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.
Jerry R,
As usual, you get half the story from chipguy, the one unfavorable to AMD.
Here is the second half. Along with 64 bit memory addressing, in AMD64 implementation you get twice the number of registers, both integer and SSE2.
With the increased number of registers, the same source code can compile to fewer instructions. So there can be a code size reduction offseting the code size increase due to bigger pointers. And, because of more registers, there are fewer pointers in compiled code to begin with.
I am not quite sure what chipguy means "program data memory". Maybe that's how chipguys talk among themselves, but there are actually 2 things, code and data that take up memory while program is running, and pointers are in code, and can also (usualy are) also in data. The pointers that are in "data" part is hard to deal with, and probably not worth the effort of creating a kludge of a workaround to save some of the 64 bits. If a lot of pointers are in "data", the program is probably a good candidate for staying in 32 bit for as long as move to 64 bit can be delayed (or forever on Opteron).
But on the code side, the registers can eliminate not only a lot of pointers, but a lot of memory accesses as well.
Where is the tradeoff? Depends on the program, and on the way it was compiled, but there should be a net benefit in performace on majority of programs.
And where there is no benefit, one can readily use the 32 bit alternative.
Joe
sgolds,
This is off topic, but if you have a choice, dump DSL and get cable. I went through all the nightmares with DSL from the very beginning, from time they Beta tested it more than 4 years ago, until I finally gave up because of endless problems. I got Time Warner cable's RoadRunner, and it just works all the time. As much as I used to hate my cable company, I have to say I am content now.
BTW, good download speed, some 2 Mbs, upload is only 100 or 200 Kbs or so, but that's enough for me.
Joe
Keith,
Thanks for the info on codecs.
Joe
chipguy,
LOL, I am certainly being held to high standards here. I say something about a feature in a microprocessor and instantly the onus is on me to give a complete rundown on which OSes support it and which don't.
It would be a fair thing to do, if you were interested in being fair, that is. While on the subject, knowing that no Windows version supports the feature you described, are there any OSs that support this feature? If so which one?
Meanwhile you don't seem to be overly concerned about even the general accuracy of claims made on behalf of AMD processors by AMD enthusiasts
I don't post here a lot, and as far as I can tell, wbmw and Elmer do a good job to correct inaccuracies of AMD enthusiasts, generally in a fair manner.
Joe
chipguy,
I told you how the P4 SMT HW
works. If MS does or doesn't support a specific feature in one
or more of its OSes at any given instant in time isn't my concern.
Finally! Even if it is not your concern, I think it would be fair to point out that the feature cannot be used in current version of Windows, if you were interested in this forum being an information exchange.
BTW, do you also think Intel was lying about the 32 bitness of its 386, 486, and P5 processors before Windows 95 came along? ;^)
First, I didn't say anything about Intel lying. I didn't see any statement from Intel advertising the feature of turning HT on or off on demand, which is not implemented in software. I only saw in your posts here, and I am not aware of you identifying yourself as a representative of Intel. So it is only your posts I am talking about, not Intel.
With that said, there is no need comment further.
Joe
Petz,
Of course, one major disadvantage of P4's hyperthreading is that you either enable it for all programs or you disable it.
Chipguy posted a theory, by which HT can be disabled per application, and I took his word for it, only to realize that it may be an intention, or something in the spec that at some point someone can implement, but at this time, no one has implemented it, and there is no way it can by used by a current PC user this way.
BTW, chipguy knowingly posted this half-truth here, and never corrected it, even after prompting to do so, which gives you an idea about who is posting here in good faith and who does not.
Joe
smooth2o,
Is there any reference where the 750,000 came from any official AMD source? All I remember seeing was a post on [H]ardOCP site.
Joe
Elmer,
One can only remember back to the stary eyed dreams of Hammer being the highest performing processor on the planet....</>
.... in its class <g>.
Joe
yb,
I heard rumors that Athlon64 based Shuttle SFF system is in the works.
Joe
Elmer,
AMD #2, reasonably successful, not very profitable
I think we have an oxymoron here.
Not really. If you are still in the business and virtually all but one of your competitors are out of that business, that's success. And there is always "potential" to be profitable.
