Opterons, shortage of silicon, and the rest

More on Opteron... Lateron
There might be a non-conspiratorial, purely economic reason for the dearth of Opteron MBs.

The Opteron version of AMD64 is targetted at servers. The chipsets available now don't have AGP. Consequently the only thing that can be built is a server board with little utility for other purposes.

The Taiwan MB guys aren't crazy. They generally build multipurpose boards to maximize their sales potential in return for the effort needed to design them and perhaps more importantly build inventory.

AMD sales into the server market have historically been a small percentage of Intel's. Further the server market is much smaller than the desktop market. So it might just be that they chose to wait for an AGP chipset to implement a multipurpose board rather than pursue a smallish percentage of a smallish percentage market.

This combined with a new CPU which has yet to be proven which inspite of the 64bit marketing still adds further risk. Recall how many bugs have been found after a million CPUs have shipped? Server customers are notoriously conservative and would be unlikely to put a AMD64 in a critical path until they have been fully wrung out, preferrably by someone else. That would add another "smallish percentage" multiplier.

Just to pick some numbers out of the air (which no doubt means they're wrong):

x86 Server units = 0.01 x86 Desktop units
AMD server unit share = 0.1
AMD64 "risk factor" = 0.2 (completely arbitrary)

That means the total multiplier might be as low as 0.0002. This in turn needs to be divided by the number of potential suppliers.

Now consider that the server TAM is more than half fulfilled by major OEMs and that whitebox servers would be the destination for these boards and the potential is further reduced. Assuming a generous 150M desktop current annual rate that yields about a 15000 unit whitebox potential for everyone to divide up.

Added to the current state of the world economy then it's no wonder to me why the MB suppliers have chosen to proceed cautiously.

jtm
Email address supplied

AMD is foolish not to be in the motherboard business themselves. They've already done the design work for the chipset and accompanying reference-design samples for other vendors.

All they need to do is contract out the production to a Taiwan motherboard company, and the rest is gravy. They get to charge premium prices because of the Official AMD branding (just like Intel does). They already have distribution channels (which they try so hard to f*sk up). Maybe this move would help them get into so-called tier one vendors, too.

Name, email supplied

In praise of PAT Gelsinger
Intel says there's no such thing as PAT

Hello,

I just read the article regarding Intel's PAT for the 875 chipset. The SimHQ interview indicates, as you suggested, that Intel "bins" the MCH to find "the fastest silicon." And, they "can then use this fast silicon to shave off a couple of memory clock cycles."

If I understand this correctly, you're only half-right, in that yes, better silicon is used. My argument is that I think it *can* be considered a "technology" in the sense that PAT can be en/disabled, effectively using more or less aggressive memory timings. From a hardware architecture standpoint though, obviously there is no difference.

This ASUS press release supports my position: http://www.asus.com/news/2003/20030528.htm It appears that PAT can be en/disabled in the BIOS, but instead of PAT, it is called "Memory Acceleration Mode." So, enabling PAT would cause the system to use more aggressive memory timings, ie shaving off a couple cycles. Note that ASUS claims they support PAT on their 865-based mobos in this press release.

Herman Poon
Email address supplied

Opteron is Lateron
Mike, I read this story: Intel strong armed vendors at AMD launch

I'd like to ask you to put up a list of the vendors who DID show up at the opteron launch. Along with a little nudge to your readers to support those who resisted Intel, and show up anyway. Naturally you couldn't be too overt with this

If people were to choose those products from vendors who resist such tactics, then other vendors may start to resist, as well. If a trend were to develope where a good number of cunsumers showed hesitation to purchase from people who use these tactics, as well as those who knuckle under to them, it might just slow them down a bit.

Craig
Email address supplied

Mike, As one of the dozen Opteron launch partners, we have been going through the same issues. We luckily got a few of the MSI boards and are waiting for a proper working Tyan board.

From what I have heard, the motherboard manufacturers had late shipments from Broadcom and even AMD for the main chipsets.

Just my 2 cents worth :)

Company name, email supplied

Have you seen this page? http://www.sis.com/products/chipsets/oa/socketa/755.htm# for Athlon 64 there is a picture and a detailed pdf document at: this address.

This pdf also has benchmarks comparing with via k8t400, testing real boards with driver versions etc.

This page says that the 755 will have an fsb of 800 MHz

talk about great deals:

newegg is carrying this this motherboard.

this ECS L7S7A2 based on the sis 746 that has usb 2.0/FIrewire and it includes a 6 in one card reader (that's about 20$ normally) for only 57$ incl shipping!! Tests at Overclockers workbench has it beating the Epox 8RDA in most of the non Memory intensive benchmarks.

