AMD Turion 64 on the Desktop

Table of Contents

It seemed to take forever, but our article on running AMD’s mobile 64-bit CPUs on socket 754 desktop boards is finally done. We examine the power efficiency, cost and implementation of Turion 64 ML44, ML40, MT40 and MT34 models, along side Athlon 64-3200+ and Intel Pentium M 770 CPUs. It’s a complicated, sprawling piece, far from being the last word on the topic (of Turion 64 on the desktop), but a good intro and overview.

Februrary 18, 2006 by Devon
Cooke
and Mike Chin

This article is one of the most demanding we’ve tackled at SPCR in quite a while. The idea for the article germinated many months ago, but it’s taken all this time to assemble the samples, necessary components and new test gear to bring it to fruition. The testing and writing took several weeks. Turion 64s work on many desktop socket 754 motherboards. What’s so hard about that? It’s what we thought too, but correctly documenting all the details turned out to be pretty challenging. In the end, it’s still not a definitive piece on the subject, but we think it’s a good introduction and overview.

– Mike Chin, Editor

Building a silent system is all about managing heat intelligently. With the
exception of hard drives, the only sources of noise in a typical PC are fans,
and less heat means it’s possible to run fewer, slower and quieter fans. Advances
in heatsinks and power supplies have made it easier to just slap together a
few parts to make a system that is reasonably quiet, but the basic problem,
heat output, still remains. Even the best heatsinks on the market struggle to
cool the hottest processors, and no power supply is immune from the efficiency
losses ° and increased heat ° caused by sustained high output.

So, how do you go about reducing heat output? The amount of heat in a system
is directly proportionate to the amount of power it consumes, so reducing heat
means cutting down on power consumption. A number of methods for doing so have
become popular at SPCR, including

  • Undervolting and underclocking: Slower, low voltage chips consume less power
  • Deliberately choosing low-end parts, as they are often slower and
    lower voltage than the latest and greatest
  • Choosing more efficient parts, with high performance to watt ratios
  • Using parts designed for the laptops (such as HDDs and CPUs), which consume minimal
    power to conserve battery life

Hardware manufacturers have also recognized the importance of reducing power;
high efficiency power supplies are now de rigeur, and Intel and AMD have added
performance-per-watt as one of their competitive benchmarks °
still a big step below the alter of high performance, but at least the parameter
is now in the cathedral. Nowhere is this more apparent than in the mobile processor
market, where low power consumption translates into long battery life as well
as low noise. Intel’s Pentium M processor (part of the “Centrino”
brand) is probably better known than AMD’s Turion 64, but both
have a lot to offer in terms of performance per watt.

It’s widely known that the Pentium M can be used to power a desktop system
thanks to a growing number of motherboards and Small Form Factor systems designed
specifically for it. The Pentium M uses a pin array that looks like it should
fit in a standard socket 478, but its configuration is different and it requires
a socket 479 board to work. Bringing Pentium M to the desktop required the interest
and sustained effort of companies such as AOpen and DFI, who did not have official
Intel support. Intel’s view for a long time was that the Pentium M and socket
479 are part of the laptop-specific Centrino platform and should be supported
by the mobile division.

There was early interest in Pentium M for the desktop right here at SPCR. The
first reference in the forums appeared back as early as mid-2003 when forum
member Beyonder wrote: “I
think I’d chew off my arm for a Centrino MB and processor solution for the desktop
segment at the moment….
” Since then, we’ve helped popularize the Pentium
M for desktop by reviewing no fewer than five products designed specifically
for this processor:

Not many people realize that the Turion 64 can also be used in this way. In
fact, it doesn’t even require a special motherboard ° many Socket 754 boards
that support desktop processors can also run Turions. Ironically, this compatibility
is one of the reasons that Turion 64 for the desktop has not been as popular
a concept: There are no special products for this application, and no special
marketing efforts by any interested companies to promote the product as there
have been for the Pentium M. Hence, there isn’t much awareness about the concept of Turion 64 on the desktop.

TURION 64 PERFORMANCE

Another reason why more people haven’t turned to the Turion 64 is because not
much is known about it. AMD has done a good job of selling the Turion 64 as
a notebook processor, so people tend to assume that it is less powerful and
less interesting than the top-of-the-line Athlon 64 and X2 processors. Sure,
it may be efficient, but why trade performance for efficiency?

Here’s an open secret: The Turion 64 is based on the same AMD64 architecture
that powers AMD’s desktop and server processors. In terms of performance, the
Turion 64 should perform identically to an Athlon 64 of the same specifications.
So how does AMD justify saying that the Turion 64 is “designed from the
ground up”? The key is in the silicon. Although the processor logic
is the same, the underlying technology is not. Turion 64 chips are built from
low-voltage transistors that have less electrical leakage, and the layout of
the chip has been optimized for low power, not high speed.

