Register for free to join our community of investors and share your ideas. You will also get access to streaming quotes, interactive charts, trades, portfolio, live options flow and more tools.
Register for free to join our community of investors and share your ideas. You will also get access to streaming quotes, interactive charts, trades, portfolio, live options flow and more tools.
And just who have you heard is going to incorporate PCI Express into the processor that is going to put K8 at such a disadvantage?
I don't think that was his argument. Intel will have chipsets that have PCI Express connections and direct memory connections, so going from a PCI Express video card to memory can be very direct. AMD is at a disadvantage because a tunnel will have to map from PCI Express to HT to get to the northbridge embedded in the processor. There is an issue, but I'm not sure it is worth a red flag just yet. How much extra latency are we talking about, and is it really a factor for the next generation graphics cards, or even beyond that.
Yes they are complimentary and do not conflict. But that's not the issue here. The problem is that Opteron has an integrated the memory controller. This is great for low latency memory/processor operations, but not for I/O systems. Opteron is pin restricted. It doesn't have a PCI Express bus. But it has a HT bus, which then must be converted into a PCI Express bus. This takes extra latency and is not efficient.
Opteron currently has 3 HT links. 2 for interprocessor communication, and 1 for communicating to the chipset. Suppose they drop the 1 HT link that is currently used to connect to chipset and replace it with a PCI express interconnect. Perhaps this is a possible plan for them.
Even without this, is translation going to be that big of an issue? Are we going to go back to the place where video cards start relying heavily on system memory again? The bottleneck today is not the pipe between the video card and the CPU. It once was, but with so much function moved to the graphics card, it is less important. Latency is not really an issue for transfering textures from main memory to video memory - bandwidth is. So even with a translation hub, I don't think AMD is going to be hurt by having an on die memory controller and a PCI Express translation hub with the next generation of video cards. I'm not sure why you think they wouldn't work at all.
Any arguement that is made for Itanium demise due to CT also applies to POWER and SPARC. Would care to explain that to IBM and SUN?
I think IA-64 was in trouble even without Intel having CT. There just isn't much room in this market anymore for multiple proprietary architectures. Uniformity is winning out. Sun obviously recognizes this and is moving to Opteron. I think Sparc's days are numbered as well. It may be IA-64 can survive in a limited space, but I doubt we'll ever see it become mainstream like Intel had planned for. It could have gone another way had Intel been more aggressive pushing IA-64 into the mainstream sooner. Now it appears it will never get there. Their own CT will prevent that. IA-64 isn't necessarily doomed, it just isn't really going to be much of a factor living in a small niche.
By the end of this year AMD should be able to sell 400 K per Q. Then it will reach the milestone of having a 1% market penetration
Well they'll have 100% of the x86-64 market for the foreseeable future. That is a threat. They have something Intel doesn't, regardless of units.
And don't count on your 400K number. AMD64 in all forms (Opteron,A64 desktop,A64 notebook) will be 4-5 million/quarter by the end of the year. It will likely surpass 400K this quarter.
I sure wish AMD would be more forthcoming about the actual number of units shipped like they used to do. Now we have to guess or rely on bogus analyst data.
OT
I had to sell QCOM
Ouch! I made some very nice money on that one.
For my last house I decided to just borrow 90% using 2 mortgages because I did not want to sell any of my stock holdings at the time. I later paid off the 2nd mortage and some of the 1st after some big stock gains. It worked out well, but probably a bit risky if you can't handle the larger payments for a while.
I have had too many posts deleted for unknown reasons and I am unwilling to continue under these conditions
Sorry to hear this Elmer.
It's still the same board that reprices options for themselves at shareholder expense.
When did they last do this? The reprice that just happened only applied to employees below the director level according to the SEC filing.
Yes there is risk but how much?
Not much IMO. I think you're getting 'free' money. I would consider this play myself if I wasn't already beyond my comfort level.
Yeah, that's exactly why I'd never be caught short puts these days.
Yeap. Selling QQQ puts today is basically like selling insurance against a bad event. Of course being long shares wouldn't be pretty either.
I guess I made it through the mess once with very few scars, so I'm not sure the fear is warranted, but I have to admit I am more cautious these days. I imagine it still weighs a bit on the market. I wonder where the market would be if 9/11 never happened...
