Originally posted by: hans007
Originally posted by: anthrax
Well, from looking at the trends and early reports on the Conroe, AMD will loose its performance leadership for desktop processors.
From observation.
Conroe offers better performance per clock. It also has a deeper pipeline at 14 stages. AMD64 has 12 stage and still can't match the Conroe.
AMD can counter with the following.
1.)AMD can increase performance of AMD 64' by the following means:
a): Increase CPU clock speed. (shortest time)
b): Increase memory B/W (short term, it has been planned for ages)
c): Increase Cache (short term)
d): Use 65nm (medium term)
e): New archtecture (long term)
2.) However, implmenting the above isn't soo easy.
a.) Increasing Clock speed on the AMD 64 is going to be tricky on a 90nm. The fastest AMD64 is 2.8GHz and those don't seem to o/c past 3.0GHz very well. Its fairly safe to say that 3.0GHz is about the limit for 90nm.
However, even saying that 3.0GHz is the limit, most current AMD top out at 2.6GHz, so its safe to say that 2.6 GHz is about the fastest you will see a 90nm AMD 64 aimed at the mass market.
b.) I don't expect the new AM2 AMD64 to bring huge performance increases.
- DDR2 is unlikely to bring huge performance increases to the A64 platform. (THG did preview of the AM2)
- Just look at the switch from single channel DDR to dual channel DDR on A64. The benifits were very small.
- DDR2 also has increased latency that will detract from the gains from increased memory B/W.
c.) AMD can increase performance by increasing Cache from 1MB per core to 2MB per core. However, this is a very very expensive option for AMD. A dual core AMD64 w/ 2MB per core will have a die size of 300mm2 if made on a 90nm process. Thats 50% more die space over the current one. Here is the comparison
AMD single core 512kb (84mm2)
AMD dual core 1048kb (199mm2) 136% increase over single core.
AMD dual core 2048kb (> 300 mm3) 50% increase over dual core.
This clearly shows in AMD 64 x2 are significantly more costly to manufacture than single core AMD64 for the above reasons and a priced accordingly.
AMD is also haveing some capacitiy issues. If it aggresively pushes Dual-Core CPUs, they will be only shooting themsevles in the foot. Dual core takes 2 the die space but doesn't sell for twice the price. Their competitor will definately exploit this and will heavily push Dual Cores at low prices.(They have more capacity and also use 65nm) So in other words, dont' expect huge drops in prices for dual core AMD 64's.
d.) Well, expect 65nm from AMD in 2007. This will allow them to increase performance with further with more clockspeed and more cache. However, its competitor is ahead.
Also, the competitor's CPU offers better performance per clock than the A64.
e.) AMD can push a new architecture. But thats going to take quite some time. At least late 2008
To sum it up. AMD won't have having a performance lead once its competitor pushes its new chip.
- AMD is actually the one that has more to loose in a huge push to Dual Core. Increase die size will decrease production and reduce revenue unless Dual core chips can be priced 136% that of single core chips. (Its competitor can easily push a new marketing push and educate users to expect "Dual Cores" on their new PC's. That will really cuase AMD some problems.)
- AMD might still retain some advantages in the server market. Xeons don't scale very well when compared to AMD64 expecially past 2 cores. (Hyper Transport vs FSB) However, AMD has very limited time to exploit this weakness. Server market chips typically have higher unit prices too. This can increase revenue. However, it must be noted that most volume servers are still 2 ways and upping the #CPU is normally costly in terms of software licences for products like Oracle, Windows Server etc. etc.
one of the biggest risks for AMD is they have no money. unlike intel they still aren't THAT profitable.
if they end up dropping their ASPs to lik e$70 a chip again (unlikely but their ASPS will probably drop once conroe is out, instead of increasing prices of X2s like they are now) they will be impacted financially.
their big strong hold is 4 socket and up servers once woodcrest is out, and tha tis why hector ruiz said he is going to possibly take production away from consumer grade stuff and move it to enterprise (to protect ASPs basically).
the thing is if AMD has to go through a 2 year span of selling chips at $70 asp again they would just not have enough money to ever upgrade their factories to match intels. they are having trouble as it is buying 65nm equipment and intel is already planning for 45nm. so i hope they pull something out of their ass because they need to stay profitable for this to even be a 2 horse race.
