Slashdot Mirror

How about an explanation of the numbering system right from the horse's mouth
mirror
from AMD's athlonXP site (doesnt' seem to be working right now)

web archive of AMD's site

Good news for the average computer idiot who wants to upgrade or buy a new machine. I think it's past time to undo the damage Intel's marketing has done with the Megahertz Myth. I'm weary of explaining it to people. It will be nice to have something more helpfully descriptive to a consumer than "cache" and "bus", or at least clarify that they don't refer to paper money and vehicles that carry children to school. :P

Here for Opteron and here for Apple.

Heh, I guess the Cray Red Storm system kind of shoots down that theory... ;-)

Actually the design of Opteron beats Itanium for HPC, and the relative number of Opteron vs. Itanium HPC design wins bears that out nicely.

Under a rock maybe?

AMD Athlon(TM) XP Processor: Now Available in Model 3200+

What dont you have a pdf-reader or are you full of shit?

Reference to SSE:
White paper on Athlon XP

As for the Intel thing. The selling point of AMD has always been they where fully "Intel Compatible". They had to be, otherwise noone would buy their product.

By the terms of the lawsuit described in that article, AMD cannot clone intel chips. They can (and do), however, implement intel patents, the latest I can think of being SSE2 instructions in their athlon-XP processor. They renewed their cross-licensing agreement in 2001 for ten years, and it cannot be broken without mutual consent.

This is a long standing agreement, dating from 1976.
intel press release

It is fairly well-known to insiders that Dave Cutler, chief software architect for Windows NT at Microsoft, approached AMD with the concept of extending the x86 instruction set for 64-bit instructions and data.

The motivation for this move was probably complicated, but Intel's slow-motion malaise regarding its IA64 strategy was no help. Microsoft needed a 64-bit platform that would gain wide acceptance before it devoted a significant amount of resources to drive Windows support on the platform to consumer-level quality.

Some even make the further claim that Cutler may have actually designed the instruction set for AMD and handed it to them intact. In other words, he approached them and said, "If you build a chip that runs this instruction set, we can guarantee NT support for it, and backwards compatibility with x86-32 will come for free."

AMD even acknowledges Dave Cutler and has a page with his information on their web site. If you do a search for articles, you'll find supposedly leaked memos mentioning builds of NT running on the new chip before it was even announced publicly (and hence before SuSe knew about it either).

You be the judge.

During the company's first years, the vast majority of its products were alternate-source devices, products obtained from other companies that were then redesigned for greater speed and efficiency

Ahh yes, this sounds familiar... In protected mode on the x86 (with segmentation enabled), far memory pointers are 48-bit, are they not? 16-bit segment selector + 32-bit memory offset into 4gb virtual address space. That is something of a problem, not being able to fit a far pointer into a general register (kinda makes indirect memory references via pointer hard, don't it?).

However, the new 64-bit architecture as proposed by AMD (I don't have Intel's specs) seem to suggest this might not be a problem anymore. The AMD64 Architecture Programmer's Manual Volume 2: System Programming has lots of information and I must admit it's a lot to read, but from what I gather would this not be an issue anymore? 32-bit memory offset (upto 4gb virtual address space) + 16-bit segment selector can easily fit within a 64-bit register, however I'm not sure if the CPU is designed to handle such things. From the Application Programming Documentation (at http://www.amd.com/us-en/Processors/DevelopWithAMD /0,,30_2252_875_7044,00.html) , AMD claims that "In 64-bit mode, the AM64 architecture supports only the flat-memory model in which there is only one data segment, so the effective address is used as the virtual (linear) address and far pointers are not needed." (page 23).

So maybe I have indeed answered my own question... segmentation as it was designed for will not easily be utilized... If so, would anyone with any sort of CPU architecture design/experience possibly give any reasons for this? Would it just be too difficult to modify the instruction sets and architecture to easily accomodate full segmentation support?

Wait just a second here. Do you mean to tell me that Intel and AMD still don't have no-execute flags for their page tables?

See AMD64 Architecture Programmer's Manual Volume 2: System Programming
5.6.1 No Execute (NX) Bit (page 173)

Maybe you're thinking of AMDs of the distant past. Current ones, and I'm specifically pointing out the server (Opteron) chip here, support thermal cutout.

Tech Sheet on the Opteron

Section 3.6 refers to the thermal cutout feature.

I don't know what "PCI clock locking" is supposed to be. All I can find about it is that it refers to overclocking - you want to lock the PCI speed so it stays at 33 or 66Mhz, and then you can do whatever you want with the FSB. If that's indeed the case, I can't imagine wanting such a thing on a server...

There have been some rumors that AMD/Opteron is a possible position for Sun, but the practical difficulty in a 8-way and up machines with the Opteron is probably a significant limiting factor in that path.

Possible postion?

Difficult 8-way?

I answer your FUD from my (Athlon) workstation at AMD...

Actually, this "feature" that Intel has installed on their chips allows the P4 to run as fast as it can until it heats up to a certain temperature, then the clock is scaled back. As a result, unless you have optimum cooling (unlikely), you are not getting the best performance out of your computer.

