The Rules, Contenders, And Test Setup
We set some ground rules for the vendors that wanted to participate in this story. The manufacturers received instruction that all platforms were to have:
- At least two LGA 2011 interfaces supporting at least 135 W CPUs to accommodate the Xeon E5-2690 processors we planned to use.
- At least two gigabit Ethernet ports (excluding any dedicated IPMI NIC.) Most server motherboards already include at least two gigabit ports. Although some vendors offer solutions with four or more on-board ports, we wanted the option for at least two.
- IPMI or similar remote monitoring and KVM-over-IP integrated or through an included add-in module. Nobody has industry-wide numbers breaking out how many servers have IPMI 2.0 and remote KVM-over-IP built-in versus provided by an external controller, though we've heard that as many as 70% of these platforms ship with these features.
- No additional SAS controller (such as an value-added LSI SAS2008). However, PCH-enabled SAS is OK. One of the major new features with the Patsburg chipset is its ability to enable up to eight SAS 3Gb/s ports using an on-die Storage Controller Unit. This is really an aggressive move by Intel, as LSI sells many of its own controllers that augment Intel-based servers. If Intel can achieve parity with LSI in terms of features, then this will become a major industry trend.
- No built-in InfiniBand or 10 GbE, as typical servers in this segment do not have those more enterprise-oriented extras as integrated features. Vendors do offer solutions with them, but in lower volumes.
- A price, including any required accessories, below $1800. This figure may seem high, but it is actually very close to average for a barebones chassis, motherboard, and redundant power supplies in the segment we're targeting.
Generally, most vendors complied with our list, which made our comparison easier. There were a few exceptions, which we'll note as we go through the story. Notice that we didn't ask the manufacturers to adhere to a size restriction, since we wanted to make sure that vendors had the flexibility to submit different types of servers. As the picture on the previous page reveals, all three vendors submitted 2U chassis. That's a pretty telling indicator of what the industry considers its sweet spot.
The other key piece of this puzzle is the processors we're using. Intel sent two Xeon E5-2690 CPUs, which are very similar to the workstation E5-2687W parts reviewed previously (except that 200 MHz of base clock rate is sacrificed for a 15 W-lower TDP, a trade-off that makes sense in a cramped 2U enclosure). The Sandy Bridge-EP-based Xeon E5 platform's new features were covered in this short video, which is worth watching as an overview.
Another important consideration was our choice of memory. Trying to be vendor-agnostic, Micron, Kingston, and Samsung memory were all used with each barebones system. Also, the systems were tested with both unbuffered ECC and registered DIMMs. After verifying that compatibility was uniform across each system, we settled on registered DIMMs for testing, since they're most common in dual-processor servers.
|Processors||2 x Intel Xeon E5-2690W (Sandy Bridge-EP) 2.9 GHz, Eight Cores, LGA 2011, 8 GT/s QPI, 20 MB Shared L3, Hyper-Threading enabled, Power-savings enabled|
|Row 2 - Cell 0||Tyan GN70-K7053 (S7053 + KGN70M1)|
|Row 3 - Cell 0||Intel R2208GZ4GC "Grizzly Pass"|
|Memory||Micron 64 GB (16 x 4 GB) DDR3-1333CL9 Registered ECC|
|Row 5 - Cell 0||Samsung 64 GB (16 x 4 GB) DDR3-1333CL9 Registered ECC|
|Row 6 - Cell 0||Kingston 32 GB (16 x 4 GB) DDR3-1333 CL9 Unbuffered|
|Hard Drive||2 x OCZ Vertex 3 120 GB in RAID 0|
|Graphics||On-board graphics from each platform|
|Power Supply||Included power supplies from each platform|
|System Software And Drivers|
|Operating System||Windows 7 Ultimate 64-bitUbuntu Linux 10.10 64-bit|
An abbreviated test suite was used for this round-up, since the performance differences between platforms ended up being very slight (in light of common processors and memory). All tests were run using a clean installation of Microsoft Windows 7 64-bit, except the Folding@Home test that used Ubuntu Linux 10.10 64-bit. The older version of Ubuntu is probably the most widely-used distribution for this application, as it produces consistently high performance.
|Benchmarks and Settings|
|Adobe After Effects||Version: CS5.5Tom's Hardware Workload, SD project with three picture-in-picture frames, source video at 720p, Render Multiple Frames Simultaneously|
|Adobe Photoshop||Version: CS5Tom's Hardware Workload, Radial Blur, Shape Blur, Median, Polar Coordinates filters|
|Adobe Premiere Pro||Version: CS5.5Paladin Workload, Maximum Render Quality, H.264 Blu-ray profile|
|Euler3D||CFD simulation over NACA 445.6 aeroelastic test wing at Mach .5|
|3ds Max 2012||Version: 10 x64Rendering Space Flyby Mentalray (SPECapc_3dsmax9), Frame: 248, Resolution: 1440 x 1080|
|Visual Studio 2010||Compile Chrome project (1/31/2012) with devenv.com /build Release|
|Folding@Home||6903 Work Unit under Ubuntu 10.10 64-bit-smp -bigadv CLI client parameters|
|Synthetic Benchmarks and Settings|
|Cinebench 11.5||CPU Test, Built-in benchmark|
I agree. Just reduce it a little bit but don't make it too hard to see
As an AMD fan, I wish we could. But while Magny-Cours was competitive with the last gen Xeons, AMD doesn't really have anything that stacks up against the E5. In pretty much every workload, E5 dominates the 62xx or the 61xx series by 30-50%. The E5 is even price competitive at this point.
We'll just have to see how Piledriver does.
Having said that I would suggest you include expected PPD for the given TPF since that is what folders look at when deciding on hardware. Or you could just devote 48 hours from each machine to generate actual results for F@H and donate those points to your F@H team (yes Tom's has a team and visibility is our biggest problem).