This article completes the series of reviews of spring-born Intel's chipsets and mainboards based on i875P and i865 line from leading manufacturers. Several i865x based boards were tested a bit earlier, and today we have gathered eleven products from famous companies plus one more model based on the top i875P chipset. Since the architecture and features of the i865x family (Springdale) were already closely examined, the theoretical section will be omitted except one moment concerning activation of the Intel PAT (Performance Acceleration Technology) in the 865 series. The situation is now much clearer though I can't say it's successfully solved.
If the cell with the board saying "Buffalo" contains an elephant...
...believe your eyes! This is exactly a buffalo, it's just very shy and hides inside the elephant :). We wrote about such a mysterious symbiosis of these animals, but the situation changed since that time. So, the function named "PAT" is really integrated into the i865x chipsets and can be activated, but it's hidden too deeply :). Today we have enough models the developers of which took the trouble of going into the heart of the chipset to get the PAT. Now we can witness a bunch of new technologies from board makers. However, in spite of variety of names the effect is very similar: all of them actually enable the PAT with following delving into the chipset (this fact makes the i865x based boards superior to the i875P).
But the i865x chipsets can actually be called "a failed version of the i875P" because Intel, as a manufacturer, doesn't test the PAT in the chips marked as a north bridge in the i865 series, or they simply fail in the tests (in certain conditions). That is why the board makers are pretty modest in evaluation of their "proprietary" technologies: "it can work, but if you have any problems, it is quite possible that they are caused exactly by its incorrect functioning in this case". But Intel has already started or about to start delivering a new revision of the i865 chipsets where the PAT will be locked to withdraw this problem. (But let's wait for the response from the mobo makers.)
Meanwhile, all the boards which support this technology have successfully passed all the tests with the PAT enabled. Does it mean that problems won't happen? It's hard to answer this question. Most probably, functioning of such a non-standard PAT can be very dependent on quality of memory modules used. We have tested two variants: our standard modules (TwinMOS 256 MB PC3200 DDR SDRAM DIMM CL2) and TwinMOS 512 MB PC3200 DDR SDRAM DIMM CL2 modules received recently. We were mostly interested in more capacious chips as they are less reliable in case of the same quality and manufacturer. To the credit of chipmaker Intel, respective board makers and TwinMOS the systems worked flawlessly even with the timings of 2-2-2-5.
OS and drivers:
The following table contains brief characteristics of all the boards tested:
All the diagrams demonstrate the results with the PAT enabled (except one diagram marked respectively), as well as with other speed-enhancing options which do not bring to failure. The effect from the PAT will be discussed later on the diagram mentioned above. It should be noted that the ASUS P4P8X has the PAT activated whatever option (Enable or Auto) is selected, in the current firmware version. By the way, the scores of this i865P based ASUS board show that all the Springdale line maintains reliable operation of the "overclocked" chipset, and the new-fledged boards based on even the lower-level chipset do not yield to the i865PE.
We consider the PAT activation on the i865x to be overclocking though a harmless
one which doesn't trigger grave consequences and, therefore, can be allowed (like
an attempt to start up the memory with the lower timings than specified in its
SPD). In contrast, the dynamic overclocking in MSI's models is harmful since it
makes an impression of easy and safe overclocking, and it doesn't satisfy experienced
overclockers anyway. The scores of the MSI 865PE Neo2-FIS2R are given in both
modes but we will estimate its performance in the normal one.
The MPEG4 stream encoding algorithm is more sensible to the bus throughput
and CPU clock, that is why the MSI 865PE Neo2-FIS2R with its dynamic overclocking
takes the lead. The difference between the other boards is not significant: the
outsider (Fujitsu-Siemens D1627-A) falls behind the leader (MSI
without o/c) only by 4.5%. The boards with the PAT enabled are mostly ahead; among
them are both ASUS and ABIT IS7-G, as well as ECS Photon PF1
(PAT off). The AOpen AX4SPE Max unexpectedly falls into the last position
(even with the PAT on) and 8PENXP from Gigabyte which usually makes
pretty efficient boards (the last ones are Shuttle's models and Canyon
9I6GM-L, which doesn't look surprising). The budget i875P is somewhere in
the middle but the scores of the participants are too close to draw certain conclusion.
The 3ds max rendering illustrates effectiveness of the Dynamic Overclocking
technology: the CPU clock on the MSI board grows up by 7% under the load, and
the same advantage is achieved here. The scores of the other participants are
almost the same.
Archiving with a large dictionary is a good test for PAT because it mostly
depends on the memory latency. Above are all models with the PAT enabled, below
are all the rest (with the Canyon 9I7PA-L (i875P) being a watershed). The
difference between the slowest board among the first and the fastest PATless models
makes an impressive mark of 5%. The DOT hardly helps the MSI - it can't get to
the first place. So, the best are ASUS P4P800 Deluxe and P4P8X,
ABIT IS7-G, AOpen AX4SPE Max and Albatron PX865PE Pro; and
the worst runners are Fujitsu-Siemens (very feeble), Shuttle and
Canyon on the i865G, and Gigabyte. You can calculate the difference
of one board over another yourselves, but the overall gap of 34%(!) between
the boards riding the same chipset looks really impressive.
We didn't expect anything new from games. Usually, the scores are easily predictable
from the previous diagrams. They just help to estimate whether the advantage some
boards may have is important; certainly, this effect should be estimated in the
most popular class of applications which does require top processors, chipsets,
video cards and other speedy components. The leaders are the same: ASUS,
ABIT, MSI, Albatron; Fujitsu-Siemens and Gigabyte
alternating with Shuttle AB60/RS and Canyon 9I6GM-L drag behind.
The ECS keeps in the middle close to the Canyon (i875P) proving
that it makes not only low-end boards. AOpen's model works too slowly,
and even the PAT is of no help.
As to the effect from the PAT, it's clear that this is the only life-boat for the AOpen AX4SPE Max - without this technology its scores are comparable only to the Fujitsu-Siemens. Another important fact is that the PAT enters the EPoX 4PDA2+ into the group of leaders, though the old firmware version didn't shine at all. Finally, ASUS and MSI gain least of all from the PAT, though in general they remain on good places.
The descriptions of the models given above will help you make the right choice if you are mostly interested in functionality, accessories and overclocking (you can briefly estimate them in the table). They are really important factors because this is exactly an integrated support for various standards of peripherals connection which distinguishes the i865PE based boards and other chipsets of the Springdale family from their predecessors. If the speed factor is the most vital for you look carefully through the test scores because the performance gap can be very great in spite of the same chipset.
The scores are so different that a situation when a board on the new i865x
loses to a i845xE based board certified
for the 800MHz FSB is quite possible. Moreover, even owners of a board riding
the top i875P can't be sure of their supremacy - for example, the Canyon board
has pretty modest achievements. That is why it's not worth looking for the inexpensive
i875P if performance is more important for you than the ECC support.
Vladimir Rybnikov (firstname.lastname@example.org)
Serguei Pikalov (email@example.com)
Dmitry Mayorov (firstname.lastname@example.org)
Write a comment below. No registration needed!