AMD Phenom 9600 Vs. Intel Core 2 Quad Q6600: Fight!

Phenom is here.

Today, you can actually go out and buy AMD "true quad core" Phenom CPUs. As Jason Cross noted in his earlier Phenom preview, AMD is shipping its Phenom 9500 (2.2Ghz) and Phenom 9600 (2.3GHz) CPUs. A Phenom 9600 will set you back about $275—not too shabby for a quad-core CPU. In fact, that $275 puts it in the same price category as Intel's Core 2 Quad Q6600, which cost around $280.

We recently received an engineering sample of AMD's latest processor. This particular sample is clock unlocked, and natively runs at 2.6GHz—so it's sort of a preview of the Phenom 9900, which won't ship until March.

But you can't buy a Phenom 9900 today. Since the price of Intel's Q6600 and the Phenom 9600 are so close, and both are available today, we decided to benchmark the AMD CPU running at 2.3GHz against the Intel CPU running at 2.4GHz

AMD's Spider Platform

Before we dive into numbers, it's worth taking a look at how you'd build a Phenom today.

AMD has been heavily touting the company's "Spider" platform, which consists of motherboards built using the new AMD 790FX chipset, the AMD HD 38xx series of graphics cards, and the Phenom processor. The company claims that this would be the most balanced solution, offering good performance and lower power utilization than Intel. Today, we'll look at two pieces of the Spider platform: CPU and core logic.

We wrote in-depth about the Phenom architecture back in October. But it's worth touching briefly on the speeds and feeds of the new processor.

Phenom 9600 Feature	Specification
Clock Frequency	2.3GHz (11.5x multiplier)
L1 Cache Size	128KB per core, split into 64KB data cache and 64KB instruction cache
L2 Cache Size	512KB per core (2MB total)
L3 Cache Size	2MB Shared
Memory Controller	128-bit wide integrated on-die
Memory Controller Clock	Up to 1.8GHz
Memory Support	Up to DDR2-1066
HyperTransport 3.0 Link	One 16-big, full duplex, up to 3600MHz effective
Transistor Count	450 Million
Process Technology	65nm SOI
Die Size	285mm^2
Nominal Voltage	1.1 - 1.25V
Rated TDP	95W

That TDP rating is essentially identical to the rating for Intel's Core 2 Quad Q6600. Power management is more sophisticated than in AMD's Athlon 64 X2 line, with CPU cores and memory controllers on separate power planes. So the memory controller can be running full bore while the CPU power state is lowered during idle and vice versa. Also, clock frequencies can be changed on a per-core basis, though voltages must remain the same on all cores.

Phenom sports a new SSE engine, with full, 128-bit wide registers and 2x128 bit loads per cycle. There are other tweaks to floating point, as well as improvements in the prefetcher and memory controller.

ASUS M3A32-MVP Deluxe

For this CPU shootout, we built a system using an ASUS M3A32-MVP Deluxe motherboard, which uses the AMD 790FX chipset, complete with socket AM2 support.

click on image for full view

The new chipset supports up to four PCI Express x16 connections, though if you drop in four graphics cards, the slots run in x8 mode. The connection to memory and from the north bridge to the south bridge are via 2GHz HyperTransport 3.0 links. AMD CPUs have direct access to memory via the processor's integrated memory controller. In theory, this can mean lower latencies when accessing main memory as compared to the more traditional external memory controller, used by Intel-based systems.

Current motherboards based on the 790FX still use the older AMD SB600 south bridge. It's certainly capable enough, with plenty of USB 2.0 ports, but only four native SATA 3GHz ports (with support for native command queuing, or NCQ.) Additional SATA ports are supplied through an external Marvell SATA controller chip.

click on image for full view

The premium edition of the ASUS motherboard also supports FireWire 400, Gigabit Ethernet, and Wi-Fi. It's a high end board costing about $100 less than the equivalent Intel X38-based motherboard: the P5E3 Deluxe WiFi Edition.

While the BIOS setup offers a rich set of features for overclocking and tweaking, it's not fully baked yet. There are still some minor issues with the system as built. For example, we could never get Adobe After Effects CS3 to run. Plus, reboots would occasionally hang, requiring a hard reset.

While these seem like typical teething problems with newer chipsets, AMD has had a long summer to iron out issue with the new chipset. We firmly hope that BIOS updates will solve these issues, however.

