Two things I wanna say to the AMD fanboys:
1. Where in heavens name did you get the impresion Penryn is going to be late to Q108??? It's coming on schedule in 2007. Read the Intel press release. Penryn was scheduled for end 2007. Intel is obviously ahead of schedule as they have booted 5 different products with multiple OSes. Which is why I'm sure they will pull it into end Q307 with such good silicon.
http://www.intel.com/pressroom/archive/releases/20070128comp.htm
2. Will everyone stop jerking off on the lack of an integrated memory controller please. I'm not an engineer but I have some common sense. This lack of a controller does not seem to change the fact that Conroe kicks AMD's ass so why the hell would it be such an issue moving forward. It's obvious Intel's engineers designed an architecture that works pretty darn well in spite of this. I can understand if the discussion is around other new features on the Barcelona architecture but for goodness sakes...stop talking about the IMC. It doesn't matter.
The way I understand this simplistically, the IMC on AMD's chips is a faster link to a smaller cache which means quick access but the possibility of the instruction coming up empty handed in the cache and hence the cache needing to be refreshed more often. Whereas for Intel they have a bigger cache (which means fewer cache refreshes) but a narrower tunnel through the FSB which could limit the number of instructions travelling through at any given time. It appears to me from the simple fact that Conroe wins all the benchmarks that Intel's engineers figured out how to optimise this since Athlon/Opteron are getting their ass whooped on every benchmark.
So...get over it!!!
Subscribe to:
Post Comments (Atom)
34 comments:
yes, the IMC is not the bottleneck for a lot of the workload that you and me are using. I have long debated this in my post last here: http://computing-intensive.blogspot.com/2006/07/imc-myth.html
and again, the 45nm part of Intel will surely come out this year as what you have said. But the AMD fanbois would not like to hear that. And the funniest thing is when AMD said they will produce the 45nm part in Q2 (never mention about the general availability), they quickly said that AMD and Intel will come out the 45nm part at 'quite' the same time.
Look at my post on your "penryn pwns" article if you want to get an idea of why we think penryn is delayed. It's not just one news source, and it's not just the inquirer.
You seem to be losing you lack of bias when it starts looking bad for Intel. I wonder if that means you're exactly 180 to sharikou!
1.Some may have gotten that impression from this roadmap:http://pc.watch.impress.co.jp/docs/2007/0202/kaigai333_02l.gif
However, if your sure they'll pull it in that's all any1 needs to know. Its all right, I'm sure your not the only one that believed they would introduce Penryn within a year of Conroe.
2.Speaking of memory, if yours wasn't as inefficient as it is, you would recall that an IMC was part of the reason that K8 dominated the 3 years after its introduction. But after an added 3 years of R&D, "Conroe kicks AMD's ass" without an IMC. You could probably drive a car without using your hands, but that doesn't make it a good idea either. But it appears to you that "Intel's engineers figured out how to optimise" their cache. Wow that simple, huh. You don't think an added 3 years of core improvements so as to increase IPC had much to do with it. If only AMD could think of simply optimizing their cache, but all is lost. They've shot their load with K8. If only they could intro a new chip. Wait they will and just about a year after Intel's new architechure debuted. Wow, what luck, such is the nature of the development cycle these Co.'s are on. That when one makes a product, the other actually follows with one of its own. You can just keep on comparing C2D to K8, but remember, you never heard of the word Conroe back when K8 was kicking ass.
Will everyone stop jerking off on the lack of an integrated memory controller please
well its hard to argue with these sort of people who changed their keyboard connection from PS/2 to USB2 because they thought they could type faster.
Man this is important, , ArsTechnica has a good review of why.
This is so important that the processor that will replace Core2 (maybe Core3) will have an IMC anyway.
To make a long story short, the IMC allows for (among others):
* Better Core/Processor Scaling (thik 4 sockets, 4 core per socket servers)
* Less latency for memory accesses.
* Better virtualization instructions.
The extra latency makes you need a fatter cache, not the other way around...
Of course, this is more an issue with big iron, and not in desktops, laptops, worskstations or small servers.....
