In spite of the micro-architecture being a year older than Barcelona, Core on 45nm is going to give AMD a hard time. I'm sceptical that Barcelona will sweep every benchmark. Intel is really pulling out all the technology stops in the last 6 months. I think Penryn will make the performance/watt playing field level between AMD & Intel and when Nehalem comes out in 2008 AMD is toast.
http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2915&p=1
http://enthusiast.hardocp.com/article.html?art=MTI2OCwxLCxoZW50aHVzaWFzdA==
I am now predicting lay offs at AMD within the next 6 months as well the high probability that the New York fab commission decision will be pushed to 2008 as AMD struggles to manage their cash & debt and continues to see their quarterly profits decline.
Subscribe to:
Post Comments (Atom)
24 comments:
:(
I tend to agree with your assessments on AMD pushing out the New York fab decision - to date, AMD has survived on gov. subsidized debt from German taxpayers and I don't think they will find anyone willing to pony up the bucks to cover the NY fab after AMD loses money every quarter of 2007 (losses to increase QoQ for the entire year as well).
As for K8L, it is pretty obvious that it will be a niche market chip designed for FP intensive operations. AMD can not release a Quad to the general market due to cost issues and it is very, very likely that if Core2 (even prior to Penryn) can continue to ramp clocks that Intel will retain a 20% performance advantage in INT.
AMD's 65nm stumbles could very well have doomed the company already. Q1's loss will tell us a lot about the state of the industry but I think Intel is finally ready for the knock-out punch before AMD is able to increase capacity which could potentially kill the entire industry.
AMD will have a big quandary with K8L. Intel will once again step on the gas with quad core to flip the market from dual to quad on the client side. At which point AMD will have to get competitive. Due to the manufacturing cost advantages of using two dual cores in a package vs a monolithic design, Intel will be able to beat the hell out of AMD on pricing. AMD will have no choice but to follow thereby sucking capacity, having to ramp their new technology onto 65nm volume which is still struggling and offer deep discounts.
This is looking like a year of significant losses for AMD. The only card they can play is if Barcelona comes in dual core to compete with Core 2 in the mainstream or basic segments. But keep in mind that it appears AMD's 65nm still has kinks as they are either to only provide the lower clock speeds at the same thermals as their 90nm parts which means if they increase clock speeds, thermals will shoot up.
Intel the technology company is back. AMD will have to get used to being a sub 20$ stock again this year.
"AMD will have to get used to being a sub 20$ stock again this year."
No, sub $10.
Heh, you sounded like Sharikou in this post :)
Looking at AMD's desktop roadmap, it looks like they're transitioning their entire line over to Rev H by the end of 2007 (See here) This is an amazingly fast ramp, 6 months to move over their entire line.
Intel's transition (here) shows that they too will have transitioned over by the end of 2007 - an 18 month transition.
Both companies at 65nm, similar IPC with maybe a slight advantage for AMD. Intel will have only a handful of 45nm chips, mostly for bragging rights. Both will have quad-core at 1 and 2P, but at 4P AMD's scaling advantage will be pretty clear with HT3 vs the quad-FSB/snoop filter monstrosity from Intel using hot, expensive FB-DIMMS. Pretty competitive landscape if you ask me. AMD's margins were hurt by server ASP reduction, but Intel's server platform is still hampered by FSB limitations are 4P, and their IPC advantage at 1 and 2P will evaporate in 6 months. By the middle of 2008 only will Intel have volume in 45nm, and both will be introducing a new microarchitecture.
Given at least equivalent performance, there's no real reason why Intel will spank AMD. AMD might lose marketshare from here until 3Q 2007, but after that I think they'll continue their slow erosion of Intel's portion.
They no longer need to completely dominate Intel in performance because the stumbling-block in OEM acceptance has been crossed. Once those doors are cracked open, they merely need to keep pace in order to continue growing.
I do think there might be high-level departures at ATI, especially given the R600 fiasco since I wrote that post.
It's interesting, but Intel has basically decided to cut their nose to spite their face. By entering a price war with AMD, matching them $ for $ at every step, they haven't been able to keep Core 2 at premium prices like it would have been in the past. Are they willing to drive prices so low that both they and AMD suffer losses? I dunno, but Gelsinger frankly seems to have quite a bit of emotion invested in this race. Cool, calm thought is what's needed in a battle of attrition. Hector Ruiz seems pretty calm - maybe too calm, the guy seems almost asleep at times - but if AMD wants to compete with Intel LONG-TERM they simply don't have a choice except to grow and be creative. This they're doing, with steady progress and some out-of-the-box thinking e.g with Fusion. Notice who sets the pace of idea-leadership, and who copies lately? It's the behavior of a leader, even if sitting here in January 2007 it's the smaller company.
