||AMD Athlon 64 FX-51 Performance Preview
September 23, 2003 Brandon Bell & Chris Angelini
Summary: After years of hype, rumor, and speculation, AMD's first desktop processor, Athlon 64 is finally here! The Athlon 64 FX-51 is AMD's high-end processor built for 64-bit computing. But how does it perform in comparison to the competition? Find out now!
| Introduction||Page:: ( 1 / 13 )|
AMD’s seventh generation Athlon processor has served the company well for the past four years. In fact, the success of Athlon saved AMD from the brink of extinction.
In more recent times however, the architecture has begun to show signs of age. While Intel has moved at a steady pace incorporating new performance-enhancing features into the Pentium 4, they’ve also been cranking up the Pentium 4’s clock speed. Increments of 133MHz and more recently, 200MHz, have been achieved. Meanwhile, AMD has been struggling to get considerably past 2GHz.
The end result is that Intel’s processor lineup has been winning an increasing number of benchmarks, including applications that have traditionally favored AMD’s Athlon. AMD has extended the life of its Athlon XP processor by incorporating more cache with Barton, and increasing the processor’s bus speed to 400MHz, but these were only stopgap measures, AMD needed a new architecture and they knew it. Their answer lie in K8, also known as the Hammer family. Not only would Hammer bring with it a new micro-architecture to fight Pentium 4, it would also introduce the world to AMD’s vision of 64-bit computing. Unfortunately for AMD, it just wasn’t ready for release.
Finally, after considerable delays and months of speculation, AMD’s first 64-bit desktop processor is here: the Athlon 64 family. AMD is working hard to make up for lost time too, as they’re bringing two products to market, the Athlon 64 and the Athlon 64 FX.
|<% print_image("01"); %>||<% print_image("02"); %>||<% print_image("03"); %>|
At the extreme high-end of the desktop PC space, AMD offers the Athlon 64 FX. This is AMD’s new flagship processor. Just below it is the Athlon 64, this is AMD’s latest offering for the mainstream consumer. Both processors are based on the same basic core, and while the initial Athlon 64 launches with the 3200+ rating, AMD claims that it performs faster than the Athlon XP 3200+. We’ve already gone over the micro-architecture in our Opteron 242 article, so lets quickly rehash the basics before explaining the differences between Athlon 64 and Athlon 64 FX.
Most importantly, AMD has designed the Athlon 64 processors to be outstanding performers in both 32-bit and 64-bit applications. In fact, it would probably help if you think of the Athlon 64 primarily as a 32-bit processor, as 64-bit applications are still few and far between. Just consider its 64-bit capability as an added bonus for the future.
Second, the Athlon 64 isn’t the dramatic difference in design that the Pentium 4 was over the Pentium III. Instead, AMD has tweaked the Athlon’s core for more performance per clock cycle, while at the same time tweaking the registers for 64-bit use.
SIDEBAR: You can check out an official webcast of the Athlon 64 launch at AMD’s website.
| New micro-architecture||Page:: ( 2 / 13 )|
The most talked about feature is probably the processor’s integrated memory controller, a function that was once left up to the North Bridge of the system chipset, an external chip. This dramatically improves the processor’s memory bandwidth, increasing performance. The Athlon 64 FX features the same 128-bit wide interface to its memory controller as AMD’s Opteron processors, while the Athlon 64 relies on a tamer 64-bit interface.
AMD is quick to point out that this new feature makes the processor’s front-side bus speed less important, as the memory data that was once tied to the front-side bus in previous processors now operates at the same clock frequency as the processor. If you recall, the Athlon XP 3200+ featured a 400MHz front-side bus. In addition, the speed at which the processor and chipset communicate is dependant on the processor’s HyperTransport link. Currently, both Athlon 64 processors can communicate to the system chipset at speeds up to 1.6GHz effective.
