64 Bit Top to Bottom
An exercise in Forward Thinking
by Josh Walrath
penstarsys.com
This artcile argues that Barton may be a sort of Clawhammer light. A Clawhammer with half the cache, no onboard memory controler, and designed to work with socket A.
Responses on Ace's
aceshardware.com
Johan is doubtful about this -
aceshardware.com
" Ok, let us look at the economical aspects:
- If true, 64 bit is no special feature of Hammer, makes it harder to market Hammer, to
convince people to move from Athlon to Hammer
- Negates a performance difference between Hammer and Athlon, in 64 bit mode, the Hammer
should be much faster in serverbenchmarks
- Athlon could eat market share away from Hammer
- AMD expects to launch the hammer in mobile, desktop, server, workstation -> the only
role that the Athlon has to play at the H2 2003 is being the budget CPU that takes on
Intel's Celeron.
- There is no need for 64 bit in the desktop market
Athlon 64 bit does not exist, neither does Intel x86-64 bit.
Johan De Gelas "
___________________-
Also from penstarsys.com -
"AMD officially announced the availability of the Thoroughbred based Athlon core as the new Mobile Athlon XP. AMD publicly stated that it would start shipping this part in late Q1, and because it is a mobile part, it is understandable why AMD didn't announce this at the end of Q1 (it usually takes notebook manufacturers about 6 weeks to integrate a new part into their lineup). AMD claims that certain markets will see this chip first, as none of the biggies have announced this one yet (expect HP and Compaq to do separate announcements, well, maybe not so separate anymore...). T-Bred is the shrunk (and partially redesigned) Athlon XP made on the .13 micron process. Apparently AMD hasn't had overwhelming problems switching to the .13 micron process (unlike other manufacturers), but it is still a time consuming and expensive process nonetheless. AMD will probably release the T-Bred next as the Athlon MP 2100+, and possibly the 2200+. From all indications the Palomino core truly is maxed out at 2100+, and the recent pictures that showed the Palomino core at 2200+ were fake. I of course could be wrong, but this is what AMD has told me, and so far they have rarely steered me off course.
Another piece of news that raised eyebrows is the new roadmap available from AMD here. Two very interesting things come out, first that Barton is no longer going to be made on SOI, and secondly it will have 512 KB of L2 cache. Many (including myself) had theorized that Barton would never be released as an SOI product due to AMD's need to produce as many Hammer chips as possible on their SOI process. It made no sense to take away Fab space from the Hammer series, which would see higher ASPs than Barton, but at the same time AMD is forced to keep its mainstream products competitive with Intel's Pentium 4 with 512KB of L2. This will allow the Performance Rating to be a bit more accurate than by just increasing clockspeed alone. What this also does is give AMD some more leverage in gaining performance vs. just increasing the clockspeed. Without SOI manufacturing for this part, it cannot clock up as high as a SOI part (some think that SOI can add up to 15% in terms of overall clockspeed). This is essentially two speed grades that AMD could not achieve. By utilizing the standard .13 micron process, and introducing 512KB cache, AMD is able to extend its mainstream performance, and keep it competitive with Intel until the Hammer achieves satisfactory market penetration.
Straight from AMD: Hammer is exceeding all expectations in terms of performance and expected time to market. Introducing a new architecture is hard, ask both Intel and AMD. It took nearly a year for Intel to adequately supply the market with Pentium 4's and quality motherboards, and it took about 8 months before there were significant numbers of Athlon systems with good motherboard support. Hammer is progressing very, very nicely. It has some distinct advantages over both the original Athlon and the Pentium 4 launch which will help bring it to market in a fairly quick manner. First it is based on the proven Athlon core, with some tweaks in there to extend its clockspeed headroom. Basing this design on the Athlon has allowed AMD to further extend the architecture without going through a massive redesign (not to say that the Hammer was a small undertaking). AMD had something good, and they felt that a few changes to the basic core could renew and expand the architecture (2 more pipeline stages, 64 bit processing, enhanced L2 cache, better branch prediction, on-die memory controller, etc.). The first Hammer samples that AMD showed were running in a very advanced state, considering how new they were. The underlying chipset architecture was also working much better than expected. AMD originally did not intend to show the Hammer running at their suite during IDF, but Hammer was running so nicely in the labs they couldn't resist carting it out and showing the press (and interested shareholders). Because Hammer is progressing so fast, AMD knew that it could not produce Barton on SOI at the same time it was trying to ramp production on Hammer to get it to market en-masse.
This is all good news for AMD. .13 production is finally hitting the streets, AMD has extended the Athlon architecture with the new Barton, and Hammer is coming along very well. I would expect AMD to deliver press systems in late August, with limited availability in September, and the full fledged market assault in early October. Motherboard support will probably not be a problem due to the Hammer requiring a 4 layer motherboard (vs. the 6 layer the original Athlon and Pentium 4 required). Not only that, but some motherboard manufacturers report that nearly 50% of their production goes to Athlon based products. These manufacturers are probably already working on AMD 8000 based motherboards, and I have a sneaking suspicion that NVIDIA will be one of the first 3rd party chipset manufacturers to release a product based on the Hammer architecture (can you think of another manufacturer that has more experience in implementing a HyperTransport based product?). Q1 and Q2 are always slow quarters, and this is a very good time for AMD to transition over to .13 in a slow and deliberate (and careful) fassion. There is not as much pressure to provide high performance parts to the market, and achieving parity with Intel in terms of performance is not so much a driving factor. It is definitely a time to slow down and take a very careful look at the entire market and how the company is addressing it. This is going to be a very long summer waiting for what comes out of the gates this fall..." |
