AMD AM2 product line (AM2 X2 5000+) Evaluation

Product: AMD AM2 product line (AM2 X2 5000+)
Author: Sean Kalinich
Manufacturer: AMD
Spelling and Grammatical editor: Carlos Echenique, Jason Hambly, Paul Mercer, Sean May.

Introduction:

AMD, Intel, Green, Blue, it seems like the war between these two CPU manufacturers has always been there. Playing a game of leap frog with CPU Speeds, power consumption etc. Recently it seems that AMD has begun to outpace the charging rhino of Intel. Starting with the release of the Opteron Series x86-64 CPUs, AMD went in a direction that most industry analysts (in their short-sightedness) thought was doomed to fail. Combining both x86 and 64-bit instructions into the same CPU, they also moved away from the traditional architecture of placing the memory controller in the northbridge, and planned ahead for dual core and many, many other items that have catapulted them into the enthusiasts and gamers hearts and PCs.
Now AMD has made another move, the move to the newer DDR2 standard. This evolutionary move is the AM2 CPU. AMD has been nice enough to send us one for evaluation. Let’s take it out on the track for a few laps.
Architecture:
220, 221 what ever it takes…  

Most of you are familiar with AMD’s architecture. For those of you who are not, you can read up on it here on AMD’s website.

Specifications:
COMMON TRAITS FOR AM2 PROCESSORS:

Manufactured:	Fab 30 in Dresden,Germany
Process Technology:	90-nanometer DSL SOI(silicon-on-insulator) technology
Packaging:	Socket AM2 (940-pin organic micro PGA)
HyperTransport technology:	Supports single HT link – up to 8.0 GB/sec per link bandwidth
Memory:	DDR2 memory controller
Effective data bandwidth:	Up to 12.8 GB/sec dual channel memory bandwidth Total CPU bandwidth: Up to 20.8 GB/sec (HyperTransport + Memory bandwidth)
Memory Speed – FX and X2:	DDR 2 memory up to and including PC2 6400 (DDR2-800) unbuffered
Memory Speed – A64 and Semperon:	DDR 2 memory up to and including PC2 5300 (DDR2-667) unbuffered
Common Features Added:	AMD Virtualization technology
Chipsets:	ATI, NVIDIA, SiS and VIA

AMD ATHLON 64 FX-62 PROCESSOR TECH SPECS:

Frequency / Cache Sizes:	2.8GHz w/ 1MB L2 cache-per-core (2MB total L2 per processor)
L1 Cache Sizes:	64K – L1 instruction + 64K – L1 data cache per core (256KB total L1)
CPU to Memory Controller:	same as CPU core frequencies
Memory Controller:	Shared integrated 128-bit wide memory controller
DDR2 Memory Supported:	Up to and including PC2 6400 (800MHz) DDR-2 memory
HyperTransport Links:	1
HyperTransport Spec:	2GHz (2x 1000MHz / DDR)
Effective data bandwidth:	20.8 GB/Sec {8GB/sec HyperTransport Link + 12.8GB/sec memory}
Packaging:	Socket AM2 organic micro-PGA
Fab location:	AMD’s Fab 30 wafer fabrication facility in Dresden, Germany
Process Technology:	90nm (.09-micron) Silicon on Insulator (SOI)
Approximate Transistor count:	227.4 million
Approximate Die Size:	230mm2
Nominal Voltage:	1.35-1.4V
Max Thermal Power:	125 W
Max Ambient Case Temp:	55-63 degrees Celsius
Max Icc (processor current):	90.4A
Min P-State (with C’n’Q):	1.0 GHz
Nominal Voltage @ min -state:	1.1V
Max Thermal Power @ min -state: >	38.0W
Max Icc @ min -state:	31.8A

AMD ATHLON 64 X2 5000+ PROCESSOR TECH SPECS:

