Hardware Upgrade - Logo
« Back   -  Print
Quad FX: The First Quad Core AMD Platform
Quad FX: The First Quad Core AMD Platform
"Inspired by enthusiasts and built for enthusiasts, we take a look at AMD's new dual-Athlon 64 FX Platform: Quad FX. The first version of a brand new platform, Quad FX's future is closely tied to K10. "
Page 1 - Introduction

“QuadFather”: up until a few weeks ago this is the informal name AMD has used to refer to their new dual-Athlon 64 FX platform. The name, coined off of Francis Ford Coppola’s famous film “The Godfather”, is meant to signify the power and respect associated with AMD’s new desktop platform. The “Quad” stands for the four Athlon FX cores that are the main attraction of the platform. Although we like the name “QuadFather” just as much, the platform is officially known as “Quad FX”.

Armed with two 1207 pin sockets and a total of four cores, the Quad FX platform supports up to 12 SATA-II hard disks and 4 PCI Express 16X electrical slots; remarkable numbers for a desktop platform. Quad FX is unique primarily because it blends workstation features with desktop priced components. It allows the creation of a personal workstation without the need to buy workstation motherboards, processors, or memory.

Quad FX is aimed towards a smaller sector of the market compared to many other AMD platforms. AMD has said on many occasions that Quad FX is enthusiast-oriented and inspired; however, this does not necessarily mean that the platform is limited to enthusiasts only. The chief feature of the platform, the support for four cores, implies that Quad FX is meant for customers who use mutli-threaded applications that benefit from multiple cores. Programs that take advantage of multiple cores range from 3D games to professional rendering applications. The decision of whether 4 cores are actually needed greatly depends on the applications that will be run on the system on a daily basis.

Every program might not greatly benefit from multiple cores, but system performance still improves as a whole. More cores increases efficiency and allow multiple background processes to be run at the same time. In addition, more applications, particularly demanding ones, can be executed parallel to each other.

slide_analyst_day.gif (41117 bytes)

A shy Quad FX, at that time mostly known to us as 4x4, was first formally introduced to members of the press during AMD’s Analyst Day in June of 2006. Small hints of information regarding their next generation architecture, internally known as K10 but more widely recognized as K8L, were also released at that time. Although the release date hasn’t been written in stone, new processors based on the K10 are expected to ship in the second half of 2007.

Quad FX is not a temporary platform from AMD. Currently, a widespread misconception is that Quad FX is just an attempt from AMD in order to keep up to pace with Intel’s quad-core processors. However, this does not actually turn out to be the case. Quad FX has been tailored specifically for extremely high performance systems and is expected to be updated when the K10 architecture is released. A key feature of Quad FX is that it allows for seamless upgrading. AMD seems to be pretty confident in 4x4’s ability to upgrade CPUs as they have already brought up the possibility of an octal core system. We will discuss more regarding this subject further into the article.

Page 2 - The Athlon 64 FX 7x Series

In the weeks leading up to the debut of Quad FX, a lot of speculation regarding the pricing of the processors that will be used in the platform has hit the web. AMD has held true to their promise that two of the FX-7x processors will be available for well below the thousand dollar mark. Here is the official pricing of the Athlon 64 FX-7x series. As always, all prices are for shipments of 1,000 processors.

Processor Clock Cost
Athlon 64 FX70 2.6 GHz 599 USD
Athlon 64 FX72 2.8 GHz 799 USD
Athlon 64 FX74 3 GHz 999 USD

Each price above is for a pair of processors. The package will also come with a heatsink/fan combination for each CPU. At the moment, AMD will only be offering the Athlon 64 FX-7x series in pairs. In addition, all future FX processors will be designed specifically for use in the Quad FX platform. This means that single socket FX processors are destined to disappear from the market. The Athlon 64 FX-62, based on socket AM2, will therefore be the last AMD processor from the FX series meant for single socket desktop systems.