This does not mean that the Turion 64 will perform slower. The
effect of the optimizations is to reduce the maximum potential
clock speed that the Turion 64 can support; clock for clock, there should be
no difference. The effect of the slower transistors can be seen in the range
of clock speeds for each processor: 1.6 ~ 2.4 GHz for Turion 64 chips, and 1.8
~ 2.8 GHz for Athlon 64 and Athlon 64 FX chips. In theory, the regular Athlon
64 chips should be better overclockers than the Turion 64 chips.

Of course, there are other factors that affect performance, like socket choice.
Turions are available only for Socket 754, which means they cannot take advantage
of dual channel memory, which is the domain of Socket 939. Another factor is
dual core. AMD’s top processors all feature two cores, but this feature has
yet to show up in their Turion 64 lineup. When it does, a change in socket
type will render the dual core Turions incompatible with desktop equipment.

TURION 64’s ATHLON 64 EQUIVALENTS
Turion 64 Model & TDP
Athlon 64 Equivalent &TDP
Clock Speed
L2 Cache
Socket Type
HyperTransport Speed
ML-44 (35W)
Clawhammer 3700+ (89W)
2.4 GHz
1 MB
S754
1.6 GHz
MT-40 (25W)
Clawhammer 3400+ (89W)
2.2 GHz
1 MB
S754
1.6 GHz
MT-37 (25W)
Clawhammer 3200+ (89W)
2.0 GHz
1 MB
S754
1.6 GHz
MT-32 (25W)
Clawhammer 2800+ (89W)
1.8 GHz
512 KB
S754
1.6 GHz

The table above should give you a good idea of what the Turion 64 chips are
capable of, performance-wise. We’ve picked the closest matches we could find between the two processor familes. The Turions should
perform a little faster and run considerably cooler, since they are based on a newer revision of the AMD64
architecture (E5) than the Clawhammer chips (C0, CG). The equivalent desktop core (E6) is not widely available for Socket 754; the ones that are available sport 512KB cache rather than the 1MB of most of the Turion 64s.

So much for the AMD comparison. But, the Turion 64 is not meant to compete
against the Athlon 64. The real question is how it stacks up against Intel’s
Pentium M. Extensive performance testing is not SPCR’s raison d’°tre,
but the short answer is it’s too close to call. The
Tech Report
and Laptop
Logic
have both done detailed performance comparisons of the two processors.
Our study of such reviews and our own extensive experience with Turion 64 and
Pentium M desktop system has convinced us that there is no compelling performance
reason to choose between the Pentium M and the Turion 64. They are both powerful,
modern processors that provide speed performance; when you factor in their power
efficiency, their performance is staggering ° sort of like a 60 mpg hybrid
also capable of 0 to 60 mph in 6 seconds.

WHY CHOOSE A TURION 64?

64-bit Computing ° One thing to consider is that if you’re interested
in 64-bit computing, your only choice is a Turion. Intel has yet to integrate
x86-64 into their lineup of mobile processors ° even the recently released
Core Duo (Yonah) chips don’t support it. AMD, on the other hand, has supported
x86-64 from day one ° no surprise, since they invented it! For most users,
however, the lack of 64-bit capability is meaningless; 32-bit computing will
continue to be mainstream for some time yet.

SSE3 Support ° Another (minor) point is that the Turion supports
SSE3 code, which could make it an attractive choice for multimedia use. The
performance boost from SSE3 is rarely large, however, so this too is for special
use only. Both x86-64 and SSE3 are if-you-don’t-know-what-it-is-you-don’t-need-it
features.

Power Efficiency ° For many people, power consumption is likely
to be the biggest factor in choosing between the Turion 64 and the Pentium M.
As always, the different ways in which Intel and AMD declare power requirements
make a direct comparison difficult. Even worse is the fact that AMD produces
two versions of the Turion for every speed class: An inexpensive, low efficiency
model, and a more expensive, high efficiency model. We’ll get to actual power
measurements later. For now, take these manufacturer’s specifications with a
grain of salt.

  • All of Intel’s current generation of Pentium M processors have a TDP of
    27W and a default Vcore between 1.287~1.400V.
  • The “ML” family of Turion 64 processors have a TDP of 35W and
    a default Vcore of 1.35V.
  • The “MT” family of Turion 64 processors have a TDP of 25W and
    a default Vcore of 1.2V.

Any power comparison is muddied by the Turion’s integrated memory controller
and HyperTransport Link. The Pentium M lacks this functionality; the memory
controller and FSB are controlled off-chip. This means that, even if the two
processors end up consuming exactly the same amount of power, a system with
a Turion is likely to consume slightly less power since there is no need for
an additional chip to control the memory. In the context of a system, this difference
is probably so small that it’s irrelevant, but it should serve as a caution
against making a big deal out of a watt or two.

What the manufacturers’ specifications do tell us is that all of these processors
are in the same ballpark when it comes to power. In fact, clock speed may be
more of a factor than which processor you choose. Just like performance, the
Turion 64 and Pentium M seem to be pretty close in terms of power. In the end,
a purchasing decision between the two may come down to price, availability,
and ease of use.