Frequency / Cache Sizes:	2.6GHz w/ 512KB L2 cache-per-core (1MB total L2 per processor
L1 Cache Sizes:	64K- L1 instruction + 64K – L1 data cache per core (256KB total L1)
CPU to Memory Controller:	same as CPU core frequencies
Memory Controller:	Shared integrated 128-bit wide memory controller
DDR2 Memory Supported:	Up to and including PC2 6400 (800MHz) DDR-2 memory
HyperTransport Links:
HyperTransport Spec:	2GHz (2x 1000MHz / DDR)
Effective data bandwidth:	20.8 GB/sec [8GB/sec HyperTransport link + 12.8GB/sec memory]
Packaging:	Socket AM2 organic micro-PGA
Fab location:	AMD’s Fab 30 wafer fabrication facility in Dresden, Germany
Process Technology:	90nm (.09-micron) Silicon on Insulator (SOI)
Approximate Transistor count:	153.8 million
Approximate Die Size:	183mm2
Nominal Voltage:	1.30-1.35V
Max Thermal Power:	<89 W
Max Ambient Case Temp:	55-70 degrees Celsius
Max Icc (processor current):	66.2A
Min P-State (with C’n’Q)	1.0 GHz
Nominal Voltage @ min -state:	1.1V
Max Thermal Power @ min -state:	31.0W
Max Icc @ min -state:	25.5A

DETAILS ON OTHER AM2 PROCESSORS:

All AMD64 CPUs	128KB of L1 cache (per core)
Athlon 64 X2:
Nominal Voltage:	1.30-1.35V for standard 89W processors
“EE” Nom. Voltage:	1.20-1.25V for energy efficient 65W processors
“EE/SS” Nom. Volt.:	1.025-1.075V for energy efficient, small form factor 35W processors
Die Size:	230mm2 for 4800+, 4400+ and 4000+
Die Size:	183mm2 for 5000+, 4600+ and 4200+, 3800+
Transistors:	~227.4 Million for 4800+, 4400+ and 4000+
Transistors:	~153.8 Million for 5000+, 4600+ and 4200+, 3800+
L2 Cache Size:	<1MB per core for 4800+, 4400+ and 4000+
L2 Cache Size:	512KB per core for 5000+, 4600+ and 4200+, 3800+
Frequency:	5000+ @ 2.6GHz, 4800+/4600+ @ 2.4GHz, 4400+/4200+ @ 2.2GHz, 4000+/3800+ @ 2GHz

Athlon 64 and Sempron:

Nominal Voltage:	1.35-1.4V for standard 62W processors
“EE” Nom. Voltage:	1.20-1.25V for energy efficient 35W processors
Die Size	103mm2
Transistors:	~81.1 Million
Athlon 64 L2 cache:	512KB
Athlon 64 Freq.:	3800+ @ 2.4GHz,  3500+ @ 2.2GHz
Sempron w/256KB L2:	3600+ @ 2GHz,  3400+ @ 1.8GHz, 3000+ @ 1.6GHz
Sempron w/128KB L2:	3500+ @ 2GHz,  3200+ @ 1.8GHz

“Modular” Design:
I think the couch would look better over by the window…

As I mentioned in my rambling introduction, AMD has great forethought in the way they design their CPUs. With the Athlon64, Turion, and Opteron line it was planning for Multi-Core. They were designed from the beginning to be multi-core; the same can be said for the rest of the design. If you look at a picture of an AMD CPU you will notice that the memory controller is offset just a little, this allows for an easy swap of this part of the CPU. Instead of a complete redesign to give you access to new memory, AMD can remove the old and snap in the new.
Evolutionary Move:
Many of you are thinking of the AM2 as a new advance in technology, while in many aspects this is correct. What it is not, is a leap forward. Instead, the AM2 is an evolution of the Athlon 64; this is similar to when AMD moved the Athlon Slot-A to Socket-A, or when they moved the Thunderbird cores to use PC133 memory. Before this point DDR2 speeds and latencies were not where they needed to be. Now with DDR2 800 and DDR2 1066 CL3 and 4 out and readily available, the move makes more sense. AMD can leverage the increased bandwidth of DDR2 and the Athlon AM2’s on-die memory controller to provide you with a better platform. AMD also moved to a 940 Pin Socket for the A64, although you will not be able to put a 940-pin Opteron in the AM2 Socket, AMD has changed the pin out to prevent the enterprising and outright stupid from attempting this.