Again, check in arstechnica, where John Stokes had more time to write the article...
Salud!
You will be eating crow once Barcelona hits.
The other crap being slung around is "well AMD will have high K on 45nm too, because IBM announced it!"
In an interview with the new VP of manufacturing/technology, he was quoted as saying they are looking at it in late 45nm cycle or 32nm.
"Late 45nm cycle" would mean ~late 2009/early 2010, 2 years behind Intel.
Also note - IBM has an extremely talented research team, however their manufacturing prowess is another thing. Look no further than IBM's SiLK announcements several years ago where they demo'd a 2 layer structure which represented a breakthrough in low K dielectrics (SiLK is a spin on Low K). Once they tried putting it into an 8 metal layer production chip it failed miserably and the technology was shelved.
It is one thing to announce a technology in research/development, however Intel's announcement was more substantial as they had a fully functional computer chip that used that process technology. The trick with High K/metal is the integration and ability to keep maintain a decent yield. Obviously the integration on the Intel side has been done, and based on their normalized yield foil they showed, the yield is on the same trend curve as previous technologies.
No one has come up with anything that is newer than the roadmaps that have been linked and are flying around on the internet saying q1 08. Obviously, unless it's newer and actually from Intel, the only idea of a launch date that is credible is q1 08. I know Intel said q3 07 a long time ago, and then everyone just reposted that when they got info on high k and 45nm, but that's just too bad, because that isn't up to date.
Also, I enjoy how you immediately label anyone who believes penryn is coming out q1 08 a fanboy. Sounds very mature and non-reactionary. Obviously the signs of someone who in no way feels threatened by these claims of a delay, and thus the signs of someone who is not a fanboy. (Note the obvious sarcasm).
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/
http://www.rubyworks.net/forumz/viewtopic.php?t=419
http://www.theinquirer.net/default.aspx?article=37338
http://www.dailytech.com/article.aspx?newsid=5921
By the way, is that enough links for you, or are you still of the opinion that there is absolutely no evidence to suggest that Intel is going to release 45nm, and thus penryn, late.
Though Core2 without and IMC beats an Athlon with an IMC, the whole point of and integrated controller isnt only performance. The current advantage that opteron has at 4P is due to the ability to scale the memory interconnect, and this because of the IMC.
Also and IMC provides with some virtualization advantages.
Having said that, the IMC is overhyped in what performance at nonserver level is concerned.
I just realized I posted AMD's youtube video, which has nothing to do with penryn, so ignore that. Also, all the rubyworks ones are to a single thread in a forum. The images come from elsewhere, again, I'm sorry if this is inflationary. I basically just looked for all the links I've posted on this and copied them back in.
I don't, in any way, think the IMC is overhyped. The IMC does not necessarily yield large performance improvements, but it does trim the fat from the interface between the CPU and the board, which increases IPC. A direct connect interface does much of the same, while allowing for interprocessor communications. While the second point doesn't seem like a big deal, in an age where the most efficient systems are increasingly the most integrated, being able to have multiple processor architectures or uarchitectures simplifies the process of creating capable systems not only for high end businesses, but for the general consumer.
I know that's a lot of generalized points, but I think my point is clear, that k8 is more forward thinking than core, even if core performs better. Obviously, in a business and product where advancing technologically is extremely important to the consumer, it's kinda nice to know there's actually a company that's trying to do that still.
Greg said...
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/
http://www.rubyworks.net/forumz/viewtopic.php?t=419
http://www.theinquirer.net/default.aspx?article=37338
http://www.dailytech.com/article.aspx?newsid=5921
By the way, is that enough links for you, or are you still of the opinion that there is absolutely no evidence to suggest that Intel is going to release 45nm, and thus penryn, late.
So, i guess you are the mike, the old crony of sharikou :) the source of the 'roadmap' is not from ruby work, and you shamelessly put it there and gave out the link as if it was from there.
anyway, if you read some other link correctly, it was saying intel will produce and ship the 45nm part in late 2007, but will only 'launch' the product in Q1 2008. While i'm not sure how true the launch date is, those sites still indicates a 2007 production and shipment.
too bad, i guess all the amd fanboy such as u, sharikou, scientia, etc, are not going to be able to take this fact. well, eat the crow when intel start to ship the 45nm part at the year end then.