Joe
RSteiner,
You can certainly underclock, but I don't know if passive cooling will cut it. You would definitely need a mega heat sink, and a good air flow in the case (but good air flow needs case fans). But once you get a good heatsink, like one of the Thermalright either 70, 80 or 90 mm, there is no reason not to put a nearly silent fan on it, like this one:
http://phamcomputer.safeshopper.com/44/87.htm?633
or one of the panaflo fans. The ADDa fan I linked is rated at 22.5 dBA, and for comparison, sound of whisper is supposedly 23 dBA. Some fans go down to 1500 or so RPM vs. 4000 to 5000 RPM many of the 60mm default fans have.
If you already have a CPU, such as XP3000, you can continue to run at full speed, with much quieter system. I actually have righ now the same ADDA fan on my Thermalright 800U (nice model, with screws now as opposed to clip of the previous non "u" 800 model). It lets me overclock my Barton 2500 to 2.2 GHz or 3200 rating effortlessly (without increasing voltage).
But I have very well ventilated case, with actually 3 more of those fans. The system is quiet, but not silent. I originally had one fan intake and one outflow, but for some reason, I had 2 IDE drives die in theat system in a short period of time. They ran very warm, almost hot. My case has a spot to place a fan in front of the hard drives, so I added a 3rd one, and now my drives are cool.
Hard Drives may be another consideration, because they like to be cool, and even if you totally silence your system, you will always hear the drives.
Joe
Jerry R.
In my opinion, I believe AMD's business model is flawed.
It could be that CPU business is more "winner take all" than most other businesses. Look back at companies making CPUs, and their performance. It is Intel at #1, very successful, profitable, AMD #2, reasonably successful, not very profitable, and the rest are either out of business or going that way.
Joe
Keith,
Is there a DIVX codec for Windows Media Player? It doesn't seem to be part of the automatic downloads. I ended up downloading DIVX player and it seems actually nicer. You can do smooth resizing of the image, rather than just 100%, 200% and Full Screen that Windows Media Player offers.
Joe
Keith,
The Inquirer has been reporting on this, and I seem to recall seeing it elswhere as well. But there has not been any official announcement AFAIK.
Joe
yb,
I see it positively that Newisys joined contract manufacturer instead of one of the OEM's.
Exactly. This is important to current Newisys customers, which is that Newisys did not go to their competition, and they don't have to buy from competition.
Joe
yb,
Does AMD get any money from Newisys sale?
Possibly. AMD was one of the investors.
Joe
yb,
I think AMD did a big mistake by introducing Barton as Athlon FX. They had to call it Athlon GX, and then make Thorton as FX as well.
Bartons are still called Athlon XP, and Thorton will be called Athlon FX.
Joe
wbmw,
just to let you know that I am not ignoring you deliberately, I have not signed up as a paying member for Ihub, so I don't have an option to reply privately (to a messages you sent earlier).
I am a little confused about some messages that I got from you and Tench in the past that were deleted. Just wanted to make it clear, sorry about wasted bandwidth.
Joe
Petz,
I hope someone will correct me if I am wrong, but I was under the impression that there was a change in functional units from Merced to McKinley, but not from McKinley to Madison. Madison was mainly about shrink to .13u, increasing cache and increasing clock speed. And if you asked me, I think they achieved these objectives. Now we will see if anybody buys it.
Joe
edgarcayce,
We believe you have a lot of shares, and that it has been frustrating to see them not go anywhere. But as far as blaming management as incompetent, well, they are not really firmly in carge of their own destiny. Intel has the pricing power, the market power, clout, marketing dollars, you name it.
AMD management (and employees) may be doing their best, or even THE best that can be done under the circumstances, and still come up short. Or they may succeed again as they did with Athon.
They are doing a hell lot better than all the microprocessor manufacturers other than Intel (since they are pretty much out of business, with exception of IBM and Sun, but Sun will go down as well soon).
my patience is running very thin. I am tired of the empty promises and the hollow looking forward statements. AMD is becoming a joke. They cannot get it right and Intel is trying to crush them with every quarter that passes.
True. Hammer delays have been highly frustrating, and the last delay, the one of the desktop from Q4 2002 to Q3 2003 has been downright crushing. But here we are, 2 months from the launch, and since I stuck around for such a long time, I am going to give it another shot.
