Each initiative is also a spike driven through competitors' Xeon and Itanium offerings.

Not that there are many of the latter to speak of anymore.

Yuri/conductor

Intel will have additional 6xx series in January. So what should PR be for an A64? Is the target the 5xx clockspeed or 6xx clockspeed?

Yuri/conductor

If it's going to break, fix it before it does!

Sun Opteron secret project 'Thumper'

http://www.aceshardware.com/forum?read=115103638 links to:

http://www.theregister.co.uk/2004/10/01/sun_thumper_revealed/

which says


Sun Thumper server plans excavated by The Reg
By Ashlee Vance in Chicago
Published Friday 1st October 2004 21:05 GMT

Exclusive Sun Microsystems has an all out Opteron onslaught planned for next year that could include attacks on both the server and storage sides of the house, The Register has learned.

For years, Sun has promoted the idea of a data center-in-a-box centered around its UltraSPARC-based servers. CEO Scott McNealy has been the leading advocate of this plan that has Sun ship pre-tested, pre-configured bundles of servers, storage, networking and software to customers. Sun has promoted this as a better deal for customers than buying parts piecemeal and then assembling the gear on their own.

Sun will extend this concept to the Opteron market in the coming months. It currently sells two- and four-processor Opteron-based servers, but plans to complement these systems with boxes designed by Sun cofounder Andy Bechtolsheim, who returned to Sun earlier this year. These systems will be configured to work with a wide range of networking gear, storage boxes and software, sources told El Reg.

Such a move, however, is not surprising. Sun is clearly looking to add its enterprise expertise to the world of x86 computing and make large hardware sales. What is surprising is a secret project code-named Thumper that goes along with the Opteron agenda.

At present, details on Thumper are proving hard to unearth. What is known is that Thumper is a specially-engineered system that combines Opteron processors, unique networking equipment and some of Sun's most-advanced file system technology. The 64-bit release of Solaris 10 also plays a central role in the Thumper project, sources said.

So far, our helpers are staying mum on what the secret sauce of Thumper really is - as Sun believes the system will be a major threat to competitors. As we understand it, tweaks done to Thumper boxes will make them well-outperform simple combinations or clusters of Opteron-based servers from a networking, processing power and software partitioning standpoint. Since Sun owns all of the parts needed to make Thumper other than the Opteron chips, it's thought the company will be able to price Thumper systems far below competing offerings. Thumper is due out in the first half of 2005.

Along with Thumper, The Register has also uncovered a futuristic Opteron-based storage system in Sun's labs. This system is still a research project and may never make it to market, but it's a brand new direction for the company.

The storage box is around 3U high and has four Opteron blades in the front-end. These servers connect into a midplane, which has 16 disk drives total on the back-end. It's basically a NAS (network attached storage) system packed full of capacity. The system connects to the network via two gigabit Ethernet ports on each blade.

Sources have revealed that Sun is looking to use this system as a type of archive box for massive amounts of storage. It's somewhat similar to the Centerra system from EMC but would be used to hold tera- and even petabytes of information at a low cost.

We'll bring more information on this gear as it arrives.

In total, Sun has clearly committed to the x86 market in a massive way. Customers could see serious benefits with Sun applying its well-regarded engineering expertise in this "commodity" market. It, however, remains to be seen if Sun can keep costs down on these systems as promised. The whole idea behind x86 systems is that they are cheap and aimed at general purpose tasks. If Sun can harness the power of Opteron and maintain decent price/performance, it could be a real threat in a market that it was slow to enter. ®


Yuri/conductor

But think. AMD can still compare to Intel. Just copy model numbers! (But change first digit to 1 higher, to annoy Intel!)

Yuri/conductor

"Total dominance" was total stupid marketing! Total waste of money. Non geeks do not know about secret rating outperformance!

See my edit of prior post for something we may agree about!

Yuri/conductor

This cannot be. 18-19mm^2 is too big for 256K of L2 at 90nm. It is 15 mm^2.

Simple math!

114mm^2 = 90nm Clawhammer
84mm^2 = 90nm Winchester

By extension: 69mm^2 for 90nm 256K L2 part

Yuri/conductor

I said 'The Celeron D is sometimes 50% faster than the Northwood Celeron, how anyone thought you could ignore this and loosen the K7 ratings considerably and not expect to be mixing it performance-wise and credibility-wise is beyond me

Who says they ignored it?