"By the way, is that enough links for you, or are you still of the opinion that there is absolutely no evidence to suggest that Intel is going to release 45nm, and thus penryn, late."
I love how spinners come in... Q1'08 is 2 years after 65nm releases which is a normal technology cycle transition time as viewed by the ITRS, which with all due respect is a little bit more authoritative than the yahoos who think they understand process technology and manufacturing.
Q3'07 would have been 6 months EARLY by a standard 2 year cycle, Q1'08 is dead on normal for a transition (especially given the disruptive integration change highK/metal gate brings).
Likewise when people compare the 65nm (and in the future 45nm) release dates between Intel and IBM/AMD, they conveniently overlook the fact that at launch, Intel is achieving the ITRS roadmap targets, while AMD is typically 3-5 quarters beyond hitting those targets. While AMD is THEORETICALLY 1 year behind Intel on the 65nm node, most people who understand process technology will understand they are really about 2 years behind as they are basically printing 90nm transistors (in fairness they are probably doing a litlle better than 90nm transistor performance on 65nm) at smaller gate lengths and have yet to implement many of the 65nm improvements.
And for those who don't believe me just look at AMD's comments on high K - yes they MAY be targetting the 45nm node as well, but their own director said it would be late 45nm (meaning ~early 2010), which by coincidence would put them 2 years behind Intel. AMD is simply playing fast and loose with their interpertation of "technology node transition".
LOL, wow, I never new anyone would figure it out pointer! But seriously, I'm not mike. I'm a5h3n on those forums, and have also posted as ashenman on these forums. I happened to switch names every now and then because of multiple blogger accounts that would randomly switch in the auto-entered fields. I was generally too lazy to change them and re-paste my post, thus the multiple names.
Also, I never once insinuated that those were originally from rubyworks. If you want to find out where they're from, ask quark, who posted them. He didn't both linking them, though I've seen them, and heard talk about them elsewhere.
After mucking through your last paragraph, I guess I have this to say. I'm not an AMD fanboy. I strongly dislike Intel as a producer of products that I tend to purchase, yet as a business, I think they're pretty cool (though for reasons that are illegal). Being that Intel said that production "begins" in the second half or 2007, it makes perfect sense that they'd just have enough volume to launch by 08.
Also, Intel can only demonstrate parts from their R&D fab right now, meaning if their 45nm fabs haven't been completed, they can't produce 45nm tech in volume until those fabs are completed and have been producing chips for some time (as wafer take quite some time to complete).
"I'm not an AMD fanboy. I strongly dislike Intel as a producer of products that I tend to purchase, yet as a business, I think they're pretty cool (though for reasons that are illegal)."
Funny funny man!
"Also, Intel can only demonstrate parts from their R&D fab right now, meaning if their 45nm fabs haven't been completed, they can't produce 45nm tech in volume until those fabs are completed and have been producing chips for some time (as wafer take quite some time to complete)."
Intel's "R&D" fab, which by the way is a development fab only (Intel has a separate fab, RP1, for research) will have >1/2 the capacity that F36 does on 45nm alone and that is while it is still producing a significant amount of 65nm capacity (again ~1/2 of AMD's F36 theoretical capacity) This is fairly significant, especially at the start of a ramp.
Before you start talking about R&D fab - do your homework! Look at the fab size of D1d vs say AMD's F36, you also might want to look at the capacity and then make your bogus conclusions... AMD does some development in F36, should we assume that is limited production too? D1d's overall capacity will eventually be >20K WSPM (I believe it is already >15K WPSM right now but I don't have any links)
Question. How many weeks does it take for a wafer start to make it all the way through the facility? Add that to mid q3 (my best guesstimate of an early start) and tell me when that means these things will start shipping to retailers.