Since we've been reviewing CPUs using the Nvidia GeForce 8800 GTS 320MB graphics card, we decided to test CPU performance with that graphics card. We'll evaluate the implications of the overall Spider platform at a later date. For more details on the new chipset, check out Jason Cross's preview.

Benchmarks and System Setup

Since the price of Intel's Q6600 is so close to that of the Phenom 9600, it's the ideal candidate for a shootout between two CPUs. Also, the clock rate differential is relatively minor: the Q6600 has only a 100MHz edge over the Phenom 9600. Perhaps a better comparison would have been a Phenom 9700 at 2.4GHz, but those parts aren't available yet.

Also unique is the DDR2 memory speed. The Q6600 is a 1066MHz FSB processor (actual FSB clock frequency is 266MHz, quad-pumped.) The ASUS motherboard could only run the DDR3 memory at DDR3-1066 speeds. So in theory, AMD should have a latency edge, and memory bandwidth should be very similar.

We'll also toss in the benchmark numbers from our Intel Core 2 Extreme QX9650 review just for reference. This is the ultra-high end version of Intel's 45nm Penryn architecture. Note, however, that the QX9650 is in a completely different price class, and we include it only as a reference.

	Model	Model
Processor	AMD Phenom 9600 @ 2.3GHz	Intel Core 2 Quad Q6600 @ 2.4GHz
Motherboard	ASUS M3A32-MVP Deluxe WiFi Edition, AMD 790FX Core Logic	ASUS Maximus Extreme, Intel X38 Core Logic
Memory	Crucial Technologies DDR2-1066 running at DDR2-1066 speeds.	Kingston PC14400 DDR3 (2 x 1GB; running at DDR3-1066 speed)
Graphics Card	PNY GeForce 8800 GTS 320MB	PNY GeForce 8800 GTS 320MB
Hard Drive	320GB Seagate 7200.10	320GB Seagate 7200.10
Optical Drive	Sony 16x DVD+/-RW	NEC 16x DVD+/-RW
Power Supply	Corsair TX750W 750W	Silverstone SST-ST75ZF 750W
CPU Cooler	Arctic Cooling Freezer 64 Pro	Arctic Cooling Freezer 7 Pro

System Setup
We tested with both Windows XP and Windows Vista, but all benchmarks were run on Windows XP.

For benchmarking, each system had a clean install of Windows XP Professional with Service Pack 2 and all updates installed. The hard drives were defragged prior to each benchmark session. Before we ran each benchmark, we executed the command rundll32.exe advapi32.dll,ProcessIdleTasks. This immediately executes all background idle tasks to completion, including tasks such as the Windows prefetcher.

We also used the latest Nvidia 169.02 drivers for the GeForce 8800 GTS, which have added support for multithreading, thus improving performance on multicore CPUs.

The Benchmarks
We used a superset of our standard benchmark suite:

• BAPCo's SYSmark 2004 SE. Based on real applications, the benchmark gives us a taste of how real applications might run on the system. Note that rather than reporting an overall score, we break it down the individual Internet Content Creation and Office Productivity scores.
• POV-Ray 3.70 beta 23. The POV-Ray freeware software rendering application has been around since the days of MS-DOS. The current 3.70 version for Windows is still a work in progress, but it does support multithreading. Beta 21 now shows the rendered scene in progress.
• 3ds Max 8. 3ds Max 8 is Autodesk's popular 3D modeling and animation tool. Note that we used both the SPECapc 3ds Max test and a couple of pure rendering tests. The new SPECapc test for 3ds Max 8 incorporates a number of new scenes. For platform testing, we've shifted to pure software mode, rather than using either OpenGL or Direct3D, to further minimize the effect of the GPU.
• 3ds Max 9. For rendering, we use the more recent 3ds Max 9. We use one new scene, Dragon, and one older scene, Radiosity, for 3ds Max 9 rendering tests.
• NewTek LightWave 9.0. This is a popular 3D modeling and animation tool used primarily for special effects in a number of television shows, plus a variety of other applications. We used the current "Radiosity-Box" rendering benchmark. This latest version of LightWave also ships with a 64-bit version in the box.
• Adobe After Effects CS3, which ran a variety of filters using a fixed script on identical content to generate a final animation. These include radial blur, frame blends, depth-of-field, and other effects-based filters.
• Maxon's Cinebench 10.0, their latest 3D content creation benchmark based on Maxon's Cinema 4D engine.
• Windows Media Encoder. We also installed Windows Media Player 10, so we could encode a fixed 127MB, WMV-HD video clip using Windows Media Encoder Advanced Profile. The encoder was configured to encode the video as streaming media. Microsoft has also released a beta version of a new Advanced Profile codec, which can encode to VC1, used in many HD DVD and Blu-ray movies.
• Mainconcept H.264 and MPEG-2 Encoders. We now use Mainconcept's encoders instead of DivX. We're now using a new file, which is a 1080p WMV-HD clip from The Magic of Flight. We use the "DVD" profile for the MPEG-2 encoding and the "iPod" profile for H.264 encoding.
• DivX 6.7 / VirtualDub 1.76. DivX 6.7 now has SSE4 optimizations built in to the codec. We normally don't use DivX, but did want to see the potential impact of SSE4.
• 3DMark06 and PCMark05. While these are synthetic benchmarks, they can yield insights into individual subsystems within the PC.
• Seven 32-bit games, including Flight Simulator X, Supreme Commander, Prey, World in Conflict, Quake Ware: Enemy Territory, Team Fortress 2 and Company of Heroes. We run at low resolution with low detail to maximize the impact of the CPU, then run at higher resolutions and higher detail to see if the CPU has much effect at more normal gameplay resolutions.

