Battle at the Top

The long-awaited Socket 939 Athlon 64 from AMD debuted on June 1 at Computex, but it has taken quite a while for Socket 939 motherboards to make their way into the retail channel. With the Socket 939 still perched at the top of the AMD product line, both in price and performance, the offerings still remain very small. We take a look at 6 new motherboards for Socket 939 from Abit, Asus, ECS, Gigabyte, and MSI.

The Asus A8V Deluxe was not included in the original roundup because Asus told us a new Revision would soon be available. The Asus A8V Deluxe Revision 2.0, with a working AGP/PCI lock, was received a few days ago and has now been added to the roundup. We have included full test results from the Revison 2 A8V and compared it to the other 5 top-end 939 boards in the roundup. DFI has also announced a Socket 939 board based on the nForce3 Ultra chipset, but the retail introduction of that board is still several weeks away.

The Second-Generation chipsets for Athlon 64 represent a genuine improvement in features, flexibility, and performance for AMD's flagship processors. As you saw in our chipset review, the nForce3 has added the features to bring the nVidia chipsets to the competitive edge of Athlon 64 chipsets. The HyperTransport speed, which was widely criticized on the 150 chipset, is now 1000 in the Ultra version of the chipset used for 939. The chipset is, otherwise, the same as the 800 and 1000FSB versions that we saw in our Socket 754 Roundup: Comparing Generation 2. The on-chip Gigabit LAN and on-chip Firewall are unique and truly useful features among Athlon 64 solutions. nVidia also fixed the problems with their implementation of PCI/AGP lock on the 150, and we are finding a working PCI/AGP lock on every nForce3-250 motherboard that we test, just as nVidia promised.

VIA's update to their excellent K8T800 chipset was less dramatic as there are not many new features, but the improvements are just as dramatic under the hood. As you saw in our review of the K8T800 PRO chipset, VIA brings 1000 HyperTransport to the PRO chipset for all Athlon 64 Sockets - 940, 754, and 939. Perhaps the most significant improvement was the addition of a working PCI/AGP lock to the K8T800 PRO. Since all PRO chipsets support 1000 HT, the chipsets used in our Socket 754 Roundup: Comparing Generation 2 are the same as you will see in our 939 roundup.

As reported in the Socket 754 roundup and the Socket 939 launch review, VIA had early problems with their PCI/AGP lock. While we continued to see problems with a working AGP/PCI lock in early 939 samples, we are pleased to report that every vendor in our roundup now appears to have the VIA PCI/AGP lock working. If you have a VIA board with PCI/AGP lock issues, our advice is to contact your motherboard vendor for the latest BIOS or possibly a revised version of your motherboard. The good news is that VIA has worked with manufacturers to fix the issues, and the issues can be solved; the bad news is that there are still many VIA K8T800 PRO motherboards in the market with PCI/AGP lock issues, and some require a motherboard replacement to fix the problem. In some cases, we are looking at a third motherboard revision before we finally receive a working PCI/AGP lock on a VIA K8T800 PRO board.

Today, we are looking at 6 new Socket 939 motherboards, representing the top performance level for the Athlon 64, which is the fastest current CPU. Consider this roundup a search for the best of the best, since Socket 939 supports the top-performing Athlon 64 processors available in the fastest Dual-Channel memory configuration. The Gigabyte K8NSNXP-939 and MSI K8N Neo2 feature the nVidia nForce3 Ultra chipset, while the Abit AV8, Asus A8V Deluxe Rev. 2, ECS KV2 Extreme, and MSI K8T Neo2 are based on the VIA K8T800 PRO chipset.

