AMD’s Phenom II X6 has just been released, bringing six-core processing power within reach of students and others without thousands of dollars to spend on Intel’s Core i7 9xx series.  But how well does the X6 do with multithreaded open source development?  This short review looks into Linux development using GCC and make with AMD’s new processor.

—

Professional tools for (pseudo-)professional developers: The AMD Phenom II X6

Lately this blog has been dead, but only because I’ve delved deep into programming and the creation of a new Linux distribution based, in true WDR? fashion, on aesthetic principles that I’ve gained through my historical and philosophical studies.  However, one thing that was holding me back was my lack of a powerful development machine.  It’s incredible to think that for the past year I’ve done paid development on a netbook whose wiring is so flimsy that the video card can be fixed (or broken) by repeatedly dropping the machine from a height of two feet.  The time was right to invest in a desktop machine, especially now that I’ll be working for AbiWord in Google’s Summer of Code 2010 competition implementing non-linear page flows.

Enter the Phenom II X6, AMD’s latest and greatest CPU.  My friend who actually works in IT and I had agreed until just a couple of days ago that the Intel Core i5 750 would be the most cost effective option, but once the X6 was released its six cores won me over.  Numerous reviews have already shown that under non-multithreaded loads, AMD’s new processor is a loser, but as soon as the threads can utilize all six or even four cores, AMD trounces the competition.  As anyone who does Linux development knows, going multithreaded is usually as simple as typing “make -jN”, where N is the number of parallel jobs that you want to execute while building your app (normally N = number of cores/processors + 1).  At $200/$300 per processor, that would make the X6 a bargain-basement priced high-power workstation (Intel’s current six-core offering, the Core i7 9xx series, is actually faster than the X6 but also costs +$1000, out of the reach of mortals and students like myself).

Specs

Thus, I headed to MicroCenter and dropped ~$730 (including tax) buying parts to build a Phenom II X6-based system.  My final system consists of:

• CPU: AMD Phenom II X6 1090T Black Edition ($300)
• Motherboard: Gigabyte GA-880GM-UD2H, 880G chipset, mATX form factor ($50, after CPU bundle discount)
• Power supply: Enermax Tomahawk 500W ($50)
• CPU cooler: Zalman CNPS7500-Cu with blue LED, all copper radial heatsink ($40)
• Case: Cooler Master Elite 341, mATX mid-tower ($30)
• Memory: Kingston HyperX 4GB DDR3-1600 ($130)
• Video: AMD Radeon HD4250 (integrated, included with the motherboard)
• Hard drive: Western Digital Caviar Blue WD5000AAKS 500 GB 7,200 RPM SATA ($50)
• CD/DVD: Some random Toshiba dual layer Lightscribe DVD+/- RW ($35)
• Wireless: Crappy 802.11B/G card using Realtek 8139 chipset ($15)
• OS: Ubuntu 10.04 “Lucid Lynx” AMD64 edition (free and libre, though following the window button debacle perhaps not quite so free as in free speech)
• Video drivers: AMD’s proprietary fglrx XOrg drivers, v. 8.723.1, downloaded from Ubuntu
  

Yes, I did go ahead and splurge on some of the components.  Cheaper RAM, a cheaper power supply, the stock CPU cooler, and perhaps even a less expensive case could have been found.  In all, I could have probably shaved about $100 off without degrading performance, but since I value longevity and quality in my components I decided to spend extra.  There’s also a Phenom II X6 1055T model that has slower core speeds and that costs $100 less which may interest some readers.  I will probably try overclocking a little, especially since the 1090T is a Black Edition part and has an unlocked clock multiplier, but as the benchmark results below show there really isn’t any need.

Benchmarks (compiling)

As I’ve mentioned, this is supposed to be a development system to help me with my GSoC and Linux distribution work, so I thought that I’d run some compiles to see how fast they would finish.  For all these compiles, I have Firefox with 10 tabs running as I’m typing into Gchat and this WordPress blog, which represents a realistic compile environment for me.