We only report the individual Internet Content Creation and Office Productivity scores for SYSmark. We also broke out our game charts into separate genres, so there are now charts for FPS games, RTS games, and one just for Flight Simulator X.

SYSmark 2004 SE Benchmark

Let's first take a look at the results from SYSmark 2004 SE, which simulates real-life workloads for both Internet Content Creation and Office Productivity. The content-creation part uses apps like Photoshop, 3ds Max, Dreamweaver, and more, while the office-productivity tests use typical office apps, such as PowerPoint, Word, and Excel.

This applications-oriented benchmark is all Intel. The Q6600 easily outpaces the Phenom, even in the less demanding Office Productivity Suite.

PCMark05 Benchmarks

PCMark05 consists of a series of synthetic benchmark suites, each designed to test an individual subsystem, such as memory, processor, and hard drive. The test auto-detects which CPU you're using and loads dynamic libraries optimized for the processor under test for each function. So an Athlon 64 would run code tweaked to run best on its architecture, while an Intel CPU running the same test would run different code optimized for that processor. Many of the tests are small enough to fit into the large L2 caches of modern processors, so those with higher clock speeds have an advantage. It's an idealized view of performance. In the real world, application optimizations can vary widely.

We had initially thought that improvements in the memory controller, plus support for DDR2-1066 memory would help Phenom in memory accesses. In this particular desktop-oriented benchmark, that doesn't seem to be the case. The Phenom only manages to beat the Q6600 in the 192KB block read test. It does manage to tie the Core 2 Quad in the 8MB block write test. Given that Intel's CPU is a dual-die design with two dies sharing a front side bus, we would have expected a greater penalty from Intel's design. The 4KB block tests are particularly poor, suggesting that the L1 cache on the Intel design is more efficient.

As for the processor itself, if you look at the PCMark 4-thread CPU test, you'll see the Q6600 generally outpacing Phenom when four simultaneous threads are running.

But then, this is just a synthetic test. What about real applications?

Content Creation: 3D Modeling and Rendering Benchmarks

Now we turn to performance using actual working applications. We'll take a look at a pair of popular 3D modeling and rendering tools: 3ds Max 8 and LightWave 9. We use 3ds Max 8 when we run the SPEC APC test, but use the newer 3ds Max 9 to test rendering performance.

Again, it's not really much of a contest, with Intel ruling the roost in these 3D rendering and applications tests.

There is one ray of sunshine here for the Phenom. The Cinebench 10 benchmark runs parts of the benchmark in both single threaded and multithreaded mode, then reports the overall efficiency of moving from single to multiple threads. Both Intel CPUs see multicore efficiencies of around 3.5-3.6 (3.53 for the Q6600.) The Phenom 9600's multicore speedup is 3.77. This is no doubt due to the integrated memory controller. But it's not enough to catch the Q6600.

This suggests that if AMD can get the clock frequencies up, it could be very competitive with Intel in multicore performance. Given what we've seen, that may not happen until AMD brings up its 45nm process, however.

Content Creation: Media Editing and Encoding Benchmarks

We've shifted from After Effects 7 to After Effects CS3, and are using a new After Effects script to check out animation and filter rendering performance. Windows Media Encoder 9, Mainconcept MPGE-2, and H.264 encoders, as well as the latest 6.7 release of the DivX codec, were used to check out video transcoding.