The phasing out of single socket FX solutions also signals the rise of Athlon 64 X2 CPUs as the new flagship AMD solutions for single socket AMD desktop platforms. It would not be too surprising if AMD’s next major announcement will concern new Athlon 64 X2 CPUs. The new X2s will replace the current Athlon 64 FX-6x series.

Taking a closer look at the Athlon 64 FX-7x processors, it is evident that the Athlon 64 FX-72 is very similar to the Athlon 64 FX-62. Both share the same clock frequency and also have 1 Mbyte of L2 cache per core. The main difference is in the sockets they use; the newer FX-72 being based on a 1207 pin design while the older FX-62 is based on Socket AM2. Since AMD will be selling the Athlon 64 FX-72 processors in pairs, by halving the pricing of the FX-72, we get a theoretical price for a single FX-72 processor of about $400. Place this figure against the $713 required to buy an Athlon FX-62 and it is plain to see just how competitive AMD is being with the pricing of their FX-7x processors.

The $999 MSRP for the Athlon 64 FX-74, which is the current top-of-the-range CPU for the Quad FX platform, is identical to the price of the Core 2 Extreme QX6700; currently, Intel’s only LGA 775 quad-core solution. Looking at Intel’s roadmap, their next quad-core processor, the Core 2 Quad X6600, will be released in January of 2007 (around a month away) at an official price of 851 USD. As a result of Intel’s choice of pricing, the FX-72, at least from data currently available, will be in the same price segment as the Intel Core 2 Quad X6600. The Athlon 64 FX-70 will compete with any possible quad-core processors released from Intel at a lower price around the 599 USD range. It is important to remember that although we are pretty certain the prices used above are final, since they are provided by Intel’s Roadmap, pricing is always open to change before the products are actually released.

cpu_1_s.jpg (40042 bytes)

 

cpu_2_s.jpg (19529 bytes)

cpu_3_s.jpg (37031 bytes)

The new 1207 pin CPU Socket is the same interface used for Opteron Revision F CPUs. Opteron Revision F CPUs made their advent into the market last August boasting support for DDR2 memory, a first for AMD server processors. Even though they do have the same number of pins as Athlon 64 FX-7x CPUs, Opteron Revision F CPUs are not compatible with Quad FX.

Page 3 - The Asus L1N64-SLI WS Motherboard

In order for us to properly compare AMD’s Quad FX platform against a Quad-core Intel system, it is necessary to factor in the cost of the motherboards. As both platforms use traditional DDR2-800 memory, the cost of memory is the same for both systems. The same cannot be said for motherboards.

At the moment the only manufacturer of Quad FX motherboards is the famous Taiwan based company ASUS. The motherboard is expected to be available for a price point of 300 USD. For Intel quad-core solutions, the price of the motherboards varies depending on the chipset. Pricing can go from $200 for a solution with the Intel 975X chipset to $300 for the NVIDIA reference board equipped with the 680i SLI.

From the information available to us right now, this means that the maximum amount of extra money consumers would end up spending buying a Quad FX motherboard instead of an Intel-based motherboard is $100. It is essential to keep in mind the price differences between the motherboards of the platforms throughout the course of this review as all factors contribute to the final verdict.

Even though ASUS is currently the only Quad FX motherboard manufacturer, the situation is destined to change soon. More manufacturers should be coming out with their own Quad FX motherboards in the coming months, and AMD should be introducing their own Quad FX chipset which will support Crossfire technology.

scheda_4_s.jpg (69014 bytes)

In order to fit in all of the features on the motherboard, smaller PCB dimensions were compromised. We weren’t too shocked about the board’s large size considering it comes with two Athlon 64 FX sockets and 4 PCI Express 16x slots integrated onto it. In addition, if four video cards is to even be a possibility, the board must have adequate space. Four DIMM slots are placed between the two CPU sockets. Each processor independently handles two DIMM slots, so with 4 modules of DDR2-800 memory installed, the theoretical memory bandwidth of the system is 25.6 GB/s. This is only a theoretical value that represents 100% efficiency. In a real-world situation, expect the actual bandwidth to be a smaller figure.