Being that Intel said that production "begins" in the second half or 2007, it makes perfect sense that they'd just have enough volume to launch by 08.
ok, that's funny. intel will ship in late 2007 and launch in Q1 according to those link, then Sharikou, Scientia amd fanbois and alike (not including you this round :)) quickly said intel's 45nm part is in Q1.
Then AMD said they will have the 45nm part in Q2 (without indication of the launch date) and claim 'HVM' in next 1-2 Q. Then again those AMD fanbois and alike quickly said AMD will have the 45nm part in Q2 2008.
if you ever wanna believe those links, it's up to you. But, comparing 2 schedule at 2 different point/milestone, it is an obvious ridiculous fanbois act.
"Question. How many weeks does it take for a wafer start to make it all the way through the facility"
Depending on the priority of the lot it is about 13-14 weeks to get a wafer out of the fab. It then has to be packaged (not sure how long this is). This should be about the same for both Intel and AMD though I don't have a direct link.
Most of Intel's 45nm equipment for initial production is already in place and qualified so it is more a matter of driving the yields up and getting the initial designs at good enough performance before starting significant wafer volumes.
Thanks anonymous, I wasn't aware of the size or exact purpose of development fab, but now I'll be able to argue more effectively. My only question is, why have a fab that large if it's not meant for long term production of sale-able chips. However, if I'm going to do my research, why would my conclusions still be bogus?
Pointer, we've always been saying availability will be in q1, based on those links, not that they'll finally start producing or shipping them. Also, I'd have to look at what links your talking about to know what you meant by AMD saying they'd have 45nm _______ by q2. It seems to me like this puts the main fault on the Intel fanbois for immediately thinking Intel would have penryn available to the masses by q2 or q3.
Well pointer, I guess we should just ignore every link ever given to us, because we don't want to or think we should believe it.
Also, I believe that scientia speculated that 45nm from AMD could arive at q3 late q2, but wasn't concrete. Maybe I just have to reread his last post.
"My only question is, why have a fab that large if it's not meant for long term production of sale-able chips. However, if I'm going to do my research, why would my conclusions still be bogus?"
Answer: because after 1-2 generations of development it is MEANT for long term production and "sale-able" chips. Why do you state that these fabs are not meant for long term production? this is categorically false and you have nothing to support this...it is just another stupid assumption by someone who has no background in manufacturing, all of Intel's past development fabs have been converted to production ("long term" to use your words) after 1-2 generations of development work.
D1a = F15 (production .35, maybe some 0.25?)
D1b = F20 (production 0.18um, maybe also 0.25um?)
D1c - think it still is called D1c but is doing volume 90nm and now 65nm production.
D1d - 65nm PRODUCTION (for the last 12 months!) and eventually 45nm PRODUCTION (after 45nm process is cert'd)
All of these fabs have and will do "long term" production. It makes sense as you don't have to rip out the pilot line tooling that the development was done on. As development is coming to the end additional equipment is added to the existing development line equipment and it is ramped to production volumes
D1d right now is producing approx the same capacity as AMD's F36 and I believe at full build out will actually exceed F36's capacity...
"However, if I'm going to do my research, why would my conclusions still be bogus?"
I think this statement speaks for itself - you have made a bunch of conclusions before getting any facts. You assume Intel's development fabs don't do long term prodcution, you assume they are fairly small... Generally one does some research and gets some data before throwing out some conclusions.
I just made the general assumption that the development fabs were only that. I'm sorry being "stupid" (or uninformed if you don't want to be hostile about it) makes it so you have to talk down to me like an idiot.
"I'm sorry being "stupid" (or uninformed if you don't want to be hostile about it) "
Please re-read my comment...
"it is just another stupid assumption by someone who has no background in manufacturing"
I did not call you stupid. And I don't think saying you have no background in manufacturing is hostile.
My apologies...
"As Intel took a technological lead late last year, AMD responded with "frantic price cuts" after a weak start to its first quarter, Needham analyst Y. Edwin Mok said in a note to clients on Tuesday.
