Multimedia and Networking Benchmarks

We ran all the configurations through a set of encoding benchmarks. We used chapter 9 from The Sum of All Fears, with a target quality of 75% and without audio. This does a single-pass encode that results in reasonable quality at the cost of file size. We also ran a multi-pass encode of a 30-second commercial, with a target size of 5 MB and with audio. The original video is 30 MB in size, and AutoGK makes three passes: a compression test, an initial pass, and the final pass. For larger files, the compression test scans segments of the video, so it isn't a full pass. With this small file, it ends up simply doing three full passes.

Encoding Performance - AutoGK 1.96

Encoding Performance - AutoGK 1.96

We didn't bother including results for each configuration, since the different graphics cards have no impact on encoding performance. The small difference in encoding rates is well within the margin of error. This comes as no surprise, since DivX and Xvid are CPU/memory limited and make very little use of the graphics card. In the future, we may see plug-ins that will use the GPU's programmable shaders to help accelerate encoding, but for now, the CPU is the biggest bottleneck. Again, the custom system slightly outperforms the HP DX5150 by 2 to 3%.

As with encoding performance, using a different graphics card has no impact on network performance. We employed our standard testing methodology using the Windows 2000 DDK. A second system, an MSI K8N Neo Platinum with a 3400+ processor, was set to communicate with each system in turn. Traffic was routed through a gigabit Ethernet switch, but that doesn't matter in real world performance scenarios - rarely does sustained network traffic exceed 500 Mb per second.

Networking Performance - Win2000DDK

Networking Performance - Win2000DDK

The networking solution in the nForce 6150 chipset is superior to that in the Xpress 200 chipset. The ATI chipset does better when sending data, but in both cases, it trails the NVIDIA chipset. The flip side is that the HP system has slightly lower CPU usage, which makes sense. Jumbo frames were not enabled in testing, which can result in higher throughput and lower CPU usage. Realistically, servers might benefit from Jumbo frames and higher gigabit Ethernet performance, but most desktop systems don't need to worry about it.

General Application Benchmarks Gaming Benchmarks – Battlefield 2
Comments Locked

48 Comments

View All Comments

  • gibhunter - Wednesday, December 14, 2005 - link

    That's not an X2 3800+. It's the standard 2.4GHZ 3800+ single core.

    Regarding these HPs, I have a few of these at work. They are realy great. For $500 and change you get an Athlon 64, 512MB of ram and a WinXP Pro. Try and put a system like that yourself and you'll spend just as much or more and that's not counting the snazzy keyboard and mouse that comes with that system. It really is a good deal.
  • JarredWalton - Thursday, December 15, 2005 - link

    Just to reiterate, the linked 3800+ is indeed an X2:

    "We actually have an X2 3800+ Smart Buy, sku # pz635ua#aba....it might be
    listed incorrectly as a 3800+, but it's an X2. I'm in the process of
    getting that fixed."

    That's from an HP representative, one of the marketing managers of the small-business division.
  • Googer - Thursday, December 15, 2005 - link

    Equally impressive for the $500-ish range is this http://e4me.com/products/products.html?prod=eMachi...">e-machine
  • LoneWolf15 - Thursday, December 15, 2005 - link

    If you're a business, e-Machines isn't equally impressive. Part of what you are paying for is the support. The system reviewed carries a three-year warranty (par for the course on business systems) and probably carries business-level support too. Most HP systems also use a fair number of brand-name parts (i.e., ASUS mainboards in most systems). I don't deny that eMachines has its place, but it comes nowhere near something that HP puts out.

    P.S. While I like most of HP's system configurations, even home ones, I haven't heard good things about home-level support. And one other thing, Jared...why does the article say this system has a Clawhammer core CPU? I thought Clawhammers went the way of the dinosaur on Socket 939 long ago. Anything this new ought to have a Venice or San Diego core chip in it.
  • JarredWalton - Thursday, December 15, 2005 - link

    Well, it does have a ClawHammer -- at least the system I have does. You have to remember that AMD only has one fab producing 90 nm parts, and they have an old fab that still produces 130 nm parts. Perhaps AMD gives them a better deal on the older chips? Or perhaps it's just that this model was made a little while ago? If it had used a San Diego core, I expect power draw would have dropped another 20 W at least.
  • mino - Saturday, December 17, 2005 - link

    You are wrong on this. AMD publicly stated sometime in the Q2 that they have converted all of their lines onto 90nm production.
    Also AMD does have only one fab - FAB30 - currently in producing CPU's. While there is FAB25 it produces flash and is part of Spansion division and there is also FAB35(or 36?) in qualification process the only fab producing AMD CPU's in volume is currently FAB30 on 200mm wafers.
  • JarredWalton - Friday, December 23, 2005 - link

    Hmm... obviously I'm not paying close enough attention to AMD's fabs. I could have sworn they still had a 130nm fab making CPUs. I would have thought 130nm would be sufficient for a lot of stuff - better to keep what you have running instead of retrofitting old fabs, right? Then again, new fabs are getting more and more expensive.
  • mino - Sunday, January 1, 2006 - link

    Well, Austin FAB25 was not suitable nor meant for smaller than 180nm process (for logic products). AMD thus made a cash cow out of it during hard AthlonXP times. Also the capacity of any FAB is measured in wafers/time not chips pre time. In other words AMD could make twice as many K8 CPU's on 90nm than on 130nm. Couple that with huge capacity constraints AMD faced in 2005 and fact they had only one 200mm FAB and it becomes clear why not to produce on 130nm. Around this time FAB35 should come online so the tight supply of the last quarter should not repeat for some time. Also AMD's 90nm SOI process is pretty good so don't expect FAB30 phase-out anytime soon(90nm is last logic process for FAB30). Shame FAB35 wasn't online in 2005, Intel would've had a way hotter year than it had.

Log in

Don't have an account? Sign up now