scheda_1_s.jpg (55545 bytes)

The cooling of the system is a major issue that ASUS seems to have taken into consideration when designing this board. There are a total of three fans on the motherboard. The two chipsets are cooled by connecting two heatpipes to a radiator that dissipates the heat.

scheda_3_s.jpg (70107 bytes)

Above are the four PCI-Express 16x electrical slots. The single 32 bit PCI slot and PCI-Express 1x slot can also be seen in the photo. Although the physical connectors remain the same, two of the PCI Express 16x slots only work at 8 lanes.

scheda_2_s.jpg (64940 bytes)

The board comes with a total of 12 SATA-II connectors. Half of them are colored red while the other half are black.

Page 4 - nForce 680a SLI: The "New " chipset from NVIDIA

The nForce 680a SLI is the name of the chipset which NVIDIA has designed for Quad FX. The 680a is based on the nForce 500 series, and so it naturally shares many of the 500 series’ technical characteristics. In reality, the chips are nearly identical save for a few exceptions.

chipset_nvidia_schema.jpg (33462 bytes)

The new Quad FX platform relies on two 680a SLI chips on the same motherboard. With double the number of usual chipsets, as we have said many times before in this article, the two 680a MCPs combine with each other to support four PCI Express 16x electrical slots, 2 of which are only half bandwidth (8 lanes) on a 16x interface. One chipset handles the 32 bit PCI slot while the other chipset handles the PCI Express 1x slot. The coupling of the chipsets also allows for 12 3GB/s Serial ATA channels and up to four Gigabit Ethernet ports.

The nForce 680a chipsets support up to 12 Serial ATA channels. This feature allows the construction of complex RAID configurations. All 12 channels can be used in a single RAID array or alternatively, it is also possible to configure 4 independent RAID arrays.

In 4x4, the Athlon 64 FX processors communicate with each other using a coherent HyperTransport (HT) link. Both chipsets communicate with the processor using two HyperTransport links. The above diagram from NVIDIA is incorrect in the sense that it might lead people into thinking there is a designated processor for which the chipsets communicate with. There is no designated processor that specifically communicates with the chipsets. More realistically, one of the two processors connects directly to one of the two chips using a HyperTransport link.

As AMD is famous for, the memory controller is integrated directly into the CPU. Because of this the 4 memory modules communicate directly with the processor. Using the coherent HyperTransport link between the CPUs, the memory controller of one processor can be accessed by the other processor, thus allowing one CPU to access the other CPU’s data in its DDR2 bank. Data from the L1 and L2 caches can also be requested from the other processor.

The 680i SLI comes with NVIDIA LinkBoost technology. LinkBoost allows the frequency between the SPP and MCP (HyperTransport Link) to be increased by 25%. In addition, this technology also increases the communication speed between the SPP or MCP and the PCI Express bus by 25%.

schema_linkboost_2.jpg (25747 bytes)

In order for LinkBoost to properly work, it is required to use a LinkBoost certified video card. Currently, the only video cards that have proper certification are the GeForce 7900 GTX and the GeForce 8800. If these video cards are detected on either the MCP or SPP, then the clock speeds will automatically be increased.

The 680a SLI also supports SLI Memory. SLI Memory is officially known as EPP, or Enhance Performance Profiles. EPP memory was developed by NVIDIA and Corsair in an attempt to make the overclocking of system memory a less arduous task. EPP memory is only for use with certified platforms.

EPP allows users to overclock both their CPU and memory at the same time. In addition, EPP also handles all of the timings and other factors such as voltage. This can be a blessing for relatively new and inexperienced overclockers.

