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.
Pleased to see KeithDust2000 has found a new home.
I hope it suits him better.
AMD's stock price has been the best vindication of his constant "doubts".
Well, well well
So KeithDust2000 is an intel employee moderating an AMD board. When I challenged him on it he deleted my posts. Then I thought he would ban me. He did.
Why don't you do something about him?
Just look at all posts he deleted.
Makes the idea of an investors board worthless. Let me know when you believe in making it OK again.
That's funny intel reminds me a lot of Enron and MCI.
". . .delayed due to the desire to utilize at least some of these advanced processes?"
I think the emphasis may be off here while the sentiment is about right.
Clearly a simple shrink from 90nm to 65nm doesn't reap the kind of benefits that were seen, for instance, in the 180nm->130nm node transition. This is well documented throughout the industry in spite of the analysts refusal to shift from the geometry metric as a measure of a semi mftr's prowess.
Wei even said that the future process performance improvements will NOT come from finer geometries. Of course yield/wafer will always benefit from smaller chips. Wei comments are confirmed by the article at RWT where it talks about whick knobs can be turned to tweak a process and how the traditional ones don't work anymore.
In the past it has often been the case that AMD changed everything at once - something rather risky. This time AMD has the luxury of shifting to 300mm at just one fab while keeps the design constant and just tweaking the process. This is something I am MUCH happier about.
The change to a Dual Strain Liner with SiGe looks to be very good news. It sounds like AMD engineers found the right knob to turn - and Intel's didn't. If AMD can realize the transistor-level improvements at a chip level it promises to put them even further ahead in the areas that are important, that is performance/watt, somewhere where they already have established quite a reputation.
In short its not that AMD delayed the shift to 65nm but rather 65nm is not the answer to what AMD wants. SiGe DSL *IS* what it wants. When more chips are needed, late in 2006/2007 then AMD can shift to 65nm and do it at a more leisurely and measured pace.
Great Intel/Ibm/Amd process review
Thanks mas, really a great read. There are so many issues that directly impact investment here. Not the least of which is the disappointing result of implementing the intel 955 EE on their 65nm process where leakage has increased (again!) and only a small improvement in power under load. Considering that the average processor is 99+% idle this is a really bad outcome for any business concerned with heat/power costs.
Seems that the P4 architecture is just too reliant on lots and lots of high performance transistors.
Presler - an expensive disappointment
Well with a few hours spare I did a quick read-around on the new 65nm Presler. $999.
Quite a disappointment, not only as a processor but also an indictment of intel's 65nm process. Just as I suspected, leakage current goes up while consumption under load falls only a little.
Opteron open boxes
Funny that's how I've seen them all the time. At shows and demonstrations they always have one open so you can touch the heatsinks and feel that they are running only just warm. I remember well seeing my first Opteron - it was a Newisys box - and how people lined up to feel how cool the Opteron was running and the amazement that pervaded.
Nice little article on code profiling w/ Codeanalyst
http://www.devx.com/amd/Article/30190
Shows how useful 1) having those events logged is & 2) why having more TLBs adds so much.
New AMD Processors
This quote from Anandtech today:
"We are expecting AMD to release a new series of processors in the very near future, but at the moment, we will not be able to discuss any of this information any further. As the time period approaches, expect us to make some comment about this."
http://www.anandtech.com/guides/showdoc.aspx?i=2655
SiGe maybe?
"out of posts for the day"
Thank the lord for that. Enough hyping of Intel's supposed prowess on this board, better to do it on the ihub intel site.
intel has been hyping its future for months, even years. If they can't make a decent itanic then why oh why does anyone buy this "coming soon, the most wonderful x86 chip ever seen". intel went from being somewhat engineering based to being a marketing company. I don't believe any benchmarks of engineering samples - there's too many things that can go wrong. Those errata can bite you in the butt and the fixes can really change things.
Fundamentally intel lost its engineering talent and I see a future of screw ups.
Capacity & shortages
I think we should keep in mind a couple of factors.
1. There may be induced shortages as AMD seeks to introduce new steppings. I guess that they may be launching the SiGe strained stepping about the same time as Yonah, i.e quite soon. Best to clear the channel of existing stock.
