a lesser decrease, but the bottom line results are
Non-Temporal Store (direct memory access for writing past the processor
cache levels) is also absolutely predictable and differs a little
only because of the difference in FSB frequency values of the contenders.
In ordinary memory write operations DDR2 gains revenge for the reading
failure outscoring i915+DDR400 almost by 13% (however the Gigabyte
mainboard didn't manage to squeeze so much from DDR2-533). i925X confidently
Pseudo-Random Read Latency (prevents D-TLB misses) is minimal in
mainboards with DDR, only i925X remains on the same level. DDR2 impairs
the results by approximately 4%, just as the integrated video.
Random read method doubles the latency, in this case DDR is beyond comparison outscoring i925X by more than 7%, and i915 mainboards with DDR2 – by 11-12%. Integrated video usage costs much less, just a couple percents slower.
On the whole it's all clear: i925X is really faster than i915 with
DDR2, but this speed advantage is hardly enough to remain on the level
with the same i915 but with DDR. However, 10% difference in synthetic
tests does not make noticeable difference between our contenders in
The first role in archiving is quite expectedly played by memory
latency, so the conclusions in the above paragraph are true. Meanwhile,
the maximum lag from the leader of the system based on i915 with DDR2
and integrated video is less than 6%.
Practically all the tests included into our public technique
for video encoding are weakly dependent on memory performance. These
results wouldn't have deserved a separate diagram (less than 2% between
the best and the worst systems) but for the sudden lag of a Gigabyte
mainboard with DDR400, which we witnessed in half of the tests (in
the other tests the mainboard was on the level with its competitors).
Unlike rendering programs (where the results are almost completely
determined by CPU), image processing in Photoshop is more critical
to the difference in chipsets and memory types. Scatter of readings
is not large again, but Gigabyte 8GPNXP Duo with DDR400 is again lagging
behind the rest of the mainboards by 8-13%. The reason for this strange
behavior is not clear, but despite the undoubted potential of DDR
in i915 mainboards we would still recommend to orient toward the models
designed for a single memory type.
We plan to publish a separate article with a comparison of the i915G integrated video with other integrated chipsets, which are its real competitors unlike one of the top video accelerators on GPU NVIDIA. In the remaining part of our roundup we'd like to note that the actual position of affairs in games is preserved (i925X and mainboards with DDR are a tad faster than mainboards with DDR2), though the difference between the contenders is getting too measly.
If you wish, you can re-read the comments about the low level comparison of the systems under review. What concerns practice, there are only several obvious conclusions.
Considering what I wrote in the beginning of this article, if it's
vital for you to switch to Socket 775, you may choose a system to
your liking. There are no reasons to give preference to any of the
memory types, so the present or future upgrade options should be considered
individually. What concerns our recommendations on specific mainboards,
it's too early to make them. It looks as if mainboards based on the
Intel 900 series chipsets can be compared directly, without paying
serious attention to their memory type and chipset. On this note we
bid you farewell until the second part of the roundup, which will
review an unexpectedly fast model from Intel, Albatron mainboards,
and many others.
Write a comment below. No registration needed!