The last three major features in the 680a SLI all deal with networking. These features have been present in later chipset version of the NVIDIA nForce 5 and 6 series. FirstPacket prioritizes packets in order to allow packets to flow smoothly in the upstream bandwidth of a broadband connection. Less latency-sensitive applications are moved down the list in priority while higher latency-sensitive applications are prioritized higher.

Obviously, not every package can be of higher priority. Users are allowed to adjust applications and games and signal what priority they belong under. Networked games are a good example of a latency-sensitive application that would be treated with higher priority. FirstPacket also reduces the amount of dropped packets, helping to prevent stutters and jitters in VoIP transmissions or video conferences.

firstpacket.jpg (16399 bytes)

Using FirstPacket, the networked video game is treated with higher priority over the FTP transmission. Generally, FTP transmissions tend to be much less sensitive compared to video games.

DualNet is the second networking technology that is featured in the 680a SLI. As the name suggests, DualNet allows two Gigabit Ethernet ports to be used placed side by side in order to handle net transmissions.

The last major networking feature included is TCP/IP Acceleration. TCP/IP includes a dedicated network traffic accelerating processor which is built into the Gigabit Ethernet Controllers. Consumers are given full control over this feature via the NVIDIA Network Access Manager.

Page 5 - Test Configuration

We tried to execute all tests using applications that actually benefit from multiple cores. The Quad FX platform has been placed side by side against Intel’s Core 2 Extreme QX6700 in all benchmarks. These are the processors used in the comparison:

  • Intel Core 2 Extreme QX6700 (bus 1066 MHz, clock frequency 2,66 GHz - cache L2 2x4 Mbytes) - Dual DDR2-800 memory
  • Intel Core 2 Extreme QX6600 (bus 1066 MHz, clock frequency 2,4 GHz - cache L2 2x4 Mbytes) - Dual DDR2-800 memory
  • Intel Core 2 Extreme X6800 (bus 1066 MHz, clock frequency 2,93 GHz - cache L2 4 Mbytes) - Dual DDR2-800 memory
  • AMD Athlon 64 FX74 Socket 1207 pin (clock 3.000 MHz - cache L2 2x1 Mbytes) - Dual DDR2-800 memory
  • AMD Athlon 64 FX72 Socket 1207 pin (clock 2.800 MHz - cache L2 2x1 Mbytes) - Dual DDR2-800 memory
  • AMD Athlon 64 FX70 Socket 1207 pin (clock 2.600 MHz - cache L2 2x1 Mbytes) - Dual DDR2-800 memory

These are the components used in the systems:

  • motherboard Asus L1N64-SLI WS (chipset NVIDIA nForce 680a SLI)
  • motherboard: EVGA 680i SLI (chipset NVIDIA nForce 680i SLI)
  • 4 Gbytes memory DDR2-800 Corsair, model: Dominator TWIN2X2048-8500C5D (4 modules each 1 Gbyte); Timings of 4-4-4-12 2T
  • hard disk Western Digital Raptor 150 Gbytes, configurato con un'unica partizione da 150 Gbytes
  • video card : NVIDIA GeForce 7900 GTX

The operating system of choice was Windows XP Professional 32bit and Windows Vista RC2 64bit. Where possible, we used applications compiled in native 64bit for benchmarking with Vista.

These are the tests we ran on the platforms:

Multimedia

Auto GK 2.27 - Divx - conversione traccia 9 "Kill Bill volume 1"; audio inglese 6ch
    Divx 6.11
    scena 6
    Qualità 75% - no audio

Itunes 7.0.1.8
    conversione traccia audio da 613 Mbytes in formato MP3 192 Kbps

Sony Vegas+DVD 7.0b
    conversione filmato H.264 1080p in Pal DV - video quality best

Sony Vegas+DVD 7.0b
    Render to Other track - PAL DV Widescreen
    traccia hdwatermellon - best quality

Sony Vegas+DVD 7.0b
    Video and Audio Benchmarking

Paint.net
frosted glass - 10 pixel

Mainconcept H.264 Encoder
    Conversione da DV a H.264 High
    Video ECS factory tour - DV 16,836 frames
    720x576

