||AMD Dual-Core Opteron Performance Preview
April 21, 2005
Summary: Earlier this month Intel debuted their dual-core Pentium Extreme Edition 840 to the public, launching the first systems based on the technology at the beginning of this week. Today it's AMD's turn, and unlike Intel, AMD is not constrained by the front-side bus. Read all about AMD's dual-core plans, including X2 and Athlon 64 FX-57, as well as the future for AMD, and of course, you'll also see how AMD's dual-core offerings compare to Intel's as well as seven other processors in our Opteron performance preview article!
| Introduction||Page:: ( 1 / 14 )|
Thatís of seemingly little consequence to AMD. Rather than target the desktop workspace with dual-core, itís going after the same market that originally embraced Opteron--servers and high-end workstations. Thereís quite a bit of chest-thumping going on the AMD camp, partly because it feels that the real market for dual-core is that higher-end of the spectrum and partly because AMDís architecture seems much better suited for multiple processing cores interacting with each other. Indeed, Pat Patla, director of server and workstation marketing at AMD is quick to point out that right from the inception of its design, Opteron was designed to accommodate dual-core processing. Of course, a lot of people seem to forget that at the end of the day, it doesnít matter who has the more elegant implementation. Itís the better performer that will take home the accolades.
Fortunately, todayís announcement and release of the AMD Opteron 100-, 200-, and 800-series of dual-core processors gives us a much clearer picture of who will be faster (in addition to whose implementation is prettiest, if that matters to you). We have at our disposal Intelís Pentium Extreme Edition 840, at 3.2GHz, which was previously previewed, AMDís Athlon 64 FX-55, the current performance champ in a majority of gaming environments, AMDís Opteron 252, a 2.6GHz equivalent to the FX constrained to registered memory and a 940-pin interface, and the Opteron 875, a production dual-core sample running at 2.2GHz and priced in excess of its weight in gold. It doesnít have the same residual value but, well, thatís another matter entirelyÖ
On a side note, both Intel and AMD lament the fact that the other announces product well ahead of availability, but itís interesting to note that they continue to do the same thing themselves. Intel is skirting the problem to some degree by making a select number of Extreme Edition processors available to Dell at excruciatingly high prices, while AMD releases the dual-core Opteron and announces its desktop lineup with an anticipated ship date months away. Opteron systems should be available at launch from preferred system builders, such as HP, but the white box folks wonít have much. Neither company is innocent here; and we should all bear in mind that dual-core is much rarer in the wild than the flood of previews would indicate.
| AMDís Dual-Core Architecture||Page:: ( 2 / 14 )|
The key, according to AMD, is that its cores communicate with each other directly, on the die, through a system request queue and crossbar, whereas the two cores on Intelís Pentium D use an 800MHz front side bus, bottlenecking performance significantly. Truly, as we saw in the Extreme Edition 840 preview, memory bandwidth does drop by a third when both cores contend for the available throughput. The immediate effect that has on performance in single-threaded software appears to be negative, which threaded apps gain despite the reduction. What remains to be seen is if AMDís architecture maintains its single-threaded numbers while posting similarly large gains in threaded apps.
From a core perspective, thereís really nothing new to report about the Opteron chips. They consist of the same 64KB of data cache, 64KB of instruction cache, and 1MB of L2 cache. Only now, there are two of them manufactured at 90nm. As mentioned, both cores attach directly to a System Request Queue and crossbar, over which they communicate with the packageís three HyperTransport links and integrated memory controller. It still supports dual-channel DDR memory at up to 400MHz and those HT pathways still purr along at 1GHz. The only difference is now there are two cores utilizing them.
If youíre worried about bottlenecking, donít be. AMD is claiming resource conflicts are rare and the impact of shared memory is a roughly 10 percent reduction in bandwidth. In cranking up the HyperTransport frequency last year, AMD helped circumvent any limitations there. Our benchmarks should ultimately reveal any real weakness, though.
Power is another viable concern, especially so given the 130W thermal design power quoted on Intelís Extreme Edition 840 and the workstation-class EPS12V power supply that shipped with our test system. Surprisingly, AMD is citing the exact same envelopes for the dual-core parts as it established previously for single-core processors. That is, Opteron falls below 95W and the Athlon 64 will be below 110W. Letís all collectively thank manufacturing advances for those numbers.
| More Dual-Core||Page:: ( 3 / 14 )|
AMDís Approach, Continued
Meet Opteron 875
Much of AMDís naming scheme for Opteron carries over to dual-core. Youíre still going to see 100-, 200-, and 800-series processors; however, rather than increment by two each time a new speed bump emerges, AMD is starting with the x65, where x represents the family designator, and incrementing by five for each 200MHz bump. Thus, you have the 865 at 1.8GHz, and 870 at 2GHz, and the 875 at 2.2GHz. Traversing families, the 175, 275, and 875 all run at 2.2GHz; they just support different multi-processing configurations.
