Slashdot Mirror


Behind the Closed Doors of AMD's Chip Production

rokali writes "Tom's Hardware is running an article on AMD's chipmaking procedure, plants, and future. Check out the pictures of Fab 36, their new plant slated to open in 2006, which will put of the next generation of 65nm chips. From the article: 'Currently, AMD's devices in Dresden are still produced on 200 mm wafers; the new APM 3.0 using 300 mm wafers won't be ramped up until Fab 36 opens. Production startup at the new facility is slated for the beginning of 2006, at which point the company will have invested an additional $2.5 billion.'"

8 of 151 comments (clear)

  1. Question by elid · · Score: 3, Insightful
    Incentives from the German government and the EU have lured a number of high-tech firms to the Saxony region of Germany, many of which have formed alliances. AMD, Infineon and ZMD work particularly closely together.

    Anyone know anything about this? What makes Dresden so interesting to AMD?

    1. Re:Question by homerj79 · · Score: 2, Insightful

      If I recall correctly from the PR fluff AMD put out a few years ago when they announced Fab 30, its due to the highly skilled workforce because of the Technische Universität Dresden (Dresden University of Technology).

      --
      SYSOP ('sih-sop) n.: the guy laughing at your typing.
  2. Re:Motherboards by ad0gg · · Score: 4, Insightful

    They used to have their own chipset and it sucked(speedwise and feature wise) compared to the VIA chipset that was out at the same time. AMD doesn't need its own chipset now since Nvidia makes a really great chipset.

    --

    Have you ever been to a turkish prison?

  3. Chipsets would suffice by toadlife · · Score: 5, Insightful

    Back when AMD still made chipsets for their own chips, the motherboards that used them were incredibly stable. I wish they hadn't stopped making them.

    --
    I don't always use unix-like operating systems; but when I do, I prefer FreeBSD.
  4. Re:New toys aren't cheap by ciroknight · · Score: 2, Insightful

    By integrating the memory controller, they've assured low latency. DDR2 simply says they can get all of the data they need all at once, and store it in a local cache. Meanwhile, the Pentium 4 beast is a memory hungry chip; most operations the chip spends its time toiling on are operations that require a lot of data (SIMD) (why do you think they hyped it for movies and such??). This means the Pentium 4 has to hit the memory bus more often, or increase its cache size. We've already seen the Prescott go up to a 2 meg cache, and the Extreme Edition go up to an L3 cache to keep local copies of even more data, but it's simply not enough to keep the Pentium 4 competitive (not to mention it drives up the thermal profile of the chip quite a bit). The fact is it simply needs to go to local memory more often. Integrating a memory controller makes it cheaper to go to local memory, instead of waiting on the North Bridge to fetch the information from memory and send it back up the pipe. This is why A64 can go to ram, pick up huge chunks of memory at once, process it, and send it back, while Intel's Netburst chips sit and wait, idling their time away.

    Processors have actually gotten too fast for their respective systems. AMD tries to get around this by bringing the memory closer to the chip. Intel's trying to get around it to bringing the memory on the chip, and Intel's approach isn't working as well as they'd hoped.

    --
    "Victory means exit strategy, and it's important for the President to explain to us what the exit strategy is." G.W.Bush
  5. Re:New toys aren't cheap by VoidWraith · · Score: 2, Insightful

    A single G5 is not in the same bracket as these new Opterons. If you want an AMD chip to compare to G5s, look at Athlon-64s. I'm not speaking about Suns, because frankly I don't know enough about them.

    Your suggestion of using more less expensive processors works for x86 processors too. Why use an opteron 8xx when you could use a few Athlon-64s? That's the sort of approach Google takes, redundant arrays of inexpensive computers.

  6. Re:New toys aren't cheap by evilviper · · Score: 3, Insightful
    A single G5 will run you about $1500,

    Okay, but the Opteron in question is a DUAL-CORE chip, for servers only (desktop chips are much less expensive), still unreleased, and it's only the MSRP, whereas the real price you can buy them for will surely be lower.

    Besides, that is the high-end Opteron. The low-end dual-core chip is the 165 for $637.

    4 XServes be cheapter to assemble, run, and maintain than the processors used to make one 8-way Opteron server.

    Yes, but people that need an 8-way system can't just use 4x 2-way systems, otherwise they'd be doing that! Just as people that bought a 64-bit system so they could use 16GBs of memory, can't just have 4GBs of RAM in 4 different systems instead...

    If they wanted the equivalent of an XServe, they'd be going with lower-end Opterons, such as the $637 one. The parent even said as much in the first sentence, which you completely ignored. Show me an 8-way G5 system, and then you can compare prices...
    --
    Slashdot gets worse every day... Pipedot: News for nerds, without the corporate slant
  7. Re:Any market for single-core-only rejects? by Fulcrum+of+Evil · · Score: 2, Insightful

    That is like saying people who plunk down a few hundred thousand dollars for a really nice (or at least expensive) car or boat will not tune it. $1600 is dirt cheap for a 4-way configuration; I expect that the overclockers will give it a lot more attention than they give current quad-cpu offerings.

    Exactly. At $300k, you're buying a pretuned car that is damn fast and difficult to improve on. A $1600/cpu 4way box starts at around $10k or more and is useless for games. I can't see overclockers spending the price of a cheap car on a computer just to crank up the speed and burn it out. That's potentially a $6400 mistake.

    --
    "We returned the General to El Salvador, or maybe Guatemala, it's difficult to tell from 10,000 feet"