Like the Athlon XP, the Athlon 64 family features a 2-way set associative 128KB Level one cache, 64K for instructions and 64K for data cache. AMD has doubled the size of the L2 cache from 512K in the latest Athlon XP processors to 1MB in Athlon 64. Increasing the size of the L2 cache is a popular enhancement CPU manufacturers implement to improve performance.
To further improve performance, AMD has improved its branch prediction in Athlon 64 while including larger Translation Look-Aside Buffers (512 entries in Athlon 64 versus 256 in Athlon/Athlon XP) with reduced latencies. The Athlon 64 also incorporates Intel’s SSE2 instructions (in addition to 3DNow! Professional and SSE), first introduced with the Pentium 4.
Finally, to help the Athlon 64 scale to higher clock speeds, two additional stages have been added to its pipeline, bringing the grand total to 12 stages in its integer pipeline and 17 stages in its floating-point pipeline. If you recall the original Pentium 4 launch, the hyper-pipelined nature of its design was one of the more controversial features.
The Athlon 64 is produced at AMD’s 0.13-micron Fab in Dresden Germany. AMD has spent a considerable amount of time refining its 0.13-micron process, which now not only features copper interconnects, but also silicon-on-insulator technology (SOI), licensed from IBM. SOI is a technique that reduces the capacitance of the millions of transistors within the processor, allowing them to run faster and consume less power. The Athlon 64 is composed of approximately 105.9 million transistors; nearly twice the amount of Athlon XP’s 54.3 million.
Athlon 64 FX versus Athlon 64
As we mentioned earlier, one of the key performance differences between the Athlon 64 FX and Athlon 64 is its wider memory interface: 64-bits in Athlon 64 versus 128-bit in the FX model. Essentially, the FX model is nothing more than an Opteron processor; the chip even uses the same 940-pin socket interface.
AMD also separates the Athlon 64 FX-51 from the Athlon 64 3200+ by running it at 2.2GHz, while the Athlon 64 3200+ operates at 2GHz.
Finally, the Athlon 64 FX requires registered DDR memory - the Athlon 64 does not. This is an important distinction for Athlon 64 FX system builders, as registered DDR memory is harder to come by; Kingston recently announced its HyperX registered DDR400 memory modules.
SIDEBAR: The Athlon 64 operates at 1.5V
| System Setup||Page:: ( 3 / 13 )|
AMD Athlon 64 FX-51 (2.2GHz)
AMD Athlon XP 3200+ (2.2GHz)
Intel Pentium 4 3.2GHz (Hyper-Threading Enabled)
Intel Pentium 4 3.0GHz (Hyper-Threading Enabled)
ASUS SK8N nForce3 Professional 150
ASUS A7N8X Deluxe nForce2 400 Ultra
ABIT IC7-MAX3 875P
1GB Registered DDR400 (Athlon 64 FX)
1GB Corsair DDR3500 CAS2 (Athlon XP, Pentium 4)
NVIDIA Reference GeForce FX 5900 Ultra 256MB
Detonator Driver 45.23
2x Western Digital Raptor 36GB 10,000RPM SATA HDD (RAID 0)
Windows XP Professional
Windows XP Professional 64-bit Edition
SiSoft Sandra MAX3
3D Mark03 v330
Comanche 4 Benchmark Demo
PC Magazine Business Winstone 2002
Quake III: Arena version 1.32 ‘fscrusher’ demo
Unreal Tournament 2003 Demo
Desktop resolution 1024x768, 32-bit color, 75Hz refresh
All tests run with power saving options disabled in Windows XP, as are the Automatic Update and System Restore services. Graphics options under the ‘Performance’ are disabled for maximum performance.
All three platforms were run with the most aggressive memory settings possible without sacrificing stability. The Athlon 64 FX-51 was set at CAS2.5-3-3-7, so perhaps higher quality modules would expose additional performance.
