Update July 2004: AMD went against their roadmap and launched the Sempron CPU without the Intel Product Name scheme. We reported on the event here, but have left the article up for historial value.

4AM and the phone rings; our Asian correspondents have another roadmap for us. Excellent. After all, who doesn't enjoy writing about AMD processor roadmaps in the indecent hours of the morning.

Fortunately our perseverance is not unrewarded. The first important information we have to report on is the impending death of Duron. While many say good riddance, many others are reminded of our recent Duron benchmarks to conclude there is a definite processor hierarchy in the low end market. AMD's new chip; code named "Value" for now will eventually replace Duron, although specifications are hazy at best. What we do know is that the new chip is designed to aggressively compete with Celeron (with an advantage) and Celeron Only. Interpret that as you will. We also have other interesting specifications; the processor will be 32-bit only, it will span three socket architectures (Socket A, Socket 754, Socket 939), it will support dual channel memory on Socket 939 and will have the same NX (No eXecute) "antivirus" support found in the new Athlon 64, Efficeon, future Intel and Sparc processors.

According to the roadmap, the new processor will essentially replace the Athlon XP for Socket 754 which has been talked about in roadmaps before this one today. AMD's new move to specifically market to and target low end "Value" sector seems contradictory to the same roadmap which prints a study claiming the sub-$600 PC market does not particularly regard processor brand as important. AMD sounds fairly aggressive with the initiative to outpace Celeron D (Prescott based Celeron, 512KB L2 cache 533MHz FSB) which will already surface this week.

But more important than what AMD will name the new processor is how they will name the new processor; recall Intel's move to product numbers? AMD will in fact move the "Value" based processors off a "PR" rating to a number based system instead. AMD already has experience with this in the Opteron line, but moving the entire value segment to mimic Intel's rating system seems just as ambitious. We can only guess that AMD is testing the water for a new naming procedure on the sub-$600 PC market that does not regard processor brand name as important. Particularly when the newest roadmap claims:

AMD is evaluating the model number methodology for the "Value" brand giving recent press articles on Intel's change from GHz to model numbering.

Undoubtedly we will see AMD shift to a similar nomenclature for their other product lines if they can similarly market their value segment accordingly to Intel. Without a doubt, AMD's numbering system bares a striking resemblance to that of Intel. According to AMD's roadmaps, these new processors appear to be based off the 256KB L2 130nm "Paris" cores for Socket 939/754 and 512KB L2 130nm "Barton" cores for Socket A.

AMD "Value" Naming
Processor Socket Launch Date Old Name
"Value" 370 Socket 939 Q2'05 N/A
"Value" 350 Socket 754 Q4'04 N/A
"Value" 340 Socket 939 Q4'04 N/A
"Value" 320 Socket 754 Q3'04 N/A
"Value" 320 Socket A Q3'04 XP 3200+
"Value" 300 Socket A Q3'04 XP 3000+
"Value" 280 Socket A Q3'04 XP 2800+
"Value" 270 Socket A Q3'04 XP 2700+
"Value" 260 Socket A Q3'04 XP 2600+

Aside from AMD's new approach to the value segment, we also have more information on new processors. According to the recent roadmap we can expect to see Athlon FX-57 in Q2'05. To no surprise FX-55 and FX-57 will only debut on the dual channel Socket 939 architecture. Most interesting, however, is the small footnote below the roadmap which claims the FX-55 and Athlon 64 >4000+ (presumably 4200+) will require 104 watts/80 amps. According to P = IV, we assume the processors must be running at 1.3V. A significant change in core voltage can only mean that AMD is using a different core for these processors-- current Clawhammer Athlon 64s run on 1.5V. So are we to believe the upcoming 90nm transition to "San Diego" and "Winchester" cores will come this fall? If so, the design of the upcoming 90nm "San Diego" and "Winchester" architectures will decrease core voltage but increase overall wattage. This sounds extremely similar to Intel's newest 90nm venture, Prescott.

All of this news comes hot on the heels of AMD's open announcements to pursue dual core technology, which in turn comes hot on the heels of Intel's February decision to do the same. Perhaps as Intel and AMD both approach the threshold of power consumption, the solution becomes more instead of faster. Wasn't it a 15th century playwright who once claimed "Many hands make light work?"