Windows Media Encoder 9 with advanced profiles
    Video ATI factory Tour; 6 min 57 sec
    320x240 - 282 Kbps

DVD Shrink 3.2
    Fahrenheit 9/11
    compresso a 2000 Mbytes

Rendering

Cinebench 9.5
    CPU Benchmark (CB-SEC)
    CPU Benchmark Multiple (CB-SEC)Cinebench 2003

3Ds Max 9
    radiosity 1920x1080
    Underwater_Environment_finished 1920x1080

Povray 3.7 beta
    benchmark

Page 6 - Consumption and Temperature

In order to test power consumption, we measured the total power intake of the system. Where possible, we used the exact same components between the two systems.

  • motherboard: Asus L1N64-SLI WS (chipset NVIDIA nForce 680a SLI)
  • motherboard: EVGA 680i SLI (chipset NVIDIA nForce 680i SLI)
  • 4 Gbytes memory DDR2-800 Corsair, modello Dominator TWIN2X2048-8500C5D (4 moduli da 1 Gbyte); impostazione memoria con timings 4-4-4-12 2T
  • hard disk Western Digital Raptor 150 Gbytes, configurato con un'unica partizione da 150 Gbytes
  • sched video: NVIDIA GeForce 7900 GTX
  • Power Supply : Be Quiet! Dark Power PRO 580 Watt

consumo.png (46565 bytes)

There is little to talk about regarding the power consumption tests and the conclusion is simple: the Quad FX platform has a very high level of power consumption under idle and full load. It is important, however, to take into consideration the different chipsets used on the motherboards and the three fans which are mounted on the Quad FX system. Despite this, as a whole the consumption of the Quad FX system is quite high compared to Quad core Intel systems.

For our temperature tests, we ran 3 sessions of POV-Ray 3.7 multitasking beta benchmark sequentially using Core Temp Beta 0.94 to report the temperatures. The test systems were not enclosed in cases, so it is likely that operational temperatures might rise when in a home environment.

temp_qx6700.gif (2036 bytes)
Intel Core 2 Extreme QX6700

temp_qx6600.gif (2103 bytes)
Intel Core 2 Extreme QX6600

temp_x6900.gif (1481 bytes)
Intel Core 2 Extreme X6800

fx74.gif (2355 bytes)
AMD Athlon 64 FX74

fx72.gif (2353 bytes)
AMD Athlon 64 FX72

fx70.gif (2275 bytes)
AMD Athlon 64 FX70

The temperatures are quite similar throughout, but the AMD solutions run slightly cooler. A factor that might contribute to Intel’s Quad core solutions having higher temperatures is how all four cores are in once processor. For AMD solutions tested, each processor has only two cores.

Page 7 - Windows XP Professional- rendering

3dsmax_1_xp.png (43126 bytes)

3dsmax_2_xp.png (44569 bytes)

cinebench_xp.png (52932 bytes)

povray_xp.png (41846 bytes)

The rendering tests show a clear performance advantage for Intel Core 2 Quad architectures. The only instance in our testing in which Quad FX is able to pull ahead of Intel is in Cinebench, a benchmarking utility based on the 3D software CINEMA 4D.

Page 8 - Windows XP - multimedia

autogk_xp.png (44271 bytes)

dvd_shrink.png (45312 bytes)

Converting a DVD film to DivX and backing up a DVD film are two scenes in which Intel greatly excels with their Core 2 Duo and Quad processors. The Intel solutions generally hold a large advantage while AMD solutions tend to lag behind.

mainconcept_xp.png (44793 bytes)

paintnet.png (43026 bytes)

Although with Main Concept the performance differences between Quad FX and Intel Quad core processors are slightly reduced, Intel still holds the lead.

vegas_1.png (46154 bytes)

vegas_2.png (43980 bytes)

wme.png (46879 bytes)

Quad FX is able to beat out all Intel solutions in the first Sony Vegas, however, in all other tests Intel Quad core solutions are able to take the lead. The margin of advantage, however, is light.

