A few days ago, AMD officially unveiled their RV670 GPU, the chip at the heart of the ATI Radeon HD 3800 series of graphics cards.  Just in case you missed it on the day of the launch, our coverage of the new Radeon HD 3870 and HD 3850 cards is available right here.  While we did thoroughly explain the RV670’s architecture and explored its performance in that article, we did not tell you that the new Radeon HD 3800 series is but one part of an entirely new platform from AMD dubbed “Spider”.

The rest of the Spider platform is comprised of not only new desktop processors, but core logic as well.  The second piece to the Spider puzzle is AMD’s native quad-core Phenom processor and the third is the new 7-series chipset.

News of Phenom and the 7-series chipset has been circulating for what seems like an eternity.  We’ve shown you glimpses of the platform and talked about its new features in a number of articles and news posts here at HotHardware.  But today we can finally give you all of the details and show you just how AMD’s new processors and chipsets perform.
 
Before we get started, some specifications are in order.  What we have for you in the table below are the main specifications and features of the AMD Phenom 9600 and 9500 processors.  Take a gander at the specifics and then strap yourself in as we take a hands-on look at AMD’s latest and greatest and even check out the Phenom 9700, which won’t be available until early next year.
 

 
AMD Phenom Retail Box

AMD Phenom Processors  Specifications and Features

At the heart of AMD's Spider platform is the native quad-core Phenom processor.  Phenom is the brand name given to processors based on the Agena core, which is virtually identical to the Barcelona core used in the recently released AMD quad-core Opterons.
 

In a quad-core Phenom processor, each of the four cores is outfitted with 64K of L1 instruction and 64K of L1 data cache, for a total of 512K of L1 cache per CPU.  The L2 cache compliment of each core is 512K, for a total of 2MB.  New to the Barcelona and Agena cores is 2MB of dynamically shared L3 cache.   Unlike L1 and L2 caches, which are exclusive to each execution core (data in Core 1’s L2 cache cannot be accessed by Core 3, for example), the L3 cache is shared among all the cores.  Also new to Barcelona and Agena is a 128-bit wide memory controller that can be configured as dual independent 64-bit channels to allow for simultaneous read and write memory operations.

Currently, all Phenom processors will be built in AMD’s Dresden, Germany facility using the company’s 65nm SOI (silicon on insulator) manufacturing process.  Each quad-core die is comprised of approximately 463M transistors (about 357M less than Intel’s quad-core Yorkfield) and is about 285mm2 in size.
 

Looking deeper into the Phenom's native quad-core, single die architecture, we've learn that the cores themselves have been revamped considerably for efficiency and performance. Here are a few of the key salient points of Phenom's new core micro-engines:

 

• A new floating point scheduler now supports 36 128-bit operations
• Support for 128-bit SSE operations, an upgrade from the previous 64-bit architecture
• Two SSE operations and one SSE move can be processed per cycle
• Processor instruction fetch has been increased from 16 to 32 bytes
• Advanced branch prediction with built in a 512-entry indirect branch predictor
• Data cache bandwidth has increased from 1 x 64-bit loads per cycle to 1 x 128-bit loads per cycle
• L2 cache / memory controller bandwidth has been increased from 64-bits per clock to 128-bits per clock
• HyperTransport 3.0 Support for up to 20.8GB/s of raw bandwidth

In addition to the enhancements listed above, Phenom processors support dynamic clock gating on a per-core basis.  Though core voltages won't be managed independently, the clock speed of each core can throttle back when idle, which could in turn provide significant power savings. And AMD's "CoolCore" technology allows for functional blocks of each core to be shut off when not in use, further improving power efficiency.  You may have heard of AMD's "Dual Dynamic Power Management" technology referred to as "split power planes" in the past.

We should note that to fully take advantage of AMD's "Dual Dynamic Power Management" technology, a next-gen platform must be used.  Users that drop a Phenom into an existing socket AM2 platform will not have support for split power planes, because current motherboards lack the necessary support.

On the surface, AMD’s new Phenom processors look just like current Athlons.  The processors use the same packaging and have the same pin configuration.  Underneath that unassuming heat-spreader, however, lays a quad-core die based on the Agena core.

