In November 2004, Short-Media ran a guide that compared the various sockets and architectures available in the Athlon/Opteron line with the promise we'd return as time passed to reevalute our predictions and hammer out some new ones. We're back to evaluate our predictions and reassess the the situation in the world of AMD. Like last time, we'll go through each CPU that represents a significant percentage of AMD's volume shipments, and give you the rundown on the CPU's history in terms of its codename, die size, market names, silicon revision, coure count, socket, frequencies, voltages and thermal dissipation. Just like our 2004 edition, our editors and analysts have put on their prognostication caps for the 2006 Athlon comparison to give you future-proofed purchasing advice, as gleaned from dozens of articles and AMD roadmaps. While much has changed between 2004 and today, AMD has put considerable effort into condensing their offerings, and into providing a more stable environment for their product lines to grow in.

The most drastic changes are the elimination of the Athlon XP, the quiet EOL for socket 754 architecture as we predicted, the sudden evaporation of Socket 939's stability as an upgrade path, and the introduction of up to four different socket technologies when all platforms are considered. On a positive spin, AMD was finally able to bring their production down to 90 nanometres to combat heat and ramp up processor speeds. Perfecting the 90nm strained silicon on insulator technology allowed AMD and IBM to realize native dual-core CPUs with the Denmark, Toledo, Manchester and Taylor cores for their respective market segments. On the horizon, we see a more stable socket outlook than our 2004 analysis, as well as a move to 65nm in the near future, and lastly the introduction of quad-core products in the form of Barcelona and Altair for the Opteron and Athlon 64/FX series of CPUs. So, let's take a look at what AMD has done in the past, and what they intend to do on through 2008.

AMD Opteron

Original Release Date: April 22, 2003

Current Sockets: 939 (Opteron 1xx only - DEPRECATED), AM2/AM2+ (Workstation), F/F+ (2P/8P Opterons)

Note: TDP and voltage ranges include the HE (Low Clockspeed/Low Voltage/Low Dissipation) and SE (Higher Clockspeed/Higher Voltage/Higher Dissipation) CPUs, which are exceptions to "Normal" qualities in a given range.

Core Revisions (Chronological - Oldest > Newest):

Codename	Market Name(s)	Core Rev.	Cores	Die Size (nm)	Socket	Frequencies	Voltages	TDP
Sledgehammer	Opteron 142-150/ 240-250/ 840-850	B3/C0 /CG	1	130	940	1400 - 2400MHz	1.15v- 1.55v	30-89w
Venus	Opteron 144-154	E4	1	90	939	1800 - 2800MHz	1.35v/ 1.4v	67-104w
Troy(2P)/ Athens(8P)	Opteron 242-256/ Opteron 842-856	E4	1	90	940	1600 - 3000MHz	1.35/ 1.4v	55-92.6w
Denmark	Opteron 165-185	E6	2	90	939	1800 - 2600MHz	1.30/ 1.35v	110w
Italy/Egypt	Opteron 265-285/ Opteron 865-885	E6	2	90	940	1600 - 2600MHz	1.15- 1.35v	55-95w
Santa Ana	Opteron 1210-1220SE	F2	2	90	AM2	1800 - 2800MHz	1.3/ 1.35/ 1.4v	103/125w
Santa Rosa (2P)	Opteron 2210-2220SE	F2	2	90	F	1800 - 2800MHz	1.2- 1.375v	68-120w
Santa Rosa (8P)	Opteron 8212-8220SE	F2	2	90	F	2000 - 2800MHz	1.2- 1.375v	68-120w
Opteron Brisbane?	Opteron 1/2/8xxx? (Est. Release Date: 1H07)	G	2	65	AM2/2+ or F/F+	2100 - 2600MHz?	????	65w (Est.)
Barcelona ("K8L")	Opteron xxxx (Est. Release Date: 2H07)	H	4	65	AM2	2600 - 3000MHz?	???- ???v	68-120w?
Budapest/	Opteron 13xx [K8L w/ HT3.0] (Est. Release Date: Q407)	H?	4	65	AM2+	???? - ????MHz	??? - ???v	??? - ???w
Shanghai	Opteron 23xx/83xx [2P/4P Budapest w/ HT 3.0] (Est. Release Date: 1H08)	??	4	65	F+	???? - ????MHz	??? - ???v	??? - ???w
????????	Opteron ?????? (Est. Release Date: 2H08)	??	4	65	AM3	??? - ????MHz	??? - ???v	??? - ???w