Sorry getting OT.
I also wrote some QQQ Sept $32 Puts
Longer term than I like, but that looks like 'free' money, unless there are terrorist attacks.
I wouldn't mind getting assigned more AMD this month (still short Feb $15 puts like you were before rolling). I'll do like I did with my large December assignment...wait for a spike up and dump. With all this news I think a run is imminent. I don't mind using margin for a short time period.
I'm also thinking of revising what to do with my long term cap gain shares. I think I will hold them another year instead of selling in April.
I would still like to hear 90nm is going well, or better yet see a 90nm product for sale. I'm suspicious at this point.
Bought some more AMD today.
Glad to see you more on the long side Elmer. Are you thinking AMD prospects look better now or just recognizing others think AMD prospects are better and will drive the stock up?
48-bit virtual addressing.
Hey thanks for that tidbit. Let's see that is about 280 terabytes. Not sure that it matters for the lifetime of the k8 family. As you point out this can be increased at anytime in future revisions or projects.
Don't play dumb, you already know what I mean,
I'm not sure I understand what you mean. You seem to think AMD64 is not a full blown 64-bit architecture, but you haven't stated why. It has been shown already AMD64 has a long mode (which 64 bit Linux and Windows use), which is a full blown 64-bit mode with 64-bit virtual addressing and registers. It even has a compatibility mode the O/S can switch into for running existing 32-bit apps natively (albeit with a small translation layer for o/s and system calls made by the 32-bit application).
If you don't mind the x86 instruction set, there is no more elegant solution than this.
edit: oops - sorry Semi - I see you changed your mind in a later post. I should read all posts before replying...
He knows the AMD64 design does not use 64 bit addressing.
It does use 64 bit addressing. Only 40 bits are extended to the physical interface for now because no more is needed until larger DRAM's are developed. The virtual addresses are still 64-bit, and that is all the matters. Arguing over who has a bigger physical interface is really irrelavant.
Can you currently outfit an Itanium machine with more physical memory than 40-bits can address? No! Why waste the pins in the meantime.
Without a recompile into a 64bit app, I would be surprised if there was much peformance *gain* for very many applications at all.
You may be right, but that was not AMD's original claim. I'll see if I can dig up a paper I read a while back stating this. Their is certainly overhead of WOW64 as you point out. Maybe AMD believed that the OS and device drivers would get a significant speedup in native AMD64 that would more than make up for the WOW64 performance hit, which may end up being true, but isn't yet. This AMD document has some details on WOW64 (google search 'AMD64 device drivers' turns this up):
http://www.amd.com/us-en/assets/content_type/DownloadableAssets/Porting_Win_DD_to_AMD64.pdf
Quote:
Porting to AMD64
Maintaining compatibility with 32-bit applications
A system call made to 64-bit Windows by a 32-bit application will require translation of its arguments such as address pointers. This translation layer is integral to the 64-bit operating system. 64-bit Windows has a layer called Windows on Windows 64 (WOW64). WOW64 is the emulation layer that enables a 32-bit application to application to operate in Compatibility
Mode when under 64-bit Windows.
WOW64 exists as a Dynamically Linked Library (DLL) that is integral to the operating system. The OS creates a separate 32-bit process to run each 32-bit application, and the thunking DLL resides within each of these 32-bit processes context. The 32-bit application is dynamically linked to the thunking layer. Each time the application executes a system call to the operating system, the thunking layer is invoked and performs the following sequence of operations:
• Translates parameters if necessary
• Transfers control to (Calls) the 64-bit kernel
• Translates the results if necessary
• Returns the results to the application
They better make sure there is no performance hit when using legacy 32-bit apps before official launch, or AMD may get some bad press.
AMD claimed in the past a performance increase for existing 32-bit apps under 64-bit windows. So I would have to agree there are still problems in the graphic drivers which is affecting gaming benchmarks. It is unfortunate this portion is out of AMD's and MSFT's hands.
edit: I imagine AMD has people working with both Nvidia and ATI on this, so it isn't entirely out of their hands.
Did everyone see this interview with Dell? Sounds like Intel made them the promise after Dell pressured them.
http://www.pcmag.com/article2/0,4149,1501767,00.asp
MM: You mentioned 64-bit. Obviously there's Itanium out in the marketplace, which is a fairly small part of the business at the moment. AMD's got an offering there. How fast do you think that's going to scale? Do you have any interest in doing something around the AMD processor?
