A recently announced new Intel platform is surely revolutional by many parameters. To upgrade to it now means to replace almost all PC innards. But the tests did not show revolutional speed rates and aroused a lot of disputes (basically on near scientific issues). Alas, people are always unconsciously expecting miracles from anything new and when most reviews rate it as «could be worse», this new thing gets rejected. This often happened before. Do you remember, for example, the release of Pentium 4 or the first AMD64 processors? These products managed to take the worthy place on the shelves only after considerable time, initially looking very strange.
As the SPEC CPU2000 hardware tests depend only on CPU, chipset, and memory, this article will mostly concern the present situation with the performance of new Intel chipsets and DDR2 memory in this test.
Note that the memory operated in dual channel mode in all the tests.
Out of the variety of combinations we shall choose and test the following pairs:
Unfortunately, results of this review cannot be considered absolute. The fact is, unlike the "old" i875, new mainboards on Intel Express may have buggy BIOS versions. Together with the lack of utilities for new chipsets to view the current memory configuration, it may lead to confusion. So in a couple of months we'll probably have to get back to testing this platform and new mainboards.
You should take into account that we didn't use the fastest (meaning - expensive :)) Pentium 4 Extreme Edition processors, but common Prescott CPUs instead. Of course, P4XE usage provides higher performance, but usually people cannot afford this processor. At the first opportunity we'll update this material with the P4XE data or will write a new review.
To check the «farther-better» hypothesis, one of the testbed configurations included CPUs operating at 2.8GHz.
Note the cited Intel data that in SPEC CPU2000 the DDR2 memory is faster than the predecessor by 1% - 3% in SPECint_base2000/SPECfp_base2000 with the CPU clock at 3GHz. But they used for tests a single i915G mainboard and the DDR memory timings were 3-3-3, so, taking into account that our DDR data has been obtained from the fastest mainboard with the i875 chipset, we can expect slightly different figures.
We used Intel compilers v8.0 (C/C++ 044, Fortran 040) optimized for the Prescott core (refer to this article for details about the compiler and its optimization switches).
So, let's start with comparing top (at least by their positioning :)) platforms.
... Apparently Intel used a different measuring approach :(. The results above show that today's new platform loses to the old one in math solving. With the CPU clock rising, performance difference noticeably decreases, and probably with the CPU clocks at 4 GHz and higher the new platform will become faster. However (judging from Intel plans) this may happen a little earlier - with the arrival of 1067 MHz bus CPUs. They will most likely not be able to work with the i875 chipset (if the mainboard makers don't overclock it), which will make Intel Express chipsets the fastest series together with these CPUs due to processors (not the chipset and/or memory) and, as always, "best of the best" lovers will have to upgrade their platforms to i925XE.
Note that this already happened before - i850(E) chipset together with RDRAM had been a long-time leader, and only dual-channel chipsets i875/i865 managed to take its place.
Conclusion - it is not worth while upgrading from i875 to i925X yet, but when making long-term hardware purchase plans, pay attention to the new chipsets (certainly with the fastest CPUs).
Let's mention once again that we used mainboards from various manufacturers, so the difference in results can also be due to BIOS differences.
Thus, in this diagram we can see that the i925X chipset is a little faster than its younger brother. Lagging in several tests may be explained by the lower i925X clock frequency (252.eon results also speak for this fact, this test always depending on CPU only). As a result, it outscores in overall characteristics by 0.5% and by 1.5% in SPECint/fp_base2000, in certain tests - up to 5.4%. Such an increase in 189.lucas is strange, because in CFP2000 it's usually 171.swim and 179.art that react to the memory speed. And such a behavior from 181.mcf was expectable.
On the whole we can say that costly products with the i925X chipset are not justified (so far?).
You may consider the following alternative of a gradual upgrade to the new chipset: you first upgrade the mainboard and CPU only (mainboards with i915 and Socket 478 are theoretically possible, so you have a chance of keeping your CPU) but you keep the old memory (we do not consider a video card upgrade due to the nature of this test :)). But a user will not gain much from this upgrade - opportunity to upgrade the old memory to DDR2 and PCI Express bus support (well, and support for faster CPUs). Let's see how well the i915 chipset performs with DDR.
Alas, it didn't work magic either. There is a little lagging of the new platform.
We surely can write off a part of the lag to the novelty of the mainboard and BIOS, but -10% in 181.mcf and almost -7% in 179.art demonstrate that this chipset performs apparently worse with DDR than the old one. (Maybe we just didn't see a mainboard from ASUS with the Intel Express chipset :)).
And the last test, carried out on the Gigabyte mainboard with the dual memory types support, will show if there is a reason in gradual upgrade of the system and in memory-universal mainboards.
The third miracle also escaped us :(. It turns out that even if you buy a universal mainboard, there still will be no sense in upgrading the memory to DDR2 (taking into account that a new CPU bus for the i915 chipset is not planned at all).
Summing up the test results:
The bottom line - upgrading to the new platform, which will be used for calculation tasks, is presently not reasonable. Concerning the future, the platform possesses some positive stuff. But at present this stuff is of theoretical nature and can become useful only in new versions of chipsets, CPUs, and memory with higher (relative to the present ones) clock frequencies. Actually, the direct relation of speed and clock frequency is obvious even without testing :).
Kirill Kochetkov firstname.lastname@example.org,
09 July, 2004
Write a comment below. No registration needed!