"Mixing-it" performance-wise is the goal!!!

The average performance of things that matter should show a Sempron 2800+ to be the equal of a Celeron-D 2.8GHz. If Sempron outperforms too much, you leave money on the table. If Sempron underperforms too much, you lose credibility.

AMD did it right. Sempron slightly outperforms Celeron-D across the spectrum of things that matter. Synthetic benchmarks do not matter. Applications do.

Yuri/conductor


EDIT: Perhaps we can agree here: AMD can drop PR ratings across all brands and switch to Opteron and Intel style model numbers in the near future. Start of year 2005 would be a good time. Both companies are heading towards having many different cache sizes, number of cores, clock speeds. Now that Intel has done this, it can be okay for AMD, especially if they are smart and choose first digit 1 more than Intel :) :)

Example: Sempron 440 for part that is equivalent to Celeron-D 340

This would give Intel fits!

That is a typo. 90nm MA64 is 84mm^2.

Yuri/conductor

Facts don't matter? Please! You did not say this in your post before! You believed the review, and said that it showed the Sempron was rated wrong against Celeron-D.

I show that the review uses bad choices of synthetic benchmarks, and bad combining methodology of averaging to reach a bad conclusion.

Then you say facts don't matter!!!

Facts show that for real-world applications, Sempron rating is more than fair against Celeron-D.

Once again:

Games, Office, Software development, Content creation, Image rendering: Sempron outperforms rating

Encoding, bogus antivirus disk drive benchmark, synthetic benchmarks: Sempron underperforms rating

Which is more important?

In the real world, Sempron PR rating is more than fair!

Yuri/conductor

To mas,

You need to be a critical reader, not just accept such poor methodology! Taking an "average" can be strange business, as can selection of tests. Let us look in detail:

Test 1,2,3: PCMark04. Who cares?! Memory speed test? Look at what that will do to averages.

Test 4,5,6: 3DMark01/03 More synthetic tests!

Test 7: Quake 3. Celeron D wins an old game
Test 8: UT2004. A recent game. Sempron ahead
Test 9: Aquamark fps: Tie.
Test 10: Aquamark "CPU" synthetic rating. Who cares? Surprise, Celeron wins even though it doesn't win in FPS!!!
Test 11: FarCry: Tie.
Test 12: Doom3: Sempron wins

Test 13,14: Business winstone, content creation. Sempron wins.
Test 15,16,17,18: 4 encoding benchmarks, 2 with HUGE average-affecting leads for Celeron, 1 other Celeron lead, 1 Sempron lead

Test 19: Compression. Celeron lead.
Test 20: Antivirus speed??? What is this test? Celeron lead.
Test 21: Image edit. Slight Celeron or tie.

Test 22, 23: Software compiling/dev. Large Sempron win.

Test 24,24,26,27: Rendering. Clean Sempron wins.

So what did AMD bring on itself?

Victory!

If you want to:
(1) Render images
(2) Build software
(3) Create content or use office apps
(4) Play any recent game

then Sempron is your processor.

If you want to:

(1) Encode and compress data
(2) Run synthetic memory tests
(3) Run synthetic graphics tests
(4) Run synthetic cpu tests
(5) Run antivirus quickly

then you want Celeron-D.

It does not get much more clear than that.

Yuri/conductor

This is old news, as has been the speciality of forbes.com lately. Any news item is repeated as news by forbes.com in about two weeks later, for unknown reasons.


Yuri/conductor

DDR400 2-2-2 memory, not 2-3-2.

Yuri/conductor

To Jhalada,

You are right, of course. If 2-2-2 timings would make no significant difference, Intel would use them, at that is the memory type everyone would use with such a high-end processor.

The other sign of the coverup is that nowhere do you see absolute scores! Such scores could be compared with public data, but then everyone would know instantly that the DDR-400 scores were a joke!

Intel does this once before, at the launch of Nocona. They provide only ratios of various benchmarks, but do not provide scores for the older, "slower" system. It turned out they chose a very slow old system, to exaggerate the improvement of going with the new system.

Blowing smoke indeed!

Yuri/conductor

Why do you post such nonsense articles? That author has been against AMD for a long time now. All he does is whine and whine that AMD will not give him parts for $50 that he can overclock by 50%. Good! AMD investors should rejoice at his misery!