MD: Well I think this idea of taking the 32-bit instruction set and extending it is good. I don't think AMD is the only company that's thought of that. In fact, I think Intel's kind of on record as talking about that. I don't think they've made a full, official announcement, but they have sort of indicated they have [it], so I think when you sort of step back and you say, what would you do with one if you had it today—I fully expect that there will be a variety of choices there, and we're going to fully participate.
It's been 2 days, give it a chance. It's shipping and in volume.
I believe you. I just don't understand why they didn't wait a couple of weeks when OEMs could announce products along with their launch. If I wanted to buy a Prescott for some reason today, I would not be able to.
I guess they wanted to just slide it in under their other product launches, which do appear to be available.
Please, get a sense of humor. I haven't looked either.
I just find this highly unusual. It isn't a matter of humor, and I'm not trying to get on your case. Why did Intel even bother to launch Prescott? They launched a 3.4Ghz 130nm P4 in 2 versions (regular and EE) that both outperform Prescott. They didn't seem to have OEMs lining up for it, and they didn't even have a launch event. Is it just a matter of needing to ship 90nm parts because they have some manufacturing lines that need to do something? I don't get it. Why not hold off the product until it is ready? I could understand if Prescott offered something drastically new (like 64-bit extensions), but otherwise it doesn't make much sense.
edit: It makes even less sense that they launched it and it doesn't seem to be available. I would understand launching if they had built up an inventory they needed to sell.
They're selling like hotcakes!
Selling like hotcakes or selling hotcakes? :)
Have any OEMs announced Prescott products or have products available? I haven't seen any, but I admit I haven't looked too hard.
mysef posted it on the Intel board and I thought I'd return the favor :)
Oh. :) I don't read the Intel board, so I missed the point. I agree with your statement here. Means very little.
Given that Rob Herb is leaving (left?) AMD, his insider sales are not too surprising. The other sales are pretty small.
who claim "this time it's different". It isn't. That's my opinion.
My opinion is, "this time it MAY be different". I'll let you know in 2 quarters whether or not it really is different. :) You may of course be right, and history plays in your favor.
I don't see any reason to sell my AMD stock now, do you? Are you going to sell your shares soon?
I expect AMD to be very late to 90nm and their roadmaps are disengenuinious.
Well I expect AMD to be shipping 90nm parts starting in the 2nd quarter for a product launch in early Q3. That is still later than the original schedule. I'm not going to sell now on any fear that they won't meet this expectation. I'll sell when they actually fail to meet it. I don't think it will be too late to sell then. Their 130nm products are performing well against the competition's 90nm parts.
So I have 2 quarters to ride it out...this should also be enough time to see how Prescott shapes up.
I'm not declaring a winner or loser. I just think AMD is a good investment for at least 2 quarters, and possibly much longer.
A64 isn't selling well because it either isn't in demand or isn't very manufacturable.
I'm not sure you or I have enough information to draw that conclusion. We don't know how many A64s AMD is trying to produce. Perhaps AMD is not being very aggressive in the 130nm A64 ramp because it is going to be replaced shortly with 90nm parts with a new socket. Their customers may even be telling them they don't want A64 until socket 939 because they don't want to have multiple products. I guess that falls in line with your not in demand theory.
I, too, think AMD will slip 90nm. They already have. If AMD can't pull off 90nm, I'll trade out of AMD. I think I have a few quarters of potential run up before needing to do that, so I'm not concerned, yet.
which means that Intel has a lot of work on its plate.
The question I have is the power a process problem (too leaky transistors) or a design problem, or a little of both?
If the design is just a power hog, then yeah, they have a lot of work. I wonder how much effort has already been put into reducing the design's power needs.
This should be a cause for concern for AMD investors.
I'm not too concerned, yet. Let's see how the next few quarters shape up. The first couple of quarters of a product ramp are always slow, and I'm not sure it is important whether it was 100K, 200K, or 300K.
My expectation was that the pipeline would be extended by only 2 stages, and with other enhancement, the resulting IPC would be 5 to 10% higher for Prescott than NW.
