As I mentioned on the podcast this week, I have been working to re-fit Damage Labs with new hardware all around. Since I test desktop GPUs, desktop CPUs, and workstation/server CPUs, I have a number of test rigs dedicated to each area. Our desktop CPU and GPU systems have been the same for quite some time now. Heck, my massive stable of 30+ CPU results dates back to the Sandy Bridge launch. However, as time passes, new hardware and software replaces the old, and we must revamp our test systems in order to stay current. Oddly enough, we've just hit such an inflection point in all of the types of hardware I test pretty much at the same time. Normally, these things are staggered out a bit, which makes the change easier to manage.
Fortunately, though, I've been making solid progress on all fronts.
The first of my test rigs to get the treatment are my two graphics systems—identical, except one is dedicated to Nvidia cards and the other to AMD Radeons, so we can avoid video drivers for one type of GPU causing problems for the other. Also, I can test two different configurations in parallel, which really helps with productivity when you're running scripted benchmarks and the like.
The old GPU rigs were very nice X58 systems that lasted for years, upgraded along the way from four cores to six and from hard drives to SSDs. They're still fast systems, but it was time for a change. Let me give you a quick tour of our new systems, and we'll talk about the reasons for the upgrade.
Behold, the new Damage Labs GPU test rig. Innit pretty? In the past, our open-air test rigs have sat on a motherboard box, with the PSU sitting on one side and the drives out front. This system, however, is mounted in a nifty open-air case that the folks at MSI happened to throw into a box with some other hardware they were shipping to us. I was intrigued and put the thing together, and it looks to be almost ideal for our purposes. I'm now begging MSI for more. If we can swing it, we may even give away one of these puppies to a lucky reader. That may be the only way to get one, since this rack apparently isn't a commercial product.
Here are a few more shots from different angles.
Nifty and pretty tidy, all things considered. Even takes up less room on the test bench.
Now, let's talk specs. I had several goals for this upgrade, including the transition to PCI Express 3.0, a lower noise floor for measuring video card cooler acoustics, and lower base system power draw. I think the components I've chosen have allowed me to achieve all three.
CPU and mobo: Intel Core i7-3820 and Gigabyte X79-UD3 - The X79 platform is currently the only option if you want PCIe 3.0 support. Of course, even after Ivy Bridge arrives with PCIe Gen3 for lower-end systems, the X79 will be the only platform with enough PCIe lanes to support dual-x16 or quad-x8 connectivity for multi-GPU rigs.
Obviously, the conversion to PCI 3.0 essentially doubles the communications bandwidth available, but that's not all. The integration of PCIe connectivity directly into the CPU silicon eliminates a chip-to-chip "hop" in the I/O network and should cut latency substantially, even for graphics cards that only support PCIe Gen2.
The Core i7-3820 is the least expensive processor for the X79 platform, making it an easy choice. Yes, we've dropped down a couple of cores compared to our prior-gen GPU rigs. That's partly because I didn't want to get too far into exotic territory with these new systems. With four cores and a Turbo peak of 3.8GHz, the Core i7-3820 should perform quite similarly to a Core i7-2600K in cases where the X79 platform's additional bandwidth is no help.
We did want to be able to accommodate the most extreme configurations when the situation calls for it, though. That's one reason I selected Gigabyte's X79-UD3 mobo for this build. Even some of the more expensive X79 boards don't have four physical PCIe x16 slots onboard like the UD3 does. Those slots are positioned to allow four double-width cards at once, making the UD3 nearly ideal for this mission.
Cramming in all of those slots and the X79's quad memory channels is no minor achievement, and it did require some compromises. The UD3 lacks an on-board power button, a common feature that's only important for, well, open-air test rigs like this one. Also, the spacing around the CPU socket is incredibly tight. With that big tower cooler installed, reaching the tab to release the retention mechanism on the primary PCIe x16 slot is nearly impossible. I had to jam part of a zip tie into the retention mechanism, semi-permanently defeating it, in order to make card swaps easier.
Still, I'm so far pleased with Gigabyte's new EFI menu and with the relatively decent power consumption of the system, which looks to be about 66W at idle with a Radeon HD 7970 installed. That's roughly 40W lower than our prior test rigs, a considerable decrease.
Memory: Corsair Vengeance 1600MHz quad-channel kit, 16GB - If you're going X79, you'll need four fast DIMMs to keep up, and Corsair was kind enough to send out some Vengeance kits for us to use. Setup is dead simple with the built-in memory profile, supported by the UD3.
PSU: Corsair AX850 - Our old PC Power & Cooling Silencer 750W power supplies served us well for years, but they eventually developed some electronics whine and chatter under load that interfered with our acoustic measurements. It was time for a replacement, and the wonderfully modular Corsair AX850 fit the bill. Although 850W may seem like overkill, we had some tense moments in the past when we pushed our old 750W Silencers to the brink. I wanted some additional headroom. It didn't hurt that the AX850 is 80 Plus Gold certified, and I think the nice reduction we've seen in system-wide idle power draw speaks well of this PSU's efficiency at lower loads. (In fact, when the 7970 goes into its ZeroCore power mode, system power draw drops to 54W.) Even better, when load is 20% or less of peak, the AX850 completely shuts down its cooling fan. That means our idle acoustic measurements should be entirely devoid of PSU fan noise.
CPU cooler: Thermaltake Frio - The original plan was to use Thermaltake's massive new Frio OCK coolers on these test rigs, but the OCK literally would not fit, because the fans wouldn't allow clearance for our relatively tall Vengeance DIMMs. That discovery prompted a quick exchange with Thermaltake, who sent out LGA2011 adapter kits for the older original Frio coolers we had on hand. Although the original Frio isn't that much smaller than the OCK version, we were able to shoehorn a Frio in a single-fan config into this system. The fan enclosure does push up against one DIMM slightly, but that hasn't caused any problems. With a cooler this large, we can keep the fan speed cranked way down, so the Frio is blessedly quiet, without the occasional pump noise you get from the water coolers often used in this class of system.
Storage: Corsair F240 SSD and some old DVD drive - The F240 SSD was a fairly recent upgrade to our old test rigs, and it's one of the two components carried over from those systems, along with the ancient-but-still-necessary DVD drive for installing the handful of games we haven't obtained digitally. The biggest drawback to the SSD? Not enough time to read the loading screens between levels sometimes.
That's about it for the specs. I'm very pleased with the power and noise levels of these new systems. The noise floor at idle on our old test rigs, with the meter perched on a tripod about 14" away, was roughly 34 dB. I'm hoping we'll be able to take that lower with these systems, although honestly, driving too far below that may be difficult without a change of environments. Our basement lab is nothing special in terms of acoustic dampening and such. We'll have to see; I haven't managed to squeeze in a late-night acoustic measurement just yet.
For what it's worth, we have considered using a system in a proper PC case for acoustic and thermal measurements, but that hasn't worked out for various reasons, including the sheer convenience for us, typically rushing on some borderline-abusive deadline, of being able to swap components freely. We also have concerns about whether a case will serve to dampen the noise coming from the various coolers, effectively muting differences on our meter readings that the human ear could still perceive. We may still investigate building a dedicated, enclosed acoustic/thermal test rig in the future, though. We'll see.
Now that the new Damage Labs GPU test rigs are complete, I'm sadly not going to be able to put them to use immediately. I have to move on to testing another type of chip first. I'll get back here eventually, though. I still need to test Radeon HD 7900-series CrossFire, and I understand there are some other new GPUs coming before too long, as well.