        
An AMD Phenom 9700 Processor

We should note that retail-ready Phenom processors will have different markings than what you see pictured above.  The CPU you see here is an AMD engineering sample 2.4GHz Phenom 9700 that hasn’t been laser etched with the same brandings as a retail-ready processor.  Retail-ready Phenom processors have markings much like current Athlons with the model numbers and other identifying information.

        
Details from CPU-Z with the Phenom 9700

The latest version of CPU-Z shows many of the details we laid out on the previous page.  As you can see, the processor is based on the Agena core with socket AM2+ packaging.  Our particular chip is stepping ‘2’ and supports all of the instructions sets listed in the aptly names ‘instructions’ section.  The processor’s L2 cache configuration is 16-way set associative, but the shared L3-cache is 32-way set associative.

Overclocking The Phenom 9700 Quad-Core Flat-Out

We know many of you are wondering just how much clock speed headroom our Phenom 9700 has left under the hood, so we spent some time overclocking our chip using a Gigabyte 790FX-chipset based motherboard.  We simply upped the processor voltage and increased the HT link frequency until our test machine was no longer stable.  A stock AMD PIB heatsink was used.

We tried various different voltages and memory settings, and unfortunately were only able to take our particular sample up to 2.62GHz – an increase of 220MHz, or 9.2%.  We must stress that our chip was an engineering sample and our overclocking results may not be indicative of actual retail product.  We hope this is the case, because as you’ll see a little later on, AMD is going to have to ramp clock speeds up considerably to compete with the performance of Intel’s high end quad-core processors.

Next up, we have some details regarding the AMD 7 series chipset.  The AMD 7 Series chipset will initially be comprised of the high-end 790FX, the 790X, and 770.  All of the chipsets are being manufactured at 65nm and according to AMD they are twice as power efficient as the previous generation of AMD chipsets.  In fact, the 790FX chipset has only a 10W TDP.  They also support PCI Express 2.0 and HyperTransport 3.0 for massive amounts of bandwidth, and exploit the clock gating and dynamic power management capabilities of AMD’s new Phenom processors.

We should also note that all initial motherboards based on the 7-series chipset will feature the ATI SB600 Southbridge chip.  Sometime next year the SB700 is due to arrive, but initial offerings will use the mature SB600 only.

The 790FX is the chipset that will power many high-end socket AM2+ motherboards.  It is outfitted with 42 PCI Express lanes that can be configured in any number of ways.  Because many 790FX-based motherboards will be touted as 4-way CrossFire capable, however, 32 of those lanes will be dedicated to PCI Express x16 graphics slots, while the rest are shared amongst other expansion slots, peripherals, and the chipset itself.

The 790X chipset is fundamentally very similar to the 790FX, but with fewer PCI Express lanes.  Whereas the 790FX supports 3- and 4-way CrossFire configurations, the 790X will support only traditional dual-card CrossFire.  The 770 takes things one step further down the line and supports only a single graphics card due to its PCI Express lane configuration.
 

Another feature common to all 7-series chipsets will be a new performance tuning application called AMD Overdrive.  AMD Overdrive is similar to NVIDIA’s nTune System Utility and Intel’s Desktop Control Center in that it gives users access to numerous settings for overclocking and performance tweaking.  Using Overdrive, users can alter the frequency of individual cores, memory timings, key voltages, etc. We have yet to test the AMD Overdrive application for ourselves, but screenshots of the interface are available in the slides above.

Motherboards based on the 7-series chipsets have actually been available at a few on-line resellers for a couple of weeks now, a couple of which are pictured below.  First up, we have the Asus M3N32-MVP Deluxe.

 

     


     


     
The Asus M3N32-MVP Deluxe

 

The Asus M3N32-MVP Deluxe is based on the high-end AMD 790FX chipset.  It’s equipped with four PEG slots and supports 4-way CrossFire, but the slots are configured in such a way that 4-way CrossFire is only supported with single-slot graphics cards; with dual-slot cards installed the fourth slot is unusable.  The board also features integrated WiFi, Firewire, HD audio and a heavy-duty chipset cooling solution.  Another interesting feature of the M3N32-MVP Deluxe is what Asus refers to as ‘Cool Mempipe’ technology.  Cool Mempipe is basically a removable, adjustable heatsink apparatus that links system memory to the chipset cooler, for enhanced cooling performance.

