• Home
  • blog
  • AMD Phenom II X4 955 Black Edition for AM3

AMD Phenom II X4 955 Black Edition for AM3

The Phenom II processor was a step in the right direction for AMD. It generated more performance per clock and allowed for higher frequency chips. Priced aggressively, the Phenom II X4 920 and 940 Black Edition were price-competitive alternatives for Intel’s Core 2 line. AMD had, in essence, caught up. The X4 920 and 940 however are AM2+ processors, and AMD is currently transitioning to the AM3 socket with DDR3 memory support. Today AMD unveils the newest AM3 members of the Phenom II family: The X4 945 matches the X4 920’s 3.0 GHz frequency while the X4 955 Black Edition becomes AMD’s new flagship CPU, clocked at 3.2 GHz.

Apr. 23, 2009 by Lawrence Lee

Product AMD Phenom II X4 955 Black Edition AM3 Processor
Manufacturer AMD
Volume Price US$245

The release of the Phenom II series of processors was a step in the right direction
for AMD. The new design generated more performance per clock and allowed for
higher frequency chips. Priced aggressively, the Phenom II X4 920 and 940 Black
Edition were economically viable alternatives for Intel’s Core 2 line. It wasn’t
a knockout blow, but AMD had in essence, caught up. The X4 920 and 940 however
are AM2+ processors, and AMD is currently in the middle of a transitioning to
the new AM3 socket with DDR3 memory support. Up to this point their AM3 processors
have lagged behind the X4 940’s 3.0 GHz clock speed.

Today AMD unveils the newest AM3 members of the Phenom II family. The X4 945
matches the X4 920’s 3.0 GHz frequency while the X4 955 Black Edition (with
unlocked multiplier) becomes AMD’s new flagship CPU, clocked at 3.2 GHz.


Our X4 955 Black Edition sample.

The new Phenoms have the same amount of cache and power rating as their AM2+
counterparts, though they are clocked much higher than AMD’s current AM3 lineup.
The X4 955 BE and 945 supplant the X4 940’s top position, and in anticipation
of this release, AMD has dropped the 940’s price making it an extremely attractive
option. The aggressive pricing makes it clear as day that AMD wants to battle
Intel on value.

Model
Freq.
Total L2 Cache
L3 Cache
Socket
TDP
Best Retail
Mfg. Price
Intel Q9650
3.00 GHz
12MB
N/A
LGA775
95W
$322
$316
Intel Q9550
2.83 GHz
12MB
N/A
LGA775
95W
$269
$266
X4 955 BE
3.20 GHz
2MB
6MB
AM3
125W
TBD
$245
X4 945
3.00 GHz
2MB
6MB
AM3
125W
TBD
$225
Intel Q9400
2.66 GHz
6MB
N/A
LGA775
95W
$220
$213
X4 940 BE
3.00 GHz
2MB
6MB
AM2+
125W
$169
$195
X4 920
2.80 GHz
2MB
6MB
AM2+
125W
$169
$195
X4 810
2.60 GHz
2MB
4MB
AM3
95W
$168
$175
X3 720 BE
2.80 GHz
1.5MB
6MB
AM3
95W
$134
$145
X3 710
2.60 GHz
1.5MB
6MB
AM3
95W
$119
$125

The new Phenom II may not be more power efficient, as they retain
the original Phenom II’s thermal design power rating of 125W. This higher power
draw will translate into higher electricity bills, possibly enough to offset
any savings gained by taking advantage of AMD’s tempting prices depending on
the amount of usage. The X4 955 BE’s direct competitor is the 95W Intel Core
2 Quad Q9550, a 2.83 GHz chip priced about $20 higher.

TEST METHODOLOGY

Common Test Platform:

Intel CPU/Motherboard:

AMD CPU/Motherboard:


CPU-Z screenshot: Phenom II X4 955 BE during load.

 


CPU-Z screenshot: Phenom II X4 955 BE idling with C&Q enabled.

