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.
Chris,
Don't worry about Elmer. It's not bad to have some contrarian info on the thread (as long as it is not infinitely reposting the same info as you know who).
As far as the leaks ased on pre-production hardware, everybody takes it with a grain of salt (I am assuming), but you get a ballpark idea of where the performance will be.
For example, I remember that there were a lot of Tbird 700 MHz floating around prior to launch, and a lot of benchmarks were posted. There were no surprises during launch, other than the fact that the speed grade that was used in test machines was not even used at launch.
Based on this, I would not be surprised if the launch speeds are 1.8 and 2.0 GHz, as some site posted recently.
I missed the posts on this BBS when X-bit labs preview was posted, the issue may have been covered extensively. My take is that 2 GHz A64 will on the broad spectrum of benchmarks be at or slightly below the performance of P4 3.2 GHz, overall, while each one will have some spectacular wins on some individual benchmarks. Well, this is assuming 32 bit code. On 64 bit code, A64 2.0 GHz should take the lead from Northwood 3.2 GHz.
Joe
Semi,
Good memory on your part.
Joe
wbmw,
let me suggest you an experiment. Start a thread where you will post your dissertations on QS. See how many people come. That will give you some feedback on whether there is any interest, or whether your post on the subject are an annoyance that you are imposing on others.
Joe
Elmer,
A64 benchmarks are already available and they aren't good
They are not great, I will grant you that. (I only considered Xbit labs review).
Joe
sgolds,
The final version of AMD64 Windows Server was announced (or announced to be planned) for end of 2003 earlier, at WinHec. So this was just re-iteration.
Lack of regular Server version (not Enterprise) is a dissapointment, IMO.
Joe
wbmw,
Used the wrong punctuation at the end of my last response? What?
Your post implied that you would give us a break with your QS spam until Athlon64 benchmarks are available. But somehow, I have my doubts that you will take a 3 month break from spamming this board with endless repetitions.
Joe
wbmw,
when's your next "absence" going to be?
Your continuous spamming of this board makes the time investment to keep up with this thread unaffordable for me.
But as long as everyone else is ok with your spam, I am not sure if I should interfere.
Joe
Paul,
Looks like Jozef may be getting his wish.
I would like to see it in 2003 thogh... If not on Athlon64, then at least Opteron 1xx with support for unbuffered PC-3200 memory.
Joe
UpNDown,
Good post. My WAG would be that x86 is higher and processor independent is lower, because most of this "processor independent" software never runs on anything other than x86, and you can't really be sure that it is processor independent.
Joe
wbmw,
LOL. I come here after a month of absence (I've been busy), go back some 3 pages of posts, click on the first post, and surprise, wbmw talking about Quantispeed.
Joe
wbmw,
You are back to gibberish.
BTW, are you trying to spam this board out of existence?
Joe
Andy,
Regarding that AMD investor, it does appear to be the same MO. Do you think he would be so foolish to pull the same crap here???
I guess it is much easier here, since you can always sign up for free.
Joe
Andy,
For instance, if you recall "reynoso" over on SI.....the long time and passionate AMD investor that always seemed to allude to impending AMD issues and problems........well that guy turned out to be Paul Engel using one of his at least half dozen aliases.
Really? How did you find out? I figured something was not kosher with that "AMD investor", which prompted me to put together this collection: http://www.siliconinvestor.com/stocktalk/msg.gsp?msgid=17041112
BTW, check out this "AMD Investor":
http://www.investorshub.com/boards/profile.asp?User=19850
I can smell the same odor.
Joe
wbmw,
Rather than rehashing things point by point, please reread this post (and you should do so before every post on the subject), which has answer to most of your questions:
http://www.investorshub.com/boards/read_msg.asp?message_id=978644
On the subject of ideal measure of performance, there is an answer to that question in that post as well.
But let me add: On top of the fact that it will never happen, let's hypothetize, and suppose it does happen, and the true performance measure is a manifold in a space exceeding 3 dimensions.
Well, the reality of PC buying is Costco or Best Buy. Even if you succeed by penetrating those skulls with at most a single number, the chances are, you will lose units of measure of that number.
Joe
wbmw,
there are many contributers to performance in the Itanium 2, but I think the primary thing is the architecture itself.
I think the architecture (other than 64 bitness) is still a very mixed blessing, and what is now propelling Itanium is brute force, combination of huge cache (and consequently large die) plus good infrastructure / chipsets, that are finally making their way to the market.