We worked briefly with the Asus M3N32-MVP Deluxe and found its BIOS to be very complete.  The board we received, however, was not the same revision that will hit retail shelves, so all of our testing was done using the Gigabyte GA-MA790FX-DQ6.

 

      


     


     
The Gigabyte GA-MA790FX-DQ6

 

Like the Asus board, the Gigabyte GA-MA790FX-DQ6 is built around the 790FX chipset.  And it too supports 4-way CrossFire, but its slot PEG slot configuration is even more limiting because all four slots are adjacent to one another.

We’ve worked with this motherboard for a couple of weeks now.  We used the Gigabyte GA-MA790FX-DQ6 to test Phenom and Athlon processors, and found it to be extremely stable and user friendly.  The BIOS requires users the press CTRL-F1 to access all of its features, but when you do there are options aplenty.  One important feature missing in the current BIOS, however, is the ability to alter CPU multipliers with Phenom processors.  We suspect that’s coming with a future revision though.

Finally, we also have some shots of an MSI 790FX based motherboard on-line.  The MSI motherboard has the most flexible slot configuration in our opinion, as it can accommodate four double-wide graphics cards.  You can see the MSI 790FX board in action right here.

How we configured our test systems: When configuring our test systems for this article, we first entered their respective system BIOSes and set each board to its "Optimized" or "High performance Defaults". We then saved the settings, re-entered the BIOS and set memory timings for either DDR2-1066 (AMD) with 5,5,5,15 timings or DDR3-1066 - 1600 with 7,7,7,20 timings (Intel). The hard drives were then formatted, and Windows Vista Ultimate was installed. When the Windows installation was complete, we updated the OS, and installed the drivers necessary for our components. Auto-Updating and Windows Defender were then disabled and we installed all of our benchmarking software, defragged the hard drives, and ran all of the tests.
 

HotHardware's Test Systems  Intel and AMD - Head To Head

System 1: Core 2 Extreme QX9770 (3.2GHz - Quad-Core) Core 2 Extreme QX9650 (3.0GHz - Quad-Core) Core 2 Quad Q6600 (2.40GHz - Quad-Core)  Core 2 Duo E6850 (3.0GHz - Dual-Core) Core 2 Duo E6600 (2.4GHz - Dual-Core)  Asus P5E3 Deluxe (X38 Chipset) 2x1GB Corsair DDR3-1800 CL 7-7-7-20 - DDR3-1066, 1333, 1600 GeForce 8800 GTX On-Board Ethernet On-board Audio WD740 "Raptor" HD 10,000 RPM SATA Windows Vista Ultimate Intel INF 8.3.1.1009 NVIDIA Forceware v163.75 DirectX Redist (November 2007)

System 2: AMD Phenom 9700 (2.4GHz) AMD Phenom 9600 (3.0GHz)  AMD Athlon X2 6400+ (3.2GHz)  AMD Athlon X2 4600+ (2.4GHz) Gigabyte gA-MA790FX-DQ6 (AMD 790FX Chipset) 2x1GB Corsair PC2-8500 CL 5-5-5-15 - DDR2-1066 GeForce 8800 GTX On-Board Ethernet On-board Audio WD740 "Raptor" HD 10,000 RPM SATA Windows Vista Ultimate NVIDIA Forceware v163.75 DirectX Redist (November 2007)

System 3: AMD Athlon 64 FX-74 (3.0GHz x 2) Asus L1N64-SLI WS (NVIDIA nForce 680a SLI) 5x512M Corsair PC2-6400 CL 4-4-4-12 - DDR2-800 GeForce 8800 GTX On-Board Ethernet On-board Audio WD740 "Raptor" HD 10,000 RPM SATA Windows Vista Ultimate nForce Drivers v15.08 NVIDIA Forceware v163.75 DirectX Redist (November 2007)

Preliminary Testing with SiSoft SANDRA XII  Synthetic Benchmarks

In our custom LAME MT MP3 encoding test, we convert a large WAV file to the MP3 format, which is a popular scenario that many end users work with on a day-to-day basis to provide portability and storage of their digital audio content.  LAME is an open-source mid to high bit-rate and VBR (variable bit rate) MP3 audio encoder that is used widely around the world in a multitude of third party applications.
 

