Preview: Intel X38 Express - BeHardware
>> Motherboards

Written by Marc Prieur

Published on September 24, 2007

URL: http://www.behardware.com/art/lire/684/


Page 1

Introduction, the card



With the high end X38, is Intel offering a new chipset that is supposed to replace the i975X? Whatís new compared to the classic P35 Express?

The differences between the P35 Express and X38 are confined to the MCH (Northbridge). In terms of the ICH, we in fact find the version 9 which is connected to the X38 via a DMI bus at 2 GB /s (1 GB /s in each direction). In its basic version, it manages four PCI Express ports, has an integrated Ethernet Gigabit controller and controls four AHCI 3 Gbits /s type SATAs. In its DH version, it manages a total of six SATAs, while the ę R Ľ allows managing Matrix Storage (RAID 0/1/5/10), as well as two innovations, Rapid Recovery and Turbo Memory on its six SATA ports. Also, note that the SATA ports of the ICH9-R are Port Multiplier compatible and can therefore be doubled via an additional chip in order to support several peripherals.


However, letís take a look at the real differences between the X38 and P35 Express. For the latter, PCI Express management is limited to a single PCI-Express x16 port. For this reason, those cards equipped with two PCI-Express x16 ports actually have one port connected to the MCH and which is wired in x16, and a second relayed to the ICH which is wired in x4. The performances of this second port are thus largely inferior and to install a good graphic card in a multi-screen or CrossFire (SLI canít be activated on Intel chipsets) configuration doesnít give the best results.

With the X38, Intel decided to shift into high gear. First of all there is management of a total of 32 PCI Express lanes, or two 16 lane ports, and also its lanes are within the 2.0 norm. In its version 2.0, the PCI-Express speed is doubled to attain 500 MB /s of data speed per lane in each direction, or 8 GB /s on a single x16 port. Note that while they should soon arrive, PCI-E 2.0 graphic cards arenít yet available. Either way, in all situations and in the beginning the performance gain will be inexistent for classic use.

The other differences between the P35 and X38 are more subtle and less documented. Intel mentions an improved memory controller and says that they have pulled the ę overspeed protection Ľ that was present on the P35. Letís get the more practical side of things and see how this chipset does in tests.
The motherboard

For this test, ASUSTeK provided us with a P53E Deluxe in its version 1.03G. The motherboard takes on the X38ís specifications and ASUSTeK as usual has added some chips: a FireWire controller provided by Agere, a Realtek PCI and Marvell PCI-E Ethernet Gigabit controllers (ASUS having decided not to use the one integrated to the chipset) and a JMicron chip devoted to management of a PATA and two eSATA ports. Itís now the trend to have passive cooling provided by a system of copper heat pipes and fins.


Taking apart the system gives us a glimpse of an innovation on the chipset level in the form of an IHS as up until now the die was ęnakedĽ. The IHS has two purposes, on the one hand to redistribute the heat given off by the chipset onto a larger surface, and secondly, to protect it from a clumsy dismantling of, for example, a waterblock. Once the motherboard is installed and started, we were impressed by the number of adjustments offered by the bios as well as their finesse. In addition to a large number of memory settings Ė maybe even too many Ė ASUS offers direct access to the FSB Strap as well as to various precise voltages. For example, for the northbridge this can be adjusted between 1.25 and 1.91v by increments of 0.02v.

The bios is compatible with Intel Extreme Memory which is an extension of SPD of DDR3 memories destined to function with higher settings. Memory manufacturers can also offer standard SPD settings for their modules and a second more aggressive set for those who like to fiddle with adjustments. However, note that this group usually does this on the memory level and the interest of such a function is rather limited in our opinion.


Page 2
In practice, Conclusion

In practice
We first looked at the performance of the memory controller tested with the help of ScienceMark 2 for latency speeds, as well as the framerate in Far Cry, and compression of files obtained with WinRAR 3.7. These two ę practical Ľ applications were chosen because they can be noticeably affected by the speed of the memory sub-system, however, this isnít always the case.


Whether itís with an FSB set at 333 MHz or 444 MHz and with processor frequency of 2666 MHz, the X38 has trouble setting itself apart from the P35. We gain 1 to 3 cycles in memory access and indeed up to 368 MB /s in bandwidth. However, in practice the gains are very limited at 0.4 to 0.9% in WinRAR and at best 1% in Far Cry.

We now move on to PCI-Express speed measured with the help of a GeForce 8800 GTX on each of the cardís two PCI-Express x16 ports in three distinct tests:

- Serious Magic, a test which carries out a download from the graphic card to the system
- SPECViewperf9 UGS Teamcenter, an OpenGL graphic load simulation and very heavy professional use
- 3DMark2006, here we took the score for the two scenes of HDR/SM3.0 in 1680x1050 AA4x


Contrary to what occurs on the P35, the performances offered by these two slots are now comparable. In addition, and despite the fact that the GeForce 8800 isnít PCI-E 2.0, we saw a notable increase in the speed measured in Serious Magic, even if this does not have an evident impact on practical performances.


Power consumption of the P53E motherboard compared to a P5K3 Deluxe in the same conditions is very comparable.

Finally, what about overclocking ? To take a look at the behavior of the chipset in this area, we used a QX6850 and tried to obtain the highest possible FSB. We remained in Strap 333, because 400 would not have given us better results. By staying with a value of 1.25V for the chipset, we were able to obtain 480 MHz of FSB, then 490 MHz by increasing voltage to 1.45V. The CPU PLL also went from 1.5 to 1.7V. This was the maximum stability we were able to reach and validated in two 30 minute sessions of Prime95. For the P35 on the P5K3 in the same conditions, we obtained 470 MHz without increasing voltage and 480 MHz by increasing the chipset power to 1.55V.
Conclusion
This first look at the X38 chipset isnít the most conclusive. We know that the P35 is lacking in terms of PCI Express management, something of greater importance for those that want to use multiple graphic cards, however, the limitation to CrossFire and absence of SLI compatibility is still problematic. Also, in terms of performances and overclocking, while there was a slight improvement compared to the P35, itís more symbolic than anything.

In short, given the current situation, the X38 is a bit disappointing. The P35 Express was already very good and itís true that it would have been difficult to do significantly better. However, given the absence of major improvements between the two cards, for example, in the bios, itís going to be difficult justify a large difference in price.


Copyright © 1997-2014 BeHardware. All rights reserved.