The Opteron has a history that spans more than three years, with at least another year of life left in the product as a name. While it appears AMD is doing the socket shuffle over the next twelve months, the outlook of the Opteron is a little more secure than the roadmaps indicate. The first thing that needs to be understood is that the AM2 socket will evolve into the AM2+ socket when AMD introduces revision 3 of its Hypertransport bus architecture; socket AM2+ is pin and processor-compatible with any existing AM2 product, however to reap the benefits of HT3.0, the CPU must also be HT3.0-compliant. The AM2/+ socket itself is poised to fulfill the needs of workstations with Opteron CPUs, as AM2/+ motherboards are cheaper to produce, certify and maintain than their SMP socket F brethren. From now until the advent of AMD's 45nm fabrication techniques in 2H08, socket AM2 Opterons can be taken to any system in the interim, unless AMD drastically changes their plans. Therefore, to put it in its simplest form, anyone looking to run an Opteron PC with two or four cores in a single CPU is wise to purchase a socket AM2 motherboard if there is the intent to upgrade.

Socket F, on the other hand, is poised to fulfill the needs of high-end workstations, but primarily servers requiring two and eight CPU configurations. Socket F is a 1207 pin LGA design, and is utilised in the Opteron 2xxx and 8xxx product lines. Starting with the 2P Santa Rosa chips, Socket F is the standard that will be employed until atleast the autumn of 2008. Socket F+, much like AM2+, is an evolution of its progenitor to include support for HT3.0. According to AMD's forecasts, purchasing a Socket F Opteron 2xxx or 8xxx CPU will leave you a consistent upgrade path all the way up to the middle of 2008. Though it might seem like AMD is letting their Socket F fall by the wayside with the sluggish introduction of the F+ socket, this is simply not the case; instead, the lag time between the workstation release of HT3.0-compliant silicon for the AM2+ platform and the Socket F-platform has to do with SMP certification.

Administrators and rackmount manufacturers like HP and Pogo must assure that their mounts are guaranteed to be stable, displaying mission-critical stability, therefore any components designed for 2P and 8P applications must be stringently-tested, and this introduces a delay of up to six months. If you are in the market for a SMP socket that will cater to your upgrade demands for the next twelve months at the bare minimum, Socket F is your choice, but you will need to purchase a new motherboard and CPU if the benefits of HT3.0 outweigh the steep cost these boards and chips will undoubtedly carry. Like the Athlon FX, AMD will surely be beavering away between the winter of '07 and '08 to produce refreshed silicon within their 65nm lines to keep pace with Intel's rapid speed upgrades planned for the Xeon 5xxx line. If there are unforseen hiccups in AMD's master plan for AM2/+ and F/+ converging with AM3 in the 2008 holiday season, we can be fairly certain that they will follow the AM2 versus F stratification that AMD has employed to further segment their stake in the multi-CPU market.

Final Verdict

A single-CPU machine with an Opteron CPU will continue on the Socket AM2/AM2+ path for a while. The only confirmed 65nm part is coming in the form of Brisbane for the Athlon 64, but we have to assume that AMD will want to herald the benefits of colder, faster, cheaper Opteron chips for the server and workstation space around the new year. Based on this assumption, we advise you to wait and see how the Santa Ana compares to the upcoming Brisbane For Opterons core stacks up, and buy the victor of the two. Couple this chip in Q1/Q207 with a Socket AM2 board, and you could theoretically drop new CPUs in this board until 2008. But on the flip-side, roadmaps for the Opteron from July indicate that the Santa Ana/Santa Rosa cores will be in the limelight until the K8L shows itself in late 2007, and this may have to do with the SMP certification we spoke of earlier. By the time AMD has proven the reliability of the Santa Ana and Santa Rosa, it may be too late for them to economically unveil a 65nm dual-core, instead giving AMD the chance to jump straight to quad with the next product debut. If you're interested in an Opteron 1xxx machine, stick to the Athlon 64 line. The outlook for cheap, fast and cold 65nm dual-cores on the Opteron platform is just too hazy.