2. In the past the grey market has been a problem with big OEMs overordering then dumping their surplus on the open market. This has been solved, in part, by the Processor-in-box. Thus big OEMs are less likely to overbuy and we may see shortages in built computers lines even though there is a ready supply of parts for the DIYer.
"It" being AMD I presume.
i.e he's saying that AMD will continue to grow market share.
You're right, AMD rarely performs "as forecast". Who could have forseen that AMD would produce a blockbuster product like the Opteron/Athlon64? And they just keep on delivering like they are the market leader.
I'm glad you agree.
There certainly is some difference in process.
And the effect is major considering that intel processors are burning hot while AMD's are barely warm.
That simple observation begs a greater difference in process #'s than we have seen here.
I also draw your attention to Wei's comment about leakage and how minimizing that dimension is not a good way to go.
Da Source:
Sorry! Thought I was reading the same data that you were, an unreasonable assumption.
Here's where I got the #s
http://babelfish.altavista.com/babelfish/trurl_pagecontent?lp=ja_en&url=http%3A%2F%2Fpcweb.mycom...
65nm process
Well alan I'm not sure you and I are looking at the same data.
Here's what I see in terms of Ion:
P N
Intel: 0.91 1.62
AMD: 1.03 1.27
Looks to me that there is a much better balance in the AMD CMOS process, the old issue of which bit to tweak. Surely improving the characteristics of N is pointless if the P channel is not up to scratch?
The metalgate process of Intel has better P Ion at 1.06 and N A 1.75 but are there issues with this process? I don't know.
Area of P-channels
By eye this looks as though the P-channels are about 2x as big as the N-channels.
So are we looking at equal size according to the paper? About a 30% size reduction?
Thanks CombJelly.
"If true, it is awsome." I agree.
It validates AMD's lead. That lead looks unshakeable for several years. intel tweaked the wrong thing.
intel looks like its in catchup mode for the foreseeable future.
Great post CombJelly
Can we have some futher thoughts?
SUNW
I agree SUNW is an interesting side play on AMD.
I did report that Sun's CEO told the troops back in April that the share price wouldn't move much before Xmas.
I'm certainly looking at that if the SOXX falls out of fashion, trouble is there are so many SUNW share out there, even at $3.80 its still quite a market cap.
October Sales Plunged
Well I can see that intel's sales plunged because they can't make enough chipsets BUT WHAT THE HELL HAS THIS GOT TO DO WITH AMD????
This has to be the most twisted piece of non-logic I have seen in a long time.
As for ASPs, yes intel has been forced into the budget/value space but I have seen nothing that supports falling prices on AMD products. As to whether ASP is falling due to a higher mix of Semprons I don't know but this must be balanced by the obvious success of the more expensive X2s. On balance I'd be surprised if ASPs have fallen and if they did it would be a matter of cents rather than $20-$30 that this Diesen (pun, get it?) guy claims.
Still this will provide a buying opportunity today.
Linux goes ballistic (with Opterons)
Piece from the Inq:
http://www.theinquirer.org/?article=27991
How RedHawk Linus is being used for defence, using systems of 8 Xeons or 8 Opterons (gee I wonder which ones they'll choose, probbaly Xeons in the Artic but Opterons elsewhere).
Rep ret
Its just a two-byte return, padding the 0xC3 with a rep prefix else the conditional jump can't predict the likely outcome.
I have a 3dNow idct routine if you're interested. Its as fast as the MMX routines but absolutely accurate. I also did one that is SSE & 3dNow for Athlons. Output quality is the same.
Mike Moy: New AMD Optimization Guide
Did you see that there is a newish guide, dated 10/25? One thing I missed before was the use of a "rep ret" combination as a target for a jump.
I happened to look at pryan.org and scanned some of your patches - thought it might be of interest.
Cheers & happy turkey day.
[EDIT] What does Mozilla do with an idct routine?
MSXML 6.0 for x64
Recently intro'd from Microsoft, redistributable of XML core set supports 64-bit Windows:
http://www.microsoft.com/downloads/details.aspx?familyid=993c0bcf-3bcf-4009-be21-27e85e1857b1&di...
Intel Flash
Oh this will be interesting. Intel will try to dump some of its under-depreciated manufacturing capacity on the new company and get Micron to buy into it. Meanwhile Mr Harrison here, an Intel employee, is running the show.