More generally, all of the models will center on a similar design consisting of 233 million transistors and a 199 square millimeter die. The die is based on revision E, meaning it supports Intelís SSE3 instructions, recognizes memory modules of different capacities through an improved memory controller, and features AMDís PowerNow! power management technology to reduce consumption during periods of light use. Even from the get-go, a lower 1.35V operating voltage helps keep the Opteronís power numbers low by virtue of manufacturing improvements.
Pricing is where the 875 really makes an impression. Each chip costs a staggering $2,649. Consider that the 800-series is intended for four- and eight-way servers and youíre talking about a lot of money. Fortunately, most desktops and workstations will realize optimal performance with a single Opteron 175 sporting two 2.2GHz cores for $999. Those processors arenít expected for a while still, though.
The only other special consideration required by the Opteron 875 is a dual-core-aware BIOS. Our original Tyan K8WE failed to take the update, probably due to its pre-production status. However, a replacement board properly recognized the 875, as did ASUSí K8N-DL motherboard, also based on the NVIDIA nForce Professional chipset with a beta BIOS from Taiwan. Switching from an older Socket 940 Athlon 64 FX to an Opteron 252 to an Opteron 875 is a matter of popping the heatsink, swapping the processor, and booting back up. Itís even that easy, much to Intelís chagrin.
| The Future and Software||Page:: ( 4 / 14 )|
Without an accompanying timeline, those claims are about as useful as NVIDIA saying future graphics architectures will be even more parallel and support larger frame buffers. Who would have guessed? Then again, at least we know AMD isnít planning on holding onto DDR forever and raw clock frequency isnít the definitive answer to delivering more performance.
Wait. Hasnít AMD said that all along? Eh hem, moving onÖ
One of the sticking points Intel used to contest the viability of 64-bit processing back when AMDís Athlon 64 first launched was software support. As it were, there werenít any mainstream operating systems thatíd expose the technology. Linux, for all of its strengths, is still hardly mainstream.
While Microsoft clearly has a lot on its plate, what, with patching the holes in Windows XP and pushing Longhorn back into 2006, it took an especially long time to get XP running in 64-bit mode, despite all of AMDís efforts to help hardware vendors write beta drivers. Only now is the operating system ready for retail consumption.
Admittedly, that was the pitch from the beginning. AMD64 enabled top performance in existing 32-bit apps while facilitating the flexibility to adopt 64-bit technology when the time came. In addition to the enhanced memory addressability and extra registers acquired by running the Opteron in 64-bit mode, both Windows XP x64 and Windows Server x64 will also be better optimized for dual-core operation through NUMA (non-uniform memory access) awareness, meaning that they assign threads from particular processing cores to attached memory for lower latency. Existing Windows XP and Windows Server 2003 versions, even with their respective service packs, recognize basic multi-core functionality, but are not NUMA-aware.
Even still, the existing crop of 32-bit operating systems will demonstrate increased speed in threaded applications and multi-tasking environments thanks to dual-core processing. Among the workstation loads cited by AMD, digital content creation, computer-aided engineering, and electronic design automation are among the most compute-intensive, so you can expect to see those titles really take off in the face of available power.
| Athlon 64 X2 and Gaming||Page:: ( 5 / 14 )|
Dual-Core on Your Desktop
So, AMD will trail a bit on the desktop side, but only by a couple of months. Its Athlon 64 X2 family is expected to be nearly identical to the Opteron, only in 939-pin packaging. Oh, and intermediate models will continue utilizing different sized L2 caches, just like todayís Athlon 64 lineup.
At the top, youíll find an Athlon 64 4800+ ($1,001), followed by 4600+ ($803), 4400+ ($581), and 4200+ ($537) models. That flagship runs at 2.4GHz and comes equipped with a cumulative 2MB L2. The 4600+ runs at 2.4GHz as well, but each core has 512KB instead. The 4400+ runs at 2.2GHz with dual 1MB caches, and the 4200+ is at 2.2GHz with the 512KB repositories. Clearly, if youíre a desktop user, itíll pay to wait for these Athlon 64 X2 models because theyíre much cheaper than the Opterons that will be available, but unfortunately you will have to wait awhile. While AMD will officially launch the X2 line in June, first shipments in both North America and Europe will be targeted for system builders. AMD doesnít expect strong retail availability for North America until Q4 of this year, so those of you looking for a deal on Athlon 64 X2 may have to wait awhile.
Complicating matters for enthusiasts on a budget is AMD's single-core plans. AMD has no current plans to release a follow-up single-core Athlon 64 processor; the Athlon 64 4000+ will remain the flagship of the Athlon 64 line and, along with the Athlon 64 FX-55/FX-57, will be the last of the single-core processors AMD produces.
As with the Opteron chips, dual-core Athlon processors require only a BIOS update to their 939-pin socket motherboards. No, thereís no 754-pin version planned, it seems. After all, the memory bandwidth hit would be too severe, weíre guessing. You do get CoolíníQuiet, though, along with AMDís Enhanced Virus Protection feature. In terms of availability
|Upcoming AMD Athlon 64 X2/FX Lineup|
|CPU||Clock Speed||L2 Cache Size||Price|
|Athlon 64 X2 4800+||2.4GHz||1MB||$1001|
|Athlon 64 X2 4600+||2.4GHz||512KB||$803|
|Athlon 64 X2 4400+||2.2GHz||1MB||$581|
|Athlon 64 X2 4200+||2.2GHz||512KB||$537|
Athlon 64 FX in the future
Notice the lack of an Athlon 64 FX version of AMDís dual-core strategy. For the time being, itís recognized that games are exclusively written for single-threaded operation and as such run better on single-threaded processors at elevated frequencies. Thus, the FX series marches on at 2.6GHz for now.