A multi-CPU machine with an Opteron will continue on Socket F/F+ well into 2008. Wait to see how the 2P/8P Brisbane for Opterons (If they'll ever exist) stack up against existing Santa Ana components, and then purchase the victor with an appropriate Socket F motherboard. If you are an enthusiast, however, we sternly recommend that you steer yourself towards the Athlon 64 FX 4x4 platform for 2P/4C designs, as the Opterons will carry a price premium even greater than that of 4x4-capable chips. Traditionally speaking, 2P Opteron boards are feature-sparse for enthusiasts, and we are assured that this will not be the case for the 4x4 platform.

AMD Athlon FX

Original Release Date: September 23, 2003

Current Sockets: 939 (DEPRECATED), AM2/AM2+ (Current), F/F+ (4x4 Platform Only), AM3 (Future - 2008 Release)

Note: Only the Socket F Windsor cores carrying the FX-70, FX-72 and FX-74 monikers are compatible with the upcoming 4x4 platform. We have also taken the libert of inserting Brisbane into the lineup for AMD's upcoming chips, because it is not certain whether ot not AMD will be using Brisbane for both the Athlon 64 FX and the Athlon 64. We imagine this is very likely, as having two seperate cores going down the production line for chips that are functionally identical is a bit absurd.

Core Revisions (Chronological - Oldest > Newest):

Codename	Market Name(s)	Core Rev.	Cores	Die Size (nm)	Socket	Frequencies	Voltages	TDP
Sledgehammer	Athlon 64 FX-51	C0/C6	1	130	940	2(200/400)MHz	1.5v	89w
Clawhammer	Athlon 64 FX-53	CG	1	130	939	2(400/600)MHz	1.5v	89 & 104w
San Diego	Athlon 64 FX-55 & FX-57	E4/E6	1	90	939	2(600/800)MHz	1.35/1.4v	104w
Toledo	Athlon 64 FX-60	E6	2	90	939	2600MHz	1.35/1.4v	110w
Windsor	Athlon 64 FX-62	F2	2	90	AM2	2800MHz	1.35/1.4v	125w
Windsor (4x4)	Athlon 64 FX-70/72/74 (Est. Release Date: Q406)	F2	2	90	F (4x4)	2.6, 2.8, 3.0GHz	1.35/1.4v	125w
Brisbane?	Athlon 64 FX-?? (Est. Release Date: 1Q07)	G	2	65	AM2/2+ or F/+ for 4x4	2100-2600MHz	????	65w (Est.)
Altair FX ("K8L")	Athlon 64 FX-?? (Est. Release Date: 2H07)	H	4	65	AM2+ or F+ for 4x4	2600-3000MHz?	???	125w (Est)
????????	Athlon 64 FX-?? (Est. Release Date: 2H08)	??	4	45	AM3?	???-????MHz	???-???v	???-???w

AMD is aggressively adopting a complete changeover to the Socket F/F+ for any system featuring two or more CPUs, and Socket AM2/AM2+ for any system running one CPU regardless of core-count. The outlook for the Athlon FX is very similar to the recommendations we provided for the Opteron platform: If you intend to stick with a single-CPU system for a while, a Socket AM2 motherboard will support current AM2 CPUs and non-4x4 processors in both AM2 and AM2+ flavors. If you intend to splurge on AMD's extravagant 4x4 platform, you will need to wait until approximately December of this year, at which time you'll be on the lookout for a pair of Socket F Athlon FXs and a Socket F board; this combo will carry you through the first half of 2008, as any codenames not on the roadmap will surely adhere to the AM2 to AM2+ / F to F+ evolution.

At the end of 2008's summer time, the future of the Athlon FX's cores and sockets becomes a bit nebulous. We understand that AMD is fixing to transition to a 45nm process as the leaves turn in the same year, and we are certain that the AM3 socket is going to come with this die shrink, but we simply just don't know what, if anything, AMD has planned for their specific segmentation within the Athlon FX line. We may see the introduction of FB-DIMMs, DDR3, and a whole host of other technologies as early as AM2+, but definitely in time with the AM3 release. In the mean time, as in the case of the Opteron, AMD will almost surely be hitting the drawing board between 2H07 and 2H08 for core revisions of the Altair FX, not warranting new codenames.

