Originally posted by: dmens
Originally posted by: Viditor
Intel is ahead in some things and behind in others...
While they started shipping 65nm first, that doesn't mean AMD couldn't...it just didn't make financial sense for them to.
Many people are under the impression that shrinking to a smaller node saves the manufacturer a lot of money, but this is hardly ever the case.
For example when Intel went from Smithfield to Presler, they were able to shrink the die size by ~21% (from 206mm2 to 162mm2). However, this die size reduction is offset by the expense of lower yields and the purchase of new equipment to create it (at least for the first few turns).
What they DID get however was something they desperately needed...a drastic reduction in power and heat!
AMD had no need for this reduction in power or heat, so by waiting they are able to tweak the ramp without the expense of going into full production on the chips. This means that when AMD starts volume production of 65nm in August, they will already be at fully ramped yields from day 1 (they have confirmed this in their CC).
As to other tech advantages from the factory, remember that:
1. Intel still can't produce SOI cost effectively
2. AMD is able to tweak production on the fly because of APM, while Intel must make their tweaks in one line, then copy that accross to all of their other Fabs and lines. This means that AMD is FAR quicker in performing changes.
3. AMD is actually ahead of Intel at the moment in strained silicon. They are able to embed strained liners (DSL) on both the NMOS and the PMOS with an SiGe layer, making the PMOS as fast as the NMOS...Intel is not quite there yet.
All of that said, Intel does have MASSIVE manufacturing capabilities, and enough cash and market power to initiate any new technology much quicker than AMD can. AMD must wait until a manufacturing decision is at it's most efficient...Intel doesn't have that limitation.
process moves always make financial sense. equipment costs are fixed (and damn cheap in the global view of things). yields/bugs are issues to be discussed only after a bad ramp, never during planning, which is when you evaluate an action's financial impact. a bad ramp is simply when the first design on the process (tick in intel parlance) and the process did not converge. that did not happen with intel 65nm.
1. While over the long term a smaller process node certainly is a good idea, the timing of that shrink is also very important. Equipment costs are indeed fixed, but a delay can decrease the outlay and can also yield benefits in improvements on design. Being the first to market isn't always the most economical move. It's a rule of financial management that the longer you can continue using existing equipment, the more you reduce the cost of that equipment...in other words, capital expenditures actually vary depending on the lifespan (even though you paid only a single price). And the cost of the equipment for a new node is in the $Billion range overall.
2. Yields and bugs generally have a target amount. For example, with Intel I would bet that their target was to have an equal output to 90nm on 65nm...mind you, the percentage yield would be lower in this case (because each die is a smaller percentage of the wafer), but I imagine that it would constitute a win for them. AMD's goal was different...they have a target for an equal
percentage yield, making the output higher than the previous node (and the cost/part lower).
even with the 65nm move, cedarmills were still hot. afaik, power was tracked, but not a critical deciding factor on tapeout. the real solution to netburst thermals was getting merom (and friends) to market ASAP. I find it hard to believe anyone can say that AMD had "no need for reduction in power and heat". That is a constant pursuit, no matter how big of a lead you currently have. Ignoring the power metric would be an extremely shortsighted move. I'm sure AMD has planners, and they knew that intel was moving to a P-M based core. Performance aside, they must have known whatever came out of IDC would consume far less power than everything AMD had. now we have a 65W B0 woodcrest kicking the crap out of 80W+ opterons in many tests.
1. It's a question of priorities...of course AMD doesn't ignore power/thermals. However, they also knew that they had enough of a lead that Intel isn't going to catch up until the end of the year (when AMD's 65nm begins shipping). Given that there is less pressure for them to decrease power immediately, they chose to maximise yields first...and they sweetened the deal by adding the new DSL SiGe process as well.
2. Calling the Woodcrest a 65W part and the Opteron an 80W part is VERY misleading! Those are TDP numbers which mean absolutely nothing to actual power use and can't even be used to compare thermals.
Intel and AMD use entirely different metrics for their respective TDPs, so it's like comparing the numbers of Celsius and Farenheit without an equivicating formula.
In
Johan's article here on AT, you can see that the 65nm Woodcrest uses more power than the 90nm Opteron.
However, that said it's also worth noting that he concludes that Woodcrest has a better performance/watt number because it performs faster...
back to the issue of "tweaking the ramp". i read that as a failed ramp. imo, the fact a ramp needs to be tweaked means the process team screwed up. the process should have been ready as soon as the design taped out. any less is lost revenue.
Both Intel and AMD are constantly tweaking the process to improve efficiency. That's one reason that different batches of CPUs will overclock differently. They get feedback from the way parts are binned and by altering the doping amounts or any one of the many other factors, they can improve the yields/performance. This is the essence of the term "mature yield"...it means that the process has been tweaked about as far as it can be.
in regards to the "advantages" you listed, they're not advantages, just design/cost tradeoffs. You can design around leakage and body effect, and SOI has its own quirks to be dealt with. I didn't read much about the strained liners thing, pretty crazy if they can make P as fast as N. imho, that sounds awfully like marketing braggery, much like the 40% increase number they tossed. but even that can be designed around. just use domino... intel has mucho experience with that topology, heh.
in any case, manufacturing is simply not AMD's strong point... they need to fight with their own assets in mind. maybe its just me, but AMD seems to be trending towards the supercomputing front. the consumer market looks to be closing up pretty fast for them.
The 40% increase is BECAUSE they can now make P almost as fast as N...
Article
I don't understand why you think that the consumer market is closing up on AMD at all...
1. Gaming - Conroe looks like it will rock...however at shipping levels of less than 10% of the mid-high range only in Q3 and 15% in Q4, the pricing and availability will allow AMD to sell everything they can make this year. Also at the very high end, while many have frowned at 4x4, what if you used (instead of 2 x FX CPUs) an FX-64 and an Aegia physics coprocessor or a GPU in the second slot...
From
ArsTechnica
"From a technical perspective, a multisocket coherent HT system that includes a dual-core processor and a tightly coupled GPU or physics coprocessor is a fantastic idea. If ATI or NVIDIA were to take AMD up on the licensing offer, such a system could make for a high-performance, expandable, and relatively low-cost God Box. Think about it: you don't have the extra production cost associated with a graphics daughtercard, you can cheaply expand the amount of DDR2 that's attached to the GPU socket, you get the benefits of a shared pool of system and graphics RAM, you have a high-bandwidth link directly between the CPU and GPU with no intervening bridge chip, and so on. It has all the makings of a killer gaming rig that, while expensive, might still be cheaper than a comparable Conroe system"
Now couple that with the continuing rumours of AMD looking to acquire or form a partnership with ATI...
2. Mobile - The basic fact is that Turion continues to gain marketshare...this will most likely continue even after Merom, but it will be relegated to the value laptop arena (the largest segment).
3. Mainstream - I must admit that this is the area where AMD will probably get stomped in the consumer lines...not from Conroe (shipping volumes are too low) but from the PD price cuts. I don't think it will effect their revenue share overall, but Intel will certainly have their Day in the Sun here. Of course, the dual core K8Ls in H1 07 may flip that around again...
Facebook
Twitter