Yuri/conductor

No, this is silliness! Such a slide falls apart under examination. DDR 400 in an enthusiast system would be using timings 2-2-2, not 3-3-3. Beyond this, the two parts are not clocked at the same speed!

Fix this, and would any advantage be left for DDR2 with the 1066 bus?

Not much!

Yuri/conductor

I did not mean one could know for sure or calculate earnings or some such thing. For one thing there is the flash business too. Just what you say in the end: these numbers look good. They are a positive sign for Q3, not a guarantee.

Yuri/conductor

I think these numbers look good for AMD Q3. Would you agree?

Yuri/conductor

The $110+, and even the Monarch $170, are for OEM parts, not boxed parts with heatsink, fan and 3 year warranty.

Monarch charges $205 ($35 more) for the warranty and heatsink/fan boxed product.

If you search pricewatch for "3200 retail" cheapest is $197

Yuri/conductor

To Elmer Phud,

I do not have an ability to send private messages, so I will respond here.

You said, regarding the Xeon throttling issue:

You might take notice the he described his processors as having 1 meg L3 cache. These would have to be NorthWood processors and they don't support the 800MHz FSB in dual mode.

I assume this was a typo and that he meant 'L2'. Other posters in that news thread said that all 800MHz FSB Xeons are Noconas. Is that not correct?

Yuri/conductor

Alan,

Perhaps it is that 3.4s simply throttle more often than 3.2s, but are almost as available?

Just kidding!

An interesting data point.

If 3.6 is not available, how can 3.8 be coming soon? Or is 3.8 only for P4 not Xeon?

Yuri/conductor

Alan,

also a strange thing that all the Xeon are specified with the same power level of 103W

I believe Intel decides this is the highest acceptable power. As a consequence, one expects to see not many 3.4, and even fewer 3.6 Xeons, which would seem to fit with observations.

Yuri/conductor

An interesting point about the case cooling. One small error however. The poster was having the mentioned throttling troubles with 3.2GHz Xeons, not 3.6GHz Xeons.

Yuri/conductor

To Alan,

Thank you for your thoughts.

You say here:

The nocona does require better cooling than some older systems provide.

But the poster with the problem did say this:

4. The CPUs, sinks, and fans were replaced as a single unit, so we
didn't have any issues of using cheap fans from the old processors, we
just had the problem of the processors shipping with crappy stock
sinks/fans.


So perhaps Intel needs to put a much better fan and heatsink with each Nocona it sells?

Yuri/conductor

It does appear though that Verisign had already purchased the Opteron servers, and is now choosing to deploy Solaris onto them. Meaning, this is not a new sale of Opteron servers. That would explain why AMD is not also announcing this.

That is not to say it isn't good news for AMD. It is.

Yuri/conductor

Those chips were basic Prescott (a.k.a. Nocona) parts. 1MB L2 cache at 90nm.

Yuri/conductor

Fanatical gamers concerned about benchmarks might well be using more expensive cooling solutions.

Yuri

To morrowwinder,

Why do you say this? The poster from the Intel newsgroup stated clearly that he used the cooling solution provided by Intel with the new processors.

Yuri/conductor.

To I_banker, that is interesting. But perhaps most gamers would not make such exacting tests, or if they do worry about benchmarks, maybe they use much better cooling?

Or it could be that server chips are set to cause throttling at a lower temperature because they must be more reliable and are assumed to be working around the clock?

Yuri/conductor

Do any of you know if Nocona throttling problems are widespread?

Here is my post on the AMD board:

http://www.investorshub.com/boards/read_msg.asp?message_id=4140408

Eventually the links lead to an Intel newsgroup, so a shortcut is here:

http://groups.google.com/groups?hl=en&lr=&ie=UTF-8&safe=off&threadm=f37bc412.0409131...

This describes a 3.2 Xeon Nocona system undergoing very bad performance because of throttling.

Could this be related to this reference here?

http://www.theinquirer.net/?article=18688

I know this is more of a tabloid than other places, but there is an interesting line that says:


Industry reports say the EM64T range from Intel, launched hurriedly this year and which compete with AMD's range of chips, isn't exactly as ready as it is willing to be ready, an important distinction.

Could these two be related? That is to say, could performance throttling because of heat problem be a widespread issue?

Thank you.

Yuri/conductor

Intel AMD64 Xeons have thermal throttling issues?

