Kentsfield is King

We've spoken at lengths about what multiple cores in a processor can do for
you in our Technology series over the past few issues. The Kentsfield-based
QX6700, fondly called the Core 2 Extreme should have hit the markets a fortnight
ago as you read this and we got one in our Labs to put to some special torture
tests and evaluations to find out just how good it is for ourselves. And, yes we
were impressed.

Intel lent us a fresh D975XBX2 motherboard to test the quad core. The
motherboard costs Rs. 13,500. To this board, we added the 2 GB DDR2 Kingston RAM
reviewed last month and installed Win XP Pro SP2 on a SATA hard disk for our
test rig. The D975XBX2 does not offer on-board graphics, therefore, we added an
NVidia 7800 GT card that had 512 MB on-board memory. Rather than running a
couple of benchmarks and analyzing the scores (what would we compare it with for
starters?), we decided to also implement a few scenarios and see how the system
did.

Price: Rs 52,000 (3 yrs warranty) Meant For: Workstations Key Specs: Four cores at 2.66 GHz each, 4MBx2 L2 caches, support for Intel vPro, virtualization, EM64T Pros: Increases your system performance, great virtualization support Cons: None Contact: Intel Technology India, Bangalore. Tel: 25075000,E-mail: saranya.rustagi@intel.com  SMS Buy 131261 to 6677

Since the QX6700 is a supposed to have better virtualization support, we
installed the latest version of VMware Player on it and ran virtual machines. We
ran programs like POVray and CineBench for processor intensive computation and
encoded MPEG files to DivX, across our host as well as virtual machines. We hope
the results we got guide you to a better understanding of how this processor
will stand up to your expectations.

Performance extreme

We used POVray 3.7 Beta 16 with its built-in benchmark and CineBench 9.5 (which
has separate single and multi processor benchmarks) for the tests. POVray
rendered its benchmark scene at 2606.79 PPS (pixels per second), which is 673.51
PPS of CPU time. This is the first time that we have benchmarked performance
using POVray but the average published score for the Core 2 Extreme X6800 is
1430 PPS and our score for the quad core is 55% more than that dual core.

CineBench 9.5 finished its single-CPU rendering with a 432 CB-CPU units
score. Under its 'All CPUs' setting, it upped the ante to touch 1327 CB-CPU
points. Multiprocessor speed up is, thus, 3.07. In the same vein, Cinema 4D
shading was 530 CB-GFX, OpenGL hardware lighted shading stood at 4023 CB-GFX
(OpenGL speed up is, thus, 7.6).

Two VMs on our Kentsfield running PCMark 05

As you can see, when all the cores are used, the performance clearly
outstrips that of the dual cores by a large margin of 2 to 3 times. This is
expected of applications that are truly multi-threaded (like RDBMS, audio/video
render tools like Photoshop and 3DSMax). This is why we prefer to bill this quad
core as a 'workstation' class processor rather than a 'desktop' one.

Productivity story

We ran PCMark 2005 with just the system and CPU test suites enabled. These check
for multi-threading performance, application start up, file decryption and some
graphics processing. Specifically, its multi-threading tests check for file
compression, decompression, encryption and decryption along with video, audio
and image trans-coding.

The QX6700 test rig scored 7413 PCMarks overall and a CPU score of 8494. Next
came 3DMark 2005 with the graphics and the CPU suites selected. The QX6700
finished up with 8083 3DMarks and 10,802 in CPU score.

These benchmarks are not capable of fully utilizing the four cores of QX6700.
In fact, the multi-threading in these benchmarks is just sufficient to load a
dual core processor but insufficient to fully occupy the two more cores we have
with the quad core. This results in unequal load division. This is apparent when
you leave the Task Manager's performance panel on when the benchmarks execute:
you see only two of the four cores reasonably occupied at a time.

Also sometimes a core has finished its tasking but remains idle for a long
time before it gets assigned something new or another core gets loaded up. As a
result, these scores are not greater than the Core 2 Duo scores (we reviewed in
October) by a large margin.

Most desktop-class applications out there today (like this word processor)
are built around the same threading and optimization models as 3DMark and PCMark.
Therefore, with these applications you would not see a marked improvement in
performance over a dual core.

Virtualization

So, how exactly do you check virtualization and its performance? We did two
things. One, we created some virtual machines and ran them using VMware Player.
The VMs were each given 512 MB of memory. Since we had 2 GB memory for the host
system, we limited the number of simultaneously running VMs to just two. We ran
benchmarks inside them simultaneously, then we ran the benchmarks in the VMs as
well as the host machine. At all times, we also took note of relative
performance of the VMs and the host systems.

The PCMark CPU score inside the VM averaged 5028 units. This is what a single
processor P4 class system would score. 3DMark as well as PCMark's graphics
tests refused to run inside the VMs citing Direct3D disagreements with the
virtual hardware. The benchmark scores for everything except the POVray and
CineBench 'All CPU' rendering remained around the values we got without the
VMs running. The two drops in scores were reasonable, as all the four cores were
no longer available exclusively and had to be shared with the VMs. Well, this is
only half the story. Next month, we hope to continue our review of the quad
core, with further testing of virtualization, other hardware and with more
applications and new scenarios.

Preview: Quad-core Intel Xeon Processor

Clovertown is the codename for Intel's new Quad core Xeon CPUs. We managed to lay our hands on a unit early enough to do a preview of it. Clovertown's features include intelligent power capabilities, Advance Smart Cache, Wide Dynamic Execution, Intel Virtualization Technology, support for Intel 64 arch, Advance Digital Media boost and smart memory access. The Clovertown we got ad 2.66 GHz clock speed, 8 MB L2 cache, 1333 MHz FSB and consumed 120 W of power. Plus it used LGA 771 socket and not the LGA 775. For testing this, we received a server machine with a SuperMicro X7DB8 Xeon motherboard, 4 GB DDR2 and 3 SCSI 72 GB HDD with RAID 0. Performance To stress the server, we first ran NetBench and WebBench, but did not notice any performance improvement. The benchmarks were not able to utilize all the eight cores (it was a dual CPU machine). Then we decided to change the benchmark and ran 64 bit POVray as the rendering farm on this server. This way we were able to load all the cores to 100 %. Moving one step forward, we configured the server to run POVray benchmark on one core and we got 557.01 PPS (pixel per seconds). After that, we kept adding a core and re-ran the benchmark until all 8 cores were used. The results can be seen from the performance graph alongside. Each core adds 500 to 600 PPS. Net month, we'll give you a detailed review of the server.1

Value for money

This Quad is priced just a couple of thousand bucks above the last Core 2 Duo
EX6800 processor, which finds a little better performance for current
applications. But if you want to be future ready, you should spend the extra
2,000 and get quad core.

BOTTOM LINE: The four cores boost performance for applications that
know to make use of them. If you have such applications, upgrade to quad core!