Is AMD in trouble?

[Furen]

Originally posted by: Spinne
I know that all three decoders are currently 'complex'. C2D has one 'complex' and three 'simple' decoders, hence it seems that AMD has an edge when it comes to deoding 'complex' instructions, if one assumes that both AMD and Intel's decoders perform equally well. Intel will have an edge if there are a lot of 'simple' instructions. Hence my statement that AMD could think about adding more decoders, 'simple' or 'complex'. I'm not a CPU designer or even an EE, so I can't judge if this is even possible without a major redesign, but if the decoders are modular, i.e. they can be thought of as black boxes (including whatever scheduler decides which decoder does what', then it doesn't sound that hard a thing to do.

Not quite, AMD's complex decoders are not the same as Intel's complex decoders because the micro-op code of the two architectures is completely different. Most x86 instructions end up being decoded as a single micro-op in Intel's microarchitectures, so most workloads are just "simple" instructions (that the complex decoder can handle, too). With AMD, on the other hand, you need complex decoders because quite a number of x86 instructions end up turning into 2 micro-ops (which are usually fused until execution). Overall, I'd say that you can consider the individual decoders in each architecture equivalent to one another, with the complex Intel decoder being the equivalent of a regular decoder plus the microcode decoder in AMD CPUs.

 

[JumpingJack]

Originally posted by: Henny
AMD's going to get a real spanking in the marketplace with Woodcrest, Conroe and Merom.

That means they'll need to cut price. However AMD's achilles heel is their manufacturing. Intel will have over half their products on 65 nm before AMD will have their first 65 nm product out. (learning curve is everything in this business). Intel also has much more manufacturing prowess then AMD. One of Intel's major 300mm fabs produces more output then all of AMD combined.

For the last several years AMD has had superior designs that could make up for their manufacturing weakness. With inferior designs and and lagging manufacturing it could be a real challenge for them.

However AMD has survival embedded into their culture and they're agile and innovative. When Intel is on top they get arrogant, complacent, and underestimate AMD. By the time they get the wakeup call they're in panic mode. I doubt that part of Intel culture will ever change.

Now the other factor is the PC OEM's. They will never let Intel get too big or AMD get too small. There will always be forces at work to keep the market competitive.

Ruiz is a much better CEO then Sanders ever was. I also think that Otellini is a much better CEO then Barrett ever was.

Bottom line is that Intel and AMD will play leap frog back and forth for the forseeable future IMHO.

However right now my portfolio is going long on Intel and short on AMD.

Well composed post.

 

[JumpingJack]

Originally posted by: Regs
AMD will have quadrupled their manufacturing capacity by mid-07 if not earlier.

Unlikely -- this is just an estimate, but logical -- a 300 mm fab gives you 2.25 200 mm fab equivalents, at 90 nm the product mix was single core about a year ago so lets say they now have 3.25 200 mm single core 90 nm equivalent fabs (as they started Fab 36 on single core Athlon's and semprons). Now, dual core is ramping into the market place faster than most expected, 1/2 point is about now. This cuts out roughl 1 of your 200 mm equivalencies. So now you are down to 2.25 200 mm equivant fabs. By the end of 2006, 75 to 80% is now dual core, so you are down about 1.75 of your 200 mm equivalent single core fabs. that brings you to 1.50 200 mm equivalents.

It is no wonder they dropped their 2x1 meg product line so quickly, dual core push into the market is eating their lunch.

65 nm will restore 2x equivalent until quad core comes on line, thus we are now back to about 3 200 mm fab eqivalents, but then this gets cut in half with quad core -- so by mid 2007 at roughly 1.50 of what they have now (not including foundry work). Then they will need to stop 200 mm fabbing to convert fab 30, so they are in deep doo doo if they don't get 65 nm going and the die size down.

 

[dmens]

Originally posted by: Spinne
Well, the C2D architecture attempts to gain performance through wider datapaths as opposed to faster datapaths. It's a very superficial resemblence, but imagine how bad it'd be for AMD if Intel had an entirely new method of increasing performance.Right now a designer at AMD can say - "let's make our CPUs even wider than they are right now", he does not have to say - "how is Intel performing so well" and then sit down with his sliderule and figure it out from scratch. Am I wrong?

