Speed Tests


I wanted to test the performance of some new AMD dual core systems I am building. These systems are designed for server use and speed is important. As a good overall test of CPU, RAM, motherboard and hard drive speed I chose to compile Qt 4.1. It is a very large project with thousands of source code file that need to be read from the hard drive, compiled and then written back to the drive. This is not a good performance test for home desktop game boxes because it does nothing with the graphics system.

Compiling Qt 4.1 on Fedora Core 5 (i386 & AMD64) and Debian 3.1 (i386) with various systems. Where possible systems were booted into run level 2 to free up resources.

The script and the source.

According to the documentation for make and gmake, if you have lots of RAM you can use the -j option to set the number of threads to compile at the same time. It seems that the number of compile threads is affected more by CPU ability then RAM or hard drive speed and that a single core CPU does the same with one thread as it does with many. According to the documentation for make and gmake, if you use the -j option but do not set the number of threads it will use as much RAM as possible. This also incorrect, in fact no mater how much RAM you have using the -j option without a number will cause the make program to continue to allocate RAM long after all RAM and swap are used up. This causes the system to lockup because the running make processes no longer have enough RAM to finish the tasks they started.

You should compile with all CPU cores in use, I am finding best results when compile is done with CPU cores + 1. The Go.bsh script will automatically detect the number of CPU cores and add 1 for best results.

Update: VMware Player 4 vs. Qemu-KVM

Top Scores
  1. Intel Q6600 quad core 2.4GHz
  2. AMD Phenom II X3 710
  3. AMD Quad Core Opteron 275
  4. Intel E7300
  5. Dell XPS M1730 Intel T8300
  6. CPU Upgrade
  7. AMD Dual Core Opteron 185
  8. AMD Dual Core Opteron 2218
  9. AMD 4400+ Athalon 64 X2
  10. AMD Dual Core Opteron 275
  11. AMD Dual Core Opteron 165
  12. AMD 3700+ Athalon 64
 Boat Anchors
  1. Dell Laptop: Intel Pentium 4 2GHz
  2. My Laptop: MiTAC Intel Pentium 4 1.6GHz
  3. Intel Pentium III 550MHz
Intel Q6600 quad core 2.4GHz
System Threads / Minutes:Seconds Comments
OS: CentOS 5 i386
RAM: 2GB DDR2-800 RAM, 2.7GB swap
HDD: 2 x SATA2 Samsung 7200RPM 16MB cache T166
 5 / 10:19
These CPU's have 2MB of cache per core and sell for about $219 as of today. Motherboard is ASUS P5N-E SLI. Looks like Intel grabed a clue from somewhere. This system was built as an Asterisk phone server but because of the lack of video on board I would not use it for a server. On a positive note this system has a fanless PCI-E video card, these are getting hard to find. I have seen many servers fail because the fan on the video card died, they are small have have a low MTTF.



AMD Phenom II X3 710 Processor 2.6GHz
System Threads / Minutes:Seconds Comments
VS:CentOS 6, 64bit bare metal
RAM: 4GB RAM
 4 / 11:13 No VM and able to use all 3 cores, one more then the VMware Player limit.
VS:CentOS 6, 64bit host: Qemu-KVM
3 Cores enabled 		4 / 14:02 with 2GB RAM
4 / 14:03 with 4GB RAM
 Not using LVM. Hard drive is an image file converted from vmdk to qcow2. Host top around 299% CPU usage. This is not the CPU usage by the compiling of Qt4, it only shows that using multiple cores takes advantage of the extra cores in the host. Older VMware Players could not do this. Extra RAM made no difference.
VS:CentOS 6, 64bit host: Qemu-KVM
RAM: 2GB RAM
2 Cores enabled 		3 / 20:39 Not using LVM. Hard drive is an image file converted from vmdk to qcow2.
VS:CentOS 6, 64bit host: VMware Player 4.0
RAM: 2GB RAM
2 Cores enabled 		3 / 22:59 AMD-V set to automatic. Host top at 200% CPU usage. VMware Player will only allow 2 cores to be used.
VS:CentOS 6, 64bit host: VMware Player 4.0
RAM: 2GB RAM
1 Core enabled 		2 / 46:42 AMD-V set to automatic