That was my expectation as well. Now with a much longer pipe they have the potential for much better frequency scaling (after fixing whatever the problem is). If I knew ahead of time that Prescott was going to be 31 integer stages, I would have guessed a more significant loss in IPC (5-10%), and much higher frequencies (4 Ghz). So Prescott as is would have beaten my IPC guess, and missed dismally on my frequency guess.
In other words, if you are going to buy a top end Intel part, you should invest in a good cooling solution, or you're going to get the performance of a much lower grade part as it throttles performance.
It is likely that most cooling solutions have some headroom, so Prescott might indeed exceed TDP very often, and would have no problem doing so as long as the temperature was below the throttle limit.
So TDP does not equal max power except in a system with a poor cooling solution. The motherboard has to be designed to be able to supply the maximum current and maximum voltage.
Those tables that were posted are pretty telling. Maybe AMD should start spec-ing with TDP. It is obvious this would be much lower than their current max power spec.
Prescott overclocks to 4 Ghz right now.
If that's the case, why didn't they launch higher frequencies? Something is keeping them from doing just that. Power seems the most obvious thing here. If that is the only problem with Prescott, that is fixable, although at a possible cost to ultimate frequency.
So I remain cautious going forward as we don't know the magnitude of Prescott problems. It would seem they are somewhat complex as they have already had lots of time to work on fixes, and this is as far as they got. I wonder what earlier revisions looked like.
Where are the Prescott OEM press releases?
Did Intel even host a launch event?
I've never seen such behavior from Intel. This is a major core release, not just some minor revision. Are they saving it for when they have the new platform?
So the fact that Prescott sucks really bad worries you more than if it had been good at launch?!
I'm worried that with a much longer pipe they have the nearly same IPC as Northwood. I didn't expect this long of a pipe. It implies they might have a lot of unseen frequency headroom that will be seen later, when the problems are worked out. If the problems boil down to being fundamental, they may never get it worked out, but there is a possibility that they are all fixable within a few revisions. It could happen quickly, so I'm being cautious. I'm not exiting my AMD position now. I'll watch for rumors of a big fix to Prescott, and probably even wait for some results. I may be waiting a long time, and if so, AMD should be doing extremely well by then.
I'm not going to parade around and say Prescott sucks. That doesn't help me make money. I'm glad AMD is in the lead for now in both performance and power (desktop/server). I'll continue to hold AMD as long as that is true.
This was my point-- how significant is the actual increase? Less than the ratio of 20 to 31 would imply? How much less? That's all.
Hard to say, and it varies by code.
What is more important is how much logic was reduced per stage going from 20 to 31. This would tell us the frequency headroom. The number of pipeline stages increased by 55%. Each stage would have a fixed overhead for clocking, so they give up some of the 55% there. A 31 stage pipeline is more complex than a 20 stage pipeline, so they give up some there by needing more logic. The die size grew (if measured on the same process), so they have longer routes going between various logic, possibly meaning longer critical paths. Lastly they may not have been able to balance the previous 20 stages of logic into 31 stages perfectly, which could give up some or all of the 55% depending on how bad the imbalance is.
We can see the result - Prescott is slower in frequency. Thinking about it more, low Prescott frequency is doubly perplexing. I would not have even expected this poor of result if Prescott came out on 130nm.
I wonder how high Northwood would clock on the 90nm process? Probably over 4Ghz easily. Makes me wonder why they didn't just extend Northwood's life by doing a shrink to 90nm as a backup plan to Prescott.
I feel like I'm looking through a fogged up window trying to understand what is going on. It could be they can fix the problems fairly quickly, and suddenly Prescott will jump to life, and it could also be that it would require major rework to fix. The first case is scary to me as an AMD investor, as it implies a wounded predator ready to pounce once it stops bleeding. There may be little warning.
It's only running at 3.2 GHz right now...
Yeap...that's why I said best case 3.4Ghz...I was being generous.
Also, I'm not sure that all pipeline stages are equal, nor that all stages are typically followed, so you can't simply compare the total NUMBER of stages, worst case, between Prescott and Northwood, and from that derive any % IPC slowdown
The integer pipeline is now 31 stages, not including the decode phases (since trace cache stores decoded instructions). Northwood was 20. The microarchitecture is similar enough between the two that they can be directly compared. Without any microarchitectural enhancements, IPC would go down significantly with this large increase in pipe stages. I wasn't trying to derive a %. I just said they made enough improvements to counter the decrease in IPC from having a longer pipeline. They did so at a cost to complexity and transistors. The complexity may have ended up costing them frequency.