Price and Availability ° When it comes to availability, Intel’s
market presence is tough to beat. Everybody knows Intel, so everybody sells
it. What’s more, the Pentium M has been on the market a lot longer and is available
in a retail box, which makes it easier for retailers to get hold of.

On the other hand, getting a Turion will cost you a whole lot less ° if
you can find one for sale. Not only are the processors themselves cheaper, but,
thanks to the compatibility with Socket 754, a compatible motherboard can be
had for acorns. Compare that to the Pentium M, which requires a special motherboard
that comes at a high premium. And, unlike the Pentium M boards, Socket 754 boards
are readily available ° you might even own one already!

MOBILE PROCESSOR PRICE LIST (FEBRUARY 2006)
Turion 64
Clock Speed
Price
Pentium M
Clock Speed
Price
Turion 64 Price Advantage
ML-44*
2.4 GHz
$354

780*
2.26 GHz
$637
$283
MT-40*
2.2 GHz
$268
770*
2.13 GHz
$423
$155
ML-40*
2.2 GHz
$220
$203
MT-37
2.0 GHz
$225
760
2.0 GHz
$294
$69
ML-37
2.0 GHz
$184
$110
MT-34
1.8 GHz
$189
745
1.8 GHz
$241
$52
ML-34
1.8 GHz
$154
$87
MT-30
1.6 GHz
$150
730
1.6 GHz
$209
$59
ML-30
1.6 GHz
$145
$64

* Note that the ML-44 and the Intel 780 are not exact clock speed equivalents. Neither are the MT/ML-40 and the Intel 770.

As of February 13, 2006, the Turion has a significant price advantage over
the Pentium M, based on the official price lists from AMD
and Intel.
For the most part, the Pentium M costs about 30% more than a similarly clocked
“MT” part, and about 50% more than an “ML” part.

A similar comparison can be made for motherboards. Online retailer NewEgg sells
a total of two
motherboards
compatible with the Pentium M; both retail for US$220 or more.
By way of comparison, they sell 83
different boards
for Socket 754, all of which are under US$100 and 23 of
which are under US$50. (Admittedly, not all of them will work with Turion 64.) Taking the motherboard into consideration, the price
difference between the Turion 64 and the Pentium M jumps significantly °
over $400 for the top-of-the-line processor.

Wide Compatibility with 754 Motherboards ° In addition to the better
variety of motherboards that is available, using Socket 754 means that the Turion
64 is compatible with a wide range of aftermarket heatsinks. This contrasts
sharply with the Pentium M boards on the market, almost all of which use proprietary
cooling solutions which can’t be counted on for quiet performance. Even better,
because most heatsinks on the market are designed for cooling hot-running desktop
chips, it should not be hard to find a heatsink that is capable of cooling the
Turion passively, thus eliminating a source of noise from the system.


Many K8-compatible heatsinks can be used with the Turion, like this heavy
Zalman 7000CU.

One word of warning: Like the Pentium M, the Turion 64 is supplied without
an integrated heatspreader to protect the CPU. Because the bare CPU die is exposed,
it is not hard to crack or chip the CPU. The surface of the CPU die is also
lower than it would be otherwise, so not all heatsinks may provide the necessary
tension. Extra care should be taken when installing and removing the CPU heatsink,
and large, heavy heatsinks are best avoided.

A TURION 64 TEST PLATFORM

Although there have been several comparisons between the performance of the
Turion 64 and the Pentium M, power consumption has not been closely looked at…
at least not closely enough for our standards. So, we set out to find out exactly
how much power the Turion requires. We know it’s not much, but how much is not
much?

First we had to get our hands on some Turions. AMD generously supplied
three Turions: The most power hungry model (ML-44), and two equivalently clocked chips
with different TDPs (ML-40 and MT-40). Two of these
(the “ML” models) are rated for 35W, the other two (“MT”)
for 25W. We also purchased a relatively
low powered model (MT-34) used from an SPCR forum member.The four Turions would go head to head against each other, a Pentium
M 770 (2.13 GHz), and a standard Athlon 64-3200+.


Our most and least power hungry Turion chips, side by side.


Two 2.2 GHz chips, two different TDPs.

With true excitement (and a little trepidation) we plugged one of the Turion 64s into
a socket 754 motherboard that we had on hand, an EPoX EP-8KDA3+, carefully
installed a Zalman 7000 HSF (more on that later), and pressed the power button:
Alas, the board wouldn’t post. We tried again after swapping out the processor
with an Athlon 64 and updating the board with the latest BIOS, but again, it
would not show any sign of life with the Turion.


EPoX EP-8KDA3+: Great board, but no Turion 64 support.

Undeterred, we turned to the web to find a board that would work, and decided
that a DFI LanParty UT NF3 250GB might be the way to go. It is an older
board based on the older nForce3 chipset, but many people reported success using
it with the Turion 64. It was also on a
list of socket 754 motherboards that people have (and haven’t) succeeded in
using with Turion 64, created by a Japanese enthusiast
. The caveat on is
that the 40 or so boards on the “works” list are not guaranteed to
have full functionality with the Turion 64.