LAME MT MP3 Encoding Test  Converting a Large WAV To MP3

POV-Ray, or the Persistence of Vision Ray-Tracer, is a top-notch open source tool for creating realistically lit 3D graphics artwork. We tested with POV-Ray's standard included benchmarking model on all of our test machines and recorded the scores reported for each.   We shoudl also note that we used the latest 64-bit beta build of the program.  Results are measured in pixels-per-second throughput.
  

POV Ray Performance Details: www.povray.org

The POV-Ray 'all cpu' rendering benchmark had the new Phenoms crushing all of the dual-core offerings, regardless of clock speed.  But the rest of the quad-core configurations all finished well out in front, including the Core 2 Quad Q6600, which as we've already noted, is clocked at the same 2.4GHz as the Phenom 9700.
 

Cinebench R10 is an OpenGL 3D rendering performance test based on Cinema 4D. Cinema 4D from Maxon is a 3D rendering and animation tool suite used by 3D animation houses and producers like Sony Animation and many others.  It's very demanding of system processor resources and is an excellent gauge of pure computational throughput.

Cinebench R10 Performance Tests  3D Modeling & Rendering Tests

This is a multi-threaded, multi-processor aware benchmark that renders a single 3D scene and tracks the length of the entire process. The time it took each test system to render the entire scene is represented in the graph below, listed in seconds.

The performance trend we've seen throughout much of our testing to this point played out again in the Cinebench R10 benchmarks.  The Phenoms outpaced the dual-core offerings in the multi-threaded test and showed a significant clock-for-clock advantage over the Athlons in the single-threaded test.  Intel quad-core offererings at every frequency performed much better than either Phenom processor, however.
  

Futuremark 3DMark06 - CPU Test  Simulated DirectX Gaming Performance

 
3DMark06's built-in CPU test is a multi-threaded DirectX gaming metric that's useful for comparing relative performance between similarly equipped systems.  This test consists of two different 3D scenes that are processed with a software renderer that is dependent on the host CPU's performance.  Calculations that are normally reserved for your 3D accelerator are instead sent to the CPU for processing and rendering.  The frame-rate generated in each test is used to determine the final score.

The trend continued with the CPU performance module built into 3DMark06.  Once again the Phenoms outperformed the dual-core processors at every clock speed, but the higher clocked QuadFX rig and Intel quad-core processors put up much better scores overall.

For our last set of game tests, we moved on to some in-game benchmarking with the Crysis SP demo and F.E.A.R. When testing processors with Crysis or F.E.A.R, we drop the resolution to 800x600, and reduce all of the in-game graphical options to their minimum values to isolate CPU and memory performance as much as possible.  However, the in-game effects, which control the level of detail for the games' physics engines and particle systems, are left at their maximum values, since these actually do place some load on the CPU rather than GPU.

Benchmarks with Crysis SP Demo and F.E.A.R. v1.08 DirectX 9 and 10 Gaming Performance

Crysis reported an edge for the Phenom processors over the QuadFX and Athlon 64 X2 4600+ systems, but the higher-clocked 3.2GHz Athlon 64 X2 6400+ was the fastest of the AMD-powered systems by a couple of frames per second.  The Intel-powered systems were in a league of their own, however.  The spread was almost identical in the F.E.A.R. benchmark, but in the F.E.A.R. test the QuadFX rig put up the best score from the AMD camp.  All of the Core 2-based systems finished with much higher framerates, once again.

We have one final data point we'd like to cover before bringing this article to a close. Our goal was to give you all an idea as to how much power each of the system configurations we tested used while idling and running under load.
  