Still, getting the performance they did out of a 31 stage integer pipe is applaudable, but it should have resulted in a much higher frequency part, which is clearly the screwed up part. They were probably thinking the same thing when the sketched it out. They may still get there eventually if they didn't totally botch up the pipeline balancing.
You can claim victory for AMD, but watch out for a potential comeback.
Yes I am majorly long AMD (long shares, long calls, short puts), and I have no current investment in Intel. But, only the paranoid...
My thoughts on Prescott-
Given the much longer pipeline, I'm actually impressed with Prescott's IPC. It must be a terrible disappointment that it is only running at 3.4Ghz, at best, right now. They clearly targeted Prescott for much higher frequencies (longer pipeline, additional cache latency cycles). A pipeline that long with an IPC about equal to Northwood could be scary for AMD if it ever works out as planned. There could be one or more of several possible problems:
1. The pipeline could be poorly balanced, meaning too much logic is clumped in one or more stages, while other stages have time to spare, resulting in a frequency no higher than the Northwood core. This problem could range from mild to disastrous to fix.
2. Prescott could be current limited. This is not the same as thermal power limited. They could have always invented a cooling solution to overcome power dissipation, but current delivery is not as easy. They could be getting large enough voltage drops on parts of the chip at higher frequencies to cause failures. This is fixable with a better power grid and more metal strapping for power hungry parts of the chip. With only 7 layers of metal they may have some trouble with this. It is also possible the package/motherboard itself is current limiting, and if so that should be resolved with the newer socket that is on the way.
3. They may have not done enough critical timing path fixes yet. It only takes 1 critical path to limit a chip's frequency. Inaccurate timing simulations can lead to missing paths that are critical. These will be fixed over time as they are found on silicon.
There are many other possibilities, but these would be my popular votes. Worst case for Intel is they completely botched the pipeline balancing early on, and this is not something that is easily fixable without major rework. I would be surprised if they made a screwup that large, but it is possible. It is more likely there are lots of smaller problems they can fix over time, and Prescott will eventually meet its higher frequency aspirations, and possibly leave AMD in the dust in the long run.
All of this still doesn't explain Prescott power problems. That is still a mystery to me. It is possible they had to target the transistors to be very fast (but more static leakage) just to get to the launch frequencies. If so, that spells even more work for them to do.
For now it looks good for AMD, but watch close for signs of breakthrough improvements. Remember P4 started out as a dog, but overtook Athlon as it matured.
Note the "INTEL CONFIDENTIAL" and the "ES" on the second line.
This doesn't necessarily mean it isn't the same as a production part for benchmarks...
I'll wait for a few more reviews. Shouldn't be long now.
Dell told Intel they were going to offer a 64-bit x86 product
This is along the lines of what I heard. Intel may have been able to hold them at bay though. I would guess these recent mumblings from Intel about an x86-64 implementation are related to this. Dell has pressured them, and Intel promising them something, "soon".
but if you have a program, which you want to compile and run, you just take gcc, compile it as "gcc -O2 -c bigproj.cc -o bigproj.o" and... get pretty poor results
I have to agree with you here. Quickly compiled code seems to run faster on x86 than on IPF. This puts IPF at a disadvantage. Not too many people these days are hand optimizing code. The hardware is changing too fast to bother.
I rolled my AMD Feb $15s out to April $14s
I'm going to see how this one plays out. I think today's selloff doesn't have much merit and will be corrected by expiration.
noticed earlier in the SEC filing that the Price on the options was to be $12 or higher, and obviously everyone will exercise those options because AMD currently trades above that price.
This is not correct. 2 things.
1. Today is the day for the strike price to be set on the newly granted options. Nobody is going to exercise these until they are worth something. I'm guessing they'll set the strike at today's close.
2. The original filing shows these newly granted options are not available for exercise 6 months from the new grant, which would be 6 months from today.
So I still say the idea that the stock is down based on the reprice is a little ridiculous.