In late 2008, 2009 there will be a glut of capacity, and somebody's going to get hurt. We'll see then how the bleeding will go. Neither company is going to come out unscathed. But the short-term pain AMD is experiencing is nothing surprising, we've known for 6 months that there's a $5bn merger, why the financial analysts act like this is something new and unexpected when the bills come due is beyond me.
I think the layoff prediction is a bit of reach. AMD seemed willing to bite the bullet in Q4 to increase unit shipments and I think they will similarly bite the bullet and avoid cutting people short term for what probably would be a very small cost benefit. (Now if K8l isn't what it's adverstised to be, then layoffs may be on the horizon)
I totally agree with you're MCM vs "native" quad core argument. Notice Intel, even with K8l on the horizon, is not being reactionary and trying to pull-in their own native quad core designs.
I have said on Sharikou's repeatably that MCM is a far better approach from a business perspective. In addition to the obvious yield benefits, don't discount the bin split factor as you can match 2 dual cores as opposed to binning at the "lowest common denominator" (worst) core in a native quad core approach. This helps for both power and speed binning.
It also gives Intel an ability to adjust dual core/quad core mix much more dynamically as decision between whether it will be a quad or dual core chip can be done after wafer processing. With quad core, once you first litho step (STI)is done you are locked into a quad core chip. As wafers take ~13 weeks to process through the fab prior to packaging this gives Intel a lot of added flexibility in determining product mix.
Also while Penryn will be good people need to be careful on the clock speed end. The 45nm process with high K will yield significant power advantages, but transistor performance doesn't always equate to clock speed.
There may be other speed path issues on the Core 2 design as it scales above 3.4 Ghz. If this is the case then you will not see the substantial clockspeed gains until Intel re-optimizes the design for 45nm (as opposed to pseudo dumb shrink)
i just bought my first Centrino Duo laptop today! i'm completely blown over by how much better this is compared to my last one - Turdion.
i give up on AMD. they're a bunch of retards! I expect them to BK in Q2'08.
I have complete faith in AMD. Barcelona will be a clean sweep against Intel.
I advise that everyone sell all their Intel stock and invest heavily in AMD before the launch. You can thank me later.
To be honest, the reason why I want you to sell INTC stocks and buy AMD is so that I can dump my worthless AMD shares so I can but INTC at a cheaper price.
I think AMD is toast with nothing to match Nehalem and Intel’s “tick-tock” 2 year new architecture strategy. Our company lawyers are compelling me to say the truth about my stock purchases, you know being a member of the AMD’s upper management.
Anonymous said...
Also while Penryn will be good people need to be careful on the clock speed end. The 45nm process with high K will yield significant power advantages, but transistor performance doesn't always equate to clock speed.
This is correct. 45nm may or may not make a difference in performance. Regardless of the outcome, the FSB bottleneck is still in place and Intel has given no indication that they plan to improve it. I have complete confidence in Barcelona.
These silly fake posts are juvenile. I'm guessing you didn't notice that you misspelled Sharikou's name.
real
after AMD loses money every quarter of 2007 (losses to increase QoQ for the entire year as well).
This statement is a projection based on several reaching assumptions. BTW, this view comes from Goldman Sachs who described Intel as their all time favorite tech stock.
As for K8L, it is pretty obvious that it will be a niche market chip designed for FP intensive operations.
Huh? Integer will improve too. Exactly how does this make it a niche product? Do you even know anything about processors?
AMD can not release a Quad to the general market due to cost issues
Completely false. The cost for quad core on 65nm is about the same as X2 on 90nm. And, this is very similar to Intel's 65nm costs for C2D.
AMD's 65nm stumbles could very well have doomed the company already.
This is true however in comparison it is more likely that I will discover a 2 carat uncut diamond in my back yard.
but I think Intel is finally ready for the knock-out punch before AMD is able to increase capacity which could potentially kill the entire industry.
This last statement is nonsense for both points. No knockout punch and no killing of the entire industry.
I think AMD is toast with nothing to match Nehalem
I'm not sure what Intel will have to match AMD's GPU processing in 2008.
and Intel’s “tick-tock” 2 year new architecture strategy.