DFI LANParty UT NF3 250GB: Turion 64 recognition, but no CnQ.

Two days later, the DFI board showed up, and we tried again. Success! The Turion
64 was recognized correctly, Windows was duly installed, and we began to take
some basic measurements. Then we noticed something strange. No matter how hard
we pushed the CPU, the clock speed never went above 800 MHz. Obviously, something
wasn’t right.

A close examination of the BIOS revealed that the chip could be manually set
to the correct maximum clock speed and voltage, but no matter what we did, we
couldn’t enable Cool’n’Quiet. A conversation with Damon Muzny at AMD revealed
why: Cool’n’Quiet for Athlon 64 and PowerNow! for the Turion 64 are similar,
but not identical. The difference is in the default power state of the processors:
The Turion defaults to the lowest clock frequency and voltage, and with PowerNow!,
the clock jumps up when needed. The Athlon 64 defaults to the highest clock
frequency with Cool’n’Quiet dropping down the speed and voltage when the processor
is idling. Naturally, CnQ and PN! have differences in their code, which is why
the default clock speed of the Turion 64 was 800 MHz on this DFI board.

The lack of CnQ functionality in the DFI board was a bit disappointing, but
it doesn’t mean the board cannot be used. Since the BIOS allows for great flexibility,
the Turion 64s could be run at many different settings. It could easily be set
to the stock voltage and speed, and left at that. However, such a configuration
would not allow CnQ to be used, which we really wanted to experiment with, so our search continued.

The next motherboard we tried was an MSI RS482M-IL. This ATI Radeon XPRESS 200 chipset board recognized the Turion 64 and treated it
as if it was an Athlon 64, with full CnQ functionality. Considering the differences
between the Turion and Athlon, and between PowerNow! and CnQ, we don’t really
know why and how CnQ works on this board, but it’s great that it works!


MSI RS482M-IL: The only board we’ve found (so far) that supports Cool’n’Quiet
on Turion 64.

METHODOLOGY

A total of six processors were chosen for testing: The four Turion 64 processors
mentioned above, an Athlon 64 3200+ (Newcastle), and a Pentium M 770. Specifications
for the six processors are listed in a table below. Together, the six processors
should help answer these two questions:

  1. How much power does the Turion 64 line require, and how does power consumption
    change across the range of models?
  2. How does the Turion 64 measure up in terms of performance-per-watt in comparison
    to other processors on the market?
PROCESSOR COMPARISON
Processor
Clock Speed
L2 Cache
Default Vcore
TDP
Supported Instruction Set Extensions
Turion 64 MT-34
1.8 GHz
1 MB
1.2V
25W
MMX, 3DNow!, SSE, SSE2, SSE3, AMD64, NX-Bit
Turion 64 MT-40
2.2 GHz
Turion 64 ML-40
2.2 GHz
1.35V
35W
Turion 64 ML-44
2.4 GHz
Athlon 64 3200+
2.2 GHz
512 KB
1.5V
89W
MMX, 3DNow!, SSE, SSE2
AMD64, NX-Bit
Pentium M 770
2.13 GHz
2 MB
1.372V
27W
MMX, SSE, SSE2, NX-Bit

Test tools included:

Two test systems were built: One for the AMD processors, and one for the Pentium
M. The same power supply was used in both to eliminate any differences in power
supply efficiency.

System

AMD Socket 754

Intel Socket 479
Power

FSP Green PS
FSP400-60GLN 400W power supply
Motherboard

AOpen i915Ga-HFS
RAM

512 MB OCZ Gold PC3200 DDR SDRAM (one stick)

1024 MB Corsair DDR2 SDRAM (one stick)
Heatsink



Zalman CNPS7000CU,
plugged into the motherboard

Stock AOpen heatsink, plugged into the motherboard
HDD

Seagate Momentus 5400.3

Hitachi E7K100

The AOpen i915Ga-HFS used here is the only full-ATX motherboard for Pentium-M that we know of. It is packed with features, such as:

  • HDTV/TV Support – YPbPr connector, Svideo connector and D4 (Japan use only) connector to attain the highest image quality on 1080i or 720p resolution.
  • Supports DVI Output – On-board, along with VGA.
  • Gigabit LAN on-board
  • PCI Express x16 Graphics slot
  • Dual Channel DDR2 533 memory support

The only downside is that it retains the small proprietary CPU heatsink mounting holes from the earlier micro-ATX i915GMm-HFS board we reviewed last year, along with a similarly small Northbridge HSF.


AOpen i915Ga-HFS is the only ATX board for P-M we know of.

All processors were measured in three states to establish the full range of
power consumption:

  • Idle
  • Idle with Cool’n’Quiet, PowerNow!, or SpeedStep enabled
  • Under heavy load using CPUBurn
    to stress the processor

Two different measurements were taken in each state:

  • Total AC power consumed by the system as a whole.
  • DC power drawn by the processor from the +12V AUX connector.


