Intel Xeon 5300 Clovertown
While it is still a bit early to give our final impressions on these next-generation server processors we have so far tested two of the Intel Xeon E5320 processors with the Tyan Tempest i5000XT. The Tyan Tempest i5000XT S2696 on our test bench is powered by Intel's 5000X Chipset and offers support for dual LGA-771 sockets, eight DDR2 FB-DIMM modules, PCI Express x16, and six Serial ATA 2.0 ports. An independent review will be written on the Tyan i5000XT in the coming days at Phoronix (our Tyan Tempest i5000XL S2692 preview can be read here). Tyan's Tempest series is designed for powerful server platforms with maximum scalability and flexibility.
Intel's whitepaper on Clovertown state that the Xeon 5300 series delivers up to 1.5 times the performance compared to the leading Xeon 5100 series and a whopping 2.5 times the performance of the AMD Opteron 2220. For our purposes today we had run the Xeon E5320 against the Xeon 5150 and Xeon 5060. The Xeon 5060 is a Dempsey-based NetBurst processor. Like the other processors we are running it is built on a 65nm process and the 5060 model operates at 3.20GHz with a 1066MHz FSB. The Dempsey offers 2MB of L2 cache per core and supports Hyper-Threading Technology to offer a total of four logical cores. This processor was chosen for comparison to represent a now low-end Xeon processor that is currently selling for approximately $350 USD. The Intel Xeon 5150 runs at 2.66GHz, 1333MHz FSB, 4MB shared L2 cache, and is a dual core part without Hyper Threading. We had run our selected tests with a single Xeon 5150 and then again with dual Xeon 5150 CPUs. Finally, we performed our benchmarks with a single Xeon Quad-Core E5320 and then again with a second E5320 to make an octal-core system.
These processors were tested with Fedora Core 6 using the Linux 2.6.18 kernel. At this time there are clocking problems with Enhanced Intel SpeedStep Technology and cpufreq, but the Xeon 5300 series can be ran at full speed by disabling cpufreq from the kernel or manually selecting the higher state. For benchmarking these processors we had used Enemy Territory, Quake 4, SPECViewPerf, LAME GCC compilation, LAME encoding, Ogg Encoding, FreeBench, RAMspeed, and kernel compilation. While a normal one would not shell out a few thousand dollars on creating an octal-core Xeon server for running games based on the Quake 3 engine, Enemy Territory benchmarking was tested due to its common place for GNU/Linux benchmarking. Quake 4 was called in for action in part due to it being one of the few games to support Symmetric Multi-Processing. SPECViewPerf was used as representation of workstation graphics tasks, while LAME and Ogg benchmarks were performed to represent additional CPU metrics. For the kernel compilation, we had measured the kernel compile time for the Linux 126.96.36.199 vanilla kernel. In the compilation benchmarks the number of jobs was determined by the number of total cores in the system plus one for the -j argument. If you have any questions about our benchmarking practices or would like to see additional benchmarks please stop by the Phoronix Forums.
Latest Linux Hardware Reviews
Latest Linux Articles
Latest Linux News
Latest Forum Discussions