AMD chose to design the Athlon to run at optimum speed under normal conditions. That is why they (we) have a thermal design guide for the heatsink. There is nothing lower quality about this solution. Furthermore, I believe that with Athlon 64, Intel is now the one trying to create equivalents...

PCI Express is an Intel design bus. Win64 is an Athlon64 OS. It could be a while before we see AMD processors on PCI-Express boards.

Of course, the specs are out for HyperTransport 2.0, which is supposed to be compatible with PCI-Express. But we still need AMD to make a next generation processor with HT2. It hasn't been anounced, but 2H04 maybe.

AMD has an approach that says they will "build smarter" than their competition. Their flagship fabs (Fab 30 in Dresden, for example) are highly automated with very tight process control, ensuring the right work gets done at the right time. The focus is on equipment utilization; reduction of tool idle time. Further, they focus on minimizing the number of non-product wafers in the line, which take tool time but don't directly produce any chips that can be sold. The management of all this is done through software.

They also have to focus on fab uptime ... since they don't necessarily have the back up manufacturing capability to allow them to recover if their fab is down. For example, AMD makes about two-thirds their revenue from processor sales according to a recent 10-Q filing. Most recent quarter for which there is data (for the period ending 12/28/2003) shows $1,205M in quarterly revenue. You can estimate around $800M in revenue from their processor lines. Fab 30 make nearly all their processors. If Fab 30 were to go down for one hour, that's one hour in the 730 hours in a quarter that they can't make chips. If they have demand that is greater than or equal to capacity, and they're running at full capacity, they would loose roughly $1M due to potential finished goods that could not be made. A cost of $1M per hour of fab down time is pretty typical in the market where AMD competes and for fabs that compare to Fab 30.

A single tool going down is a problem. The entire fab going down is a huge problem. Things that can bring an entire fab down include utilities (electricity, water, gasses, etc.) contamination of facility-wide services like vacuum line, DI water, and various gasses, labor strikes, natural disasters, fires, and plant-wide software.

When you rely on software to manage your manufacturing to the degree that AMD and other high-end semiconductor manufacturers do, you tend to pay a lot of attention to the software.

This was announced on January 20th.

Remember the cooling rigs that rivalled icebergs?

You mean the Pentium IV heatsinks which weighed nearly 1lb and required a reinfoced motherboard and case to support it properly?

The low price is only possible because they're not fabricated to as high standards as Intel.

Well now you really need to define this, because when you're dealing with chips it isn't as though you can employ a factory of immigrant labour to assemble them by hand. If AMD wern't producing their CPU's to a high standard they'd have a very high failure rate, which would cost them a lot of money. One can assume from the fact that they are turning a profit and keeping the channels filled with chips probably means that they do not have a failure rate any higher than Intel does, so they must be manufactured to an equally high standard.

About the only thing I can think that you might possibly mean is that AMD chips do not carry a thermal diode to protect them in the case that they would overheat. If you think a component which would cost a couple of cents per CPU and which actually has dubious advantages means that AMD are not manufacturing their components to a high standard then you should study the differences more closely.

Notice that I didn't even bring up FAB 36 Oh, whoops.

Haven't seen any problem with AMD processors. It's necessary to follow the Cooling Guidelines, of course.

Make sure you have a good power supply. We use KingWin 350 Watt supplies that have two fans. (Ignore the language, "Extreme Series". That's there just to appeal to gamers, who expect every product to include some reference to violence or games. There is nothing extreme about them, and they are reasonably priced.)

Note that power supply manufacturers sell power supplies that have 100 Watts more rated power for sometimes close to twice the price. That's to take advantage of the "more is better" people.

Haven't seen any problem with AMD processors. It's necessary to follow the Cooling Guidelines, of course.

Make sure you have a good power supply. We use KingWin 350 Watt supplies that have two fans. (Ignore the language, "Extreme Series". That's there just to appeal to gamers, who expect every product to include some reference to violence or games. There is nothing extreme about them, and they are reasonably priced.)

Note that power supply manufacturers sell power supplies that have 100 Watts more rated power for sometimes close to twice the price. That's to take advantage of the "more is better" people.

AMD has made deep cuts in their CPU prices, probably pre-emptively.

Ah hah, you've cleverly chosen secret option D) I don't understand my own hardware.

AMD FX 64
Mobile AMD Athlon 64

You aren't talking about the same chips at all.

Ah hah, you've cleverly chosen secret option D) I don't understand my own hardware.

AMD FX 64
Mobile AMD Athlon 64

You aren't talking about the same chips at all.

I think AMD's naming makes a lot of clueful people a bit uncomfortable, but seems justifiable in a market dominated by a world-class bullshit artist like Intel.

Actually, AMD is as much to blame for the "megahertz myth" as anyone else is.

Remember that it was AMD that was so eager to engage in the great clock speed race. When AMD was the one who had the faster clock speeds, it was their PR that went on and on about the so-called "gigahertz era".