A fancy power meter from Extech helped us keep track of AC power.

AC power was measured to obtain a power profile of each system as a whole.
By design, this includes power lost in the power supply itself during conversion
from AC to DC. Most power supplies become less efficient as they approach zero
output. At the low power loads of these systems, the power conversion loss may
account for as much as 50% of the total system power. Measurements for AC power
were read off of the digital display on the Extech power meter.

However, the DC power measurements are a bit more involved. A high precision current sensor is plugged directly into the +12V AUX connector on
the motherboard, so that all power through this connection must pass through
the power meter. The line voltage (nominally +12V) and the current are measured
with multimeters, and multiplied together to get the total power running through
the connection. Assuming that the CPU only draws power through +12V AUX connection
and nothing else does, this tells us the amount of power consumed by the
CPU plus whatever power is lost to inefficiencies in the voltage regulators on the
motherboard. This assumption was confirmed by using the Fluke 36 clamp meter, which showed that no other voltage lines showed significant increases
when the CPU was under load.


An custom-built shunt featuring a LTS 25-NP current sensor allowed us…


…to measure voltage and current on the +12V AUX connector that powers
the CPU.

The final accuracy for this power calculation is better than °1W, maybe as good as °0.1W.

To repeat, the DC power measurements do not take the efficiency of the voltage regular
module (VRM) on the motherboard into account.

TEST RESULTS

Ambient conditions during testing were 20°C and 117~119VAC, measured with
the Extech power meter.


Turions on the test bench.

Troubles with Voltages

After running a complete test on all of the Turion chips, we discovered a
slight issue with the core voltage that our carefully selected motherboard
was supplying to the CPU. Specifically, the board seemed to be accidentally
overvolting all of the Turions by exactly 0.1V. Our Athlon 64 chip did not
display this problem, and ran properly at its specified voltage.

Voltage Discrepancy
Processor State
Target VCore
Actual VCore
Cool’n’Quiet (all Turion chips)
0.9V
1.0V
Stock Speed (“MT” Models)
1.2V
1.3V
Stock Speed (“ML” Models)
1.35V
1.45V

The problem is not limited to this motherboard. A little digging revealed that several laptop owners also reported this problem. One user on
AMD’s forum even suggested
that the problem was caused by confusion surrounding the way AMD specifies
the stock voltage
.

The severe lack of technical documentation available to the public makes
it very difficult to discover exactly what is going on. In fact, AMD does
not list the stock voltages for the Turion anywhere on their web site. The
proper voltages can be discovered by decoding the various part numbers (OPNs)
using the
formula found on this page
.

At this point it is impossible to know exactly why the discrepancy occurred,
and why it seems to be so widespread. There are a number of possibilities:

  1. The MSI board is detecting the stock voltage correctly, and the voltages
    specified by the OPN are being interpreted incorrectly.
  2. AMD has changed the stock voltage without updating the OPN for our processors.
  3. The MSI board and many other boards are misinterpreting the CPUID string
    on the CPU, and thus applying the incorrect voltage.

The first option is plausible enough. The lack of technical documentation
about the Turion means that it is quite likely that someone made an informed
guess about the proper OPN formula, and got it wrong. And, once that wrong
information appeared, it could easily have been carried around the net. With
no other reliable sources of information, there is no way to confirm whether the voltage derived from the OPN is correct.

The second possibility seems highly unlikely, given the tech industry’s penchant
for keeping strict records about version and revision numbers.

The third possibility seems most likely to be correct, if only because it
is very implausible that the stock voltage for the “ML” models is
so high. 1.45V is higher than the voltage required by any of AMD’s current
E-stepping CPUs. It seems very unlikely that a chip designed specifically
for low power would run at such a high voltage.

We eventually decided to give AMD the benefit of the doubt and assumed that
the lower voltages were correct. We then proceeded to re-test all of the processors
again at their correct voltages. Voltage states were modified using CrystalCPUID
(which also detected the “higher” voltages).

What a difference 0.1V makes! The drop in voltage caused all of the Turion
64 chips to react similarly to the MT-34 profiled below. The average change
in power consumption was 18°3% across all processors and power states.
This didn’t amount to much at idle (18% less of not very much is still not
very much), but the difference under load was as much at 8W! That’s a significant
difference that could tip the scales in a comparison with the Pentium M.

AMD Turion 64 MT-34 (1.8 GHz)
Processor State
CPU Power @ Wrong VCore (1.3V)
CPU Power @ Stock VCore (1.2V)
Percentage Change
Idle (CnQ)
3.1W
2.5W
19%
Idle (No CnQ)
5.9W
5.0W
15%
Load (CPUBurn)
26.5W
21.9W
17%


Note that the above power measurements includes losses via the motherboard voltage regulators.
The power drawn by the CPU alone is lower.