Measurement and Analysis Tools

  • CPU-Z
    to monitor CPU frequency and voltage.
  • CPUBurn
    K7

    processor stress software.
  • Prime95
    processor stress software.
  • FurMark
    stability test to stress the integrated GPU.
  • Cyberlink
    PowerDVD
    to play H.264/VC-1/Blu-ray video.
  • SpeedFan
    to monitor temperature and fan speeds.
  • Eset NOD32 as
    an anti-virus benchmark.
  • WinRAR as an
    archiving benchmark.
  • iTunes
    an audio encoding benchmark.
  • TMPGEnc
    Xpress
    as a video encoding benchmark.
  • PCMark05
    as a general system benchmark.
  • 3DMark06
    as a 3D benchmark.
  • Seasonic
    Power Angel
    AC power meter, used to measure the power consumption
    of the system.
  • Custom-built, four-channel variable DC power supply, used to regulate
    the CPU fan speed.

Benchmark Test Details

  • Eset NOD32: In-depth virus scan of a folder containing 32 files of
    varying size with many of them being file RAR and ZIP archives.
  • WinRAR: Archive creation with a folder containing 68 files of varying
    size (less than 50MB).
  • iTunes: Conversion of an MP3 file to AAC.
  • TMPGEnc Xpress: Encoding a 1-minute long XVID AVI file to VC-1 (1280×720,
    30fps, 20mbps).

Our testing procedure is designed to determine the overall system power consumption
at various states (measured using a Seasonic Power Angel). To stress Intel CPUs
we use Prime95 (large FFTs setting) to maximize heat and power consumption.
For AMD CPUs we use CPUBurn K7 as it seems to tax AMD processors more. To stress
the IGP, we use FurMark, an OpenGL benchmarking and stability testing utility.
We also performed a short series of benchmarks featuring real-world timed tests
and synthetics.

Cool’n’Quiet and/or Intel SpeedStep were enabled (unless otherwise noted).
The following features/services were disabled during testing to prevent spikes
in CPU/HDD usage that are typical of fresh Vista installations:

  • Windows Sidebar
  • Indexing
  • Superfetch

TEST RESULTS

The Intel setup we have for comparison is comprised of a modestly priced P45
motherboard with DDR3 support, the Asus P5Q3, and an Intel Q9650 processor.
While the Q9550 would have been the ideal candidate to do battle with the X4
955 BE, we unfortunately did not have one at our disposal. The Q9650 has a
higher clock speed and thermal design power (125W vs. 95W). To make a little
more fair, we also tested it underclocked to Q9550 speeds (2.83 GHz). For our
AMD platform, we used 790GX motherboard, the Asus M4A78T-E, which happens to
have a similar price and feature-set compared to the P5Q3 (except for onboard
graphics). Both testbeds utilize 2x2GB of Corsair DDR3 memory, a GeForce 9400GT
graphics card, a WD VelociRaptor and an OEM Seasonic power supply.

Test Results: General Power Consumption
Test State
Q9650
X4 955 BE
X4 810
2.60 GHz

X3 720
2.80 GHz
3.00
GHz
2.83
GHz
3.20 GHz
3.20 GHz
UV*
Idle
68W
68W
73W
73W
79W
76W
VC-1
Playback
86W
86W
99W
99W
99W
97W
CPU Load
(half cores)
122W
121W
157W
138W
127W
N/A
CPU Load
(all cores)
142W
141W
201W
172W
159W
153W
CPU + GPU
Load
156W
154W
217W
187W
173W
168W
*CPU undervolted from 1.350V to 1.225V. C&Q left
enabled.

Off the bat we see that the Q9650, despite its high TDP used much less power
than the newest Phenom II’s, and to a lesser extent, the 2.6 GHz X4 810 and
even the tri-core X3 720 BE. Underclocking it to the same speed as a Q9550 resulted
in negligible power savings. If we assume Intel’s TDP ratings are accurate (compared
to the rest of their own processors), we can safely guess that a real Q9550
would use even less power.

The X4 955 BE used 5W more when idle and as load increased, so did the gap,
topping out at a difference of about 60W when heavy CPU stress was applied.
When we tested the X4 955 BE at the minimum voltage at which it was stable (1.225V),
power consumption improved significantly, cutting the difference in half. Still
a sizable amount, and we haven’t even taken into account performance yet.