Abit AV8: Features and Layout
POST A COMMENT

83 Comments

View All Comments

  • harsaphes - Sunday, October 10, 2004 - link

    just set up my a8v board. no go on firewire, will not see ipod or external firewire drive. any idea?...bad board maybe? Reply
  • kd4yum - Thursday, August 05, 2004 - link

    Reply
  • Fender - Wednesday, August 04, 2004 - link

    The first words in this review confirm that you should never buy a VIA product before they revised it! (KT266 to 400 saga anyone?)
    Also, it's oddly that Hyperions used here are 8 months old...
    It could be interesting if you include, in your tests, any possible issue concerning OS installation or updating drivers (from the CD included in the box to the updated drivers from the manufacturer's site) because this is what happens to most buyers out there. Remember that you're testing a 64bit CPU with a 32bit OS, and drivers development will be a further support insurance.
    Reply
  • Wesley Fink - Tuesday, August 03, 2004 - link

    #79 - Page 13 is corrected. That error slipped past 3 proofreaders. Thank you for alerting us. Reply
  • Sidewinder0010 - Tuesday, August 03, 2004 - link

    the k8t neo2 overclocking page has a typo that was throwing me off
    "much lower than the 290 on the K8T Neo2"
    That should be changed to k8n neo2
    Reply
  • Compddd - Friday, July 30, 2004 - link

    Wesley, did Asus say when the A8V Rev 2.0 will be hitting Retail Stores like Fry's, Best Buy, etc? Reply
  • thebluesgnr - Friday, July 30, 2004 - link

    How does the lower bandwidth (4900Mb/s) affect real world performance?

    One more thing, perhaps the MSI K8T Neo2 Platinum references should be changed to MSI K8T Neo2-FIR? That's the name of this board on MSI's site, there's no mention of it being a Platinum board. And here's a link of all the boards in the Platinum series: http://www.msicomputer.com/pressrelease/platinum.a...

    Off-Topic: Wesley, will there be reviews on AnandTech of Socket A mobos based on the nForce2 Ultra 400Gb and VIA KT880 chipsets? They offer the exact same features of the mobos on this roundup, and with the new Semprons and the good XP-Mobiles I suspect a lot of your readers are still interested in this socket.
    Sorry for the off-topic.
    Reply
  • Wesley Fink - Friday, July 30, 2004 - link

    #75 - You are confusing boards as MSI had TWO boards in the roundup. We had no problems at all with the first or second K8N Neo2 (based on nForce3 Ultra) we tested, and that is the board that received the Gold Editors Choice. The MSI K8T Neo2, based on the VIA K8T800 PRO, was the problem board and we definitely did NOT give the K8T Neo2 an award.

    #74 - On page 5 we talk about the memory bandwidth differences in 1T and 2T command rates: "The best performance is at a Command Rate of 1T, and the Abit AV8 was completely stable at a 1T setting with 2 DIMMs. Standard memory bandwidth measured with SiSoft Sandra 2004 SP2 shows a 6000 MB/s bandwidth with 1T Command Rate compared to a 5000 MB/s bandwidth with a 2T setting." This is also mentioned on page 11: "While a full memory comparison of the nVidia and VIA chipsets is beyond the scope of this roundup, we did run several SiSoft Sandra 2004 SP2 runs of the memory test module. At default settings, and the aggressive 2-2-2-10 timings on the FX53, the nF3-250 Ultra showed memory bandwidth in the 6100 range for FPU and Float. The same test on the VIA K8T800 PRO boards showed memory bandwidth in the 6000 range. Performance of both chipsets at the 1T setting was very similar. 2T Command Rates, with everything else the same, generated bandwidths of 4900 to 5000Mb/second."

    Reply
  • Z80 - Friday, July 30, 2004 - link

    Considering your statement "Our concern is based on the fact that we went through 3 K8T Neo2 boards before we got one that really worked" I'm surprised that you went ahead and gave the MSI board your gold award. I see that MSI is being sued for intentionally using capacitors that were made with an improperly-formulated electrolyte solution. My personal experience with MSI quality assurance was never good at least back in the day it wan't but maybe they have changed like you say OCZ has? I'll stick with Asus and Abit, thank you. Reply
  • SignalPST - Friday, July 30, 2004 - link

    quote from the article:
    "We even found that all six of the tested motherboards performed at the fastest timings available and a 2T Command Rate with 4 DIMMs on board, so even that is a non-issue."

    Does that mean that if 4 sticks of ram are installed, there won't be a performance hit and it will work just as fast as 2 sticks installed?
    Reply

Log in

Don't have an account? Sign up now