What's brewing in Damage Labs

— 1:51 PM on November 21, 2006

The short week here, interrupted by the Thanksgiving holiday, has made it all but impossible for me to polish off an article and publish it right away. I've had a number of projects brewing at once in Damage Labs, and none of them have come to fruition just yet. That's true in part because I've been working on pulling together some new benchmarks and revised testing methods, and such things take time. I'm excited to unveil some of our new tests, though. I think many of you will appreciate them, and some of them do a nice job of stressing how well multicore systems scale.

With my personality type, it's very difficult to accept the ebb and flow of work here in Damage Labs, which goes by a simple principle that becomes ever truer as TR has more success: there will always be more work to do, and more hardware to review, than there is time to spend on it. This is, of course, a wonderful problem to have if you consider it with proper perspective, especially since it's miraculously my job to play with PC hardware all day. Yet I can't shake the constant, nagging sense of guilt over things that haven't yet gotten the attention they deserve. That list currently includes a number of interesting and worthy items, including some things I'd pledged to write about in past Etc. or blog posts.

An obvious, big-ticket item is Nvidia's G80 GPU, better known as the GeForce 8800. I feel like my initial review covered about 50% of what I'd like to do with this puppy. Among other things, I owe you guys some SLI test results. Fortunately, those are indeed coming, with both the 8800 GTX and GTS models. The real trick in testing this, for me, is to find games and image quality settings that can properly take advantage of such a killer graphics subsystem without resorting to a $$$$ 30" LCD display with 2560x1600 resolution. I may just have to fork over the cash to buy one of those displays eventually, but I'm still resisting.

Anyhow, I do have more G80 coverage coming, although it's taking longer than I'd expected.

The project that's been pulling my attention away from the G80 is a look at performance and power use in server processors, including the new quad-core Xeons. Several readers have offered their help with building some new multithreaded benchmarks based on real-world applications and real data sets, as well. Very exciting. I've also been looking into some of the SPEC tests, and I may have one or two of those to use.

Incidentally, if you happen to work on an application in an academic or business scenario and think you might be able to build a good benchmark for us, by all means, let me know. We are not a performance testing service for the private use of your company, but we are always looking for good benchmarks from those who are willing to share openly. Contact me if you have questions.

The G80 and new CPUs have stolen my attention away from another intriguing product, Bigfoot's Killer NIC. We've been slow to review this product in part because we got a slow start on it. I will admit that my first reaction to the marketing hype on this one was very negative, largely because there was nothing being said about the actual technical merits of the product—and the idea that a NIC would improve lag in games substantially was counter-intuitive. As a result, we didn't even contact Bigfoot for a review unit right away, until some of you guys started asking for a review in the forums. (My bad, I know. I just had too much on my plate at the time.) Once I did talk to the guys at Bigfoot, though, we had a startlingly frank and productive conversation, free of hyperbolic marketing hype and full of juicy technical details. I came away with the impression that what they're doing could well have tangible benefits—not for general network performance, but for certain games that speak to the network in a particular way. That advantage is almost entirely a client-side one, and the only appropriate way to test it is with networked games themselves. I intend to do so, but like I said, my attention has been siphoned away by other projects. The Killer NIC is here, though, patiently waiting its turn.

Perhaps the longest lingering project that's failed to materialize into an article is my HTPC build. I ordered the parts for a new home theater PC last December, and I got it built a month or so later. I never did a write-up about it, though, partially because the system was plagued with instability during the first few months of its existence. What's the point in trumpeting your slick new PC build if the system doesn't work right? I wanted to iron out the problems first. (Also, surprisingly enough, I found that a "production" HTPC system is rather difficult to disassemble, photograph, and the like because, well, it's a production box. Odd how the stakes seem high when disconnecting it.)

