System Performance Comparisons, 2005
(Written by Sean Reifschneider, tummy.com, ltd., November 27, 2005)
I've been testing out a variety of hardware over the last several
weeks. This includes a dual Opteron 2GHz, dual Xeon 3.2GHz, Dual-core
Pentium D 3.0GHz, and a single P4 3GHz. There are plenty of places that
run systems through various benchmarks. You find a lot of speculation
about the power consumption and head on the Opterons. However, I was
surprised by the amount of heat put out by the Opteron system as a whole.
So, I decided to take some measurements of these systems to see how
they compared from the power and heat standpoint.
The measurements were taken using an inductive power meter on the
input to the power-supply. Temperature values were taken using a
temperature probe hooked up to my multi-meter, testing at the main air
outlet on the system, the power-supply fan output, and the directly at the
CPU heat-sink, in the center of the down-wind side of the air-flow.
I've also run a simple benchmark on system performance to get a rough
idea of how the system performs. For this I downloaded the 2.6.14.2 kernel
source, did "make oldconfig" and used the default answer to all questions,
and then did "time make -j 5 modules bzImage".
Performance Benchmark
| Processor | Elapsed time |
|---|
| Dual Xeon 3.2GHz | 8m43.371s (Fastest) |
| Dual Opteron 246 (2.0GHz) | 9m0.341s |
| Pentium D 3.0GHz | 12m9.387 |
| Pentium 4 3.0GHz | 22m48.522s (Slowest) |
Power Consumption
| Processor | At Idle | 100% Utilized |
|---|
| Dual Xeon 3.2GHz | 1.4 Amps (Highest) | 2.4 Amps
(Highest) |
| Dual Opteron 246 (2.0GHz) | 1.2 Amps | 2.1 Amps |
| Pentium D 3.0GHz | 0.9 Amps | 1.5 Amps |
| Pentium 4 3.0GHz | 0.7 Amps (Lowest) | 1.2 Amps
(Lowest) |
Temperature at Idle
| Processor | System | Power-Supply | Heatsink |
|---|
| Dual Xeon 3.2GHz | 95 degrees F (Highest) | 101 degrees
F | 94 degrees F |
| Dual Opteron 246 (2.0GHz) | 81 degrees F | 107
degrees F (Highest) | Unknown |
| Pentium D 3.0GHz | Unknown | 99 degrees F | 86
degrees F (Lowest) |
| Pentium 4 3.0GHz | 80 degrees F (Lowest) | 90 degrees
F (Lowest) | 100 degrees F (Highest) |
Temperature at 100%
| Processor | System | Power-Supply | Heatsink |
|---|
| Dual Xeon 3.2GHz | 110 degrees F (Highest) | 106 degrees
F | 117 degrees F (Highest) |
| Dual Opteron 246 (2.0GHz) | 97 degrees F | 119
degrees F (Highest) | Unknown |
| Pentium D 3.0GHz | 88 degrees F | 99 degrees
F | 104 degrees F (Lowest) |
| Pentium 4 3.0GHz | 86 degrees F (Lowest) | 90 degrees
F (Lowest) | 115 degrees F |
Observations
People have recently been speaking about the Intel dual-core CPUs
using excessive amounts of power. However, compared to the dual-CPU
systems, the Pentium D is quite reasonable. It will be interesting to
see how it compares with the dual-core Xeons.
Despite what everyone says about the performance of Opterons and their
low power-consumption, they performed similarly and were within 15% on
power consumption. While the Opteron system was cool as far as the system
goes, the power-supply generated vast amounts of heat.
While the cost of the Opteron CPUs may be slightly less than the Xeon,
but the cost of the whole systems seem to be slightly the opposite. The
big issue with the Opterons is limited selection of hardware options. The
Xeon hardware is quite prevalent, with many choices available.
The Pentium D has a lot of bang for the buck. The dual-CPU systems
give better than 2x performance of the single CPU, with solid 2x more
power consumption and therefore a likely doubling in the power dissipation.
The Pentium D is nearly double the performance but with only 20% more power
consumption.
Pretty much all of the current crop of CPUs are 64-bit capable. While
the Opteron has a 32-bit mode for running software, I found CentOS 4.2
had serious driver issues when running in 32-bit mode on the Opterons. I
also had driver issues with the Opteron system in 64-bit mode as well.
Fedora Core 4 seemed to do a bit better with the hardware, probably because
of newer driver versions.
The Intel-based systems, with EM64T, all seem to run with no problems
on CentOS 4.2, in 32-bit mode.