The good thing is that the quarter ended up being not as bad as it could have been. I think Hector is learning how to manage the analysts and the expectations. Actually, that's the most important role of a manager. To manage the expectations, and then to meet them and to exceed them. The analysts like to be managed this way, since they don't really have to do any analysing on their own.
The problem with AMD is that Intel has as much to do with what happens with AMD as AMD management, so I am slowly learning to take increasingly bigger grain of salt with any forward looking statements.
Intel can hammer AMD in 2 ways:
- When Intel has performance lead, such as now, with their 2xPC3200 + 800 FSB chipsets + P4(disregarding low volume Opteron for now), Intel can "Celeron" AMD, while Intel is cruising
- When performance is roughly matched, intel can start a price war in every segment where AMD competes, hopefully getting some profits from monopoly segments to stay in black
- When AMD has performance lead, Intel can't crush AMD and AMD can make money
Now the big question is: Will Athlon 64, Hammer, mobile Athlon64, Mirrorbit give AMD performance lead? There are no guarantees, but there is a chance of that, and subsequently of AMD making money, which can have very beneficial effect on the stock price.
Joe
wbmw,
McKinley was a completely rearchitected core. The x86 portion of the block diagram also has evidence of a complete rehaul. Surely, they would have improved performance, while they were at it.
I am talking improvement per clock. Do you have any evidence of this? I didn't see any claims of improved x86 performance. Do you think Intel wants to keep it a secret?
BTW, Palomino -> Tbred had a significant redesign, with no improvement in IPC.
Joe
wbmw,
Actually, there was an enterprising poster on SI who called IBM, expressed interest in one of the Piii 1.13 systems, and got into a friendly conversation with one of the sales people, and he managed to get the info of how many IBM was receiving. I seem to recall that it was either 4 or 40 per week. There was another source based on which Dell had, or sold only 100s. I don't want to invest a lot of time into this, but I think Dan is fairly close, IMO closer than the 10,000 that you posted, which itself is nothing but a guess of one analyst who just guessed, without any support for his guess.
BTW, back on the 1.13, I seem to recall that IBM and Dell were the only ones selling it.
Joe
Elmer,
I couldn't find that table at either of the links you provided
You have to actually download the software, install it, and one of the options brings up that table.
the reference is to an old Itanium running x86 mode. Not a Madison and not a McKinley.
I am not sure how much changed other than clock speed. Do you have any refernce about improved x86 performance?
While Madison has a huge cache, which at some point can gain an order of magnitude performance (due to no cache misses), Itanium's problem is in the hardware implementation of x86, and they are so severe that cache plays no material role.
If you compare other processors with small caches such as Duron, Celeron, Willamette, Athlon, they are all in 4 to 6 range (needing 4-6 CPU clocks for 1 Flop) Itanium needs 38.9, while it has fewer cache misses
To come up with some actual performance, the highest speed Madison has 1,500M cycles per second so it should be able to process 38 MFLOPs per second, the average of the other processors is approximately 2 GHz able to process a flop per evey 5 clocks resulst in 400 MFLOPs per second (which I can confirm with the actual result of my Seti run right now).
So current processor can process more than 10x of what Itanium can process is x86 mode.
Now Intel has a plan B, which is to come up with software emulation, but nothing has been released, and no performance increase (if any) is known.
Joe
Elmer,
Link please.