The shame of it all is that despite the glitches, this system turned our very, very well. Andy, long-time TR staffer, saw the thing and went and built himself a similar system, which is as strong an endorsement as one could give. The system is in a big, black Silverstone enclosure that looks right at home in the living room, and it has a remote, a wireless keyboard, and two wireless gamepads. All of the basic HTPC hardware requirements are addressed properly, including a dual-core CPU, a large hard drive, dual TV tuners, a DVD RW drive, and a video card with component outputs. On top of that, I chose components that were quiet, even replacing a couple of them along with way with less noisy alternatives. Turns out that building an HTPC this good is by no means cheap, but it is very satisfying.

Let me pause to give a hearty endorsement to Windows XP Media Center Edition, which anchors the whole thing. Microsoft, bless their shriveled hearts, nailed this one. Nothing else I had tried could come close to supplanting our old TiVo in daily use, but MCE did so immediately. My wife and I both took to it and never looked back, even with the stability problems the box had.

Speaking of things lingering in Damage Labs, the resolution of my HTPC problems came in the form of convergence with another neglected project. After months of experimenting, I finally determined that my HTPC's stability problems were mainly caused by resource conflicts between its two TV tuner cards, an ATI TV Wonder Elite and a Hauppage PVR-150 MCE. These two devices just didn't get along well together in the same system, even if I moved them around into different PCI slots. Fortunately, some time earlier, Nvidia had sent us one of its DualTV MCE cards out of the blue, hoping for a review. Despite my best intentions, the thing had just sat in its box unopened as I worked on other things. After reading this very impressive review at HTPCnews of the DualTV MCE, I decided to give it a try in my own HTPC box. I was very pleasantly surprised, for three reasons:

  • First, the HTPCnews review's assessment was correct. The DualTV MCE's image quality is at least as good as ATI's excellent TV Wonder Elite, if not superior. The card has strong reception on both tuners and avoids an annoying compression artifact on the ATI card where faces seem to "pause" and not move immediately as the camera moves. Scrolling text, deinterlacing, and the like are all handled well, too.

  • Next, putting in a single card with two tuners resolved the stability problems. Hallelujah!

  • Finally, having two of the same tuner prevents some undesirable behaviors for an HTPC. With the ATI and Hauppage cards, the signal reception, picture quality, and audio volume levels varied way too much between the two cards, and despite making an effort, I couldn't get them tweaked up to match.
For these reasons, I'm now a big advocate of dual-tuner cards for HTPCs and of Nvidia's DualTV MCE in particular. You're going to want two tuners, so save yourself trouble by getting them both on one card.

I've also neglected to mention something interesting I learned when we converted to new power supply units for our test rigs here. For quite a while, we used OCZ's PowerStream 520W units for all of our test systems, and they were always very solid. However, with quad SLI and the like on the horizon, we needed more power to be 100% sure we were ready to test the latest stuff. And when OCZ announced its GameXStream series with big, quiet 120mm fans, I couldn't resist the prospect of a whisper-silent Damage Labs running tests at full tilt on multiple systems. OCZ generously agreed to outfit us with their new 700W GameXStreams, and we converted our test rigs to them.

I expected these PSUs to be more power efficient, especially with active PFC, so I set up a quick test involving a watt meter and a Pentium Extreme Edition 965-based test rig. Have a look at how the GameXStream 700W compared to the PowerStream 520W.

Very impressive. We have been using the GameXStreams for some time now in all of our testing, and we'll try to continue keeping an eye on efficiency when selecting PSUs in the future. By the way, the GameXStreams are also as quiet as I'd hoped. I don't know why anyone would buy a PSU with a smaller fan on it when units like these are available from OCZ, Seasonic, and others. Crazily enough, though, we may need to switch to new PSUs soon for testing, because we need four six-pin PCIe aux power connectors for 8800 GTX SLI and the like.

I have even more projects brewing here, believe it or not, but I should probably draw this already lengthy post to a close—and get back to work on them.

Tip: You can use the A/Z keys to walk threads.
View options

This discussion is now closed.