Look at the 62W Deerfield variant, for example. I think Intel will have done an excellent job if they can produce a product with about half the power of their performance model using what is essentially the same design.
If Dierfield can make it into a mainstream, and is competitive with Opteron and Xeon, that would be more of an indication of a potential of the architecture, but I think this time around, it will still be the brute force of Madison that will make or break Itanium (plus Intel's staying power).
Of course it will help that HP / Compaq will clear the field by killing Alpha and HP's chip, and the fact that Sun will not be able to keep up.
So, IMO, the scenario for Itanium success is that all RISC competitors commit suicide (except IBM), plus Itanium takes off before Xeon suffocates in the 32 bit space.
Chances of Opteron are obviously that Xeon suffocates in 32 bit space before Itanium takes off, and may be significantly improved (deteriorated) by the speed of the clock speed / performance ramp. And of course that AMD's financial condition doesn't get much worse.
Joe
wbmw,
I also know that you are running out of ways to defend it, which is why you are trying to turn the argument back on me.
You are feeling very confident today. I haven't started yet.
I know that the measurement is worthless.
Ok, let me review our last argument on the subject from about a year ago:
--------------------------------------------------------------
wbmw: Model numbers only makes the Athlon XP look better with respect to the Pentium 4 from a marketing perspective. It has no relevance to performance, and in fact, it's even less accurate.
http://www.siliconinvestor.com/stocktalk/msg.gsp?msgid=17144936
jh: Now how is MHz accurate, if every processor has different performance at given MHz?
I agree about marketing angle. Here, AXP model numbers reflect their performance vs. P4 more accurately than MHz.
SpecInt Base Peak
P4 1.7 573 587
P4 2.0 640 656
AXP 2000+ 697 724
P4 2.0A 722 735
wbmw,
No doubt Itanium 2 performs better. It's because Itanium 2 architecture is highly parallel, not because it processes data in 64-bits or accesses memory in a single flat address space.
I think Itanium can be a good database server for:
1. flat 64 bit address space
2. large L2/L3
Parallelism may have a potential, but it first has to offset lower clock speeds before it can add anything.
Joe
wbmw,
QuantiSpeed is one so-called standard that has always rubbed me the wrong way, and I believe I have enough good reasons to argue against it in every way.
Ok, at least you now admit to being an instigator of these debates, rather than saying things such as "Whenever I see QuantiSpeed questioned", since you are most of the time the questioner. Well, that's progress.
Some people have convinced themselves that QuantiSpeed is the only way for AMD to market their processors, but they say so without even trying to think of something better. Abolishing megahertz is not the answer. Coming up with a rating that looks like megahertz is not the answer, either. Performance is a complex picture that must be boiled down to a consistent marketing message - a monumental task, for sure, but one that is necessary. You can't create an arbitrary measurement when an existing measurement doesn't tell the whole story.
Yup, exactly from the playbook. AMD needs to discover origin of universe and the really, really true measure of performance. It is truly a monumental task, as you say. A task that. BTW, will never be achieved and agreed on.
At the same time, it is fine with you that Intel pushes MHz as a measure of processor performance. You have not found any reason to argue against this. For every processor that is sold with QS measurement of performance, there are 5 that are sold with MHz as a measurement of performance.
You spend 100% of your time on a minor problem, 0% of your time on the problem 5x bigger. A curious devotion of resources on your part, wouldn't you say?
Joe
chipguy,
Curious how DEC wasn't demonized for selling 500 MHz processors that were roughly equivalent in performance as the 200 MHz processors from HP that they competed against in the market place.
They competed in a completely different market place, in market place of MS and BS in CS and MBAs, non in Walmart and Costco.
I'll type this slowly so maybe you can follow: M H z i s a m e a s u r e o f c l o c k r a t e n o t pe r f o r m a n c e.
You may want to let Intel marketing know about this discovery of yours, since they are not aware of it. They have been spending billions to sell MHz as a measure of performance.
Joe
Tenchusatsu,
speaking of FUD, why not tell me what has come out of AMD's so-called "True Performance Initiative"?
You are trying to hold AMD to perfection, while Intel spends billions of marketing dollars pushing empty MHz.
If AMD does not achieve this perfection, it is labeled garbage by one of the deep thinkers on the Intel side (I forgot who it was).
Of course, you say that MHz is not perfect (which is only a pretention, to hold the high ground on the BBSs), but every day, millions of dollars are being spent to market this very same MHz that you seemingly denounce, but you never denounce it. And you accuse deception?
