Intel's 925X Platform - D925XCV Desktop Board Review

[Dave Baumann]

http://www.beyond3d.com/siteimages/b3dsmall.gif (http://www.beyond3d.com/reviews/intel/925/)A few weeks back, Intel introduced their latest Desktop Chipsets and motherboards based on them. With these new chipsets Intel are introducing numerous platform changes, signalling their intent to make the PC an even more high performance, hence versatile, platform than it is already. Here were are taking a look at Intel’s D925XCV Desktop board and the technologies the 925X chipset it’s based on brings.<blockquote>"PCI Express's serial point-to-point nature requires few sideband signals, meaning the the bandwidth per pin can be very high in comparison to PCI Conventional. The upshot of this is that a x1 connector can be a fraction of the size of a PCI Conventional slot, whilst still providing twice the bandwidth on its first implementation. Bandwidth can be further scaled up by increasing the number of lanes running two and from a device and the PCI Express standard allows for x1, x4, x8 and x16 slots."</blockquote>Read the full review here (http://www.beyond3d.com/reviews/intel/925/)

[Anonymous]

Strange, no GMA900 Benchmarks. Why?

Will you do an additional Review to compare the GMA900 with the IGP9100-Pro and the 64bit and 128bit versions of the R9200 and FX5200?

If so, will you look also about the image quality of the GMA900 in 2D and 3D?

As extremetech had found out the extreme graphics 2 solution had an really bad image quality.


Manfred

[Dave Baumann]

So far we've only been supplied with the 925X board, which lacks any integrated graphics. We've asked Intel to see if they have any 915 boards available and if so we'll take a good look over the graphics and system performances.

[Rodéric]

http://www.beyond3d.com/reviews/intel/925/index.php?p=2

Supporting 3.60 MHz PIV ?

:lol: :roll: :oops:

http://www.beyond3d.com/reviews/intel/925/index.php?p=10
Ahtlon is faster yet it has -20.3% !?!

http://www.beyond3d.com/reviews/intel/925/index.php?p=12
for the HardDrive test you changed the color convention, why ?
(Athlon isn't "beige")

Lack of units in the graphics is EVIL.

[Tokelil]

Seems like it shouldn't say "ns (lower is better)" on the harddrive graph on page 12.

[Dave Baumann]

http://www.beyond3d.com/reviews/intel/925/index.php?p=10
Ahtlon is faster yet it has -20.3% !?!

No, 925X has a 20.3% performance deficit to the Athlon.

Lack of units in the graphics is EVIL.

This has always been the case with the fill-rate graphs as the sclae is fairly arbitary (number of pixes on screen). The datapoints represent each resolutions tested. If you put a scale on it, I believe Excel screws up these graphs anyway (IIRC).

Others are fixed, thanks.

[Rodéric]

Others are fixed, thanks.

You're welcome.
Interesting Article, thanks.

[mczak]

IMHO the new platform is slightly disappointing from a performance point of view. I'd bet if intel would have introduced FSB1066 along with it, the reviews would have been quite a bit more favorable - the "slow" FSB800 (6.4GB/s) is likely a major reason why dual-channel DDR2-533 (8.5GB/s) isn't faster than dual-channel DDR-400 (of course, the higher latencies are a major factor, but the higher possible transfer rate could possibly compensate that easily).

Just a small nitpick, the introduction page mentions "Intel Extreme Graphics 3", but afaik this name does not officially exist, it was canned in favor of GMA 900. And I'd be really interested in a review of it too, from what I've seen in other reviews it is VERY competitive performance-wise with both ATI's IGP 9100 and Nvidia's nforce2, beating 64bit stand-alone low-end graphic cards, slaughtering the previous Extreme Graphics 2, while providing the best feature-set of all integrated solutions (even more features than ATI's current low-end graphic cards).
Not to mention it's the most modern graphic chip(-set) with open-source linux drivers...

[Dave Baumann]

I'm not sure the FSB is the main issue at the moment, afterall, it appears that the memory bandwidth is hardly able to reach the maximum. I think the use of DDR2 is one of the main issues at the moment, but hopefully there may be some more performance to come through tuning as it is the first Intel chipset that uses it.

[mczak]

I'm not sure the FSB is the main issue at the moment, afterall, it appears that the memory bandwidth is hardly able to reach the maximum. I think the use of DDR2 is one of the main issues at the moment, but hopefully there may be some more performance to come through tuning as it is the first Intel chipset that uses it.
I believe this theory is severly flawed. I would strongly suspect that the reason why DDR 400 and DDR2 533 post almost exactly the same stream memory bandwidth is because of limits of the FSB (its theoretical limit is 6.4GB/s, its practical limit might be well below), not because the respective memory controllers can achieve the same buffered memory bandwidth. For instance, look at these numbers, http://www.anandtech.com/memory/showdoc.aspx?i=1849&amp;p=4 DDR355 with FSB1066 results in higher memory bandwidth than DDR400 with FSB800 (yes, cpu clock is different, but it shouldn't matter much for the memory benchmarks)! Thus, extrapolating from these results, it seems memory bandwidth with DDR 400 would get a boost of around 11% from switching from FSB 800 to FSB1066, but DDR2 533 _might_ gain quite a bit more (not saying it definitely will, but I think the possibility is there).
Maybe someone will do a overclocked FSB review to back it up, all it takes is some P4 unlocked engeneering sample and a board which allows for FSB overclocking of 33% despite that stupid clock-lock...

[2senile]

So far we've only been supplied with the 925X board, which lacks any integrated graphics. We've asked Intel to see if they have any 915 boards available and if so we'll take a good look over the graphics and system performances.

Hi "Wavey", seems the fault on the 915 ICH6 has been corrected &amp; Intel are shipping them again.

In regard to the fault/recall, it seems to have affected orders for DDR2.

http://www.digitimes.com/news/a20040708A1001.html

[mczak]

Maybe someone will do a overclocked FSB review to back it up, all it takes is some P4 unlocked engeneering sample and a board which allows for FSB overclocking of 33% despite that stupid clock-lock...
Tom's Hardware has just done such a review, and you can clearly see that my theory is indeed true - the "slow" FSB indeed is the limiting factor for at least streaming read bandwidth, not DDR2-533. Some of the other benchmarks also suggest that FSB800 is indeed quite a limiting factor. http://www.tomshardware.com/motherboard/20040719/intel_925xe-17.html