Page 9 - Windows XP - megatasking

For us, it was difficult to find personal productivity applications, except for maybe pure rendering programs, that could take advantage of 4 cores. For this reason we decided on three different scenes to test multitasking in Windows XP Professional. The first scene of megatasking has all four of the following applications running at the same time:

  • Mainconcept H.264 Encode: Conversion from DV to H.264 High definition video ECS factory tour - DV 16,836 frames;
  • DVD Shrink 3.2: backup of film Fahrenheit 9/11; compress to 2000 Mbytes;
  • Sony Vegas+DVD 7.0b: Video and Audio Benchmarking;
  • POV-Ray 3.7: CPU rendering benchmark

 

megatask_1.png (54066 bytes)

megatask_1_perc.png (47323 bytes)

The second graph shows the improvement the systems had with parallel execution of the programs. It is interesting to see how much Quad FX systems were able to benefit when performing the tasks simultaneously.

For this scene of megatasking, we conducted the tests using a different method. Applications were executed in sequence one minute after the application before it. This was done in order to simulate the behavior of a user. We used the following applications in this scene:

  • Conversion AutoGordianKnot - converrt to Divx 6.11 backup of Fahrenheit 9/11;
  • 1 minuto wait- test with Photoshop: convert from pdf to 127 jpeg images at the resolution of 2816x2112 pixel;
  • 1 minuto wait- zip 12,912 files in windows, total file size is 1.79 Gbytes;
  • 1 minuto wait- antivirus scan of 12,912 files, total size is 1.79 Gbytes;
  • 1 minuto wait- itunes conversion from MP3 192 Kbit to 34 file Wav, total file size of 1.46

megatask_2.png (53978 bytes)

megatask_2_perc.png (48021 bytes)

In the case of the second megatasking scene, scalability for the most part was aligned between the processors. The 1 minute start after each application before the next played in favor of dual cores, however, it still did not have as much performance improvement as 4 cores when switching to parallel execution.

The third and final megatasking scene was without doubt the most complex. In the background we left playing an H.264 video, therefore always stressing the core and memory subsystem. The applications were run at the same time for our testing:

  • conversion AutoGordianKnot - convert to Divx 6.11 backup of Fahrenheit 9/11;
  • Mainconcept H.264 Encode: Conversoin from DV to H.264 High definition video ECS factory tour - DV 16,836 frames
  • Zip compression of 12,912 files, total file sizei 1.79 Gbytes;
  • Antivirius scan of 12,912 files, total file size 1.79 Gbytes;
  • Sony Vegas+DVD 7.0b: Video and Audio Benchmarking

megatask_3.png (55368 bytes)

megatask_3_perc.png (44488 bytes)

The third scene turned out to be extremely interesting. The larger the clocks get for the Quad FX platform’s processors, the higher their performance gets. The two Intel Quad core solutions perform the tasks in the same time they would have running all of the tasks alone. Obviously the H.264 film running in the background was a major source for decreasing performance in the tests.

Page 10 - Windows Vista RC2 - rendering

3dsmax_1_vista.png (43300 bytes)

3dsmax_2_vista.png (43983 bytes)

cinebench_vista.png (53565 bytes)

povray_vista.png (42506 bytes)

In our rendering tests run under Windows Vista RC2, we were able to run all of our applications in native 64bit. This was responsible for developing some interesting performance differences, especially for all AMD Athlon 64 FX-7x processors.

Page 11 - Windows Vista RC2 - Multimedia

dvd_shrink_vista.png (45794 bytes)

mainconcept_vista.png (45027 bytes)

paintnet_vista.png (43639 bytes)

vegas_1_vista.png (46931 bytes)

vegas_2_vista.png (44092 bytes)

Running our multimedia applications under Windows Vista produced mixed results. In some instances we were able to receive reductions in time while with others we ended up increasing the time of our benchmarks. The applications used, however, were all 32 bit and therefore identical to the ones used in our Windows XP testing.