Answer to Anthrax:
1. If you look back on history, both marketshare and revenue share don't change much for at least the first year when a speed crown goes to the other guy. It takes at least a year for the momentum to shift (and even then it's very slow), and that means that the short term will include K8L and possibly the K10. The only exception to this was when the original Athlon was introduced...but the reason there wasn't performance. The reason for the quick shift at the time was
a.) Intel made the worst platform mistake(s) in their history in addition to underestimating Athlon's performance
b.) AMD dumped all of their K6 inventory in Asia at cost
2. Rev F CPUs (AM2) will have something else in addition to DDR2...they will run at lower power. If you look at AMD's roadmap again, this will become evident in hints like the 35w X2 3800+... Remember that roadmaps can change quickly if they maintain the same architecture for OEMs. In other words, changing the TDP and adding a couple of speed grades at the last minute isn't unheard of (though it is unusual). The lower power profile means that higher clocks should be much simpler than on current chips.
3. Hector Ruiz stated in recent JP Morgan conference that volume production of 65 nm begins sometime near August this year. He also stated in the January CC that
"Our 65-nanometer data, which is already coming out, looks very encouraging. As a matter of fact, we have already microprocessor products built on 65 nanometer that are really looking as planned. Excellent at this point in time, and are confident that our ramp beginning the second half of the year will go well...what motivates the change from one node to the other, frankly, is that we able to meet our customer needs and demands are better as, and how rapidly we can make that transition". This means that AMD
could bring 65nm early if it sees it as necessary.
4. While I agree with you that Intel will probably take the speed crown on the desktop (though I doubt it will be 20% by the time it's released and available), remember that this is only one small piece of the pie...
a.) Servers - The Opteron will almost certainly maintain it's lead here, especially on the Enterprise servers. The main reason for this is obviously HT, but in addition the planned Intel quad core is an MCM (like Smithfield or Pressler) and it also requires all data to go through the FSB.
b.) Value market - Though I'm sure plans could change at Intel, we have yet to see anything in their roadmap that competes with Sempron...certainly not Celeron. This is the bulk of CPUs shipped.
c.) Mobile - This is an interesting sector because we still know so little about it. I was quite surprised that Intel didn't show a working Merom chip at IDF! Mobile was Intel's fastest growing sector (though
recent reports show that it has reduced drastically). I think we will have to withold thoughts on this sector until the Turion X2 is released and we can get benches on Merom...the reason is that we really have no data on either the Merom or TX2 as far as power and performance goes (applying Conroe's few benches performed on an Intel machine just doesn't cut it, and TX2 will be using the Rev F lower power cores...).
5. AMD's new architecture is due in late 2007/early 2008 according to Phil Hester (AMD's CTO) in
this interview. Also, AMD will be 100% 65nm by mid 2007...
To Hans:
AMD is quite profitable and has almost $2 billion in cash...of course it's nowhere near what Intel has (almost $13 billion), but because of the inroads they've made over the last 2 years they're no longer vulnerable to a cash squeeze like they used to be.
I don't think Intel can afford a "Price War" for a number of reasons...
1. They must be very careful of their pricing and marketing because of the lawsuit. Wells Fargo estimates that the final judgement on that will be worth in the neighborhood of $4 billion, and that AMD has ~75% chance of winning (this was pre-Skype). The whole "Skype Fiasco" hasn't helped their case one bit, and I imagine that Intel legal is presently crawling into every orifice that Marketing has right now...

2. Once you discount the price of your chips, it's near impossible to bring them back up again. The reason that the Pentium D was priced so low is that Intel knew it was a short-term chip because of Conroe.
3. While discounting chip prices can get you some short-term marketshare boosts in that sector, you have to be careful that you're competitive in ALL sectors first. At the moment (as you point out) there are sectors where Intel just isn't competitive (Enterprise and value), and still won't be even with NGMA. It's similar to what Intel did to AMD many years ago with the Celeron... The P2 just wasn't competitive to the K6 at the time, so Intel introduced the Celeron at very cheap prices. Since they already owned the high end (Xeon), Intel used the profits from Xeon to subsidize the loss on the P2 revenue and let Celeron overwhelm the market. If AMD has their back to the wall, they can flood the market with sub-$100 dual-channel AM2 Semprons to gain marketshare, and maintain profits with their Opterons (currently their highest revenue anyway)...Intel knows this because they invented it!