Core i7 and Xeon 5500 CPU Benchmarks
Original Article Date: 2009-04-30
Intel's new Xeon 5500 series "Nehalem" processors were released about a month ago
- the most significant new technology release in the server and workstation market
in the last
four years. You can
read my article on the Nehalem release here.
Five weeks on, I have integrated a number of systems using the new technology, and
performed benchmark suites on these systems using Everest from Lavalys.
In comparing these benchmarks to archived results made on previous systems, I now have the first indication of performance gain available with this new technology.
Let's start with the most significant architectural change with the new Nehalem
technology - memory. As I expected, the new chips lived up to the promise of dramatically
improved memory bandwidth. Take a look at the chart below, which shows composite
scores for memory read, write and copy performance.
This chart indicates that the new Nehalem CPUs significantly outperform all previous
models, both AMD and Intel. The performance gain from like-for-like Intel technology
is 100%! For example, the single-processor Core i7 series has double
the memory performance of the previous Core 2 series, whilst the 5500 series CPUs
have double the memory performance of the 5400 series. There appears to be no competition!
Also note the significant jump in bandwidth between the "X" and "E" versions of
the 5500-series Xeons. The "X" (I'm guessing for "eXtreme") models support 1333MHz
DDR3 memory speed, whilst the mainstream "E" models support only 1066MHz DDR3. The
memory bandwidth between these series is seen to scale almost linearly with
RAM speed. So choice of CPU is significant when considering a system
that heavily relies upon system RAM.
This is very significant for the workstation market specifically, as applications
in this sector, such as scientific and engineering applications,
and special types of image-manipulation software,
are usually very memory-intensive. Expect BIG performance improvements in going
to Nehalem if you're currently using a workstation for this type of work.
CPU Integer Performance
This an area that Intel has done well in traditionally. The chart below shows composite
scores from four Everest CPU Integer Benchmarks (Queen, ZLib, Photoworxx, AES).
The chart shows very strong performance from both the new single (Core i7) and dual-processor
(5500-series Xeon) CPUs, improving clock-for-clock on the previous generation Core 2 and
Xeon 5400 series respectively, by about 30%. Note also how well
the Nehalem dual-processors scale over Nehalem single processor - near double the
performance. This is expected, since the Intel Quick-Path Interconnect (QPI), which
is present in both the Core i7 and Xeon 5500s, was designed from the beginning to
work in multi-processor configurations.
Integer operations are significant for all types of server applications,
including databasing, and video editing, compositing and special effects
CPU Floating Point Performance
Lastly, let's see how well the new CPUs perform on floating point. I'm a little
skeptical here, as the floating point benchmarks varied wildly from CPU to CPU, so a composite graph (as shown below), might not tell the actual story. In short,
floating-point performance will depend on your application. Nonetheless, a general
guide to capability can be gleaned from these results. The graph below is a composite
of the Everest Floating-Point Fractal Benchmarks (Julia, Mandel, SinJulia).
The chart shows the Nehalems doing well again, improving on their previous generations
by 25-30%, and also shows excellent scaling of the Nehalem technology, with the
5500-series 2P Xeon's performance being double that of the Core i7.
Floating-point performance is important in workstations or compute clusters, where any application is
modelling real-world phenomena (such as engineering analysis or
typically any custom scientific application), as well as 3D-Modelling utilities.
I know benchmarks aren't everything, and can often be a distortion of real-world
performance. But they do hint at overall trends in power, particular if you're comparing
a new CPU to its previous generation on the same set of benchmarks.
The above charts clearly show the trend of increased performance over previous Intel
CPU generations, especially on memory bandwidth and speed.
The data also suggests that Intel is signficantly ahead of AMD, even
for price. But
it's mentioned often on discussion forums that Intel usually outperforms
AMD on benchmarks, but in real-world applications, AMD performs better. Again, benchmarks
aren't gospel, but they offer clues.
The one thing that is clear from this set of results is that if you're using an
application that is very memory intensive, you need to definitely
consider upgrading to Nehalem now if you want to save on time or improve productivity,
since in either dual or single processor configurations, the performance improvements
are stellar indeed!
Chief Systems Engineer