Socket 940 Roundup: Motherboards for the Athlon64 FX
by Wesley Fink on December 18, 2003 1:39 PM EST- Posted in
- Motherboards
High End Workstation Performance
The VIA K8T800, in general, and the Asus SK8V, in particular, dominate the Workstation Performance benchmarks. While the biggest improvement in nF3-150 PRO performance with the 11/03 platform drivers is in Workstation Performance benchmarks, the nF3 still lags behind the K8T800 on these high-end boards. The SK8V once again dominates, outperforming the other three Socket 940 boards in every benchmark.
We first ran the SPECviewperf benchmarks several weeks ago. At that time, the SK8V performed very well, but it did not dominate the Workstation Performance tests as we now are seeing. Asus is making some significant improvements in performance with recent SK8V BIOS updates. The enhanced performance has not come by sacrificing stbility, however, as the SK8V has remained stable in all benchmarks.
To understand why the VIA chipset boards were performing better on the A64 FX, we ran several benchmarks at overclocked speeds on both the nF3-150 PRO and the K8T800 boards. The difference that we are seeing at stock speed gets wider at overclocked speeds. This suggests that the increased bandwidth of the A64FX and the demands of overclocking take its toll on the capabilities of the 600HT and 8-bit uplink used in the nForce3-150 PRO. If this is true, it might also explain why we find no difference at all in the performance of the lower-bandwidth mainstream Athlon64 when run on the nF3-150 or K8T800.
Facebook
Twitter
RSS
10 Comments
View All Comments
AnonymouseUser - Saturday, December 20, 2003 - link
Since this review is for the Athlon64 FX motherboards, shouldn't the links for the "Anandtech Deals" (just below the title) be for Athlon64 FX (socket 940) instead of the non-FX 3200+ (socket 754)?O_o
Wesley Fink - Saturday, December 20, 2003 - link
#7 -The scores with the 11/03 nVidia platform drivers combined with Catalyst 3.9 and the latest BIOS' we tested have dropped the GunMetal 2 benchmarks to those reported in this review. We have discussed the very unusual GunMetal scores we got in the past with Yeti Studios who is looking into the scores.
At this point, we are concerned that the GunMetal 2 bechmarks are really telling us very little about the performance of the boards and systems we are testing. Unless Yeti can update or explain what we have been seeing in Socket 940 scores, we will likely drop GunMetal 2 from our benchmarks.
We apologize for the confusion regarding GunMetal 2 bechmarks, but we have shared with you over several reviews our growing skepticism over their validity in benchmarking FX and Opteron.
TrogdorJW - Friday, December 19, 2003 - link
#7, if you look at those benchmarks in question, the results are HIGHLY questionable in the original benchmarks. They even mentioned it at the bottom of the page:"The astounding scores in GunMetal 2 by the Dual-Channel Opteron and Athlon64 FX51 are difficult to explain, since they are not duplicated by our single-channel Athlon64 benchmark. We were convinced that these scores on the original Opteron must be a fluke until they showed up again in our tests and retest of the K8NNXP-940 Dual-Channel."
My bet is that the earlier versions of the GunMetal benchmark were in some way flawed. Perhaps it was a driver issue, and the game was really only rendering about 2/3 of the screens that it was reporting. Given that all the other systems appear to be close to maxed out on frame rate by the graphics card, the FX and Opteron scores were initially incorrect and have now been fixed.
justly - Friday, December 19, 2003 - link
Wesley Fink, I have had issues with previous Anandtech articles and I thought (or at least was hopefull) that they would happen less often with some of the new staff. I now regret being so hopefull as I am still seeing the same problem.What I would like to know is what would cause the gun metal benchmarks on the Gigabyte K8NNXP-940 to drop 25% or more since the review of that same board on 9 Oct (there was even a link to this article on page one).
I realize that the motherboard and video drivers have changed along with some hardware, and BOIS updates mentioned on page 1 (stating that they "offering improved performance and added features"). The thing is that none of these changes should lead to this kind of preformance hit. What is the story here, was there a mistake in benchmarking, if so what article is correct, if not how do you explain this since most of the other benchmarks on this board varied (an estimated)5% or less.
Icewind - Friday, December 19, 2003 - link
Doubtful #5 as there is no BIOS option to enable or disable it for the VIA boards.bex0rs - Friday, December 19, 2003 - link
The integrated LAN on the SK8N is 10/100 only, not gigabit as mentioned several times.http://www.asus.com/prog/spec.asp?m=SK8N&langs...
http://www.realtek.com.tw/products/products1-2.asp...
Also, would there be any way to run the HT bus on the VIA boards at 600 to make a determination if that is the limiting factor on nV's implementation?
Wesley Fink - Thursday, December 18, 2003 - link
#1 - You are correct, and page 4 has been corrected. The SATA ports for the SK8N were correctly stated as 2 in the Feature listing for the 4 motherboards.Icewind - Thursday, December 18, 2003 - link
Unless im mistaken #1, is that one right next to the CPU cooler itself in the picture below? Hard to judge from the contrastIcewind - Thursday, December 18, 2003 - link
Best to wait for the 939 pin socket without the unregistered memory modules. I know I will. Paired with a possible PCI Express, SATA 2.0, ATi's 420, 2004 is gonna be a freaking expensive upgrade but better get the best before I finally move outa my folks house.adipose - Thursday, December 18, 2003 - link
http://anandtech.com/mb/showdoc.html?i=1936&p=...On this page you state:
The IDE connectors, IDE RAID, and 4 SATA connectors are all in good locations. They should present no problems in most case designs.
But I believe the SK8N only has 2 SATA connectors, and I can only see two on the image.
-Dan