Power at Idle

With the confusion over voltages out of the way, we can finally get down
to some actual comparisons.

Processor Power Consumption: Idle (CnQ)
Processor
Clock Speed
Vcore
CPU Power* (DC)
System Power° (AC)
Pentium M 770
800 MHz
0.734V
1.0W
40W
Turion 64 ML-40
800 MHz
0.92V
2.2W
36W
Turion 64 MT-34
800 MHz
0.92V
2.5W
37W
Turion 64 MT-40
800 MHz
0.92V
2.6W
36W
Turion 64 ML-44
800 MHz
0.92V
2.8W
36W
Athlon 64 3200+
1.0 GHz
1.12V
4.4W
37W


*Note that the CPU power measurements includes losses via the motherboard voltage regulators.
The power drawn by the CPU alone is lower.
° The system power measurement includes losses in AC/DC conversion
within the PSU.

At such low loads, the loss could be as high as 50%.

For the most part, none of the processors really differentiated themselves
when Cool’n’Quiet was running (or SpeedStep in the case of the Pentium M).
Even the Athlon 64 did not draw significantly more power than the mobile chips
unless it was loaded. This was especially true when the power saving features
were enabled and the processors were all running at reduced clock speeds.
Yes, the Pentium M did draw 2.5 times less power than the Turion chips, but
in absolute terms the difference was so small that it was irrelevant.

Even though the power consumption of the Pentium M was lower, the
total system power was higher than every other CPU tested, even
the Athlon 64. This suggests that the differences at this low power level
are due to differences in the chipsets and motherboards, not the CPUs. On
this assumption, it would appear that the AOpen board in the Pentium M system
drew 3~4 watts more than the MSI board in the AMD system. Perhaps the extra
power is needed for the AOpen board’s memory controller / northbridge chip. These are not needed for the socket 754 board, those functions being integrated in the die of the Turion / Athlon 64 processors.

Processor Power Consumption: Idle (No CnQ)
Processor
Clock Speed
Vcore
CPU Power* (DC)
System Power° (AC)
Turion 64 MT-34
1.8 GHz
1.21V
5.0W
39W
Turion 64 MT-40
2.2 GHz
1.23V
5.5W
39W
Turion 64 ML-40
2.2 GHz
1.38V
7.9W
43W
Turion 64 ML-44
2.4 GHz
1.38V
8.5W
43W
Pentium M 770
2.13 GHz
1.334V
9.9W
50W
Athlon 64 3200+
2.2 GHz
1.53V
12.8W
47W


*Note that the CPU power measurements includes losses via the motherboard voltage regulators.
The power drawn by the CPU alone is lower.
° The system power measurement includes losses in AC/DC conversion
within the PSU.

At such low loads, the loss could be as high as 50%.

Without Cool’n’Quiet, the low powered Turion MT proved to be slightly more
efficient than the ML versions, and both versions of the Turion bested the
Pentium M. Once again, the differences are very minor, and even the old Clawhammer core Athlon
64 is not totally lost in this comparison.

One point of interest is the fact that the core voltage for the Pentium M
was almost identical to the Turion ML models. The slightly higher power draw
suggests that, at the same clock speed and core voltage, the Pentium M is
not quite as efficient as the Turions.

Once again, the AC Power draw for the Pentium M system was higher than any
of the AMD systems, even the Athlon 64. In fact, even though the differences
in processor power were small, the differences in system power were not. The
difference in power between the Turion MT systems and the Pentium M was 11W
° a difference of more than 25%! The difference is split evenly between
the processors and the motherboards themselves; each reduced the total power
draw by ~5W.

Power at Load

Processor Power Consumption: Load (CPUBurn)
Processor
Clock Speed
Vcore
CPU Power* (DC)
System Power° (AC)
Turion 64 MT-34
1.8 GHz
1.21V
21.9W
58W
Pentium M 770
2.13 GHz
1.300V
23.3W
65W
Turion 64 MT-40
2.2 GHz
1.22V
26.4W
64W
Turion 64 ML-40
2.2 GHz
1.36V
34.7W
73W
Turion 64 ML-44
2.4 GHz
1.36V
37.0W
75W
Athlon 64 3200+
2.2 GHz
1.51V
53.8W
92W


*Note that the CPU power measurements includes losses via the motherboard voltage regulators.
The power drawn by the CPU alone is lower.
° The system power measurement includes losses in AC/DC conversion
within the PSU.

Under load, the battle between the Pentium M and the Turion MTs remained
too close to call. The Pentium M did improve its standing compared to its
performance at idle (without SpeedStep), but this is probably due mostly to
the inexplicable drop in core voltage that it saw under load. In any case,
neither processor distinguished itself as a clear victor in terms of power
consumption. A slight variation in the workload could completely reverse the
results, especially when the relative performance of the processors comes
into play. A task that the Pentium M is particularly suited to may end up
consuming less power simply because the processor doesn’t need to work as
long to complete it. Ultimately, the differences are probably too small to
draw any conclusions ° the margin of error is simply too high.