Will be matched by AMD's modular upgrade strategy.
Maybe only you guys are missing this, but 45nm intel parts don't come out until AMD's due to do the same (according to Intel as well). I'd post the link, but it's from digitimes, and the article is old enough to be members only now. However, the article was reposted on rubyworks, and I'm sure it's on the inquirer. So as for Barcelona getting smoked, you'll have to wait a while.
http://www.rubyworks.net/forumz/viewtopic.php?t=419
Nice argument, too bad this completely invalidates it.
I'd also like to point out, that AMD will have immersion lithography at this node, while Intel will have high-k dielectrics instead. However, AMD will eventually place these in their 45nm products as well. Also, the glut will come in q2 07, and die slightly in q3, then come back up again as fab 30 becomes fab 38.
"The cost for quad core on 65nm is about the same as X2 on 90nm. And, this is very similar to Intel's 65nm costs for C2D."
Scientia - I have heard you state this several times now (the 2 65nm processes being similar in cost), and this is completely unsubstantiated on your end with any facts and links. As you do not have the semiconductor processing background, let me point out a couple of things for you in terms of 65nm process differences:
1. SOI (or SSOI) vs bare Si - SOI adds at least 10% to finished wafer cost vs bare Si due to the cost of the substrates. SSOI cost is even more than SOI
2. 10 metal layers vs 9 - AMD/IBM 65nm process has an extra metal layer beyond what Intel uses. This also adds cost to finished wafer (means additional tooling for ILD dep, etch stop, barrier/seed, Cu electroplate, Cu polish, 2 oxide etch steps)
3. Triple Vt vs dual Vt process - AMD uses 3 Vt's for each MOS (n and p) this means 2 additional implant and lithography steps.
4. 4 strain (AMD) technologies vs 2 (Intel) - the 2 extra strain steps also require additional tooling.
These are just a few of the differences...this is putting aside yield and die size for the time being. In the case that yields were equivalent (there is no public data on who's yield is better); Intel would have overall better yield due to the MCM vs native quad approach.
Just curious - do you have any actual data or knowledge that the 2 65nm processes are similar in cost or are you just speculating?
Also the cost on a 65nm quad core IS NOT about the same as 90nm dual core. For one thing the area scaling from 90nm-65nm does not account for the die size increase. Also the 65nm process costs more than 90nm - some Sematech benchmarks I've seen put each successive tech node as adding ~15% to wafer cost.
Again do you have any background on this or are you just speculating?
Penix: "45nm may or may not make a difference in performance. Regardless of the outcome, the FSB bottleneck is still in place and Intel has given no indication that they plan to improve it."
This is NOT WHY there may only be modest increases early on on 45nm. The real reason why is that with a basic shrink (like Penryn is with some tweaks), simply speeding up the transistor switching may or may not account for 1:1 speed increase due to the fact that there are other areas of the chip beyond the transistor switching that may be the critical speed path. A simple shrink may introduce some RC delay issues which would only be addressed through re-optimizing the layout (which is not done on a simple shrink), speeding up the transistor switching wont matter if you have an interconnect delay elsewhere.
Why do people keep thinking that implementation of immersion technology early is a good thing?
Immersion litho tools cost ~50% more than dry litho tools, are immature, and introduce new potential yield issues.
The fact that Intel can extend conventional dry litho is an ADVANTAGE. I'll repeat it is an advantage, because these are proven tools that have been in production for 2 generations, are cheaper, and allow for more equipment re-use from one generation to another.
For those unfamiliar with process technology the pitch and CD's with the AMD and Intel 45nm process will be similar (assuming both are driving to the ITRS roadmap) so there is no process advantage using a more expensive, less mature tool to print the same size features.
Another thing I think I'll mention is that estimates are putting AMD's 45nm process at q2. If AMD can hold the crown over core2 for an entire 2-3 quarters the market is likely to stabilize in their favor.
I'd also like to point out that penryn is only coming out at 3 ghz according to Intel. Overclockers.com believes that this means that Intel is having problems with its 45nm process due to the very large delay in release, and the much slower clock speed.
I'll say this both because I said it about barcelona, and because I believe it is a strong possibility for both architectures. Penryn may be at lower clock speeds simply because Intel believes it doesn't need to saturate its process gains quite yet. If Intel believes penryn will outperform barcelona enough, they'll simply release it at lower clockspeeds and tdp, and simply up those when they get any pressure. This way they can make higher ASP on what are essentially lower bin chips.