One thing that perplexes us, however, is the lack of Rev. G silicon on the roadmap between now and Q307 for the Athlon 64 FX series. As you'll see below, AMD has planned Revision G chips in the Brisbane for the Athlon 64, which brings much lower temperatures with similar clockspeeds thanks to the 65nm fabrication technique. However, when looking at the roadmaps for the Athlon 64 FX, Brisbane is not mentioned in the progression from 90nm to 65nm -- in fact, it functionally skips from the 90nm 4x4 Windsor to the 65nm quad-core Altair. We can, though, reasonably assume that the Brisbane will be making an appearance with premium clockspeeds and/or enabled ccHT links to warrant the price premium of the line, and to support the 4x4 platform.

Final Verdict

If you're sticking with one CPU, wait for the new Athlon 64 5400+, 5600+ and 6000+ Windsors and see how they compare to the Athlon 64 Brisbane in speed and overclockability. When the dust settles, purchase a Socket AM2 motherboard and the winning processor core. If you're interested in the 4x4 platform, wait to see how the FX-70, FX-72 and FX-74 Windsors do against Brisbane. Sadly, though, we're not entirely sure if Brisbane is coming to the Socket F platform -- it is a reasonable assumption, so we recommend waiting for the dust to settle. Buying any 4x4-ready Socket F motherboard will allow you a smooth upgrade path into 2008 unless AMD throws a curveball.

AMD Athlon 64/64 X2/64 X4

Original Release Date: September 23, 2003

Current Sockets: 939 (DEPRECATED), AM2/AM2+ (Current), AM3 (Future - 2008 Release)

Note: At this time, no Athlon 64 (X2/X4) CPUs are slated to be compatible with the 4x4 platform - this privilege is restricted solely to the Socket F Athlon 64 FX chips. The Orleans and Windsor-class TDP/Voltage ratings also include the values given to the Small Form Fact/Energy Efficient chips of the line which have very low, outlying thermal envelopes and voltage requirements.

Core Revisions (Chronological - Oldest > Newest):

Codename	Market Name(s)	Core Rev.	Cores	Die Size (nm)	Socket	Frequencies	Voltages	TDP
Clawhammer	Athlon 64 2800-4000+	CG	1	130	754/939	1800- 2400MHz	1.5v	89w
Newcastle	Athlon 64 2800-3800+	C0 & CG	1	130	754/939	1800- 2400MHz	1.5v	89w
Winchester	Athlon 64 3000-3500+	D0	1	90	939	1800- 2200MHz	1.4v	67w
Venice	Athlon 64 1500-3800+	E3 & E6	1	90	754/939	1000- 2400MHz	1.35/1.4v	51-89w
Manchester	Athlon 64 3200/ 3500+	E4 & E6	1	90	939	2(000/ 200) MHz	1.35v	67w
San Diego	Athlon 64 3500/ 3700/ 4000+	E4 & E6	1	90	939	2200/ 2400MHz	1.35/1.4v	67/89w
Manchester (X2)	Athlon 64 X2 3800/ 4200/ 4600+	E4	2	90	939	2(000/ 200/400)MHz	1.35/1.4v	89/110w
Toledo	Athlon 64 X2 3800- 4800+	E6	2	90	939	2000- 2400MHz	1.3/1.35v	89w/110w
Orleans	Athlon 64 3000- 3800+	F2	1	90	AM2	1800- 2400MHz	1.2/1.4v	35/62w
Windsor	Athlon 64 X2 3800- 6000+	F2	2	90	AM2	2000- 2600MHz	1.2-1.35v	65/89w
Brisbane	Athlon 64 X2 4000- 5000+ (Est. Release Date: 1Q07)	G	2	65	AM2/2+	2100- 2600MHz	????	65w (Est.)
Altair ("K8L")	Athlon 64 X4 ????-????+ (Est. Release Date: 2H07)	H	4	65	AM2+	????- ?????MHz	????	125w (Est.)
????????	Athlon 64 FX-?? (Est. Release Date: 2H08)	??	4	45	AM3	???-????MHz	???-???v	???-???w