Page 12 - Windows Vista RC2 - megatasking and scalability

For testing megatasking under Windows Vista we have used the same method we used previously with Windows XP Professional. The only difference is the applications we are testing with:

  • Mainconcept H.264 Encode: Conversione of DV to H.264 High def video ECS factory tour - DV 16,836 frames;
  • DVD Shrink 3.2: backup of film Fahrenheit 9/11 compress to 2000 Mbytes;
  • Sony Vegas+DVD 7.0b: Video and Audio Benchmarking;
  • Povray 3.7: CPU rendering benchmark .

megatask_1_vista.png (54649 bytes)

megatask_1_perc_vista.png (45596 bytes)

The results are similar to what we saw with Windows XP Professional. The dual core solutions placed last while the Quad core solutions all scored one after the other. The quad core Intel processors performed their tasks in the least amount of time while the Athlon 64 FX-74 completed its tasks around 30 seconds later than the QX6600.

32_vs_64bit.gif (4596 bytes)

The above table shows, in percents, the increase or decrease in performance the processors had in tests going from Windows XP Professional with 32bit applications to Windows Vista RC2, using the same applications but this time in 64bit. The Athlon 64 FX-7x processors are able to obtain greater performance increases in these tests, while the Intel solutions all lose performance in the POV-Ray benchmark.

Page 13 - The Quad Core K10: The Innovation of 2007

In the course of 2007 AMD will release their Quad core architectures based on the K8L architecture, which is also known as K10. In addition to quad cores, the processors will introduce a number of innovations and refinements to AMD technology.

The CPUs are based on the same socket platform as Opteron Revision F CPUs. Although the Athlon 64 FX-7x series are based on the same socket, Opteron revision F CPUs cannot be used for Quad FX.

The AMD K8L architecture will be initially released as a Quad-core Opteron processor. Subsequently the architecture will be adopted for use in Turion 64 and Athlon 64 processors. By the end of next year, K8L should have made its way into all market segments that AMD has solutions in. The AMD K8L architecture is not just meant for quad core CPUs. In time, the architecture will also be used for AMD’s next generation dual-core CPUs.

schema_1.jpg (45479 bytes)

Each core will have an L1 cache of 128 Kbytes which will be split into two independent 64 Kbytes blocks for data and instructions. The L1 cache will be placed side by side with an L2 cache of 512 Kbytes for each core.

In addition to the L1 and L2 cache, AMD has integrated into these processors a third level cache, or L3, that at first will be 2Mbytes. This L3 cache will be shared by all four cores. AMD will increase the size of the L3 cache in future CPUs based on the K8L architecture.

schema_5.jpg (33654 bytes)

AMD has put a considerable amount of effort into optimizing the power consumption for their new quad core architecture. At the moment, AMD is confident enough to say that their initial TDP values are equal to that of dual-core solutions. Because of this, it will not be required to upgrade to a new motherboard, change the PSU, or even change the cooling system if upgrading from a Socket 1207 pin Opteron that is already available on the market to a Quad-core processor. The energy saving method implemented on the new quad core processor will allow each Core to have its operational frequency be independent from the other three cores. The voltage, however, will remain identical between the four cores.

schema_2.jpg (46554 bytes)

These are the main architectural characteristics that have been changed for the K10 in regards to K8:

  • New SSE4 extensions
  • 32-bit instruction fetch
  • Dual 128-bit SSE dataflow
  • Dual 128-bit loads per cycle
  • Instruction fetch bandwidth is doubled from 16 to 32bytes per clock
  • HyperTransport Version 3.0 will be supported for successive versions of the K8L architecture