Energy Efficient models available at 65 Watts max power:Power:
Power is everything in PCs. If you build a system that is power starved you will have an unstable system, yet with the increasing power demands of today’s hardware comes increased heat. To strike a balance, AMD has been doing some real work on reducing the power needed to push their CPUs. Announced, in conjunction with the AM2 release, is a series of Energy Efficient CPUs that operate at 65 watts max.  Certain CPUs will be available in 35 watt max models. This does not mean that they run around 65 watts unless you push them hard, this means that if you have them running flat out you might top out near 65 watts. The AM2 X2 5000+ that we tested was not an energy efficient model, yet even so it still runs at a maximum of 89 Watts. AMD has said that all future X2 CPUs will run at this wattage. With the rising cost of electricity, the improvements in performance-per-watt offered by the AM2 are a welcome change. The Energy Efficient Models are listed below:

AMD Athlon 64 X2 processors:

4800+, 4600+, 4400+, 4200+, 4000+, and 3800+

Energy Efficient models available at 35 Watts max power:

AMD Athlon 64 X2 processor:	3800+
AMD Athlon 64 processor:	3500+
AMD Sempron processors:	3400+, 3200+, and 3000+

Heat:
Yeah but it’s a dry heat…

As mentioned above, with increased power you have increased heat, the opposite is true for reducing your power. With the lower wattage CPUs, AMD has produced a CPU that does not double as a small furnace. These CPUs run comparatively cool and as can be seen below, run cooler than their older siblings run. The X2 5000+ that we are testing here today runs off of only 1.3v.  That is around .05v less than the 4800+ and is 200MHz faster by clock speed. The AM2 maintained a cool 38c under load in our testing.  When we added water cooling, it stayed below 34c. Ambient temperature was 23c. By comparison, the Opteron 170 at stock, on air cooling tops out, under load, at 44c and the XE840 comes in at an egg frying 72c under load, using the stock cooler, and 64c with water cooling. For cooling, AMD sent the same type of cooler that comes with the current FX series CPUs. For those of you with third party cooling solutions there is good news and bad, if your hold down bracket is single tab-per-side, it should work. The three tab-per-side brakets will not. AMD has also changed the placement of the screws on the mounting bracket. the current two screw solutions will not work as there are now four screws that hold the mounting bracket to the board.

Performance:Hardware Virtualization:
This is a huge leap forward, for those of us that run multiple operating systems or who test software and would like to limit the number of actual computers we work on, hardware virtualization is a godsend. This technology allows you to run virtual operating systems at the hardware level instead of through a software emulated layer. Although the current virtual OS applications do not support this feature, it will happen. The days of losing performance and productivity because the emulated hardware could not keep up appear to be a thing of the past.

This is what you all want to know about, how well this CPU will perform.  Will it be worth the money?  Should I start taking lunch money from the other kids for this?  …well maybe not that. I know that with all of the talk and rumors around the Internet, performance is what you want to know about.

Methodology:
For this evaluation I wanted to test the CPUs with as many real world tests as I could. There will be very few synthetic benchmarks and no time demos in this evaluation. They do not serve to show off real performance.  For comparison I have chosen a competing CPU as well as an older CPU from AMD. I would have liked to run these tests against the new Conroe CPUs, however at the time of writing, no samples were available.
For testing I used the fastest Intel CPU available to me; the Extreme Edition 840 which was run overclocked to give it the faster 1066 FSB and ran at 3.72 GHz (266×14). For my AMD choice I used an Opteron 170 that runs smoothly at 2.6GHz.