Ive been wondering howcome there aren't many tests of sis 746 based boards outthere??

I suspect that ECS has been trying a new strategy - Lower Price, Solid product =>Longer Product cycles & Greater Profit and Marketshare. look at how long the K7s5a has been going on for - im sure they are still making tons of profits from that 3 yr old board, after covering all their development costs etc.

Name, email supplied

Stop Saying AMD's x86-64 technology! It's “AMD's AMD64 technology” or “the AMD64 technology.”

“X86-86” is dead, dead, dead.

Name, email address supplied

Microsoft dragging feet on PNG transparency
Mike,

Is there a business reason why Internet Explorer still doesn't display PNG alpha channels? Mozilla, Opera etc will, but IE6 displays the transparency as pastel blue. You can get it to display the images properly, but with one of MS's custom CSS filters. See http://www.alistapart.com/stories/pngopacity .

The only mercurial reason for MS to drag the chain on this is Compuserve's patent on GIF.

James H
Email address supplied

Vole and its Updates
Hey, Mike

I've written to you before, and I just thought you should know that Microsoft updated the information on their web site on May 28 in regards to the Q811493 update. Here's a link.

I think the important part is, “This bulletin was originally released on April 26, 2003. It has been revised to reflect that an update is now available for users of Windows XP Service Pack 1.” I'm in IT and this issue was quite a head scratcher until I remembered reading something at the Inquirer about an MS patch and some very slow computers.

We had quite a few users that were ready to toss their computers out the window, had windows been available to those employees we keep in the basement.

If your ego hasn't been inflated enough, let me also inform you that an IBM desktop we had with a faulty hard drive was diagnosed and taken care of very quickly because of your report on defective Fujitsu drives done some months back. That information I'm sure led to my level of tenacity on getting an immediate replacement when discussing the matter with IBM customer service.

Thankfully, that machine was unique in our facility and we are otherwise Fujitsu-free. (I believe the replacement was a Maxtor.) There you have it. Keep up the good work and stay healthy so that the Inquirer's many readers can continue to enjoy all that you do. God bless.

Sean Quinn
Email address supplied

Storagereview has HDD stats
Hard drive makers' stories start unravelling

Can't you quote some www.storagereview.com stats in summary? With 15000+ drives in their database from 6k+ users, theirs is, I believe, the definitive guide to HDD reliability.

The tests done by the manufacturers simulate reliability by baking in an oven and calculating how this related to real world failure rate using mathematical modelling.

That doesn't simulate real life knocks during installation and use, poor quality power supplies, varying handling before delivery etc. All these affect the end user, and drives intended to work in the real world need to be built to handle these extra stresses.

According to storagereview stats:

Deskstar 75GXP : more reliable than 17% of other drives
Deskstar 120GXP : 81%
Deskstar 180GXP : 18%

DiamondMax D540X : 68%
DiamondMax D740X : 62%
DiamondMax Plus 9 : 7%

(both charts ordered by date of release).

Looks like we are entering another era of drive unreliability.

Will Smith
Email address supplied

Slimming Windows
Windows 98 fits in 16MB flash memory

We have been offering a service to install windows98 into 16MB flash chips for 3 years now.

Our common solutions are actually 32 MB disk on module or disk on chip units running web browsers and other client software. Our website http://www.embeddingwindows.com has some details.

At present this is a consulting service we offer our customers now. A toolkit is also in the works. Our installations do not involve compression rather we componentise the OS down to the Kernel level and build upwards.

Cheers
Shane Brooks
Email address supplied

Graphics, Futuremarks and ATI
Who is the culprit here?

I've been shopping for a graphics card, and trying to make some sense of Futuremarks Project Search and Compare to evaluate performance of various cards with my CPU. Not so easy and I'm annoyed. In short, Futuremark gives you much more choice of 'flavours' of Nvidia cards than they do of the ATi ones. For example.. you can select Nvidia's 5600 or 5600 ultra, however, you can't compare an ATi 9500 with the 9500pro. !! In fact - when it comes to ATi - the 9500 and 9700 series are all lumped together. ??

You can look at the card frequencies to try to determine which result is which. But this is tedious to sort through a number of 9700 and overclocked 9500 results to get information on the 9500. Very few people are kind enough to name their benchmark report with an indication of exactly which card they have.

I can't imagine that ATi would intentionaly muddle the waters by NOT allowing their cards to be identified specifically, as it only seems logical to me that this would deter sales. So I lay my suspicions on Futuremarks doorstep until I learn otherwise. They control the database and don't they then determine groupings for access of data !! ??

