Slashdot Mirror
For AMD Success Means Problems
An anonymous reader writes "AMD's success with its dual-core Opteron and Athlon processors has created something of a happy problem for the company. It can't make its products fast enough to meet demand. Just the same, with the Intel price war heating up and new 65-nanometer manufacturing technology being implemented in its factories, AMD has a lot of balls in the air right now." From the News.com article: "AMD's current pickle is the result of its success, which makes it a little easier to swallow for company executives. Demand is high, but the company's dual-core processors still use its 90-nanometer manufacturing technology. Intel's chips, on the other hand, are built using the smaller transistors provided by its 65-nanometer manufacturing technology. Not only is AMD using larger transistors, but its dual-core Opteron and Athlon 64 processors contain two processing cores integrated onto a single piece of silicon, or a die. This design has given AMD great performance during the past few years, but resulted in processors that were almost twice the size of its single-core chips."
Bah, I was hoping with only like three posts up I could be the first to mention that. My PCs use AMD processors almost exclusively and I have been using AMD since my bloody 286. My main work machine, however, is now a Mac (a PPC one currently).
Since Apple first announced Intel I thought that it was pretty obvious why they went that route rather than AMD. This, right here, was one of the main reasons. Supply programs have haunted Apple for quite some time. Why switch to a new architecture just to get more of the same?
AFAIK, this has always been AMDs problem: my earliest recollection is when they bought NexGen's K6 and sold it to Compaq in the sub-$1000 segment in 1995. Since then, anytime the get a good product, they blow it on production, leaving Intel to fill the void they created.
It is where they have failed again and again and again. I can't believe they haven't learned yet.
https://www.accountkiller.com/removal-requested
when I took microelectronics courses in university about 15 years ago the lower limit for our process was around 2um (if I remember correctly) and my professor several times seemed to strongly believe that the lower limit for gate length was around 0.6-0.7um for various reasons. Nowadays we're way smaller than that, and it's getting even smaller as time goes on: is there a website somewhere that details exactly which theoretical advances have been made during the past 10-15 years to enable processes to continue getting smaller?
-- the cake is a lie
Sure, especially when you consider that one of the problems Apple had with IBM/Motorola is that they were always in short supply of chips. Apple was constantly delayed in releasing new models because their processor suppliers weren't meeting their estimates.
Yep, that huge technological lead of a few months. This is the first time since really K5 that AMD has had production issues. And, the production issues described in the article are really the good kind. The early ones were the bad kind where the chip design itself caused the production problems. The good kind is where yields are good, speeds are good and the design is manufacturable. Your only issue is figuring out how to push more die through the line.
Fab 36 will be online soon with 300mm wafers and 65nm. Just going to 300mm wafers pretty much doubles capacity. Going to 65nm gets you another say 50% (anyone got a confirmed number). Getting FAB 36 and FAB 30 going doubles capacity again. So, by my calculations that is 2 x 2 x 1.5 or 6x cpacity increase for AMD in the next couple of months.
Did Intel switch to 65nm and 300mm sooner than AMD? Yes. Did they switch to copper and low-K dieelectric before AMD? No. Did their 90nm production even work quite right for Intel ever? Not sure. When the 90nm P4s used more power than their 130nm brethren you have to wonder.
Of course Intel has something like 6 processro FABs all over the world that are likely larger than AMDs. Doesn't take much R&D just to build more capacity especially when you are the 800lb gorilla.
Basically, Intel and AMD, at this time, are quality processor maufacturing operations. Intel tends to make technology switches before AMD, but they also get to deal with first adopter issues. And, when they both buy there equipment form the same semiconductor equipment manufacturers like Applied Materials, Novellus, and others. How much is AMD benefitting from Intel working out the bugs.
No kidding. Thanks to the price war I scored a 2.4Ghz Core 2 Duo really cheap, and have it happily overclocked at 3.06Ghz. I wanted to go AMD, but I didn't because:
:)
1. They are no longer cost-effective in comparison to Core 2
2. Compatibility issues; athough the chipset I chose is not 100% supported on the kernel rev I'm running, it's still a far sight better than getting an ATI or NForce chipset to run acceptably well. Also, The AMD-ATI merger does not bode well for Linux users, given ATI's abysmal track record. I refuse to buy ATI products and am now avoiding AMD until I see whether or not ATI cleans up its act. (insert a rabid "fuck ATI" right about here)
3. at stock clock speeds, it is 80% faster (according to benchmarks) than the Pentium D I sometimes use at the office, and well over 100% faster than the Pentium 4 (the other box) my primary box at the office. At 3.06Ghz, it's (obviously) much faster than even that.