On the SK8N, Serial ATA RAID is configured on the onboard Promise controller. On the A7N8X Deluxe, a Silicon Image controller is substituted. The 875P platform employs the ICH5’s native RAID support.
Finally, it is interesting to note that both the Athlon XP and Pentium 4 platforms are noisier than the Athlon 64 FX. The reference heatsink/fan combination used by AMD is significantly quieter.
SIDEBAR: NVIDIA also announced its nForce3 line of desktop and mobile products for the Athlon 64 today.
| SiSoft Sandra MAX3||Page:: ( 4 / 13 )|
SiSoft Sandra MAX3 – Synthetic
Sandra MAX3 is the latest creation from SiSoftware, which incorporates support for the AMD64 initiative. Its web site indicates that the benchmark supports Athlon 64, but gives no indication with regard to optimizing for the process. Perhaps that is why the Athlon 64 FX-51 fares poorly in the Arithmetic test, scoring below its predecessor in the ALU section and marginally better in the FPU part. The incorporation of SSE2 support certainly helps, though. It is also possible that, because the Athlon 64 has two extra pipeline stages, it isn’t able to execute as many instructions per cycle as the Athlon XP 3200+, which runs at the same frequency.
The multimedia test again demonstrates a gain attributable to SSE2 instruction set support. However, the Athlon 64 isn’t able to come anywhere near the Pentium 4’s score.
One specification that the Athlon 64 FX and Pentium 4 share in common is support for dual-channel DDR400 memory modules. The significant difference, of course, is that the Athlon 64 FX incorporates its memory controller on the processor die. Thus, the bandwidth benchmark isn’t much of a surprise – the Athlon 64 FX enjoys 14 percent more real-world throughput than its closest competitor.
SIDEBAR: The Athlon 64 has a die size of 193mm2
| 3D Mark 03||Page:: ( 5 / 13 )|
3D Mark03 – DirectX 9.0b
The first 3D Mark03 test is a composite score gleaned from the four game tests, all of which run at the default resolution of 1024x768. While it is known primarily for measuring graphics performance, the 3D Mark03 score shows that processor performance plays a role in graphics as well.
Intel’s 3.2GHz Pentium 4 maintains a distinct advantage, while the 3.0GHz version pulls a second place finish. The Athlon 64 FX-51 takes third place, besting the Athlon XP 3200+ by a little more than four percent.
Then, as if to contradict the gaming score itself, 3D Mark03’s CPU benchmark shows AMD’s Athlon 64 FX-51 taking a first place finish, besting the 3.2GHz Pentium 4 by a full 21 percent. It holds a 34 percent advantage over its predecessor, the Athlon XP 3200+.
SIDEBAR: The HyperTransport link within Athlon 64 provides up to 6.4GB/sec of bandwidth.
| Comanche 4||Page:: ( 6 / 13 )|
Comanche 4 – DirectX 8
The most impressive gains demonstrated by the Athlon 64 thus far are apparent in Comanche 4, a traditionally processor-bound simulator. At 800x600, a resolution hardly affected by graphics card performance, the Athlon 64 FX-51 is 30 percent faster than the Athlon XP 3200+ and 11 percent faster than the 3.2GHz Pentium 4, earning itself another first place finish. At 1600x1200, a more realistic resolution, albeit one more tightly tied to fill-rate, the Athlon 64 FX-51 is still 23 percent faster than the Athlon XP 3200+ and seven percent faster than the Pentium 4 3.2GHz.
SIDEBAR: The Athlon 64 FX-51 utilizes ceramic packaging.
| Business Winstone 2002||Page:: ( 7 / 13 )|
PC Magazine Business Winstone 2002
Business Winstone 2002 is a productivity metric that focuses on tasks you’d normally see executed in an office environment. It compliments the Content Creation 2003 test; however, we have experienced numerous issues with the latter benchmark and have excluded it from our suite.