AMD is just seeking a measure by which to compare its chips of one architecture to the dominant architecture in the marketplace (of a different architecture), that's all.
Right now, the dominant architecture is P4 in form of Northwood and Celeron.
To accept this would require some fairness, which is hard to come by when a joicy opportunity to spread FUD is at hand. And it seems that you are taking these cheap shots over and over.
I know you can concoct some seemingly rational reason to convince yourself that you are being fair and honest, but let me tell you, you are not.
Joe
Tenchusatsu,
Ok, so I was wrong. You are also using the argument for FUD purposes.
Joe
wbmw,
Thanks. Your kind words have inspired me to write at least another dozen posts on the subject....
You think you can limit it to a dozen? Nah.... I don't believe that.
Joe
greg,
Yes, I am quite familiar with that argument. Set up an unreachable goal for performance measurement, something that everybody knows will never happen, and in the meantime, market dilluted MHz, and spread FUD about QS.
Tenchusatsu often uses it, with the only difference is that Tenchusatsu maybe means it but other posters who use it just for FUD purposes.
Joe
There goes another QS post by wbmw. Like a clockwork.
It's far too complex to define the performance of everything, but one thing that should not be done is to try to confuse matters by reducing it all into a single number, which is what QuantiSpeed attempts to do.
That's deep. I am making notes. I would have never though of it without your assistance.
Until now, I was a just dumb AMD fan boy, but you have truly enlightened me.
Joe
greg s,
#msg-972321
That's your best contribution? Another QS post? LOL.
Joe
wbmw,
as long as people continue to respond to my opinions here, I can only assume that they want to hear what I think.
Yes. We can't get enough of your words of wisdom. We want more. What was it about a single number not being enough to represent the true performance? Why don't you repost that dissertation. I can't get enough.
Joe
Another value-less post by greg s. Have you ever generated any value added post on this board?
Joe
wbmw,
Constantine, you have arrived a little late in the conversation to be bringing back the performance argument. QuantiSpeed is an approximation that either overestimates or underestimates performance.
Hey, great idea. How about explaining all of your VALUABLE thoughts to every new alias that appears on this board. That's going to be a lot of fun.
Joe
There goes abother wbmw QS post. This discussion keeps me on the edge of my seat. I haven't had this much fun since the Dukakis presidential campaign.
Joe
chipguy,
I am sure that when Intel stops confusing the market with diluted MHz, and adheres to a more objective measurement of performance, AMD will follow suit.
Joe
wbmw,
I have given multiple suggestions on what AMD should be doing. These suggestions have been among my other constructive criticisms.
Since you are an expert on everything Intel, how about you give all these valueable and deeply interesting suggestions of how to measure processor performance first to them. After all, Intel controls 80+ percent of the market, and your valuable comments would do so much more there.
When you are done convincing Intel than MHz alone, without detail explanation about how it does not represent performance, is not an accurate measurement of performance, we will welcome you here with open arms, and we will try to convince AMD to follow suit and adhere to more objective measurement of performance than either MHz or QHz.
Joe
wbmw,
I find it amazing the lengths to which people choose to fight this issue.
I find it amazing how far YOU.
Yes, I said YOU,
as in YOU PERSONALLY go to argue this issue.
Please don't say other people when it is YOU who keeps starting tis issue OVER and OVER
Joe
wbmw,
Whenever I see QuantiSpeed questioned
Very "diplimatic".
Shouldn't you say "whenever I question"? Since you seem to be a busy bee on this new and refreshing subject of QS. It is so new and refreshing, every other day you start it that I am having hard time staying awake.
Why don't you try an experiment. Stay away from commenting on QS for one year and see how often QS is questioned on the boards that you participate in.
Joe
Doug,
re: Osbourne effect
But there isn't any Athlon64 for sale currently. Even if there were, at this point, so early on in a new processor family, I'd think the demonstration of headroom would do more good than harm. They could do with restoring some confidence in their ability to execute.
Confidence of whom do you think AMD needs to regain?
- Geeks on the BBSs? They are worth something, but not a whole lot.
- OEMs? For OEMs, AMD does not need to do public demonstration (with its negative Osborne effect). AMD can demonstrate things behind closed doors, under heavy NDAs.