Mok contends AMD has cut prices three times in three weeks to spur demand."
http://www.ibtimes.com/articles/20070214/amd-intel-prices.htm
Looks like Q1 may be bloody as well, although it looks like this time the price war has been initiated by AMD.
The rest of the article is also interesting - it vaguely references impact of Dell as well as AMD tries to win more business there - I'm not sure why AMD would do this if they are capacity constrained now (?)
Kind of funny how it seems only Intel is "blamed" for the price war.
Apologies accepted, and I apologize if I took undue offense to your response. I simply read too much into the tone of your response.
It is always annoying to see someone attributing an idea to me which is not mine. The main reason Pointer is eager to post here (and the reason he is mentioning me) is because he has pretty much destroyed his credibility on my blog. Is he still claiming access to inside industry information while posting nothing more technical than AMD jokes on his own blog?
The roadmap says Q1 08 however I suppose it is possible that Intel could hit Q3 07. The problem is that this would only be production from D1D. Intel can't really get volume out until it can produce on FAB 32 and this won't happen until late Q4 at the earliest. D1D is capable of producing a nice amount of server chips but not enough for desktop or mobile. Nevertheless, Intel should have a real volume of 45nm with FAB 32 and FAB 28 before AMD ships a single chip.
D1C used to be an R&D FAB but it was converted to general production. D1D however is not a general production FAB and is only capable of producing about 1/3rd of what Intel's regular FABs can do. Once a regular FAB moves into production, the output of D1D is quickly dwarfed. AMD devotes 10% of the FAB 36 wafers for testing. This ratio is a lot higher at D1D.
BTW, running an OS on the chip is not proof that it is ready. When a chip has serious errata you run a patch program from the BIOS at startup. This keeps the chip from crashing but robs peformance.
The IMC is not much of a factor on single and 2-way. Intel's dual FSB chipset takes care of this, although, we'll have to see if this becomes a mainstream product. The IMC pays off at 4-way in spite of Intel's proposed quad FSB northbridge. BTW, you need to understand that having an IMC for Woodcrest in 2007 would drive the last nail in Itanium's coffin. Intel is looking at early 2009 for X86 with IMC on a new socket. Itanium will get the new socket first.
Native quad core versus MCI was not mentioned although this is another area that is debated. It appears though that Intel is getting most of the benefit of native quad core with the hybrid bus on Tulsa. So, I would have to assume that if Barcelona does show any big gains with native quad that Intel will release a hybrid version of Clovertown.
"D1D however is not a general production FAB and is only capable of producing about 1/3rd of what Intel's regular FABs can do. Once a regular FAB moves into production, the output of D1D is quickly dwarfed."
While your latter statement is true, about other fabs (combined) eventually dwarfing production, D1d AT THE START OF THE RAMP will be capable of >10K WSPM (which I think is ~1/2 of F36?) - this is sufficient for a significant amount of initial production. Oh and while D1d is doing this small amount of 45nm, it will only be producing a mere >15K WPSM of 65nm production.
Why can't people understand that D1d is a large fab that does both development AND production?
The 1/3 # is completely pulled out of the air. (I didn't notice any semblance of fact or link to accompany this made up fact).
If you had any background you could just look at the cleanroom square footage to get a very rough idea of relative capacity. Do you have any idea how much capacity a development line takes up? You clearly have no background in this and are just throwing out your own misperceptions about what development is.
"Intel is looking at early 2009 for X86 with IMC on a new socket."
You have said this many times now - this is your opinion. The comments out there by both Intel and "questionable sites" like the INQ have stated this will be on Nehalem and is scheduled for 2008 release. I think the INQ now has a specific link to an Intel manager on this schedule.
"D1C used to be an R&D FAB but it was converted to general production."
By the way any ideas of what this "conversion" entails?
Basically the development line tooling is brought back to the copy exact configuration and Intel changes the name of the fab.
People talk about a "fab conversion" for the "development" fabs as if there is something involved...NOTHING SIGNIFCANT is done to the fab itself, the SW automation, the factory automation... the additional fab tooling is brought in at the end of the development cycle and is simply qualified like any other tool at any other fab.