This is what I have found on the Aces Hardware forum.

http://www.aceshardware.com/forum?read=115102596

http://www.aceshardware.com/forum?read=115102607

If this is true, it is quite bad, because this is only a 3.2 speed Xeon part. One can assume the effect would be much worse with faster parts.

Intel cannot seem to break free from bad news.

Yuri/conductor

HP Spurns Intel

This linked from the Aces Hardware forum here: http://www.aceshardware.com/forum?read=115102551

Article: http://www.fool.com/news/mft/2004/mft04092718.htm?logvisit=y&source=estmarhln001999&npu=y&am...

Add a chapter to the sad tale of Intel's (Nasdaq: INTC) sorry 2004. Today, longtime partner HP (NYSE: HPQ) admitted it would be dropping the beleaguered, 64-bit Itanium chip from its workstation lineup. Instead, HP will sequester Itanium chips to its high-end servers. At the lower end, it will use a mix of AMD's (NYSE: AMD) 32-64 bit combo Opterons and the similar, backwardly compatible Intel chips.

If that sounds like a small indiscretion to you, remember that HP spent years shacking up with Intel to work on this very chip. HP will continue to use it, but the decision to play the field on this particular line of machines sounds like the beginning of a long, messy affair.

The writing was on the wall some time ago, when AMD's 64-bit chips began selling like crazy because of not only their earlier appearance on the market but also their compatibility with the existing 32-bit software that still dominates the market. In other words, buying Opteron meant businesses didn't need to switch all their system software right away.

In explaining the HP decision, a spokeswoman told The Wall Street Journal that customers prefer this approach. As they say on the playground, "Well, duh."

The Itanium's reception was so frosty that Intel was forced to copy AMD's approach back in February, but by then, server makers such as IBM (NYSE: IBM) had already decided to start using the AMD chip. And it's a good thing, since Microsoft's (Nasdaq: MSFT) 64-bit Windows won't be arriving anytime soon.

While an Intel spokesperson countered today that the workstation market was never that important for the Itanium, it sounds a bit like doubtful, self-soothing posturing of the spurned. After all, in addition to servers, IBM, Sun Microsystems (Nasdaq: SUNW), and others currently offer workstations built on the Opteron.

Intel has already bulked up inventory while watching AMD gain market share. So the financial fallout from today's announcement may be minimal, but investors need to wonder whether the litany of goofs, gaffes, and the recent revenue whiff will be stopped anytime soon.


Yuri/conductor

Hewlett-Packard to Drop Intel Chip for Workstations

Here is the Street.com article:

In another setback to Intel (INTC:Nasdaq - commentary - research), Hewlett-Packard (HPQ:NYSE - commentary - research) will stop offering workstations powered by the chipmaker's Itanium processor, according to a report published in The Wall Street Journal.

The move is all the more striking given that H-P originally teamed up with Intel to design the chip.

The news comes on the heels of a number of well-publicized product delays and one recent recall at Intel. At its most recent technology forum in August, top executives 'fessed up to the problems and said they are focused on fixing them.

Itanium's star-crossed development, beset by high expenses and delays, earned it the moniker "Itanic" early in its life, and so far it has failed to elicit significant interest among customers. Intel maintains that the silicon will catch on over time as more software is written to take advantage of its high-end features.

Meanwhile, a server chip designed by rival Advanced Micro Devices (AMD:NYSE - commentary - research), known as Opteron, has proven so popular that Intel was forced to crank out a chip with similar features, in effect cribbing from its much smaller underdog competitor. H-P said back in February that it would begin offering high-end computer hardware running on Opteron chips.

Intel spokesman Robert Manetta said he could not comment on any potential financial impact because the company is in its pre-earnings quiet period. But he said the server market is a more important market for Itanium than workstations.

H-P offers servers running on both Itanium and Xeon chips from Intel.

An H-P representative was not immediately available to comment.

Intel shares were recently down 7 cents, or 0.4%, to $20.06. H-P was down 15 cents, or 0.8%, to $18.43.


Symbolic value. That's what I was saying earlier this weekend.

Yuri/conductor

I must admit I did find the Reuters headline to be quite amusing. They are clear in the text of the article, but I would agree that it is an unfortunate title as far as the Itanium is concerned.

Yuri/conductor

Yes, it ran all the fans at maximum speed when it was not required.

Yuri

Itanium unsuitable for workstations, Intel says

It continues.

Here is the link: http://www.theinquirer.net/?article=18688

There is also some interesting comment about industry sources saying Intel AMD64 Xeon products not being quite ready. This is intriguing. Are there delays in shipping products which were announced this summer?