Power Characteristics Processors and Platforms

 
Please keep in mind that we were testing total system power consumption here at the outlet, not just the power being drawn by the processors alone.  In this test, we're showing you a ramp-up of power from idle on the desktop to full CPU load.  We tested with a combination of Cinebench R10 and SANDRA XII running on the CPU.
 

While idling and under load, the Phenom powered test systems drew more power than all of the Intel powered systems; only the Quad-FX rig consumed more.  Due to the relative immaturity of the platform, however, we suspect the AMD Athlon 64 X2 and Phenom systems' power consumption numbers are skewed a bit on the high side.  What's interesting to note is that with double the execution cores and much more cache, the Phenom 9700 draws 30 more watts than the Athlon 64 X2 4600+ while idling an 53 more watts while under load. 

Performance Summary: There are a few interesting performance characteristics to summarize in regard to AMD’s new Phenom processors.  First, in comparison to similarly clocked Athlon 64 X2 processors, the Phenoms showed some significant performance gains in the neighborhood of 5% - 9% in single-threaded testing.  And in the multi-threaded tests, Phenom’s two additional execution cores obviously allow it to pull well ahead of any Athlon 64 X2 processor, regardless of frequency. AMD's IPC enhancements are for real.

Secondly, in comparison to AMD’s previous desktop pseudo quad-core offering, the QuadFX Athlon 64 FX-74 system, Phenom also showed relatively strong performance. In a few tests, namely Crysis, Sony Vegas, and Kribibench, despite a 600MHz – 700MHz clock speed disadvantage, the Phenom systems outpaced the FX-74 setup while consuming much less power.

Finally, we have to compare Intel’s quad-core offerings to Phenom.  In all but a few of PCMark Vantage’s individual tests, the 2.4GHz Core 2 Quad Q6600 outperformed the 2.4GHz Phenom 9700.  Although the Phenom did exhibit measurable IPC improvements over Athlon 64 processors, it seems those improvements aren’t enough to catch the Core 2.
 

AMD’s Phenom processor, and Spider platform as a whole, obviously took longer to bring to market than the company would have liked.  And even though they are officially "here" now, we’re sure AMD would have preferred to unleash Phenom on the world at higher frequencies than 2.2 (Phenom 9500) and 2.3GHz (Phenom 9600).  Although we showed you the performance of a 2.4GHz Phenon 9700 throughout this article, that CPU won’t arrive for a few more weeks.

 

APPROXIMATE AMD PHENOM PROCESSOR PRICING:
AMD Phenom processor model 9700 - <$300/ea.
AMD Phenom processor model 9600 - $283/ea.
AMD Phenom processor model 9500 - $251/ea.

 
According to the most recent pricing information we have on Phenom, it’s clear that AMD is positioning early Phenoms in-line with the Core 2 Q6600. Based on the performance we’ve seen from the processors, that’s the right place to be, but overall the Q6600 offers better performance and street prices for Intel’s offering are somewhat lower than the Phenom 9600 at the moment.

At this juncture AMD is in a tough spot.  Their Radeon HD 3800 series graphics cards are looking very strong.  And although we’ve only done limited testing to this point, we also think the 7-series chipset is in good shape.  But the company badly needs to ramp up frequencies with their Phenom processors.  Looking back at the performance, it’s clear that the architecture performs relatively well and is an improvement over the previous generation clock for clock.  Either AMD’s 65nm manufacturing process doesn’t agree with the architecture or there are issues with the design that are keeping it from hitting higher clock speeds.  Whatever the case, we hope AMD has a breakthrough soon.  The company has stated that they will offer an unlocked (Black Edition) 2.3GHz Phenom later this quarter and they intend to offer the Phenom 9900 (2.6GHz) priced below $350 and Phenom 9700 (2.4GHz) priced below $300(1KU) in Q108, with a 3GHz part following sometime in the first half.  The question is, where will Intel's offering be at that point?
 

Higher IPC Than Athlon 64 Native Quad-Core Good MT Performance Competitive Pricing 790FX Chipset	Only 2.3GHz At Launch Intel Faster Clock-for-Clock Limited Overclocking