According to documents from AMD, thatíll be the first time multiple FX model persists, as there are plans to keep the FX-55 in production to meet perceived demand. Both chips target gamers exclusively, pandering to view that AMD is better at listening to the enthusiast community.
| System Setup||Page:: ( 6 / 14 )|
AMD Opteron 875 (2.2GHz)
AMD Opteron 252 (2.6GHz)
Intel Pentium Extreme Edition 840 (3.2 GHz)
Intel Pentium 4 570J (3.8 GHz)
Intel Pentium 4 540J (3.2 GHz)
AMD Athlon 64 FX-55 (2.6 GHz)
AMD Athlon 64 4000+ (2.4 GHz)
ASUS K8N-DL nForce Professional Motherboard
Intel D955XBK 955X Express Motherboard
Intel D925XECV2 925XE Express Motherboard
ASUS A8N-SLI nForce4 SLI Motherboard
1GB Corsair DDR-400 CAS 2 Registered Memory (2x512MB)
1GB Micron DDR2-667 CAS 5 (2x512MB)
1GB Corsair DDR2-533 CAS 3 Pro Series Memory (2x512MB)
1GB Corsair DDR-400 CAS 2 Xpert Series Memory (2x512MB)
ATI RADEON X850 XT
Seagate Barracuda 7200.8 400GB
Windows XP Professional with Service Pack 2
Desktop resolution 1024x768, 32-bit color, 85Hz refresh
All power saving options were turned off, as were the Automatic Update and System Restore services. Graphics options under the ĎPerformanceí tab were all disabled for maximum performance.
As youíll see, weíre again doing a few things differently. Baseline benchmarks now begin at 1024x768. Although itís certainly relevant to show off performance at 800x600 or 640x480, where youíd typically see the greatest differences between processor and platforms, those settings are really falling by the wayside and in no way reflect actual game play. Youíll be surprised nonetheless. Even at 1024x768, thereís plenty of difference between competing chips.
Windows Media Encoder 9
SiSoft Sandra 2005
| Windows Media Encoder 9||Page:: ( 7 / 14 )|
Microsoft Windows Media Encoder 9
| 3DMark05||Page:: ( 8 / 14 )|
| PCMark04||Page:: ( 9 / 14 )|
| Cinebench 2003||Page:: ( 10 / 14 )|
Maxon Cinema 4D Cinebench 2003
| Doom 3||Page:: ( 11 / 14 )|
id Software Doom 3
| Half-Life 2||Page:: ( 12 / 14 )|
Valve Half-Life 2
| SiSoft Sandra 2005||Page:: ( 13 / 14 )|
| Conclusion||Page:: ( 14 / 14 )|
Itís interesting to note, first off, that although AMD was forced to take a small step backwards with regard to the clock frequency of its architecture, the x75 at 2.2GHz is no slouch even when it comes to single-threaded software. The improvement in the few threaded tests we used were promising, to say the least. Windows Media Encoder 9 in particular turns out some incredible performance numbers. Cinebench also demonstrates notable gains, especially compared to Intelís Hyper-Threaded dual-core Extreme Edition 840, which operates on twice the number of threads.
Come June, AMD will augment its appeal by releasing the 2.4GHz Athlon 64 X2 4800+, which will be even faster than the model examined today. And although the company says dual-core isnít for gamers quite yet, perhaps it is, only in a different usage model. Alan Dang and I were discussing processor benchmarking moving forward and he came up with the idea that we donít run compute-intensive tasks in the background today because we think they canít be done. However, if a dual-core processor enables a DVD encode while youíre playing Half-Life 2: Deathmatch, thereís a good chance that the way we think about demanding tasks may change. Even though games arenít currently threaded, the background processes a dual-core processor enables may very well catapult the technology into favor with game enthusiasts.
One aspect that wonít win AMD any favor with enthusiasts however is X2 pricing. With Athlon 64 X2 models starting at $537, AMDís asking price is pretty steep. In addition, AMD also admitted to us that retail availability of X2 CPUs will be pretty limited until Q4 of this year. With Intel offering a wider range of desktop Pentium D dual-core processors, and at lower prices than AMD, Intel could enjoy some favorable PR after enduring quite a bit of criticism at the end of last year.
AMD could offset some of this by offering a follow-up product to the Athlon 64 4000+, or introducing lower-cost X2 models, but so far AMD has no plans to do either of these, most likely due to manufacturing constraints (after all, AMD maintains that theyíre selling every 64-bit processor they can make).
In any case, AMDís dual-core Opteron processors are here now, and from a performance perspective, theyíre quite impressive. First shipments will be dedicated to the high-end 8-series line, with 1xx and 2xx CPUs following in May.