We've shifted our WME tests to use the latest Windows Media 9 advanced profile codec, which now encodes to VC1. The advanced profile adds more functionality for encoding WMV files, including de-noise, interlaced, and progressive encoding options.

We also use DivX 6.7 to see what impact SSE4 has. Note that the VirtualDub shell application we used doesn't recognize our WMV-HD file we normally used, so we fell back to an older DVD rip from the movie The Rock. Given that this was a DVD resolution movie, we didn't expect big things in the transcoding.

When we updated the BIOS to the most recent one, version 0603, released on the ASUS Web site, we were able to get After Effects CS3 to run and complete the benchmark. The end result is pretty slow, compared to the Q6600. But the stability problem we initially encountered looks to be a thing of the past.

On the video transcoding side, things look a little brighter for the Phenom, but not much. It's a near dead heat between the Phenom 9600 and Intel Q6600, but the other benchmarks are owned by the entry level Intel quad-core CPU.

PC Gaming Benchmarks

The Core 2 Duo redefined what users expected from a CPU for games. Previously, AMD had ruled the roost in PC Gaming. In fact, you could argue that gaming was AMD's strongest suit, though the Athlon 64 X2 line of CPUs generally outpaced Intel in the majority of other benchmarks prior to the release of Core 2 Duo.

We run one set of benchmarks with resolution and detail cranked down, to minimize the effect of the graphics card. Then we run another set of tests at a higher resolution, with more playable graphics settings, so we can better gauge the gain in performance with more realistic settings.

We used six games plus 3DMark06 CPU tests to check out game performance. The games include Supreme Commander, World in Conlict, Company of Heroes, Team Fortress 2, Quake Wars: Enemy Territory and Prey. All make fairly heavy use of the processor and memory subsystem.

At low resolutions with low detail, we can see the impact of the CPU on a game more readily. There's little to show for Phenom in these tests. Even scaling up resolution and detail levels, Phenom generally can't keep up.

Final Thoughts: Phenom not so Phenomenal

Since we're using a motherboard with an early BIOS and an engineering sample of the Phenom CPU, we decided to put off testing power draw for the time being. Even if the AMD 790FX chipset is more efficient than the Intel X38, we doubt the overall power draw will be that different. We will look into this in more detail in the future, however.

The following paragraph was written before the release of the recent 0603 BIOS update from ASUS:

Stability is an equally important axis in computer performance, and the story isn't particularly robust yet. Earlier, we noted the inability of the platform to run Adobe After Effects CS3. In our game testing, the Supreme Commander benchmark crashed several times, while Quake Wars: Enemy Territory locked up once. Thermal issues didn't seem to be the culprit here. It's very likely another BIOS update or two for the motherboard will be needed to fix these issues.

We installed the 0603 update, but had to reset the memory timings to "auto". Previously, we set the memory speed manually to DDR2-1066 speeds, but with the new BIOS update, we started getting memory errors in memtest86 (this didn't occur with the beta BIOS.) However, with the memory timings set to auto, stability improved dramatically in our game tests. Also, After Effects CS3 now runs properly.

Read our review of Intel's QX9770.

Overall, we're fairly disappointed with both the Phenom processor and this particular motherboard. The whole affair doesn't seem quite fully baked. There's not much AMD could have done at this stage to pump up performance of Phenom. The processor is what it is, and it's priced about right for its performance level. On the other hand, AMD has had a good few months to tweak the core logic and motherboard, we're surprised at the issues we ran into in terms of compatibility and stability.

We're getting in other 790FX motherboards, and will be checking out Phenom on those boards. We hope that as the motherboards and supporting BIOSes mature, the issues we've seen will disappear. Performance is another matter, and we may have to wait until the 45nm die shrink to see improvements. The real question is whether or not AMD has the luxury of time to get to 45nm.

Product:	AMD Phenom 9600
Company:	Advanced Micro Devices
Price:	$275
Pros:	Four cores on a single die; potentially efficient multicore scaling.
Cons:	Lower performance than the equivalent Intel CPU.
Summary:	Phenom is finally here, but seems six months late and still a work in progress.
Rating:	click on image for full view

Product:	ASUS M3A32-MVP Deluxe WiFi Edition
Company:	ASUS
Price:	$240
Pros:	Rich feature set and potentially good tweaking capability.
Cons:	Seems a little pricey for CPUs that are still only modest performers.
Summary:	New BIOS updates cure stability issues, making the M3A32-MVP a solid choice for Phenom users.
Rating:	click on image for full view