I don't think that Micron will be so stupid as to take assets at book but I do forsee a bitter row breaking out.
"PathScale's product currently plugs right into Hypertransport slots on Opteron-based servers only."
How long before we get video cards that plug in to HT slots I wonder?
Actually I see that Intel balance sheet has $13.93bn on it so a $25bn buyback would make quite a hole.
A 9W 1500+ Athlon64 is a very interesting product.
That's what I need for a media PC in my lounge.
I wonder when they will officially announce the OPN.
I don't make mistakes over trivial stuff.
That graph is trivial. You just accidentally messed up your own argument and now it hurts doesn't it.
Here's laughing at you.
I don't read graphs wrong.
TLB caches.
Be careful when you look at the cache diagram of the K8 that you present. The 32-entry fully-associative cache units are for TLB - Translation Lookaside Buffers (pointers to page table entries).
You cannot take the data for main memory miss ratio, that is the source for the first graph you present, and apply that to the dynamics of caching TLBs.
Yes it does, just follow the fully-associative line, its flat from 64K onwards. I made a mistake in my post at first, I used direct instead of fully-associated, then updated it. Perhaps you read an early version?
"Absolutely false."
Actually the data you present shows him to be absolutely right.
Though the graph you present doesn't show any set-associative > 8-way it can be seen that the fully associative miss-ratio is essentially flat from 64K to infinite cache and the 8-way is flat from 128K onwards. (to my eye). 16-way and 1023-way must fall between the direct-mapped and the 8-way.
It seems reasonable that the posed 1023-way line would be very, very close to the fully associative case, if so then there is no advantage to going above 64K and the miss ratio would be the same in a 512K 16-way cache as a 64K 1023-way.
I've never heard of a 1023-way, is this a typo for a 1024 or is there a subtlety when one gets to high-associativity caches?
Intel's dual-core delay really hurts
Tom Yager's really down on Intel
http://news.yahoo.com/s/infoworld/20051031/tc_infoworld/70770;_ylt=ApWwK0THMOXimoZX18oIPjUjtBAF;_ylu...
Google goes Opteron
Well this made my day. Thanks Keith.
If anyone missed it please DO give it a read:
http://money.cnn.com/2005/10/14/technology/techinvestor/tech_biz/index.htm?section=money_latest
Lots to like, I picked this bit:
"By adapting tricks it has learned from a decade of making 64-bit servers using its own Sparc chips, Sun has pulled off some technical feats that Dell (Research), Hewlett-Packard (Research), and IBM have yet to match. The x4100, for example, manages 50 percent more performance, according to Sun, while consuming a third of the power and a quarter of the space of a comparable Dell server. It does this primarily by using the latest dual-core chips from AMD, while Dell uses less-efficient single-core chips from Intel."
Can't you read? Are standards of hiring at Intel so low?
I believe any claims by Intel that they "invented" it are as credible as claims that Al Gore did likewise for the Internet and probably do more harm than good to the corporate image with corporate buyers.
Where oh-half-blind-one does it say they made the claims? Did Al Gore ever claim that he invented the Internet?
What a waste of space.
I had an Orinoco aka Wavelan aka Lucent, recently Proxim Access Point (AP1000) more years ago than I can recall - at least 4 - in my home. It failed last week and it turned out to be the power supply which is modular but built into the frame.
The date of manufacture on the power supply is 5-1-1999. There's no obvious date on the logic unit.
The AP1000 wasn't the first Wavelan manufactured AP.
It probably cost $1500 new, I replaced it with a $49.95 Linksys WRT54G.
I still use the original Wavelan cards. These units were in common usage when Centrino came out. The .bin files in the AP manager (not an early version) are dated 12/99.
While consumers may have caught onto WiFi about the same time as Centrino I believe any claims by Intel that they "invented" it are as credible as claims that Al Gore did likewise for the Internet and probably do more harm than good to the corporate image with corporate buyers.
That was for mmoy, sorry
Drop me a PM or somewhere I can talk privately.
The licence issue will not be a problem - I assure you - and Sun might be very happy to see the plugin and know who you are.
jump through a few hoops
Let me know if you have a problem. I can definitely get an answer on that for you. I will be in Europe for a few weeks from next weekend tho but will try and track your messages.
Will be having dinner at my Sun friend's house there! The source issue will be conversation over brandy.