AMD test platforms:
AMD AM2 X2 5000+
Asus M2N32-SLI Deluxe
1GB Corsair XMS2 DDR2-1066 (2×512) at DDR2-800 4-4-4-12
2x Leadtek PX7900GTX TDH 512MB Extreme (in Single and SLI for game testing)
2x 74GB Western Digital Raptor 10,000 RPM SATA 150 HDDs (RAID0)
Sony 16x DVD-ROM   <p”>SilverStone ST75ZF Zeus 750Watt Quad Rail PSU Microsoft Windows XP Professional and Windows XP x64 Edition

Opteron 170 (clocked to 2.6 GHz)
Asus A8N32-SLI Deluxe 1GB GeIL One DDR 400 @217MHz  2-3-3-5
2x Leadtek PX7900GTX TDH 512MB Extreme (in Single and SLI for game testing)
2x 74GB Western Digital Raptor 10,000 RPM SATA 150 HDDs (RAID0)
Sony 16x DVD-ROM
SilverStone ST75ZF Zeus 750Watt Quad Rail PSU
Microsoft Windows XP Professional and Windows XP x64 Edition

Intel Platform:
Intel Pentium Extreme Edition 840 @3.72 GHZ (266×14) Hyper-Threading enabled
Asus P5WD2-E Premium
1GB Corsair XMS2 DDR2-1066 (2×512) at DDR2-1066 5-5-5-15 2T
Leadtek PX7900GTX TDH 512MB Extreme
2x 74GB Western Digital Raptor 10,000 RPM SATA 150 HDDs (RAID0)
Sony 16x DVD-ROM
SilverStone ST75ZF Zeus 750Watt Quad Rail PSU
Microsoft Windows XP Professional and Windows XP x64 Edition

My Choice of Intel platform was due to an unfortunate occurrence. My two Intel SLI boards were unavailable. One died a tragic death due to a bad PSU and the other was on loan to another site for testing. This left only the Asus P5WD2-E Premium as an acceptable mainboard for comparison. Unfortunately, I was unable to conduct any SLI tests on this board and only single card scores are shown below. Once either of the two Intel SLI boards that we have in-house is available, I will add in the SLI scores from that board.
Settings for the 7900GTX, unless otherwise noted, were global profile for all games with Transparency AntiAliasing set to Supersample.
In addition to the software used for testing I also wanted to simulate a real world environment, for this I installed the following commonly used applications:
Microsoft Office 2003
SpeedFan
Diskeeper 10
Avast Home Edition
Acrobat Reader
A HighSpeed PC TechStation was used for all testing.
Corsair’s Nautilus500 was used for cooling the overclocked Opteron and the Intel XE840.

The Tests:

Gaming: These four game titles have heavy physics and particle effects throughout the game. They are in many cases capable of bringing even the fastest system to a crawl.

Need For Speed Most Wanted:
Although this is an older title, the physics involved with high speed driving that the game emulates can have a serious impact on performance and frame rate. Add into all of this physics fun, multiple cars with their own physics, lighting effects, etc., and you not only have a visually impressive game, but also one that can put a hurting on your CPU. For testing I ran the Challenge test number Nine, this track is very winding with many scenery changes, and in many cases the weather changes during the run altering the lighting and shadows.

Here, as in most games, the AMD AM2 is simply the faster CPU. The feel of the game was smoother and more responsive on the AM2. Considering the AM2 was run stock against two overclocked CPUs and still kept around a 5 frame per second advantage.

F.E.A.R.
If you have not heard of this game you need to put down the Funions and get out of the house. F.E.A.R. is a Monolith Productions Game. Monolith has an inclination towards the darker side of gaming with titles like Blood and Blood 2.  They incorporate a supernatural and creepy feel into an immersive gaming experience. F.E.A.R. combines a good story line, impressive physics, startlingly intelligent AI and heavy particle effects into a great game. For my F.E.A.R. testing I sampled frame rates in the Interval -02 – Insertion level from the entry at the gate to the point where you are blown out of the window by the little girl. This section has quite a few combat sequences with some good AI routines. The enemy clone soldiers perform complex fire and maneuver tactics. Visual setting were 1600×1200 with all other settings at Maximum, Volumetric Lighting and soft shadows were also enabled, Computer settings were at Maximum and Difficulty was set to Extreme , this allowed from more complex AI routines and more cunning enemy movements

Our F.E.A.R testing again reveals the AM2’s ability to handle complex gaming instructions faster.