- Consumers? You want consumers to buy what's for sale today, which is Tbred, Barton or Opteron. If you show what may be for sale 6 months from now, you will most likely lower today's sales. Demonstrating say Barton 3500 that may be for sale in Q4 2003 would not do any harm, since it is expected. But a 64 bit desktop CPU is something that some people would delay their purchases, because it is more of a quantum leap that say Barton 3000 vs. Barton 3500. I think big percentage of PC purchases of computers go to people upgrading. And if you are not hurting too badly, you could consider waiting 6 more months.
- Investors? Maybe. But intestors have seen so many things from AMD that were always just around the corner, that such a demostration would probably not do a great deal. Cutting losses significantly in Q2 and profit in H4 will do a lot more for investor confidence.
Also, a timely shipment of Opteron 246 in good quantity will also be an extremely welcome sign, that can be immediately translated to bottom line.
Joe
chipguy,
Thanks. Makes sense.
Joe
Doug,
I don't buy the "Intel would focus harder" line. There is no benefit to sandbagging. You'd think they would know this by now.
But there is the Osborne effect.
Joe
yb,
Question about socket 754 (Athlon 64): how many HT links this socket can pass through?
I was looking at the pins in the middle of the Opteron Socket 940 (in the Opteron Tech Docs), and there are some pins that are kind of questionable, which would most likely not be in Socket 754. These are some power pins for LDT1 and LDT2 (LDT0 being the first one). So I think there is only 1 HT port in Socket 754.
Curiously, the pins for full 128 (or 144) bit data path seem to be there in socket 754.
Joe
Elmer,
If a process is set to not use iHT, why can't the OS still use iHT when running the first with a second unrelated process?
First, it was a news to me that HT can be enable or disabled per process (thread), I am just thinking about implications.
Implication of the preference of the process to run in non-iHT mode means that it doesn't want to share the physical CPU with other processes. Actually, to say it better, the non-sharing process instruct the physical CPU not to act as 2 logical CPUs. The other process, the one that wants HT then has no say, because the requests are contradictory (IMO).
There is one small way out, if the machne has more than 1 CPU, the process that wants all of the CPU can be on 1 CPU, and other processes can be on other CPU(s). But the non-sharing processes can quickly take over all CPUs as they spawn threads, or if multiple instances of the process run concurrently.
Joe
wbmw,
Microsoft will release an x86-64 compiler at some point, but I wouldn't hold my breath.
There are 2 parts: C++ compiler and .NET engine (which enables all .NET apps to become 64 bit apps). My expectation would be that both would be released in approximately the same timeframe as the AMD64 OS. Probably within a year from now.
They are notoriously slow at working in architecture optimizations into their compiler.
If there was a huge demand for architecture optimizations, I am sure MSFT would have responded. But the issue is complicated since the target machines out there range from Pentiums (P5), P6, P68, K6, K7 (all flavors) of each, some Cyrix / Via, each requires different optimization. So it is a huge can of worms for vast majority of developers, and they rather keep it closed.
But a for a small percentage of developers, the issue is extremely important. Intel is already serving this boutigue crowd, showering the product with subsidies. I don't think MSFT sees any threat here from Intel, so the optimization market is safely ignored by MSFT.
As for their x86-64 support, they will do the minimum, plus any low hanging fruit that happens to fit within their schedule.
I don't expect any more, but even the minimum is a lot, since it involves efficient use of the extra the extra registers (the biggest performance booster). Also, compiling for AMD64 target, it there is an assurance that the machine has all of the MMX, 3DNow, SSE, SSE2, so these instructions can be used readily, rather than being a can of worms. AMD64 is one architecture, X86-32 is a dozen of different processor architectures.
I don't expect MSFT to go out of their way, but I am assuming (my assembly skills are somewhere from rusty to non-existent) that any floating point instruction can be dispatched to processor through SSE-2 mechanism rather than FP, which is more efficient. So the AMD64 compiler can completely eliminate FP instructions. (correct me if I am wrong).
By the way, my previous post had absolutely nothing to do with SPECint.
That was a confusion on my part.
Joe
chipguy,
Wrong. It [hyperthreading] is selectable on a per process basis. An OS that properly supports HT should allow you to set an attribute on each application as to whether you want to allow it to run in multi-thread mode or force single thread mode.
How does that work? Suppose you have 2 threads, one says it doesn't want HT. Than the other says it wants HT. Suppose the OS runs the HT first on one logical CPU, than starts the non-HT. What happens? The 2 logical CPU turn to 1 logical CPU and the HT is disabled for the entire system while the non-HT thread is active?
Joe