Is it bad for AMD if merom is fast because of new methods? AMD can come up with their own tricks which suit their uarch better. Ideas which suit merom may not necessarily work well on a K8.

And the "width" resemblence is meaningless. Widening the forward path is just one method to gain performance (a very costly one as well). Nobody is abandoning the frequency approach either... that is still very important.

 

[zsdersw]

In addition, the changes for K8L are far less radical than they are for C2D...which means that testing can be done quicker at the start of the ramp.

If C2D presented Intel with a process shrink as well as an architecture change, I might agree.. but it doesn't.

 

[Henny]

Originally posted by: Viditor

Originally posted by: coldpower27

Originally posted by: classy
I get a real kick out stuff like this. For one the market share AMD has won will not be lost. Here's why. One the late arrival of Core2. For Intel to even see any impact at all it would need at least 2 quarters preferably 3. The problem with that is K8L will be here and there is no reason for me to believe that it won't outperform or equal Core2. Two even right now all is speculation with Conroe and the true place of importance, the server market, it looks like AMD will still win 4 server and up setups and I believe will compete very well with the AM2 deriative Optys, not to mention higher clockspeeds with 65mm will arrive in Q4 as well.

Intel waited too long and folks are still waiting. Core is damn good and would have put Intel clearly back in the driver seat if it had been around late 3rd quarter or so last year and at the very latest earlier this year. They just waited too long. I am going to buy a Conroe, but it ain't gonna change thing for Intel's fortunes unless K8L is a flop. I see no chance at all of that happening. Holding your load is only profitable in certain situations .

K8L is not slated till H2 2007, and these are optimistic projections.

K8L QUAD CORE is not slated till H2...Dual core is still H1.

And if your going to claim Core 2 is somehow now going to make NOT make a significant impact for 2 quarters, then how do you expect K8L to make an impact when it first launches? It will take AMD time as well to switch gears to K8L, as it would for Intel to switch to Core 2 Duo.

AMD is certainly able to "switch gears" much faster than Intel, because of the way the 2 companies do manufacturing. Intel uses "Copy Everything" and has to convert far more Fabs, while AMD uses APM and dynamically uses ~10% of their production for new processes.
In addition, the changes for K8L are far less radical than they are for C2D...which means that testing can be done quicker at the start of the ramp.

The 65nm Brisbane desktop processor is not looking to bring increased clockspeeds at the present time, and looks to be like a Newcastle to Winchester transistion. To cut cost but no clockspeed ramping from the start of the process. It's not likely AMD will ramp clockspeed right when the clockspeed is introduced.

Not sure why you say this...do you have inside info about Brisbane that hasn't been released?

The server market has importance, but you don't ignore the desktop and mobile market either. The server market especially the 4P and up system also represent the lowest volume. So AMD is looking to lose market share because they aren't going to be as competitive as Intel in the large marketshare areas.

I agree that AMD will lose marketshare (probably no matter what they do), but because of the high profit of the Enterprise sector, I don't think they will lose any of the more important Revenue Share...

Ha, Core 2 Duo will stop the bleeding and continually regenerate Intel overtime. K8L is not that close.

If you say so...

I seriously doubt K8L could ramp faster then C2D. C2D was a new design using existing proven 65nm technology (ie Pressler).

AMD doesn't even have a 65nm HVM process yet. They're claiming very late this year.

Putting a new design Ie (K8L) on a new process (AMD's 65nm) is not a recipe for a quick startup.

 

[Vee]

Sure AMD is in trouble. Their position has always been precaurious.
BUT: AMD's position a year ago was better than it has ever been. AMD's position today is better than it was a year ago. And my guess is that AMD's position in 6 months is going to be better than it is today.

 

[thegimp03]

No. AMD is not in trouble and not even close to being. First, this is one processor. AMD has dominated the past couple of years with price/performance. Many of its supporters would never buy Intel again because they remember the days Intel would charge $400 for a Pentium chip while AMD would charge $100 for comparable performance. AMD being in trouble shouldn't even be an issue. Its single core A64 platform will continue to crush whatever Intel has to offer in that market. Dual core from AMD will be beat for awhile, but dual core processors are still more for an enthusiast market. I don't think the majority of computer users can rationalize buying a dual core processor right now.