Xeon 5500 Series, Nehalem 2.9GHz Processor

The Nehalem based Xeon processor -Xeon 5500 Series is the first processor

from Intel to have Native Quad Core support. We received a 1U rack server from

Intel with two 2.9 GHz Nehalem processors and 24GB RAM. We ran quite a few tests

on it, and as expected, the results were mind boggling. We will discuss the

results later. First, let's take a quick look at some of the key features of the

processor. The basic features are pretty much similar to the Core i7 desktop

architecture and offer the following functionalities:

Native quad-core



Till now, all previous Xeon processors with more than two cores were built

using multi-chip modules of dual core processors. So they were essentially two

or three dual cores modules fixed in one chip creating quad or hex core CPUs.

With Xeon 5500 series, there is native quad core design. This is similar to

AMD's Phenom X4 CPUs. The same feature is also available with Intel's Nehalem

based desktop processors, the 'Core i7'.

The advantages of having a native quad-core over an MCM (Multi core modules)

are significant in terms of processor energy efficiency, performance, and

dynamic scalability. We will see some of these in our benchmark results.

Inclusive level 3 cache



First showcased on earlier Xeon server chips and then on desktop Core i7

CPUs, the Xeon 5500 family of CPUs feature up to a massive 8MB of level 3 cache

(shared between all four cores) as compared to 2 MB of Phenom X4. The cache is

also described as an inclusive level 3 cache. Intel claims, an inclusive cache

is more efficient than an 'exclusive' cache design, even if it does mean that

1MB of Nehalem's 8MB Level 3 cache is taken up by storing a copy of the 256 KB

Level 2 cache inside each processing core.

Integrated memory controller



By modularizing the design of the CPU and the Northbridge, the memory

controller has been brought to the Nehalem CPU die. The separate processing

cores and caches are linked to the on board memory controller via a new bus

standard called the QuickPath interconnect, replacing the conventional front

side bus. As QuickPath replaces the Front side Bus (FSB), it also takes over the

role of allowing the CPU to connect to other system components, buses and

controllers such as the PCI Express controller and DDR3 memory, reducing latency

and improving performance considerably.

This shows time taken by different problem sizes on 2 socket Nehalem.

Hyper-threading



Another feature worth mentioning is Hyper-threading. Using spare resources

of a core to execute a second process thread, Hyper-threading enables a

quad-core Nehalem processor to accept and process eight threads simultaneously,

making it even more massively parallel and powerful than the current Core 2 Quad

CPUs.

Performance results



We ran three benchmarks on the server -LINPACK, SunGard and Cinebench. Plus

we also recorded its power consumption in different levels. For running all

these benchmarks, we used Windows Server 2008 as the OS. Here is what all we

got.

A dual socket Nehalem server is able to show 16 processors due to its Hyper-Threading capabilities.

Linpack



The test was really exciting with some really interesting results.

Undoubtedly, this gave the best result when compared with Intel's Harpertown or

Dunnington processors. But surprisingly, it even gave better performance than

Dunnington with 24 Cores, even though it only had 8 Cores and 16 threads. The

final result we got was a whopping 76 GFlops, which was 14 Gflops more than the

24Core (6Core * 4 Socket) Dunnington. This result was achieved with a problem

size of 50000 and 16 threads in Linpack.

Another interesting observation is that, due to Hyper Threading, the

processor was really getting an edge over its predecessor. When we ran the same

problem on 8 Threads, which was equal to its actual number of cores, it gave

much lower performance.

SunGuard



We used SunGard Adaptive Analytics as a component of SunGard's Suite of risk

management products. More precisely, it is the stripped down version of the

actual product. This benchmark utilizes Monte Carlo method financial engine to

predict the future of a fictitious portfolio. It requires two different files to

run. The first one contains sample data that represents the actual market

condition and the second file contains the sample customer's investment

portfolio. The benchmark scores are calculated on the base of time in seconds.

So lesser the time it will take to run, the better thel performance. In this

test, our server was able to finish the task in 130.5 seconds. If we compare it

against Dunnington, with 3 times the number of cores, Dunnington was able to

finish the same test in 105.9 seconds which is just 20% faster than Nehalem. If

we multiply both results with the number of cores available in each server, we

get 1044 for Nehalem and 2541 for Dunnington. If we see per core performance of

both servers, Nehalem gave 2.4 times better performance than Dunnington. This is

indeed a brilliant score.

CineBench



And finally we ran CineBench 10 x64. This benchmark measures the performance

of processor and graphics card. This test consists of two parts: first is

processor intensive and second is graphics intensive. Initially it makes use of

a single CPU for running the test whereas the latter part uses all cores. In the

graphics test, the test runs inside a 3D window. An animated scene is played

starting with a low demand for graphics which is increased later. Finally a

score is generated, when the processor works on maximum speed for the scene to

be displayed properly. The higher the scores the better the server performance.

The score we got for a single CPU was 4429, which was again 30% better than

Dunnington, which gave around 3266 CBCPU with one CPU. With all CPUs, the score

of Nehalem was 28667 CB-CPU.

Bottomline: If you are planning to scale your datacenter and want to buy

servers which can cope up with mission critical virtualization and parallel

processing, then no doubt this architecture is for you.