Not since the Athlon 64 FX was on a Socket 940 interface has the Athlon 64 been any different from its bigger brother, which has meant performance without a premium for the rest of us; roughly speaking, since the tenth of October in 2004, both the Athlon 64 FX and the Athlon 64 had shared the 939 and AM2 sockets and the same silicon with San Diego, Toledo and Windsor. History has demonstrated that, at every turn, the Athlon 64 CPUs can achieve the same clockspeeds with the same voltages as the Athlon 64 FX, supporting the belief that the FX CPUs were naught more than premium speedbins of Athlon 64 silicon, even without knowing codenames to prove this fact. This will continue, in part, with the Socket AM2/AM2+ Athlon 64 FX chips being nothing more than superior bins of the Socket AM2/AM2+ Athlon 64s with Windsor now and Brisbane in 1H07, but AMD is finally preparing to give the Athlon 64 FX a purpose. With the introduction of the 4x4 platform, which will require 1207 pin Athlon 64 FX chips, the Athlon 64 will no longer be able to match the Athlon 64 FX in speed and features, as the Athlon 64s from today and beyond will not feature the necessary ccHT links to enable SMP computing. To put it simply, if you're interested in one CPU with two or eventually four cores, the Athlon 64 will continue to shadow the Athlon 64 FX's development and deliver identical clockspeeds and voltages, but if you're looking to move into 2P/4C rigs, you must finally abandon the Athlon 64 CPU.

Final Verdict

Our recommendation for the Athlon 64 parallels that given for the Athlon 64 FX: If you're in the market for one CPU, the Athlon 64 is the place to be, and we advise you to see what happens with Windsor and Brisbane and then buy the best AM2 board of the time. If you're looking for two CPUs you must finally drop the cash on an Athlon 64 FX or even an Opteron CPU if you want more than two.

The future of AMD is clear, but not necessarily bright. The debut of the Intel Core2Duo series, with the new microarchitecture, has spelled serious misfortune for AMD in recent months, and will probably continue to do so. The refreshed face of the Xeon line, an already-established 65nm process, and an engineering team that will beat AMD to 45nm and quad-core by more than a year has spelled disaster for the chaps in green. Right now, AMD is in scramble-mode to accelerate the production of 65nm and quad-core chips to get back in the race for supremacy which Intel has firmly stolen for now. All told, however, AMD has done an admirable job at solidifying their offerings in the upcoming years; giving a true weight to the Opteron line in the last 16 months, and granting the FX line a role other than the oft-overlooked big brother of the Athlon 64 were necessary and welcomed steps. Finally, and most importantly, AMD has outlined a clear socket path: AM2 to AM2+ with an Athlon 64 for people who want one CPU, F to F+ with an Athlon 64 FX for people who want two CPUs, and F to F+ with an Opteron for people who want more than two CPUs, with everything converging on AM3 in 2008.

Despite AMD's efforts to keep up with Intel, the 4x4 debut this yule season will be a token gesture for people who have read this guide, and are aware that impressive heat and cost-saving die shrinks are coming to the Athlon FX line in the first half of 2007. No one would seriously consider a Windsor Athlon 64 FX when the 65nm successors are expected to be cheaper, run cooler, and run faster. With that in mind, we must bid a temporary but sorrowful farewell to the age of AMD supremacy, and wait until the first half of the new year for AMD to stabilize its cores and solidify the 4x4 line. From that juncture, AMD must race to release the K8L and their 45nm technique just to stay in the game, which even then may not be enough.

Despite all the doom and gloom spread far and wide by pundits and analysts, AMD will have one thing going for them, if only by necessity: Price. AMD will have to undercut Intel if they hope to sway Core2Duo and Core2Quadro ogglers back to the banner of green and white, which means we may have only begun to see a revival of AMD's CPUs going for the tidy sum of change found beneath the sofa. Intel being no fool, has the profit margins and capital to fight back, so keep your eyes peeled for a price war like we have not seen since the Pentium 4-C and the Athlon XP.