One slight discrepancy between the Pentium M and the Turion 64s is how well
the empirical power measurements match up against the specified TDP for each
processor. The Pentium M was operating well below its specified TDP of 27W,
while the Turions seemed to be closer to their specified limits. However,
with the efficiency losses of the motherboard VRMs taken into account, it
is highly unlikely that any of the chips exceeded their specifications. Judging
from this, it may be that CPUBurn runs more efficiently on the Pentium M than
it does on the Turions ° or perhaps Intel’s estimate is more conservative
than AMDs.

The biggest differences in power consumption were not between AMD and Intel,
but between the different models in AMD’s lineup. The differences between
the MT/ML Turions and the Athlon 64 finally became visible under load. Just
as AMD says, the difference between the MT and ML chips was about 10W. Whether
or not this extra 10W is worth the difference in cost is up to you.

The Athlon 64 performed surprisingly well under load. Our sample uses the
less efficient Clawhammer core, rated at 89W, but we were pleasantly surprised
to see that the combined power demand of this processor and the motherboard’s VRM came in at only 54W. That was still much higher than the next
highest contender, but it’s still better than we expected from a supposedly
“hot” Athlon 64.

The ABCs of Desktop Turion 64

If you choose to run a Turion 64 on the desktop with a minimum of fuss, there are a few steps to follow. It’s slightly more involved than using a Pentium M for a desktop system, because there are much fewer choices for the latter, which are pre-packaged specifically for Pentium M.

1. Buy a Turion 64 processor. Unlike Pentium Ms, they are not sold in normal retail packaging. Only OEM models are available, mostly from system integrators who use them to build white-box laptops. (Note that retail package Pentium M processors don’t include heatsinks, either.)

2. Get a socket 754 board that recognizes E-stepping Athlon 64 processors. Generally any 754 board that has a BIOS dated at least mid-2005 should have this. Check latest BIOS updates and technical information for the motherboard to be sure. (A gotcha: If you have to update the BIOS in order to get E-revision processor recognition, you will need an earlier socket 754 processor to do so. So it’s best to use a board that already has the E-stepping processor support.)

If E-stepping A64s are supported, the board will power up with the Turion 64. It may run at the lowest speed by default, but any modern board will let you manually set the clock speed. Just set it to your Turion 64’s rated speed and you will be off and running.

3. Get a K8 heatsink/fan that will accommodate the thinner profile of the Turion 64 die, which does not have a heat spreader. Safest are the models that use a captive spring for each of the two mounting bolts. This allows you to apply the pressure slowly and evenly (by going back and forth between the two bolts as you screw it down) so that the base of the heatsink does not tilt under pressure and risk chipping the edge of the CPU die. Most heatsinks apply quite high pressure for a standard A64 (with heat spreader), so when the bolt is fully engaged, there will be plenty of pressure between the Turion 64 and the heatsink. With the Zalman 7000cu heatsink that we used, there were no springs, so how tight to engage the bolts is a judgement call that calls for a bit of common sense. (Note: The 7000cu we used is a very heavy all-copper heatsink which is not ideal for use with a bare-die CPU. Do as we say, not as we do; a lighter 7000alcu is a better choice.)

Other Considerations

The ability to adjust Vcore down manually in the BIOS is a very nice feature to have with the Turion 64, simply because it appears most boards probably feed it too high a voltage, which will cost you a bit of extra power at load. Bring the Vcore down to the 1.35V for ML or to 1.2V for MT ensures lower power consumption. Many of these processors will also run at lower than full rated voltage so that you can get even cooler operation and lower power consumption. For these reasons, manually adjustable Vcore in the BIOS is really great to have.

But if the board doesn’t have BIOS Vcore adjustments, it’s not critical, as the system will still run much cooler than any equivalent standard desktop system. If you want to lower Vcore lower anyway, it can be done with a couple of Windows utilities ° CrystalCPUID, which we used extensively for our testing, and also RM Clock. These should allow Vcore manipulation with just about any socket 754 desktop board that will run Turion 64.

In our view, Cool’n’Quiet support for the Turion 64 is not that important on the desktop because you’re not worrying about battery life, and when run at the correct Vcore, even without CnQ, the power efficiency is already so high. With low CPU power demand, motherboard VRM inefficiencies eat up a big chunk of any additional power savings; the power reduction in our test systems was only 3W to 7W at the wall outlet, depending on processor type and clock speed.

Finally, with the CPU generating so little heat, you will not need much airflow across the CPU heatsink. You can probably use a quiet fan at reduced speed for the lowest possible noise. Choose your other components carefully, as they will impact the overall noise of your system much more than the CPU. This includes the power supply, hard drives and video cards.