First, my compiler’s information:

ersin@istanbul:~$ gcc -v
Using built-in specs.
Target: x86_64-linux-gnu
Configured with: ../src/configure -v --with-pkgversion='Ubuntu 4.4.3-4ubuntu5' \
  --with-bugurl=file:///usr/share/doc/gcc-4.4/README.Bugs --enable-languages=c,c++,fortran,objc,obj-c++ \
  --prefix=/usr --enable-shared --enable-multiarch --enable-linker-build-id --with-system-zlib \
  --libexecdir=/usr/lib --without-included-gettext --enable-threads=posix \
  --with-gxx-include-dir=/usr/include/c++/4.4 --program-suffix=-4.4 --enable-nls \
  --enable-clocale=gnu --enable-libstdcxx-debug --enable-plugin --enable-objc-gc --disable-werror \
  --with-arch-32=i486 --with-tune=generic --enable-checking=release --build=x86_64-linux-gnu \
  --host=x86_64-linux-gnu --target=x86_64-linux-gnu
Thread model: posix
gcc version 4.4.3 (Ubuntu 4.4.3-4ubuntu5)
ersin@istanbul:~$

Next, the actual tests:

AbiWord (using SVN r28860 from trunk after having run ./autogen.sh; configured with default settings, only plugins were opendocument and msword; no ccache):

ersin@istanbul:~/abiword$ time make -j7
...
real    1m45.630s
user    6m33.760s
sys    0m50.130s
ersin@istanbul:~/abiword$

Blazingly fast, I’d say!  My single-core Athlon 64 L110 netbook takes 30-45 minutes to do the same compile.

buildroot toolchain (2010.02 after having configured with make menuconfig and downloaded sources with make source; no target packages selected, only the toolchain with all compilers set to be built, including target utils and full GCC suite — gcc, g++, gcj, gcc-objc, and gfortran; 7 parallel jobs selected through make menuconfig; neither gdb nor tests built; no ccache):

ersin@istanbul:~/vorpo/build/buildroot-2010.02$ time make
...
real    19m15.274s
user    42m33.760s
sys    11m39.160s
ersin@istanbul:~/vorpo/build/buildroot-2010.02$

This is perhaps the more impressive of the two benchmarks.  An entire toolchain in under 20 minutes. In a $730 mATX system that I put together in my kitchen in an hour.  Ponder that for a second.  This same compile easily takes me 3+ hours on my netbook.  Throughout this compile, I checked up on my computer’s memory usage with GNOME’s System Monitor, and at no time did it exceed 1.5 GB memory, which includes the desktop and all the rest of the system.  CPU utilization on all six cores was 100% while compiling and fluctuated while running configure.

How about one more test for completeness’s sake?  Here’s GIMP, a complex GTK+ program with many dependencies and a good benchmarking test:

GIMP (2.6.8, after having run ./configure; no ccache):

Building GIMP with prefix=/usr/local, datarootdir=${prefix}/share
Desktop files install into ${datarootdir}

Extra Binaries:
 gimp-console:        yes
 gimp-remote:         no (not enabled)

Optional Features:
 D-Bus service:       no
 Language selection:  yes

Optional Plug-Ins:
 Ascii Art:           no (AA library not found)
 Help Browser:        no (WebKit not found)
 LCMS:                no (lcms not found or unusable)
 JPEG:                yes
 MNG:                 no (MNG header file not found)
 PDF:                 Using PostScript plug-in (libpoppler not found)
 PNG:                 yes
 Print:               yes
 PSP:                 yes
 Python:              yes
 Script-Fu:           yes
 SVG:                 yes
 TIFF:                yes
 TWAIN (MacOS X):     no
 TWAIN (Win32):       no
 URI:                 yes (using GIO/GVfs)
 Windows ICO          yes
 WMF:                 yes
 XJT:                 yes
 XPM:                 no (XPM library not found)

Plug-In Features:
 EXIF support:        no (libexif not found or too old)
 GNOME UI:            no (libgnomeui-2.0 not found)
 GNOME keyring:       no (gnome-keyring-1 not found)

Optional Modules:
 ALSA (MIDI Input):   no (libasound not found or unusable)
 Linux Input:         yes (HAL support: no)
 DirectInput (Win32): no
 Color Correction:    no (lcms not found or unusable)
 Soft Proof:          no (lcms not found or unusable)

ersin@istanbul:~/vorpo/build/gimp-2.6.8$ time make -j7
...
real    2m59.657s
user    9m7.560s
sys    2m24.910s
ersin@istanbul:~/vorpo/build/gimp-2.6.8$

The numbers speak for themselves.  Best yet, the finished binary runs flawlessly after make install.