In the mean time, barcelona is going to be a pretty amazing product, seeing as you'll essentially be waiting a year for anything with a monolithic core from Intel! AMD made the branch predictor much better, most likely improving latencies. Increasing cache speed a lot, having 4 megs of cache, having a shared l3 cache, having a lot of other IPC improving features, and having ht3.0 means that it is very unlikely that AMD's 40% increase in performance over competing Intel products is specifically religated to fp. So the niche market comment is completely misplaced.
In fact, if you look at their floating point comments in reference to their own processors, they should have about a 400% increase in performance over Intel's processors in terms of floating point (give or take a few percent).
Here are the links, if anyone thinks I'm BS'ing this.
http://www.rubyworks.net/quarkimages/inteldtrd.gif
http://www.rubyworks.net/quarkimages/inteldie.jpg
http://www.youtube.com/watch?v=G_n3wvsfq4Y
http://www.overclockers.com/tips01102/
Greg - there is no delay in Penryn. The Digitimes article only refers to a couple of 45nm sever products. Not the entire line up which is meant to be staggered across Q407/Q108.
Secondly - read the Intel press release. It's still 2007.
Third - if Barcelona has a 400% lead in any benchmark over the equivalent Conroe part, I will apologise on the front page of this blog to you. But if not, tell me what you are going to do? Because if you believe a 400% lead on anything, I'm hoping you're willing to back it up mate.
AMD already held the lead over Intel in floating point benches by a vast margin. AMD is even saying that barcelona's fp should be about 1.6-1.8 times faster than dual core AMD products and that's on a core/core basis. On the other side of things, we have 4 core amd systems beating 8 core Intel systems in fp benches.
What does all of this mean? Not much to most people posting on here. FP is mainly used in supercomputing, and we don't get much use out of it as average users. However, apparently more apps are beginning to use it? I'm not quite convinced that they are, but maybe.
180, the charts I linked said nothing about which processors would be released when, just which cores, which means server proc or not, penryn wont exist till q1 08. I had other links in there if you'd bother to look at them all. Also, here are a few more.
http://www.theinquirer.net/default.aspx?article=37338
http://www.dailytech.com/article.aspx?newsid=5921
Since the Intel press release is relatively old, I'd have to assume its info is not as up to date as new organizations. Also, the graphs I linked you to (again, only referring to cores and not specific packaging) were Intel releases if I remember correctly.
Just curious - do you have any actual data or knowledge that the 2 65nm processes are similar in cost or are you just speculating?
he is obvious has no related nor engineering background. he is capable of doing analysis of his own way, and when you disagree, you are wrong.
Anyway, another thing that Intel might be enjoying more than AMD is the economy of scale. it is always cheaper if you produce larger quantity (and you must be able to sell them of course)
Greg - have a look at this presentation made by Intel on 28th Jan, 2007 at the 45nm announcement;
http://download.intel.com/pressroom/kits/45nm/Press%2045nm%20107_FINAL.pdf
The launch of 45nm in 2H07 is mentioned 8 times. It also details 2 fabs ramping in 2H07 and one more in 1H08.
I looked at the links. Then I looked at the fact that Intel continues to commit 2H07 and the sillicon is in good shape and drew my own conclusion. Intel will pull in Penryn to Q3 and they will not allow AMD to enjoy a 40% across the board lead let alone 400% anywhere. We will see my friend who is right.
180, Intel obviously wont care that much about fp, as they have lagged extremely far behind AMD for a very long time, and have done nothing to improve it. I'm not saying, that in a large number of applications, AMD's chip will be 400% faster than Intel's, I'm just pointing out the fact that assuming AMD's 40% performance increase will be specific to fp is extremely illogical, seeing as they already hold a near, if not greater than, 200% performance advantage to Intel in fp.
I'd also like to point out that Intel said product begins in 2h07, not that the product will ship or become commercially available. This is a huge difference, as k8l has already been in production for some time and has actually already been "sold" to a supercomputing project that places it side by side with the cell (a very good use for torrenza). What we may be seeing is a discrepancy with Intel's fab architecture. That being, when Intel is able to actually to produce a chip from the R&D stage, it does not mean they can produce it in large scale, as they have to wait until their 45nm fabs are actually finished being built and tooled. This may be the source of the delay, if it really is one.
Greg - you will see product in 2007.
Post a Comment