CPU-Z screenshot: Phenom X4 955 BE: minimum undervolt at stock speed.

Performance

Benchmark Comparison: Q9650 vs. X4 955 BE
Test
Q9650
X4 955 BE
@ 3.00 GHz
@ 2.83 GHz
Time
Power (AC)
Time
Power (AC)
Time
Power (AC)
Stock
UV*
NOD32
2:23
88W
2:54
88W
2:27
128W
118W
WinRAR
2:51
98W
2:54
98W
3:05
128W
116W
iTunes
3:20
92W
3:32
91W
4:34
137W
123W
TMPGEnc
3:07
118W
3:16
116W
2:52
167W
149W
3DMark2006
3329
3322
3315
PCMark2005
9061
8667
9004
*CPU undervolted from 1.350V to 1.225V. C&Q left
enabled.

Our real-world benchmarks reveal that the general performance of the X4 955
BE is more or less equivalent to that of the Q9650 underclocked to Q9550
speeds. The X4 955 BE was noticeably faster during the NOD32 anti-virus test
and when encoding VC-1 video with TMPGEnc. The Q9650 at 2.83 GHz had a slim
margin of victory when archiving files with WinRAR and and thrashed the X4 955
BE when encoding AAC with iTunes. 3DMark was close while PCMark favored the
X4 955 BE by only 4%.

Power consumption was higher than the Intel. While the X4 955 BE
performs at a level similar to its Intel rival, it does so with higher
power draw, between 30W and 51W more during our timed tests. Undervolting cut
this difference by a fair amount, but the Intel setup still had a sizable advantage.

Benchmark Comparison: X4 955 BE vs. Other Phenom
II’s
Test
X4 955 BE (3.2 GHz)
X4 810 (2.6 GHz)
X3 720 (2.8 GHz)
Time
Power
Time
Power
Time
Power
Stock
UV*
NOD32
2:27
128W
118W
3:05
109W
2:47
112W
WinRAR
3:05
128W
116W
3:32
104W
3:16
111W
iTunes
4:34
137W
123W
5:38
112W
5:13
118W
TMPGEnc
2:52
167W
149W
3:29
137W
5:08
130W
3DMark2006
3315
3288
3273
PCMark2005
9004
7756
7738
*CPU undervolted from 1.350V to 1.225V. C&Q left
enabled.

Compared to the AM3 processors in AMD’s lineup, the X4 955 BE is not surprisingly
the fastest. The X4 810 loses out in most of our timed tests to the X3 720 BE,
a result of the higher clock speed and the fact the majority of our timed tests
are not multithreaded. Our test suite is representative of “general”
use; most applications still don’t take advantage of the extra
cores.

Benchmark Power Consumption (Watt-hours)
Test
Q9650
X4 955 BE 3.2 GHz
X4 810 2.6 GHz
X3 720 2.8 GHz
3.0 GHz
2.83 GHz
Stock
UV*
NOD32
3.50
4.25
5.23
4.82
5.60
5.20
WinRAR
4.66
4.74
6.58
5.96
6.12
6.04
iTunes
5.11
5.36
10.43
9.36
10.52
10.26
TMPGEnc
6.13
6.32
7.98
7.12
7.95
11.12
*CPU undervolted from 1.350V to 1.225V. C&Q left
enabled.

Multiplying the average system power draw by the amount of time it took to
finish our timed tests gives us a rough approximation of how much energy was
actually used. The X4 955 BE system used more power than the 2.83 GHz Q9650
system, even during the tests it won, and even when the X4 955 BE was undervolted.
For a system that is in moderate to heavy use for lengthy amounts of time, the
number of extra watt-hours will certainly add up. With the increased
power consumption also comes heat — the Zerotherm heatsink we utilized for
both platforms felt much hotter when cooling the X4 955 BE.

AMD Overdrive: Smart Profiles

AMD
Overdrive
is a useful utility that allows for CPU, voltage, and memory tweaks
to be performed directly from the Windows desktop, without having to enter the
BIOS at all. The newest 3.0 beta version introduces a new feature that may help
improve performance without increasing power draw: Smart Profiles.


AMD Overdrive: Smart Profile screen.