Comments Locked


View All Comments

  • AtaStrumf - Tuesday, May 18, 2004 - link

    I will miss Durons as well! Bought three of them and one is still running beautifully at 2,0 GHz (1,65V) and it cost me less than half of my Barton 2500+ (OC-ed to 2.2 GHz @ 1,65V) and since I unlocked the extra cache its giving me allmost the same performance. You just gotta love 'em.

    I just hope AMD gives us something else to play with. All their new chips are just too expensive to be as much fun as Durons.

    As for the new roadmap, it's obvious AMD is having the same problems Intel is hence they're both taking the same path --> 90 nano dual cores. But I hate this new naming scheme. It'll take me a week to figure out what the hell a 3/5/7/?/?xx is. Talk about following the leader :(
  • HammerFan - Tuesday, May 18, 2004 - link

    Kristopher, how about this for a multi-core hybrid?

    Integrate to seperate cores onto the same package, with the links between the two cores contained in the package and not the die itself. The could eliminate the yield issue and keep latencies lower than two completely independent chips.
  • acx - Tuesday, May 18, 2004 - link

    Integrating multiple cores onto one chip most likely allows a lot more optimization tricks for synchronizing a MP system than multiple chips on a board. The architect controls everything on a single chip and the latency to get from one chip to another is a lot less. Besides, there is more than enough transistors available for 2 cores (sans cache) so why not?
  • HammerFan - Tuesday, May 18, 2004 - link

    I like the idea of multicore CPUs..as for heat issues, why can't AMD stretch 130nm SOI far enough to allow them to produce multi-core CPUs. I'm sure there are plenty more tweaks to make on SOI to get it to run faster than it does now.
  • KristopherKubicki - Tuesday, May 18, 2004 - link

    Zap: In my opinion, several smaller cores on one chip would yeild better performance but several chips on one board would yeild higher (thus costing less?). Maybe the solution is a little bit of both?

  • ZapZilla - Tuesday, May 18, 2004 - link

    Interesting that a few years ago multi-core chips were forcast as a result of "hitting a wall" in the ability to decrease process size.

    Yet, today we find that multi-core chips are needed due to heat density issues of the ever-larger number of transistors on chip.

    Alas, there is no money it seems in moving from one process generation to the next with constant or small growth in the number of transistors.

    So which would be better, a single multi-but larger-core chip or several single, smaller-core chips (on a smaller process) plugged into seperate sockets?
  • gherald - Tuesday, May 18, 2004 - link

    Anemone, you are 100% clueless. *Everything* you just said was wrong.
  • Anemone - Tuesday, May 18, 2004 - link

    Ok I'm liking the FX-57, and I'd consider it if 90nm, but I'd have to see pci-express boards on the NF3 250gb before that would be a mixable combination.

    Not sure how I feel on this but its quite surprising and pleasantly so, that AMD would be appearing to come back into the game with a sooner release of the FX-57 rather than later. If I have kept up on the news this could be a 90nm part that also would incorporate the use of up to 500mhz DDR and potentially SSE3 (not that SSE3 over SSE2 is a showstopper).

    Because AMD delays so often, I'll probably have to believe it when I see it, but I think its the right direction they need to move, sooner rather than later to keep the momentum they've gained so far going.

    Now we need promised dual core desktop cpu's on 939 socket sealed in a promised date. Otherwise, dual cores are questionable as to socket needs and timing and certainly whether they will offer dual cores to the desktop (at least FX pricing level products).

    Good, but we need more, and they can't delay. Faster, sooner, is good and will help convince the market to use them, not just the enthusiasts.
  • Zebo - Tuesday, May 18, 2004 - link

    Just to follow up. At the business level if I were to build/buy, say 100 machines for my office workers I could save $5000-$6000 right there building machines with a Duron over a barton. You think Mr's Smith plugging away on word and email cares or would notice any difference? No. But 5K is noticeable.

    Even for enthusiasts I doubt they'd pass an a/b test w/o using benchmarks.

    For overclcokers a Duron was a steal easily overclcoking to 2.3Ghz, better than all bartons OC but the mobiles, which carry thier own premium in price.

    Nope it's a really sad day to see this chip go.
  • Zebo - Tuesday, May 18, 2004 - link

    Aww poor Duron the best bang for the buck chip EVER! So unappreciated so sad to see it's demise.:(

    #7 show me one benchmark, any test, of a $100 Barton performing twice as well as a $50 1.8 Duron and you may have a point, but thier within 20% of one another accross the board!! Making the Barton a relative rip-off just like all pentiums are.

Log in

Don't have an account? Sign up now