Yuri/conductor

It was a bios problem, see the followup post.

Yuri/conductor

HP to stop using Intel Itanium chip - WSJ

Sun Sep 26, 2004 11:03 PM ET
NEW YORK, Sept 26 (Reuters) - Printer and PC maker Hewlett-Packard Co. (HPQ.N: Quote, Profile, Research) will stop offering desktop workstations based on Intel Corp.'s (INTC.O: Quote, Profile, Research) Itanium microprocessor, The Wall Street Journal reported on its online edition.

The chip was designed with help from HP and was billed as a successor to the Intel X86 chip technology, but the first Itanium chips did not arrive until 2001 and did not handle customers' existing software as well as expected, the Journal said.

Intel had decided by then to target Itanium for high-end servers and workstations, designed to exploit 64-bit programs that tap into larger pools of memory than earlier 32-bit chips, it said.

HP and Intel could not immediately be reached for comment.

The Journal said rival Advanced Micro Devices Inc. added 64-bit capabilities to 32-bit chips, allowing customers to use both old and new software on the same computers. Intel was forced to emulate that approach earlier this year, it added.

HP decided to stop offering Itanium-based workstations because customers prefer the extended 64-bit approach, the Journal said, citing an HP spokeswoman.

Intel spokesman Robert Manetta played down the impact of the decision, the Journal said.

"The workstation market has never really been a major focus for Itanium," it quoted him as saying. "We continue full speed ahead on the server side."


Link from Reuters: http://yahoo.reuters.com/financeQuoteCompanyNewsArticle.jhtml?duid=mtfh65451_2004-09-27_03-03-33_n26...

Now maybe those who disagreed with my point about the important symbolic value of the HP cancellation will reconsider their positions?

Here we see the prediction of chipguy, which did not occur:

but the non-tabloid tech press will probably just make passing mention.

What has happened is that Reuters and Wall Street Journal have both picked up the story. It has become mainstream business news.

Yuri/conductor

Yourbankruptcy, Interesting. While the 130nm Sempron 3000+ is known to be a K7 Barton, perhaps it is possible that they would have a 90nm 3000+ K8 Sempron as well as 90nm 3100+ K8 Sempron.

Sempron 3100+ K8 with 256K L2 runs at 1.8GHz

If Sempron 3000+ runs at 1.6GHz, perhaps that fits with the pattern of Low Power Mobile A64:

1.8GHz = 2800+
1.6GHz = 2700+

Only 100 PR difference chosen there.

If this report is true and AMD (or a "main engine board merchant" ?) also reports 90nm will scale to 3GHz, this is good news.

One might then anticipate the following.

939 2.4GHz 512K L2 = 3800+
939 2.4GHz 1MB L2 = 4000+
939 2.6GHz 512K L2 = 4000+ ?
939 2.8GHz 512K L2 = 4200+ ?
939 3.0GHz 512K L2 = 4500+ ?
939 3.0GHz 1MB L2 = 4700+ ?

Here is the english translation of HKE:

Receives document which sends out by AMD, they will be able to plan in November promote 90 憟� rice Winchester core Socket 754 Sempron processor, believed could drop the Socket 754 Sempron cost thus is lowered again the price. The model will be able to include Sempron 3000+ and Sempron 3100+, can appear "BA in the CPU first line of writing the" inscription, the representative will be REV D the Winchester core, moreover according to the main engine board merchant expression, the NewCastle core ultra frequency ability which old will have approximately is 2.6Ghz, but the new core ultra frequency ability actually might reach 3.0Ghz, believed could hot promote once again the AMD ultra frequency

Yuri/conductor

H-P Ends Itanium at Low End In Setback for Intel's Strategy

http://online.wsj.com/article/0,,SB109622434518428079,00.html?mod=yahoo_hs&ru=yahoo" target="_blank">http://us.rd.yahoo.com/finance/external/wsj/SIG=11vps9ort/*http://online.wsj.com/article/0,,SB109622...

This is a Wall Street Journal article that requires a subscription for the text.

I think the title demonstrates the important symbolism of the HP cancellation.

Yuri/conductor

Thank you for the filter suggestion. I hasten to point out that it is not to my thinking an issue of "standing the heat", but of manners, and also of following the rules. Are the rules generally not enforced on this forum?

Yuri/conductor