Call of Duty 2:
Another game with quit a bit of action, Call of Duty 2, is a bar fight of a game, with enemies around every corner and allies that jump right in front of you while you’re shooting. The scenery around you is quite well done, with many smoke and particle effects, this game can put a serious dent in a system with a weak CPU. Again for testing, I chose settings that I feel would impact the CPU the most. 1600×1200 with everything set to its highest level and difficulty set on Veteran. The level used for testing was the Demolition portion of the Winter War level. This was played from start to the destruction of the German building.

Call of Duty 2 shows what the other game tests have already shown, even against an overclocked CPU the AM2 is able to maintain its lead.

Oblivion:
ElderScrolls IV Oblivion is a game that everyone is talking about (and not all of it good). This game has a way of just plain putting a hurting on your system. Both ATi and nVidia have had to patch their drivers just for this game. There is also a heavy toll on your CPU as the game seems to stutter and drag on slower CPUs. Memory is also important, and either not enough or not fast enough memory, will impact your gameplay. The display settings for Oblivion were 1600×1200, with all other detail settings on their highest setting. AA was disabled as Oblivion does not support HDR and AA at the same time. The testing level I used was the opening level where you are attempting to escape through the Emperor’s escape route. I followed this path until I reached the second wooden door.

Oblivion was a pain to test; there is no true support for XP x64 or the 7900GTX, so the auto hardware detection would not work properly, requiring the settings to be manually set. In addition, because there is no support for the 7900GTX I had to manually enable the SLI function. Without this I saw no benefit from SLI at all. Once I had the settings correct, the game had an annoying habit of cutting to a blank screen. The menus would work but there was no display. I was able to tweak some of the settings, and setting the renderinfo.txt to “read only” helped with this. Once I had this problem out of the way I was able to complete my testing.

Again, as in the other tests, the AM2 has a slight advantage in speed against the overclocked Opteron and XE840.

Cinebench:

A new test at Planetx64 Cinebench is a nice tool that can be used to gauge the rendering power of your system. Cinebench is a product of Maxon, the same people who developed Cinema 4D. According to Maxon, Cinebench is a real-world test for rendering power and is able to accurately show off a particular system’s ability render in real-time. Cinebench is especially CPU intensive, and is able to show scaling from one to up to 16 CPUs. I chose this test to show off not only the raw rendering power of the AMD AM2 CPUs but also the benefit of the x64 instructions built into all AMD Athlon 64s. (See our Lightwave 3D 8.5 review for more information.)
Here, the raw power of the AMD CPUs are evident, easily outperforming the overclocked XE840 in the rendering tests. The AM2 still keeps its head above the overclocked Opteron 170, even with the smaller cache size.

SuperPi:
SuperPi is a test that just about every PC enthusiast and overclocker knows about and uses. This test relies heavily on your CPU and CPU-to-memory speeds to calculate Pi out to some impressive decimal places. For these tests, since all of the CPUs tested are Dual Core, I ran two tests simultaneously. This shows the stability between the cores.

	XE840 @3.72		Opteron 170 @2.6		AM2 X2 5000+
	Core 0	Core 1	Core 0	Core 1	Core 0	Core 1
Windows XP Pro	29:29.046	29:33.234	29:45.348	30.54.297	31:37.127	31:24.419&
Windows XP x64	38:50.734	38:32.766	31:45.078	31:45.969	30:36.828	30:33.735

The scores here show the benefits a larger CPU cache can have. With the 1066 FSB, the XE840 is able to outperform the AM2 and the Opteron 170. The XE840 has two megs of cache, one meg for each CPU core, just as the Opteron 170 does. The relatively small instructions for the SuperPi calculations are easily stored in cache for execution. However it still requires a 500MHz OC on the XE840, bringing it to 1.1GHz faster than the AM2, to achieve this win. An interesting issue arose in our 64-bit testing on the XE840. It would seem that the XE840 is not as nimble in XP x64 when running SuperPi.

Easy DVD to DivX SVCD converter:
For DivX encoding, I used Easy DVD to DivX SVCD converter, this application is capable of using multiple CODECs for both audio and video encoding. DivX 6.2.2 with CD Quality Audio was used for all encoding.