AMD Quad Core Opteron 275
System Threads / Minutes:Seconds Comments
OS: Debian 3.1 i386
RAM: 8GB RAM, 3GB swap
HDD: 2 x SATA2 Western Digital 7.2kRPM SW RAID1
 5 / 16:30 Smokin' fast! This is the Tyan Thunder K8SD Pro motherboard with two Dual core Opteron 275's. This system was first tested with 2 cores AMD Quad Core Opteron 275.



Intel E7200
System Threads / Minutes:Seconds Comments
OS: Mandriva 2008
CPU: 2.53GHz
RAM: 4GB DDR2 RAM (Mandriva can only use 3.2GB)
HDD: IDE Samsung SP0802N
 3 / 17:16 3.03GHz (over clocked)
3 / 20:36 2.54GHz (after BIOS update)
3 / 29:55 1.9GHz (with buggy BIOS)
Mandriva only supports PAE for more then 3.2GB of RAM if you install the server version of Mandriva. This motherboard is from the Q6600 test above. I am using it now as my Hydra system with dual PCI-E video cards, each have dual output. I have to use Mandriva 2008.0 for this four monitor display because this was the last version of Xorg that still used Xinrama. Newer Xorg installes use a buggy Xrandr that only supports two monitors, it truly sucks to be me.



Dell XPS M1730 Intel T8300
System Threads / Minutes:Seconds Comments
OS: Ubuntu 8.10
CPU: 2.40GHz
RAM: 4GB DDR2 RAM
HDD: SATA ST9200420ASG
 3 / 20:26 (from internal SATA HDD)
3 / 25:52 (from external USB IDE HDD)
Installed Ubuntu 8.10 on to the internal SATA drive and it is much faster now.
Old test from external USB IDE HDD
This system can go faster. The Ubuntu 8.04 used for the speed test is in the USB mobile drive case just to the side of the laptop. It is very fast as it is but the interal SATA drives are way faster. The main OS on the internal drives is Vista and I am waiting for Ubuntu 8.10 to come out in a few weeks so I can format the drives. This system is great for gaming so I will keep a Vista partition for dual booting but the display is why I bought it. The LCD is 17" wide screen at 1920x1200. I LOVE IT!!!



CPU Upgrade
System Threads / Minutes:Seconds Comments
OS: CentOS 6.2 x86_64
RAM: 8GB RAM
 Dual-Core E5200 2.50GHz
3 / 20:38 		This is a bare metal test before I upgrade to the the CPU to a Quad core Xeon.




AMD Dual Core Opteron 185
System Threads / Minutes:Seconds Comments
OS: Debian 3.1 i386
CPU: 2.66GHz
RAM: 2GB RAM
HDD: Old IDE Western Digital 10GB WD100BA
HDD: New SATA Hitachi 500GB HDP725050GLA360
 With IDE drive
3 / 24:29


With SATA drive
3 / 24:28 		I wanted to test the difference a new hard drive would make on the performance of the speed test. It looks like it made no difference. The CPU's are still pinned at 100% usage so that seems to be the bottleneck.
Western Digital WD100BA
  • Formatted Capacity: 10.0GB
  • Interface: EIDE
  • Data Transfer Rates: 66.6MB/s (UATA)
  • Physical Pins of Data Connector: 40pins
  • Average Access Times: 9.5ms
  • Buffer Size: 2MB
  • Rotational Speed: 7200rpm
Hitachi Deskstar P7K500
  • Formatted Capacity: 465GB
  • Interface: SATA-300
  • Seek Time: 8.5ms
  • Buffer Size: 16MB
  • Rotational Speed: 7200rpm
Same system with SATA drive over clocked
CPU: 2.76GHz		 3 / 23:33 I was hoping for better. RAM is about 1.5GB used on this test.
Same system with SATA drive over clocked
CPU: 2.80GHz		 3 / 22:56 I tried to overclock to 2.86GHz but the value RAM could not keep up and the system died. Looking for better RAM.
Another Opteron 185 system
CPU: 2.873GHz		 3 / 22:51
Not much better then running at 2.80GHz



AMD Dual Core Opteron 2218
System Threads / Minutes:Seconds Comments
OS: Debian 3.1 i386
CPU: 2.60GHz
RAM: 2GB ECC RAM
HDD: Old IDE Samsung 20GB SV2042H
 3 / 24:49
This system needs more RAM.