The moment quad core Athlon 64 FX processors are made available on the market, Quad FX will be able to upgrade from four to eight cores. This will not only double the amount of cores, but will also make the system benefit from all of the above architectural innovations.

quad_fx_schema_2006.jpg (37831 bytes)
architetture Quad FX 2006: due processori dual core

quad_fx_schema_2007.jpg (41445 bytes)
Architecture of the Quad FX 2006: the quad core processor

In the same period AMD will also roll out a new version of their Quad FX platform. The platform will be based on a 1207+ pin socket. Physically the socket will be the same, however, the type of performance increase users will experience going from Socket 1207 to “1207+” is not clear at the moement. One new feature of the revised platform will be that it is able to use HyperTransport Version 3.0. On paper, the new HT version will increase performance by allowing users to take better advantage of the quad core processor. The currently released Quad FX platforms will still be able to work with the new Quad core Athlon 64 FX CPUs, but with the limitation that they will not be able to take advantage of the bandwidth of HyperTransprot 3.0. In order to use HyperTransport 3.0, it will be necessary to upgrade the motherboard.

Page 14 - Conclusion

FX_SLIready_DSDC.jpg (39852 bytes)

From the testing we have conducted today, we have been able to make some key conclusions:

  • In spite of the presence of four cores, the Quad FX platforms do not succeed being faster than the Intel Core 2 Quad solutions in applications that benefit from the presence of multiple processors. The primary reason Intel’s quad core solutions are able to perform better than Quad FX solutions is because of the stronger Core 2 architecture which has already proven itself capable of outperforming the current K8 architecture.
  • In 2007 AMD will release their Barcelona Core, their first quad core CPU. In addition to allowing for a total of 8 cores with Quad FX, the architectural innovations of the new K10 architecture will also be brought to the platform.
  • With 64bit, Athlon 64 FX CPUs have shown stronger scalability and performance compared to current Intel quad core solutions. This is evident particularly in complex megatasking scenarios.

In terms of pure performance, the Intel Quad core platforms are at the present moment faster. With 64bit applications, however, the AMD Athlon 64 CPUs are able to achieve greater margins of performance scalability.

The use of two processors and a complex motherboard has repercussions on the total power consumption of Quad FX systems. The Intel solutions tested today had significantly less power consumption compared to the Quad FX platform. Even though many enthusiasts might consider power consumption secondary to performance, it is still an important aspect to keep in mind.

When looking at the Quad FX platform, it is necessary hold into account the fact that the platform is meant to be a “technology showcase” more than something that is targeted to mainstream users. Quad FX will first launch as a partner exclusive and will only be available for retail through system manufacturers. In time it will become available to DIY users.

At the current moment it is not much of a surprise that Quad FX isn’t able to perform better than Core 2 Quad solutions. The Core 2 architecture has already beat the current Athlon 64 architecture on numerous occasions, and so naturally adding another CPU will not change the picture. It is in the second half of next year we will begin seeing major changes in terms of performance for AMD processors. This is primarily because of Barcelona. With its arrival, the performance gap between AMD and Intel processors will most likely decrease significantly.

With Barcelona, the owner of a Quad FX platform will be able to upgrade their system with two quad core CPUs, adding up to a total of 8 cores. Naturally, the architectural benefits of the new K10 architecture will also increase the performance of Quad FX.

The wisest move AMD made with Quad FX was to aggressively price the new AMD Athlon 64 FX-7x series. In our testing, the Athlon 64 FX-70, which is priced at 599 USD for a pair, obtained performances in multitasking, rendering, and multimedia for the most part always better than the Intel Core 2 Extreme X6800, the fastest dual core CPU Intel currently has to offer. The price of the Intel Core 2 Extreme X6800 also happens to hover around 1,000 USD. Taking this into consideration, it is plain to see that a system equipped with an Athlon 64 FX-70 is able to offer good performance at a relatively lower cost. If more performance is needed at a later time, consumers could simply upgrade to Barcelona in the second half of 2007 when the processor debuts.