The Athlon XP is the historic favorite in the Business Winstone series because its shorter execution pipeline is more congruent to processing integer code. Even though the Athlon 64 employs a slightly longer pipeline, its onboard memory controller is able to compensate for any loss in IPC, resulting in a 14 percent improvement. The Athlon 64 FX-51’s lead over the 3.2GHz Pentium 4 extends to a full 20 percent.
SIDEBAR: Meanwhile, the Athlon 64 754-pin sports AMD’s organic packaging.
| Quake III: Arena||Page:: ( 8 / 13 )|
Quake III: Arena v1.32 ‘fscrusher’ Demo – OpenGL
Quake III is an old benchmark with a historic slant towards Intel’s Pentium 4; we were using it to test as far back as the Athlon 800MHz review early in 2000, though. However, its viability as a test lies in the number of games that utilize the engine behind it. Most recently, Brandon developed the fscrusher demo file, which does away with the 300+ frame per second scores we’re used to by incorporating a small map and mass quantities of processor-controlled bots.
The Athlon 64 FX-51 launch marks the first time we’ve seen an AMD platform dominate a Quake III benchmark. At 800x600, the Athlon 64 holds a six percent advantage over the 3.2GHz Pentium 4 and a massive 23 percent improvement over the Athlon XP 3200+. That lead is greatly diminished at 1600x1200, but AMD’s Athlon 64 FX-51 still maintains its leader status.
SIDEBAR: What do you think of the Athlon 64’s heat spreader?
| SPECviewperf 7.1||Page:: ( 9 / 13 )|
SPECviewperf 7.1 – OpenGL
AMD classifies the Athlon 64 FX line as its response to enthusiasts who want an elite processor, distinct from the more mainstream products. Fundamentally, though, Athlon 64 FX is a retagged Opteron running at 2.2GHz. Given the similarities, we’d expect that Athlon 64 FX might find its way into high-end workstations as well as the gaming systems AMD is anticipating. SPECviewperf measures performance in several real-world workstation applications using viewsets developed by the ISVs themselves.
Considering the Athlon 64 FX’s workstation heritage, it is surprising to see the processor bested by Pentium 4 and Athlon XP alike in every test except for one. VIA’s K8T800 platform is said to improve some of these scores, but at least as it relates to the platform at hand, Athlon 64 FX takes a small step backwards.
SIDEBAR: The original codename for Athlon 64 was ClawHammer.
| Unreal Tournament 2003 Demo||Page:: ( 10 / 13 )|
Unreal Tournament 2003 Flyby – DirectX 8
Unreal Tournament 2003 Botmatch – DirectX 8
In sharp contrast to the SPECviewperf 7.1 workstation tests, Unreal Tournament heavily favors the Athlon 64 FX-51. At 800x600, it beats the 3.2GHz Pentium 4 by 24 percent in the flyby demo (fill-rate bound) and 25 percent in the botmatch test (more influenced by processor performance). The Athlon 64 FX-51 holds onto its lead through 1600x1200, where it continues to dominate the fastest Pentium 4 by 15 percent in the botmatch demo. Those scores are a definite improvement over the Athlon XP, to be sure. The Athlon 64 FX-51 is a full 35 percent faster than the old 3200+ in the flyby demo at 800x600.
SIDEBAR: Next year, AMD will introduce its 0.09-micron Athlon 64 processor, codenamed San Diego.
| Dual-Channel versus Single-Channel Memory||Page:: ( 11 / 13 )|
SiSoft Sandra MAX3 and Unreal Tournament 2003 Demo
AMD is unveiling two processors; one is the Athlon 64 FX-51 that we’ve discussed thus far, and the other is its Athlon 64 3200+ running at 2GHz. Not wanting to overshadow the Athlon 64 FX, AMD isn’t distributing review samples of the 3200+. The only difference between the two 940-pin processors, however, is their memory configurations. Athlon 64 FX features a dual-channel (128-bit) controller, while the Athlon 64 is limited to a single channel. Since we didn’t have a 2.0GHz Athlon 64 3200+, we pared our FX down to a single channel by populating two memory slots on a single channel. It isn’t a scientific comparison because the Athlon 64 FX-51 runs at 2.2GHz, but at least we can report on the performance differences between dual- and single-channel platforms.
