Multicore technology pays off for IBM's eServer 326m

Multicore technology pays off for IBM with its 326m server. Cramming as many dual-core Advanced Micro Devices Opteron processors into the smallest space possible, the eServer 326m provides enormous amounts of CPU muscle.

One quick caveat: This eServer model requires quite a bit of tweaking if used for a graphics application. It took some time to find the right combination of drivers to make the 326m server do reasonable graphics for SUSE Linux Enterprise Server 10, Red Hat Enterprise 4.0 and Windows 2003 Enterprise Server, run individually or as members of a VMware installation.

How we tested IBM's eServer 326m

Archive of Network World tests

Subscribe to the Network Product Test Results newsletter

The 32MB of dynamic RAM (DRAM) is a weakness - we had to use the proprietary (non-GPL and non-open source) ATI drivers for on Linux or suffer graphics problems. This was the only thing we could find anemic on an otherwise stellar system.



Price:Starts at $1,829; $3,222 as tested.
Pros:Blazes in performance for a 1U; cluster ready PCIe 8x bus.
Cons:Weird video glitches, pedestrian enclosure.
The breakdown
Performance 40%5Scoring Key: 5: Exceptional4: Very good3: Average2: Below average1: Subpar or not available
Features 25%4.5

Manageability 20%

Serviceability 15%4


The server fit easily into our 19-inch rack space - while there wasn't any especially charming or useful rail kits included with the 326m, slamming it into the rack was simple. Getting inside the machine wasn't as easy or fun as an HP-designed server, but the differences amounted to a few seconds.

The server included two dual-core AMD Opteron 285 processors clocking at 2.6GHz. The system included 2GB of DRAM (eight sticks of 512MB double data rate RAM), as well as an astonishingly small Seagate Barracuda 80GB 7200 RPM Serial Advanced Technology Attachment interface hard drive (larger sizes are available). The two PCIe 8x (PCI Express bus) connectors in our test unit begged for Fibre Channel storage-area network and/or InfiniBand connections for clustering. An alternate configuration is available with one PCIe 8x slot and one PCI-X 64-bit slot.

The unit includes three handy USB hubs, two on the front (USB 2.0) and two independent slots on the back for a keyboard, mouse or other device. Independent USB branches also let a bus with USB 1.1 devices run independently and not slow down the remaining bus connections, and are connected to the South Bridge of the motherboard, whereas the high-speed bus adapters are connected to the North Bridge. The eServer 326m worked well with our USB hardware (see How we tested IBM's eServer 326m).

Two Broadcom chipset-based Gigabit Ethernet ports are included, and other than the previously mentioned graphics oddities, our test operating systems found all hardware correctly. IBM also supports Intelligent Platform Management Interface (IPMI) management, an API set that lets IBM-based and third-party applications send or receive information on hardware characteristics that are operating system-independent. Everything from CPU loads to internal temperature can be ready by IPMI-aware applications - including open source management applications.


We used LMBench3 to run our performance tests, and compared the eServer 326m with the Polywell 2200s, which has two AMD Operton processors. We ran several tests, including TCP latency, operating system execution efficiency, file write bandwidth, socket bandwidth and pipe bandwidth (see table, below).

eServer 326m performance testsList of tests performed on the IBM eServer 326m and a Polywell 2200s system.
LMBench 3 TestPoint of testIBM e326m with two dual-core AMD Opteron 285sPolywell 2200s - two single-core AMD Opteron processors
TCP LatencyMeasures the time it takes to get a local host connection and gauges potential network efficiency through network drivers and kernel. A lower number is better.21.3 microsec26.9 microsec
Process fork + execveThis is a basic unit of operating system execution efficiency. It causes the kernel to span a fork and then execute a process. A lower time is better.289.9 microsec549 microsec
File write bandwidthMeasures how much can be written to a file in a specific period of time and gauges file system bandwidth. A higher rate is better.41.1M byte/sec25.5M byte/sec
Socket bandwidthTests RPC to IPC, which indicates efficiency of atomic interprocess communications bandwidth. This is an averaged rate and a higher score is better.27.2M byte/sec17.27M byte/sec
Pipe bandwidthTests memory movement speed between the kernel and user space. A higher score is better.1904.4M byte/sec1606.1M byte/sec

IBM's 326m server has been the fastest 1U performer in the lab, beating all four core or less systems. The host operating system was SUSE Enterprise Linux 10.1/2.6.16 Linux kernel.

The bottom line

The dual-CPU design and motherboard design appear to be key to our positive performance benchmark results. The dual-core chips, coupled with the DDR synchronous DRAM and a PCIe 8x bus, made for a potent combination.

Henderson is principal researcher for ExtremeLabs; Szenes is a researcher/analyst. Henderson can be reached at

NW Lab Alliance

Henderson and Szenes are also members of the Network World Lab Alliance, a cooperative of the premier reviewers in the network industry, each bringing to bear years of practical experience on every review. For more Lab Alliance information, including what it takes to become a member, go to

Learn more about this topic

IBM helps data centers to keep their cool


IBM to market Sybase Linux database


New IBM server to support dual-core Opteron


Join the Network World communities on Facebook and LinkedIn to comment on topics that are top of mind.

Copyright © 2006 IDG Communications, Inc.

SD-WAN buyers guide: Key questions to ask vendors (and yourself)