I'm sure if you have one of your graphics guru's dig into this it will make a nice follow up article on top of all the hullaballo about who is 'cheating' vs. 'optimizing' :)) I enjoyed also reading Simon Nash's opinion today: "Nvidia cries foul as hand caught in cookie jar". Here's the link to the Futuremark Project Search and Compare that I've been looking at. Here.

Looking forward to hearing from you

David Doyle
Toronto, Canada
Email address supplied

PS My apologies to Futuremark if it is the nature of the ATi beast that causes the "lumping" of their cards together :))

Mike,

Just read your article on AMD/Intel and benchmarking. You make the point in the article that Intel benefits performance-wise from SSE2 programs and you speculate that Intel keeps licensing technology to AMD in spite of the obvious competitive issues because of the US DOJ. I have an interesting perspective having worked at AMD many former lives ago.

In particular, you neglect to notice a huge driver on Intel's licensing decisions: Microsoft. Originally, when Intel developed MMX, Bill Gates actually called Andy Grove and told him to license the technology to AMD, otherwise Microsoft wouldn't code for it or support it in any way. Intel grudgingly accepted. The fact is, Microsoft holds the dominant power so far in the Wintel duopoly. And Microsoft wants hardware to be cheap, cheap, cheap. Microsoft can't afford to have its hardware become expensive as that would erode the value of the PC platform versus alternatives like Apple. As a result, Microsoft uses AMD to help keep Intel's power in check.

The other dynamic is with Intel and AMD also supporting Linux in order to keep Microsoft in check. This is a newer dynamic, however, and one that is still building. It was interesting to see how much Opteron support AMD got from the Linux community, however, and this is shaping up to be an interesting influence in the Microsoft-Intel-AMD triad.

Dave Roberts
Co-Founder, VP of Strategy, Product Management, and Marketing
Inkra Networks
Email address supplied

Nvidia's Problems not to do with Loch Ness Monster
just FYI,

I don't think Nvidia's problems are due to the Loch Ness monster. I landed her with a whip of 700lb monofilament line on a bamboo pole last time I was across the pond. Catch and Release, of course, so I guess she's still a suspect.

Cheers!

Chuck
Email address supplied

Somewhere Kurt Goedel is laughing hollowly
Robots write the news
"the system crawls a series of Web sites, downloads articles"

Articles written by whom???

Recursive: see "recursive" 8-)

Seriously, I am always amused by efforts on the part of PHBs and "entrepreneurs" (read: those seeking something for nothing) to replace the human mind with computer power.

Until they can do that, they will remain at the mercy of those of us who actually have minds. I am glad to report that all efforts to date (e.g. "knowledge management" 8-)) have proved futile.

Somewhere, Kurt Goedel is laughing hollowly.

Tom
Email address supplied

Nvidia... again....
"Nvidia did well, not badly"
"Nvidia cries foul as hand caught in cookie jar"

" sense bias. I also sense that the author decides to attack the people supporting nVidia's actions instead of actually attacking the specifics of the facts."

I sense the writer of this letter did not obtain enough facts of his own.

"No where did I read that the changes made were so drammatic as to drastically alter the image in 3dmark. No facts were presented in this editorial about how drammatically different they were. If it took the developer edition and examining the image from multiple angles to see the difference then I thinksomeone's pulling gnat-shit out of pepper."

This is where the greatest lack of understanding lies. It is little to do with how drastically nVidia altered the image, it is more important that the "cheats" they made at driver level were dependent on the path the camera takes through the demos. Should the camera be moved off of Futuremark's pre-defined course, drawing errors occured on the screen. Had this been a game rather than a fixed camera benchmark, this would be totally unacceptable.

Can you imagine playing a game such as Unreal Tournament with clipping errors and refresh errors because the driver was "optimised" for the flyby timedemo? Obviously nVidia wouldn't dream of doing this - the same should apply to 3dMark03.

As this benchmark is used to draw conclusions of how graphics hardware will perform under game conditions this is indeed underhand cheating!

Not to mention that the principle of this benchmark is to test different hardware under the exact same conditions. If nVidia write specific code that changes the visual output of this test, they are no longer working to the same conditions as ATI and other hardware manufacturers, and it can no longer be used as a valid benchmark.

"If ATI was cheating (only a little) and nVidia was cheating a bunch more, it is still cheating."

As mentioned in the original article, ATI optimised their code with no effect on the visual output. That means, should the camera be moved off course with an ATI card, all scenes would still be rendered correctly. The fact that ATI are withdrawing this optimisation in future drivers purely shows their understanding of how important the benchmarks are to their products and the whole idea of cheating ultimately leaves their customers with a sense of mistrust.

Cheers
Ben Langridge
Email address supplied