AMD 4400+ Athalon 64 X2
System Threads / Minutes:Seconds Comments
OS: Fedora Core 5 i386
RAM: 4GB RAM, 6GB swap
HDD: SATA Raptor 10kRPM 4.5ms seek
 1 / 47:18
2 / 24:54
5 / 25:04
20 / 25:08
40 / 25:06 		The number of CPU cores seems to be the limiting factor here.
OS: Fedora Core 5 i386
RAM: 2GB RAM, 1GB swap
HDD: SATA Barracuda 7.2kRPM 8ms seek
 2 / 24:26
 As with Opteron 165 Raptor vs. Barracuda test below these number indicate very little performance difference between them. I think the reason for this is that the source files are very small and both the code and the compiler are highly optimized to make as little use of the hard drive as possible while compiling.
OS: Fedora Core 5 AMD64
RAM: 4GB RAM, 6GB swap
HDD: SATA Raptor 10kRPM 4.5ms seek
 1 / 48:06
2 / 25:44
20 / 26:09 		The 64 bit OS is just a little slower then the 32 bit. Does the 64 bit OS need to compile more code then the 32?



AMD Dual Core Opteron 275
System Threads / Minutes:Seconds Comments
OS: Debian 3.1 i386
RAM: 4GB RAM, 500MB swap
HDD: SATA Barracuda 7.2kRPM 8ms seek
 5 / 28:44 This system will soon be upgraded to a 2x2 core with 8 GB RAM. I cannot wait to see the numbers it gives then. This system was retested with 4 cores AMD Quad Core Opteron 275.



AMD Dual Core Opteron 165
System Threads / Minutes:Seconds Comments
OS: Fedora Core 5 AMD64
RAM: 1GB RAM
HDD: SATA Raptor 10kRPM 4.5ms seek
 2 / 29:19  



AMD Dual Core Opteron 165
System Threads / Minutes:Seconds Comments
OS: Fedora Core 5 AMD64
RAM: 1GB RAM
HDD: SATA Barracuda 8.5ms seek
 2 / 29:28 I expected a bigger performance gap between the Raptor and the Barracuda.




AMD 3700+ Athalon 64
System Threads / Minutes:Seconds Comments
OS: Fedora Core 5 i386
RAM: 1GB RAM, 6GB swap
HDD: IDE Barracuda 7.2kRPM 8ms seek
 1 / 74:28
2 / 74:15
5 / 74:26
20 / 287:19
40 failed to complete 		The 20 threads is using swap for RAM. And the 40 threads runs out of swap, ran out of resources and crashed the system.



Dell Laptop: Intel Pentium 4 2GHz
System Threads / Minutes:Seconds Comments
OS: Fedora Core 5 i386
RAM: 512MB RAM, 100MB swap
HDD:
2 / 95:58 Informaiton in cat /proc/cpuinfo calls it a 2GHz but reports the CPU speed as 1200MHz



My Laptop: MiTAC Intel Pentium 4 1.6GHz
System Threads / Minutes:Seconds Comments
OS: Debian 3.1 i386
RAM: 512MB RAM, 200MB swap
HDD: IDE Toshiba 5.4kRPM 12ms seek
 1 / 347:47 My laptop sucks. Dude, I should have got a Dell. Bummer.



Intel Pentium III 550MHz
System Threads / Minutes:Seconds Comments
OS: Fedora Core 5 i386
RAM: 128MB RAM, 400MB swap
HDD: IDE Fujitsu 5.4kRPM 10ms seek
 1 / 362:56 Look at how my P4 laptop is just about as slow as a P3 desktop.