CONCLUSIONS

The Pentium M and the Turion 64 are very similar in terms of power consumption.
According to the benchmarks from other sites, they are also
very similar in terms of performance. Both can easily be used in desktop systems, with a few specialized motherboards for the Pentium M and a huge range of socket 754 boards for the Turion 64. Where they differ dramatically is in price, the Turion being at least $50 or 25% less costly than the equivalent speed Pentium M, and with the top models, some $300 less (50%). That’s for the processor alone, but factor in the >$200 price of Pentium M motherboards and the <$70 price of typical socket 754 boards, and you can easily get more than $400 savings by going with a top speed turion rather similar clock pentium m.

Aside from the Shuttle SD11G5 SFF barebones system, which comes with a quiet, well-designed, integrated CPU cooler, most of the other Pentium M boards have a basic problem: They require the use of a non-standard heatsink and/or HS mounting system, none of which are ideal for quiet cooling. So even after spending big money on a specialized platform and the more costly Pentium M, you still have to fiddle and tweak for a truly quiet setup. For silencers, the wide range of high performance, low noise heatsinks for the socket 754 really tilts the advantage to the Turion.

Why isn’t the Turion 64 more widely recognized for desktop use? Many factors are involved, but there are three big ones.

Consumer Awareness: To deliver Pentium
M to the desktop, AOpen and DFI had to build the market from scratch by providing
the appropriate motherboards ° and the marketing to sell them. Hence, there is consumer awareness of Pentium M as a desktop-viable product.

AMD, on the other hand, has had little interest in selling Turions to desktop users.
From what we can gather, they’re having enough time keeping up with demand in the mobile sector! So,
even though it is perfectly viable to run Turions in desktop machines, nobody
has put any marketing dollars behind the concept to make it fly.

Lack of Technical Documentation on the Turion: The relative lack of market awareness is exacerbated by the absence of technical documentation for the Turion from AMD. Without such information, even technically savvy experimenters have to tread slowly and cautiously. “Yes, the Turion works on many 754 boards” ° that statement would be a fair representation of the typical PC enthusiast’s knowledge on the subject.

Time / Availability in the Market: Pentium M has been on the market for nearly three years; the Turion perhaps a year. Retail availability for the Pentium M processors has been pretty good for a while. The Turions have been much more difficult to find until quite recently.

Many system integrators have been offering desktop systems based around the Pentium M for some time, usually at a significant price premium over similar performance standard desktop systems. Now, quiet-oriented system integrators could also offer extremely quiet, high performance, and power-efficient Turion-based desktop systems at prices very competitive with equivalent performance Athlon 64 systems, and much cheaper than equivalent Pentium M systems. It’s may only be a matter of time before some adventurous retailer begins to offer Turion-based desktops.

Like most products in the tech world, Turion on the desktop will not enjoy that long a run. There are two spoilers on the horizon:

  1. AMD will release a new socket (named S1?) for their dual-core Turion 64. It will not be compatible with any other sockets.
  2. Sockets 754 and 939 will be replaced with a single new socket called AM2, which will become AMD’s universal socket/pin configuration for all their desktop CPU models.

Both of these new sockets and processors for them will likely be released by mid-year. They signal the beginning of the end of 754-pin Turion 64s and socket 754 motherboards.

AMD’s position on 754 socket processors (Turion 64, Athlon 64 and Sempron) is that as long as there is enough demand from customers, they will keep making them. This is the key to the continued production and availability of socket 754 motherboards as well. Given market realities, it seems safe to say that Turion 64 on the desktop will be doable for at least the rest of 2006. (After all, socket A motherboards are still around in retail, years after cessation of development on processors in this form factor.)

The low power computing scene will undergo many other significant changes this year. Intel will release
Conroe, their upcoming low power desktop CPU, and AMD will probably offer
reduced power consumption versions of Athlon 64s and X2s. Specialized desktop motherboards to take advantage of Intel’s and AMD’s dual-core mobile processors may also be developed by the likes of AOpen, DFI and Shuttle. High efficiency computing
is coming to the masses sooner or later; the Turions get you there
a bit sooner at relatively low cost.

Many thanks for the generous assistance of:

AMD for the
Turion 64 samples
AOpen for supplying
the i915Ga-HFS motherboard
DFI for supplying
the LanParty UT NF3 250GB motherboard
NCIX for supplying
the MSI RS482M-IL motherboard
Intel for supplying
the M770 sample and the LTS-25 current sensor
Corsair and OCZ for supplying
the RAM used in the test systems

*

Additional resources and further reading about Turion 64:

The Tech Report – Intel’s Pentium M 760 versus AMD’s Turion 64 ML-44: Mobile Goliath meets would-be David
Laptop Logic – Clash of the Titans: Dothan vs Turion

Mobility Guru – The Turion 64 Inside Story
AMD – Turion 64 Competitive Comparison

SPCR Forum – Turion supported desktop motherboards
2ch BBS – Turion 64 Motherboard In/Compatibility List
A DIY Report – Building a C’n’Q Home Theater PC with a Turion 64

* * *

Discuss this article in the SPCR Forums.

Silent PC Review is reader-supported. When you buy through links on our site, we may earn an affiliate commission. Learn More

Leave a Comment

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