Intel's 10-core Westmere-EX Server CPU

The Westmere-EX is Intel's first server CPU to be based on the 32-nm fabrication process. This allows the CPU to be manufactured at higher frequency, without an equivalent increase in die-size or thermal footprint. There are 18 different skews available, under the brand name Intel Xeon E7 and E3 series. The 4U rack-mountable server unit, called QSSC-S4R that we received came fitted with four Intel Xeon E7-4870 CPUs on a LGA 1567 socket. This particular CPU runs at 2.4 GHz and supports multi-threading, thereby allowing each core to run two threads each. That translates to 80 threads running in a single server. Moreover, the processors share 30 MB of L3 cache. Please read the server CPUs article under the Tech Trends section to know more technical details about this CPU.

This sort of a server unit is meant for high performance computing applications like engineering design automation, geo-physical modelling, and other applications requiring large memory footprint. Plus of course, with 40 cores in the same box, the unit is great for doing server virtualization. So basically a single unit like this can replace more than 15 dual-core servers in your data center. Imagine all the space, power, and productivity you'll save by using one of these. This one factor would of course be a key criteria in deciding whether to buy a unit based on this CPU or not. So, while the initial cost of the unit might be high, this should be one criteria you should use to evaluate the RoI.

Some other highlights of the CPU include Advanced Reliability Technology, which provides automatic detection and correction of errors. It supports dynamic reassignment of workloads across CPUs, interconnect error detection/recovery, and individual virtual machine recovery in virtualized environments. All this is supposed to translate into better performance and higher reliability--something that only the high-end RISC CPU world always boasts of.

About the Server Unit

Power and cooling: The four CPUs are cooled by rather large 3U high heat sinks. The system is cooled by 8 hot swap system fans in a redundant (7+1) configuration or four hot swap fans in a non-redundant configuration. The system is powered by 4- 850W high efficiency power supplies in a redundant (2+2 or 3+1) configuration. The PSUs are hot-swappable too.

Storage: QSSC-S4R supports up to eight 2.5-inch hot-swappable SAS II /SATA II hard drives, delivering up to 4TB of internal storage for demanding data needs. This is the ideal solution for datacenter applications, enterprise database, data warehousing analysis, streaming media, messaging platforms, etc. The test system we received housed two 120 GB Solid State Drives.

Click on the image to enlarge

RAM: Our test system shipped with eight memory daughter boards, each having two memory buffers and four DIMMs per memory buffer. The system supports DDR3-1066 and DDR3-1333 Registered DIMMs (with ECC) running at 800, 978 and 1066 MHz. The memory buffers consume less power and support low power (1.35V) DDR3 ECC DIMMs. The server can accept up to 64x32 GB Load Reduced DIMMs (LR-DIMMs), which totals up to a massive 2 TB of RAM. Not long ago, systems having that amount of storage would be considered massive. Our test system came equipped with 128 GB of RAM.

Indicators for management (troubleshooting): The front panel of the server houses the system power button, system reset button and the various status LEDs which light up orange to indicate a problem with the particular part of the system. 3 USB 2.0 ports are also present on the front panel to connect the desired I/O devices. The system is missing a front panel LCD display which is found in many of the modern day server systems.

Performance

We installed Windows Server 2008 R2 (64-bit) on the server and ran CineBench and SunGard benchmarks. The former is a financial risk analysis application, which allows you to select the number of threads to use, which essentially controls how much CPU power to extract from the system. The application uses a Monte Carlo method financial engine to determine the future value of a fictitious portfolio. We ran this test with 8, 16, 32 and 80 threads. The time taken to determine the future value of the portfolio reduces significantly as the number of threads increases. The system took 227 secs for completing the processing of the workload when 8 threads were working. This number fell sharply to around 120 secs when the full computing capability of 80 threads was utilized for the process.

We also ran CineBench on the server and compared it against another 4-way, system--A six-core Dunnington processor based server, with 24-cores running. The Westmere-EX gave a score of 19.59 CPU points against Dunnington's 18.58 CPU points. The higher score is despite the fact that the Dunnington CPUs were running at higher clock frequency of 2.66, compared to 2.4 GHz of Westmere-EX. Though the score is marginally higher, it's not very impressive considering that the Westmere-EX had 80 cores running. Interestingly, CineBench reported far lower results for a 12C/12T AMD Opteron based system.

Bottomline: Given the high number of cores and the fact that vendors can build servers with even higher number of sockets, it makes sense to go for it if you're considering server consolidation and virtualization in your data center. Plus of course, given its 32 nm design, servers based on this CPU would consume lower power as well--something that most data centers strive to achieve these days.