Here is the link to Seti@home download page:
http://setiathome.ssl.berkeley.edu/download.html
The chart is in one of the more popular addons called SetiSpy:
http://cox-internet.com/setispy/
Here is the chart (formating may be screwed up)
Processor L2 Cache CpF*
Intel Itanium 2 256 KB + 1.5 MB L3 1.3
Intel Itanium 92 KB + 2 MB L3 1.9
Alpha 21264B 8 MB 3.1
Alpha 21264A 4 MB 3.2
Intel Pentium III Xeon 2 MB 3.2
AMD Opteron 1 MB 3.6
Intel Pentium III Xeon 1 MB 3.6
HP PA-8200 2 MB 3.7
Intel Pentium M Banias 1 MB 3.7
Intel Pentium II Xeon 2 MB 3.8
Intel Pentium II Xeon 1 MB 4.0
Intel Pentium III Tualatin 512 KB 4.1
Alpha 21164 8 MB 4.2
AMD Athlon XP Barton DC-DDR 512 KB 4.2
AMD Athlon XP Palomino/T'bred DC-DDR 256 KB 4.3
AMD Athlon XP Palomino/T'bred DDR 256 KB 4.5
Intel Pentium Pro 512 KB 4.5
Sun UltraSPARC-II 4 MB 4.5
Intel Pentium II Xeon 512 KB 4.6
Intel Pentium II/III 512 KB 4.6
Intel Pentium Pro 256 KB 4.6
Sun UltraSPARC-IIi 2 MB 4.6
AMD Duron Morgan DDR 64 KB 4.7
Intel Pentium III Xeon 512 KB 4.7
Sun UltraSPARC 512 KB 4.7
Intel Pentium III (Dual CPU) 512 KB 4.9
AMD Athlon Classic 512 KB 5.0
Intel Celeron 128 KB 5.0
Intel Pentium IIIE Coppermine 256 KB 5.0
AMD Athlon Thunderbird SDR 256 KB 5.1
AMD Athlon XP Palomino/T'bred SDR 256 KB 5.1
AMD Athlon XP Palomino/T'bred DDR (Dual CPU) 256 KB 5.1
PowerPC RS64-III 4 MB 5.1
Intel Pentium III Tualatin 256 KB 5.2
This processor (instantaneous) 512 KB 5.24
AMD Athlon Thunderbird DDR 256 KB 5.3
PowerPC G4 1 MB 5.3
AMD Duron Spitfire DDR 64 KB 5.4
AMD Duron Spitfire SDR 64 KB 5.5
Intel Pentium III Tualatin (Dual CPU) 512 KB 5.8
Intel Celeron II Coppermine 128 KB 5.9
Sun UltraSPARC-IIi 256 KB 5.9
Intel Pentium 4 Northwood RDRAM 512 KB 6.0
Intel Pentium 4 Northwood DC-DDR 512 KB 6.0
MIPS R5000 512 KB 6.0
Intel Celeron (Dual CPU) 128 KB 6.1
Intel Pentium 512 KB 6.1
Intel Pentium IIIE Coppermine (Dual CPU) 256 KB 6.1
Intel Pentium 4 Willamette 256 KB 6.2
Sun UltraSPARC-III 8 MB 6.2
Sun UltraSPARC-IIe 256 KB 6.3
Intel Pentium 4 Northwood DDR 512 KB 6.6
Intel Xeon Prestonia 512 KB 6.7
AMD K6-III/III+ 256 KB 7.4
AMD K6/K6-2 512 KB 7.6
Intel Celeron III Tualatin 256 KB 7.7
PowerPC G3 512 KB 7.8
AMD K5 512 KB 8.9
AMD K6-2+ 128 KB 9.0
Intel Xeon Prestonia (Dual CPU) (2 clients/CPU) 512 KB 10.1
Intel Celeron 4 Willamette 128 KB 11.2
Cyrix M2 / IBM 6x86MX 512 KB 12.2
Weitek SPARC Power µP 1 MB 13.7
Intel 486DX 256 KB 21.3
VIA C3 64 KB 21.8
AMD 5x86 256 KB 30.6
Intel Itanium (x86 mode) 92 KB + 2 MB L3 38.9
AMD 386DX 256 KB 59.5
sz,
If running 32 bit software is so important for 64 bit servers, why did Newisys go out of business for lack of Opteron server customers?
Excuse me for not following your "logic".
Joe
Jerry R,
It could mean that (i.e. higher binning parts). It could also be interpreted as their fastest transistor switching speed in volume production. Whether this allows (or will allow) for faster K8 bins remains to be seen, given the necessary speedpath work on subsequent steppings.
Good point. Based on some rumors, K8 is currently limited by layout and speedpaths, and not silicon/speed of transistors.
Joe
subzero,
By the way, thank you for listing all the 16 and 32 bit software that can run on the Opterons but not the itaniums. I'm sure that is a comfort for all those 64 bit server customers.
I bet these customers will be comforted when they realize that when used as a web server under Windows (ASP.NET), that super-duper Itanium will run at speed of between 386 and a Via C3.
Joe
Elmer,
You're not dumb. Why make a dumb post? You know perfectly well that all those OSs run natively on Itanium.
Technically they do, but you are not dumb to know that practically, by 2003 standards, Itanium does not run any of these OSs.