	XE840 @3.72	Opteron 170 @2.6	AM2 X2 5000+
Windows Xp Pro	0:43:42	0:57:48	1:02:35
Windows XP x64	0:39:29	1:01:12	0:40:21

Again, in encoding, the larger cache combined with the faster bus and memory speeds helps the Opteron and the XE840. The faster RAID solution on the Intel i975x also helps the XE840. These score are only possible when the XE840 is overclocked. At stock speeds the XE840 is not able to keep up at all.

Pricing:
The pricing on the new AM2s has been on the Internet for a while now. Some of it is accurate, some is way off of the mark. I have included below the pricing direct from AMD:

AM2 PRICING:

FX-62:	$1,031
X2 5000+:	$696
X2 4800+:	$645
X2 4600+:	$558
X2 4400+:	$470
X2 4200+:	$365
X2 4000+	$328
X2 3800+:	$303
Athlon 64 3800+:	$290
Athlon 64 3500+:	$189
Sempron 3600+:	$123
Sempron 3500+:	$109
Sempron 3400+:	$97
Sempron 3200+:	$87
Sempron 3000+:	$77

AM2 PRICING (ENERGY EFFICIENT MODELS):

65 WATTS:
X2 4800+	$671
X2 4600+	$601
X2 4400+	$514
X2 4200+	$417
X2 4000+	$353
X2 3800+	$323
35 WATTS:
X2 3800+	$364
Athlon 64 3500+	$231
Sempron 3400+	$145
Sempron 3200+	$119
Sempron 3000+	$101

Conclusion:
AMD has proven themselves again and again as a manufacturer that can deliver innovative products, performance, and stability.  With the AM2 we see another indication that AMD is not only thinking ahead, but also ensuring that they provide the consumer a quality product. The decision to wait to implement DDR2 support until the technology matured was a solid one. This is what we have come to expect from AMD, good solid products.
By comparison Intel has chosen to address the latency issues inherent in their designs, not by improving on the original concept but by stuffing as much cache into the CPU as possible. It seems they have also ignored the cost of adding this extra cache in terms of power and heat. Intel says Conroe will be a departure from this, but Conroe is not here yet.

Is the AM2 a revolutionary new CPU that will amaze you? Not really, if you are looking for that, prepare to be disappointed. What AM2 represents is an evolution of the Athlon 64 line. AMD has done quite a job, reducing power requirements, heat output, and changing the way the CPU performs. This evolution of the AM2 combined with new chipsets and improved DDR2, IS a revolutionary change. The performance that can be had from using the new AM2 platform is going to impress.
In our testing we found that the AM2 X2 5000+ was able to outperform our overclocked Opteron 170 and XE840; this is with ½ the cache of those two CPUs. This is no small feat. Generally when you compare two CPUs of the same clock speed but different cache sizes, the one with more cache will win. This was not the case in with the AM2. Also, DDR2 RAM still does not have latencies that come close to DDR. The DDR2-800 we tested runs at 4-4-4-12 while our DDR-400 ran at 2-3-3-5 and was capable of 1.5-3-3-7. So with less cache and a higher latency the AM2 X2 5000+ we tested still was faster than our overclocked CPU by a small margin. The difference between the XE840 and the AM2 X2 5000+ was even more apparent.
With static instructions to work with (such as in SuperPi and DivX encoding) the XE840 was able to leverage Netburst and the larger cache to run faster. However, this was only in those two tests, and then only with the overclock to the 1066 FSB. This is in no small part due to the work AMD has put into the AM2, and it also ties in with the new platforms from both nVidia and ATi. Although at the time of writing, the new mainboards from ATi were not available, they will be here soon. And I would not be surprised to see the same increase in performance.
Should you upgrade to AM2? You will have to decide that for yourself. I can say that with the across-the-board improvements I have seen in my testing, I would say the AM2s will be well worth the upgrade. To put it plainly, AMD has given us (the enthusiasts) another rock solid platform for us to pound away on.

Acknowledgements:
I would like to thank Damon Muzny at AMD for providing the AM2 X2 5000+ and other hardware for this evaluation.