And by the way is is a "D" (development fab) not an R&D fab, the research at Intel is done elsewhere (RP1). You'll say this is a minor point but it is significant as a lot of the tooling and processes tested during the research phase never makes it into manufacturing (or may get pushed out to a later tech node); the development line STARTS at a point that is very close to what the final manufacturing line will look like.
Thus when development is done there is not a lot of "conversion" to be done, contrary to popular (or at least Scientia's) opinion.
Rant over...
"RUIZ: The one thing that's hard to do is trying to correlate the stock price to anything a company does. It is very difficult. We're disappointed that our stock isn't performing better, but I won't correlate it to anything other than it's just one of those things"
This is the f@%$^@! CEO of AMD?!?
While stock price doesn't track 1:1 with company performance, this isn't a small cap stock with limited volumes and/or limited analyst coverage, nor is it a short term price blip. Perhaps, since there is no correlation to actual company performance, the price is dropping because AMD has been doing so well over the last 9-12 months and its forecast looks so bright?
For him to say it is just one of those things and to refuse to correlate it at all to company performance is irresponsible and shows he is in denial.
These are the same folks who met with analysts in Dec and made no mention of earnings shortfalls and gross margin drops, and then a month later warned of this on a Friday, after hours just prior to the Q1 earnings release? Did they not know in Dec that margin was dropping by 20% points? No indications of net profit shortfalls >2/3 of the way into the quarter.
Perhaps the stock price gets beat up when games like this are played by the top mgmt of the company. As it isn't because of company performance, Hector must figure out some excuse, no?
The main reason Pointer is eager to post here (and the reason he is mentioning me) is because he has pretty much destroyed his credibility on my blog.
go ahead with your zero credibility things. Share what i posted in your blog and see what the zero credibility means.
The jokes that you referred to, 2 were related to what their marketing ppl folks said, which i think it is funny. While AMD fanbois are claiming Intel to be a marketying company, I just wanna post them there to let people see what AMD did in marketing. I didn't create the jokes, AMD did. The last one is about AMD's megatasking, which is indeed a joke by itself.
Through out your blog you have shown a consistent behaviour: agree with what you said or being argued as being wrong or the post being deleted. You were trying to speak as with authority which you clearly don't have.
While Sharikou is an obvious AMD fanbois, You are an AMD fanbois claiming not to be one. Not sure if you have any Chinese friend, if you do, ask them about the JingYong's kongfu novel book. There is one character that matches you entirely: Er-BuQun, Ling-HuChong's master.
scientia said It is always annoying to see someone attributing an idea to me which is not mine
ok, i took the liberty to check on the statement. Well you didn't say that. My mistake on this and i'm sorry for this.
Though, it doesn't change the fact that you are AMD fanboy claiming not to be one and not able to intelligently discuss when people are not agreeing with what you said. You would turn to name labeling, post deleting, making fun of what the poster said or claiming the poster has no necessary background, etc. And using my post as example, you would use an unrelated statement, trying to argue that you are right on the other statement. Please do not try to speak as with authority, when you clearly don't have.
You are good at doing some analysis with your own way, you are good in spinning defending it too. Not like Sharikou, you will put more facts in between, to make up a statement. Putting a 1+1=2 statement, would not make the 1+2=4 statement becomes true though.
I don't i need to put more words into describing your blogging behavior, they just need to read through your blogs' comments (some are deleted) to see what I mean.
Sorry an additional addendum:
In addition to the insider selling AMD insiders were recently granted ~150,000 restricted stock units.
Oh and all of these filings, including the intent to sell an additional 750Mil shares, were conveniently filed on a Friday afternoon, much like the recent earnings warning for Q4'06.
Ahhh... integrity is such a fungible thing, no?
Will one count the heat and power contributed by IMC within an Opteron as part of the processor's ? If this fair unless Intel also include the heat/power contributed by its external memory controller. So will comparing the heat/power contribution of an Intel or Opteron server as a whole be a fairer gauge ? Is this an advantage to Opteron for having an IMC ?
Post a Comment