Benchmarks (graphics)

Since this is not a gaming rig, I couldn’t care less about FPS.  But what’s a review without some graphics test?  Here’s fgl_glxgears (AMD’s proprietary glxgears):

ersin@istanbul:~/vorpo/build/buildroot-2010.02$ fgl_glxgears
Using GLX_SGIX_pbuffer
2221 frames in 5.0 seconds = 444.200 FPS
2466 frames in 5.0 seconds = 493.200 FPS
2779 frames in 5.0 seconds = 555.800 FPS
2943 frames in 5.0 seconds = 588.600 FPS
2950 frames in 5.0 seconds = 590.000 FPS
2946 frames in 5.0 seconds = 589.200 FPS
2943 frames in 5.0 seconds = 588.600 FPS
2912 frames in 5.0 seconds = 582.400 FPS
^C
ersin@istanbul:~/vorpo/build/buildroot-2010.02$

I downloaded Nexuiz 2.5.2 and played the first level (DM1) with the FPS meter on.  I set all the quality features in the ATI Catalyst Control Center to full (4x AA, “quality” adaptive AA, 16x anti-ISO, “quality” mipmap) and set Nexuiz’s resolution to 1280×1024 and its “Effects” to Normal, Good, and Ultra.  At Good and Ultra, running around empty halls got around 20-35 FPS while fighting one bad guy was around 20 FPS, dipping down occasionally to 11 FPS.  With only one baddie, this was playable, but online in a deathmatch it would be unacceptable.  Normal quality was perfectly acceptable: empty halls were 40-60 FPS and fighting one bad guy was around 30 FPS.  I didn’t bother turning the ATI Catalyst quality settings down at all, and I’m sure that the speed would increase significantly.

I also did a test run of Quake Live (Q3 running inside your browser) and everything worked flawlessly, except when I was compiling and the screen started to flicker.  Lesson: don’t play 3D games while compiling (duh).

Noise

Noise was a secondary concern for me, since I don’t intend to have this machine on 24/7, and subjectively on a scale of 1-10 with 1 being my fanless Nokia N800 internet tablet and 10 being my mom’s overclocked Pentium 4 Prescott machine with a Thermaltake Volcano 5, I rank my machine at 6 when idling and 7 at full load.  It’s clearly audible from the other room, nearly 10 meters away.  When I get up close to the computer, it’s unclear whether the bulk of the noise is coming from the case’s stock system fan or the Zalman CNPS7500-Cu (I think that the Tomahawk is fairly silent).

A word on the Zalman cooler.  The CNPS7500-Cu comes with the Zalman Fanmate, which allows for manually adjusting the fan’s speed.  According to its review on FrostyTech, the CNPS7500-Cu when tuned down with the Fanmate to its minimum speed comes in at 32.5 dB, which is practically inaudible and nearly the quietest CPU cooler on their list as of July 2007 when the review was published.  However, I’ve been running the fan at full speed, which is somewhere around the middle of the pack at 50.9 dB, still quieter than the ear-splitting stock fan that comes with AMD’s older AM2 socket processors (the AVC Z7U7414001, 63.0 dB).  AMD seems to have changed the stock fan on the Phenom II to the AVC Z7UH40Q001, which I have not tested and do not intend to test now that my CPU is set up.  My primary concern is not noise, but rather that my CPU stays as cool as possible to maximize its life; that said, the Zalman CNPS7500-Cu should be a fine choice if you want to run its fan at a lower speed.

Why choose the CNPS7500-Cu, then, if I wanted to buy an aftermarket cooler?  I was originally going to buy the OCZ Vendetta, which has been on sale at MicroCenter for $15 and is very highly regarded, but since I’ll be moving soon I wanted to build a small mATX system that wouldn’t be able to fit the Vendetta’s enormous heatsink.  Rather, the Vendetta would fit (barely), but Zalman’s radial designs are more horizontal and allow for better airflow in such a small case as the Elite 341, not to mention that they keep a lower center of gravity and that I was terrified that the Vendetta would snap off (I’ve read on Newegg that it even needs special mounts for some boards).  On a purely subjective level, I enjoy that the heatsink is all-copper and feels almost as heavy as a bowling ball in my hands.

Conclusion

The Phenom II X6 is wicked fast for compiling on Linux, and GCC/make does a phenomenal job (no pun/s intended) at splitting builds into parallel tasks to leverage all six of the X6′s cores.  AbiWord, a modern full-fledged word processor, compiled in less than 2 minutes and buildroot, a system for creating cross-compiling toolchains, finished in less than 20 with the full GCC suite.  Moreover, at $200/$300, the X6 has no competitor in its price range, the nearest being Intel’s Core i7 9xx series, which costs +$1000.  I highly recommend the X6 for any open source Linux work that uses GCC and make.

Aftermath: what review is complete without photos?

The AMD Phenom II X6 1090T reviewed (for Linux developers) by Ersin Akinci, unless otherwise expressly stated, is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 United States License.