Indeed, SiSoft Sandra MAX3 shows that real-world bandwidth is greatly affected by the drop to a single channel of DDR400 memory. Compare that to the Unreal Tournament 2003 Demo scores, though. Because the chip’s memory controller is on-die (and very low-latency), it doesn’t suffer a huge performance penalty. In addition, we expect that the large 1MB L2 cache helps, too.
SIDEBAR: AMD next generation workstation/server part is codenamed Athens.
| 64-bit versus 32-bit||Page:: ( 12 / 13 )|
Unreal Tournament 2003 Demo and Comanche 4
When the Opteron launched, AMD lauded the fact that 64- and 32-bit operating systems would run flawlessly. In a 64-bit environment, the processor would even accommodate 32-bit applications without a massive hit in performance.
At the time, we weren’t able to quantify these claims because we lacked the platform support and operating system availability to run the necessary tests. Fortunately, AMD understands the importance of substantiating its claims and has arranged for a pre-beta version of Windows XP 64-bit Edition to be used for benchmarking. Because the operating system won’t emerge for another several months, and because we anticipate most enthusiasts will persist with their 32-bit configurations until there is real reason to make the transition, we only ran a couple of benchmarks. Take them with a grain of salt, of course.
There is a small overheard associated with running 32-bit code on a 64-bit operating system, according to Damon Muzny of AMD. However, the performance deltas we see here are more likely due to a combination of the immature OS, un-optimized drivers (a 64-bit OS requires 64-bit drivers, and NVIDIA’s Detonator family hasn’t yet been optimized for performance), and the small overhead AMD mentioned. These scores will certainly improve over the next few months, but at least we’ve seen today’s 32-bit applications running stably on tomorrow’s 64-bit platform.
SIDEBAR: The Athlon 64 FX-51’s memory interface is actually 144 bits wide.
| Conclusion||Page:: ( 13 / 13 )|
As you’ve just seen in our performance results, the Athlon 64 FX-51 is a dramatic improvement over the Athlon XP 3200+. But the million-dollar question is, how much faster is it than the Athlon 64 3200+? Officially, the Athlon 64 FX-51 sells for a whopping $733, almost $100 more than the Pentium 4 3.2GHz (which officially sells for $637) and nearly twice the price of the Athlon 64 3200+, which is priced at $417. We’ve seen today that the Athlon 64 FX-51 is faster than the Pentium 4 3.2GHz in most applications so it’s certainly priced competitively with Intel, but what about the Athlon 64 3200+? Is the added performance worth roughly double the price? Right now it’s just too early to tell.
AMD has already seeded retail and OEM distribution channels with processors, so the Athlon 64 FX-51 should not be a paper product, but quantities of this chip will be limited throughout 2003. Instead, the Athlon 64 3200+ will get the bulk of sales. Motherboard manufacturers have also been conservative in their product plans for the Athlon 64 FX’s 940-pin socket, preferring instead to wait for the 939-pin version coming next year. Fortunately, AMD plans to continue to produce 940-pin Athlon 64 FX throughout 2004, so end user’s should have an easy upgrade path if they chose to upgrade their Athlon 64 FX-51 processor sometime in the immediate future.
The question now is, will consumers be willing to shell out the $1,000+ for a new processor, memory, and motherboard. AMD has had a loyal following among gamers, but that’s a steep price to pay to stay on the bleeding edge of technology. The performance is certainly there, but will enthusiasts flock to this processor like the original Athlon?
SIDEBAR: What do you think of the Athlon 64 FX-51? Is AMD on the road to profitability or is this chip just too expensive. Chat with others in the news comments!