Smart Profiles allows individual CPU core frequencies as well as affinity to
be adjusted when a designated program is launched. As an example, we set up
a profile for WinRAR. As a single-threaded program, it makes little sense for
all four cores to increase to its nominal speed. So we set WinRAR to use only
Core 0 with the maximum CPU multiplier. The other core multipliers we set to
minimum, so they will stay at 800 MHz. That’s right — independent core
frequencies! You can also use this to overclock each core, as long as you have
an unlocked processor.


Smart Profile at work.

Once AMD Overdrive detected WinRAR.exe as being active, the first core ramped
up to 3.2 GHz, while the other three stayed at 800 MHz. Running our WinRAR benchmark
with this Smart Profile active resulted in an average power savings of 2W during
the operation. Using a new profile, we set it to overclock the first core to
3.4 GHz — this which wiped out the power difference, but resulted in our benchmark
finishing 5 seconds faster.

While this may seem like an easy, free way of getting a bit more performance
without using any extra power, it still needs work. Firstly, and most importantly,
the moment we loaded AMD Overdrive, Cool’n’Quiet disabled, resulting in higher
idle power draw. After we exited the utility, we had to restore the power plan
we were using in Vista to its default settings to re-enable Cool’n’Quiet. If
saving power is the goal, this absolutely kills it.

The way profiles operate needs serious improvement as well. Once a profiled
program was loaded, the clock frequencies changed accordingly but Overdrive
didn’t distinguish between an idle application, and one that was actually in
use. For example, you could set it to overclock all four cores heavily when
a video encoding application is running, but even after the encoding is finished,
the clock speeds will remain overclocked because the program is still open.
Also, it does not handle multiple profiles well. If Program A is launched first,
the Smart Profile for that program activates, even if a second profiled program,
Program B, is loaded. When Program A is shut down, the settings revert to default,
rather than switching to Program B’s profile. For it to work properly, you can
only use one profiled program at a time.

FINAL THOUGHTS

From a pure price/performance standpoint, AMD is certainly competing hard.
The X4 955 BE closely matches Intel’s Q9550, but with a smaller price-tag.
One must also consider the motherboards available especially if gaming is not
an issue. Full-sized AM3 ATX motherboards with decent ATI IGPs are widely available
and modestly priced; for example the Asus M4A78T-E we used for testing is priced $137~160. The same cannot be said of Intel socket 775; there are few full ATX boards with good IGPs.

DDR3 memory, which is still more expensive than DDR2, has
come down to the point where its not such a painful upgrade. Perhaps the best
thing to come out of the release of these new processors is the accompanying
price cut to the X4 940. We found the Phenom II X4 940 online for as low as $169 already, making it an
truly excellent value, especially for those unwilling to switch to DDR3.

Phenom II, even on the new AM3 platform with DDR3, is still significantly power
inefficient compared to Intel’s offerings, at least for the range of tests we performed. Idle power was only 5W higher, but
once we started running real tests, the difference grew large.
The Q9650 which we underclocked to simulate a Q9550, is rated 30W higher than
an actual Q9550, and it still well undercut AMD’s latest in power consumption.

Undervolting the X4 955 BE to its lowest stable voltage did make up half the difference, to about 20-30W depending
on the type of load used. That’s not too bad, and depending on how much electricity
costs in your area and your usage, it may still make a prudent economical choice.
For example, at SCPR’s location, an extra 25W running 12 hours a day for a year
would cost about $8.60 CDN, which is peanuts for anyone considering a US$250 processor. Still, if the desire for lower power consumption
in a high performance PC is environmentally-driven, then Intel remains the obvious choice.

Our thanks to AMD,
ASUSTeK, and Corsair
for the various product samples used in this review.

* * *

Articles of Related Interest
Asus M4A78T-E AM3 motherboard
MSI KA760GM: 780G Minus the HD
790GX Showdown: Gigabyte vs. MSI
Phenom II: AMD pulls closer
Zotac GeForce 8200-ITX WiFi: A Compact
AM2 Solution

Asus P5N7A-VM: Geforce 9300
IGP

* * *

Discuss this article in the SPCR forums.

Leave a Comment

Your email address will not be published. Required fields are marked *