I wanted to go AMD, I really did, but with Intel's quad core processor coming out Real Soon Now and with my board's already being certified to run it, it was the logical choice. Quad core upgradability was the clincher.
The Christian Right is Neither (Christian nor right). See: Matthew 23, Matthew 25, Ezekiel 16:48-50
This really begs the question. I'm sure there's a reason somewhere but...
Why DO they use round wafers to create the chips from. Certainly there would be less waste with square wafers (though less and less so as the chips get smaller and smaller, of course). Are there other savings in the manufacturing process that make round wafers more cost efficient in the grand scheme of things than square ones?
Supply programs have haunted Apple for quite some time. Why switch to a new architecture just to get more of the same?
Except that it wouldn't be the same. Apple was dependent on Motorola and later IBM because they were the only suppliers of the PPC chips they needed for the Mac. However, since AMD and Intel both make x86 chips that function the same, they wouldn't be dependent on AMD if they went that route because they could switch over to Intel chips at any time without much trouble (and vice-versa I suppose - Apple could switch over to AMD right now if they wanted to).
AMD used to make Pentium clones. Now, though, the AMD architecture is completely different from Intel's although they both will run the same software. The 64-bit AMD cpus seem to have fewer software faults when running Windows XP compared with the Intel P4s. This is an observation based on only a few systems and a LOT of things besides the cpu can affect that but I wonder if anyone else has noticed this (or maybe the opposite)? The comparisons between cpu architectures are always based on speed and benchmarks but not stability. Has anyone ever compared the different designs for how many GPFs they throw off, other things being equal? I was thinking maybe that's one of the reasons why the AMD systems are still selling so well, even though the new Intel Conroe is faster.
When you can't make your product fast enough for all the demand, you're not charging enough. If you charge more, you can use that to increase manufacturing capacity. I'm sure someone at AMD understands that, so maybe they were caught off guard and are backfilling orders and have decided just to not reduce the price as early as they would have.
Also, The AMD-ATI merger does not bode well for Linux users, given ATI's abysmal track record. I refuse to buy ATI products and am now avoiding AMD until I see whether or not ATI cleans up its act. (insert a rabid "fuck ATI" right about here)
Naively, since AMD, the larger company, bought ATI I would figure AMD's culture of openness would dominate. At least, I hope this is the case, because I'm sick of having only one choice for graphics cards (though I'm used to it; ever since 3dfx).
Anyway, it doesn't seem fair to refuse to buy AMD's processors (as opposed to graphics cards/chips) because of ATI's poor linux support when AMD has been a huge supporter of free software, in particular of gcc and to a lesser extent the Linux kernel.
Though I will still not buy AMD/ATI (or DAAMIT as the Inquirer calls it) graphics products until they do actually improve. And I will continue buying AMD processors until they start becoming more like ATI.
The enemies of Democracy are
It's really not that simple. The process size is usually given as the minimum transistor length. Other parameters (minimum metal trace width, minimum distance between active regions, etc) may not shrink by the exact same amount. So you can't just scale your 90nm design by 65/90 and expect it to pass the design rules. I'd say that 1.5 as many chips is a reasonable guess. If I wanted a better guess, I'd look up the die sizes of previous chips that made process transitions with a minimum number of architectural changes to get an idea what real scaling factors you can expect.
On the other hand, the smaller your chip the better wafer utilization you get because the square chips better fit into the edges of the round wafer. This isn't a huge effect, though, unless you're talking really huge chips vs really small ones.
The big deal is that AMD's 65nm process is also coming with 300mm wafers. That's 2.25x the number of die/wafer right there, plus a little bit from being able to again pack chips to the edges of the wafer better.
Fab 36 is going to mean a hell of a lot more capacity for AMD. They're probably estatic to be having this "problem" right now.
The enemies of Democracy are
Unfortunately, a hexagonal die is a lot harder to design, and harder to cut out of the final wafer, than a square one. The second problem is probably solvable relatively cheaply. The first is not.
I am TheRaven on Soylent News
You can work out AMD's yields by how they price their parts. The ones near the pricing sweet spots are likely to be the ones they get the biggest yields of.
I am TheRaven on Soylent News
Process steps move forward every 2 years.
45nm in 2008
32nm in 2010
more generally:
0.7 * gate_length(N) in year(N) + 2
Thermal limits will hit hard in 10+ years. No exponential can go on forever.