FYI, in Seti@home processor efficiency chart (# of cycles / FLOP), while Itanium is #1 in IA64 mode, it is second to last in x86 mode, beaten by such powerhouses as AMD 586, Via C3, Intel 486.
It's ahead of the 386 though. <g>
Joe
subzero,
AMD's biggest Opteron customer, Newisys, is going out of business.
Now that is a lie. Newisys was just bought today.
They have no customers of their own.
Another one, unless you did not know that some of the Opteron systems are based on Newisys design.
Joe
subzero,
You have no data whatsoever to justify this latest lie
So then, by definition, it is not a lie. A lie is saying something you know to be not true.
Joe
subzero,
How about some OS support - what OS runs on an Opteron that does not run on an Itanium?
DOS, Windows 3.0, 3.1, Windows for workgroups, Windows 95, Windows 98, Windows Me, Windows NT, Windows 2000, Novell Netware, 32 bit versions of Windows XP Pro, Home, 32 bit version of Windows 2003.
Why did Hector dodge the question about Opteron revenue and unit shipments at the conference call today?
For the same reason Intel does routinely. Don't get me wrong, I am not happy about not knowing various breakdowns. I was happier when AMD was disclosing these numbers in the past.
Joe
Paul,
It was encouraging to hear that SOI gave them their fastest silicon yet, if I understood correctly.
Good catch. That means > 2.2 GHz, I guess.
Joe
edgarayce,
"shipped more Opteron so far than all of Itanium in 2002"
There are not that many positives about AMD business, but this is one of them. So why not say it? Haven't you listened to million CCs by now? The presenter always stresses all the positives.
"gross margin improved to 34% up for 31% q on q"
Hello Bob and Hector have you guys calculated what AMD's margins must be to break even and show a penny profit?
You may be talking about net margins. Gross margins are based on sales and cost of sales.
"3rd quarter cash balance of 919 million, we expect minimal Q3 cash burn"
Where does this money come from Hector? Thin air? A money tree? We all know AMD will be in the 650M range of cash by then.
As part of consolidating the new company, AMD will assume some of its cash and some of its debt. So both cash and debt will go up.
lastly...
"we will become number 1 in flash"
Tell me Hector do you think you are Jerry Sanders now? Quit lying! Hector you've got 3-4 quarters or we the shareholders will ask for your resignation.
Well, it is a possibility, depending on Mirrorbit, demand (plus price premium} for high density flash and the NOR flash suppliers (AMD, Intel) fending off the NAND suppliers (mainly Samsung).
It looks like a 3 way race between Intel, Samsung and Spansion. If you think Hector is lying, maybe you can tell me the order of these top 3 a year from now. I, for one, am not sure. All 6 combinations can plausibly happen.
Joe
Keith,
With A64´s launch on September 23, did you actually expect 90 nm sales in Q1 ?
Well, yeah, paper launch in Q1, real shipments in Q2.
BTW, the original Athlon did not exist very long under the .25u. It moved quickly to .18u.
Joe
Wafer production starts on 200mm 90nm Fab 30 in first half 2004.
Bummer. I guess this means no 90nm sales in Q1 and most likely Q2.
Joe
Tenchusatsu,
Non sequitur
I thought this was a BBS for FUD. Is it not?
Joe
chipguy,
Perhaps you should take off your ideological blinkers.
Perhaps you could lead by example.
BTW, how is it with the HyperThreading and OS support (or lack of thereof)?
Joe
chipguy,
Well, Intel has not delivered to them for years, and we will see if what Intel is promising for next week will do it for them.
Their commitment to Itanium has made SGI (once a high flyer), a penny stock.
Joe
wbmw,
If that's the case, then future systems will have the same limitation.
I am not sure. It seems that this board had very limited overclocking options.
Joe
Paul,
Just assume for the sake of argument that cache defects are not a substantial issue, and that the huge majority of the Thortons would have cache disabled for marketing/economic reasons rather than for defects? Would it make any sense?
I would guess that some of them are salvaged Bartons with half cache bad, so the alternative is the dumpster, and from that point of view, the cost is free.
It would not make sense to switch the entire production to Baron Thorton, unless this core has some advantages. Possibilities are better bin splits, the s2k disconnect issue that supposedly cuts power consumption.
Joe