And since this was before the days of the P4, Intel's x86 CPUs
had more processing power clock-for-clock than AMD's.

If the "megahertz myth" is indeed damaging AMD in the marketplace, then it is largely due to their own shortsighted PR department.

And finally, note that Intel as least has been consistent in the naming of their consumer products. A 1.6 GHz Pentium-M is more capable than a 2.0 GHz Pentium 4. But Intel does not play fancy numerical name games. They tell you the architecture brand name, the clock speed, and the FSB speed of the chip. And those are the distinguishing characteristics. When I see a 2400+ Athlon on sale at Fry's, it's much obvious to a casual consumer like myself where it stands in the AMD family tree.

Here is the official patent from the USPTO. It was originally filed in 1998, but IC's have been around much longer than that, so I'm sure there's some prior art somewhere. This next quote could almost have come from the depths of the SCO complex:

"'Our Main focus is the IP [intellectual property] business now,'" he [CEO Jeff Wallin] said."

Kinda sounds like Rambus and look where they've gone.

Amigori

I think you will find them right here

"The Microsoft(R) WOW64 (Windows(R) on Windows) subsystem will allow most 32-bit applications to launch seamlessly on 64-bit Windows," said Brian Marr, Windows product manager. "WOW64 is designed to provide interoperability and great performance on AMD64 processors across the 32/64-bit boundaries. As customers migrate to 64-bit Windows XP and Windows Server, they will have a code base that will support both 32- and 64-bit applications."

While it's wonderful Linux understands multiple ABIs natively, Windows does not, and utilizes WoW to seamlessly launch 32-bit applications on 64-bit builds of Windows.

to cool their new Prescott blast furnaces. An Intel D875PBZ w/Prescott 3.2GHz burns 69W more power at maximum load than an ASUS K8V Deluxe w/Athlon 64 3400+, and lacks the Cool 'n' Quiet power-saving modes of the AMD chip. The Prescott burns 61W more than its Northwood predecessor. Comparing an ASUS P4 board to an Athlon 3200+ the results are even worse! AMD's upcoming desktop Athlon 64 CG core will increase that gap (.pdf file, see page 9).

Good grief, California should ban these things before they overload the power grid, between the power wasted by the chips themselves and the additional air conditioning they'll require. All this for performence that's rarely better and frequently worse!

BTX: Big Thermal eXhaust

" The way it works is that an XP2600+ is 2.6 times faster than a 1GHz Duron, a 3000+ is 3 times faster than a 1GHz Duron, etc.
This is according to "PC Hardware in a Nutshell" 3rd edition (O'Reilly).
Can anyone back this up with a reference from AMD?"

Nope, but I can happily dispute it. According to this document from AMD's site, it's based on the performance that would come from a 2.6GHz Athlon processor. The specific core to which it's compared is not detailed, however it would only make sense from a marketing perspective to compare it to the last core used prior to PR implementation, which was the Thunderbird core. The Thunderbird core, as any other, will not necessarily scale in performance in a perfect linear fashion simply by increasing clock frequency. Thus, it is more correct to state that an AthlonXP 2600+ performs similarly to a Thunderbird-core Athlon at 2.6GHz, if such a creature were to exist. To say that it is 2.x times faster is to belie the reality of diminishing returns.

Are there any apps that are 64 bits? Is there any reason at all to go 64bit?

You mean like this portland compiler

Actually I wouldn't say that gcc produces particularly bad code on all computers, it's sorta average, but not bad. Certainly the 3.3.x series are alot better than 2. Pretty good at number crunching and it is more standards compliant than most.

AND you can get the best of both worlds because data sizes do NOT have to be 64 bit just to use the 64bit registers. In fact here is a quote from an AMD manual on driver porting for 64bit-Windows on AMD64 platforms:
*Stupid freaking AMD engineer made the PDF world accessible but then went and encrypted it so that I can't cut and paste, print it, or do anything with it*
Well basically INT's and LONG's remain 32 bit while Pointers, LONG LONG's and Floats are 64 bit by default.
The paper can be found at AMD's website

A hammer isn't the right tool for every job as much as you'd like it to be.

Ah, but can't a Hammer also execute 32-bit code natively? ;)

Here's what I've discovered: If you're on a Windows platform, invoke Start | Run | winmsd.exe, then click on System Summary to get the Family, Model, and Stepping of your K6. [It beats the hell out of unscrewing the case, prying off the heat sink, and reading it directly from the processor.] Anyway, it looks like we have [and these are PDF documents, by the way]

Family 5, Model 7 == K6
http://www.amd.com/us-en/assets/content_type/white _papers_and_tech_docs/21641.pdf

Family 5, Model 8 == K6-2
http://www.amd.com/us-en/assets/content_type/white _papers_and_tech_docs/21641.pdf

Family 5, Model 9 == K6-III
http://www.amd.com/us-en/assets/content_type/white _papers_and_tech_docs/22473.pdf

Anyway, thanks again!

Here's what I've discovered: If you're on a Windows platform, invoke Start | Run | winmsd.exe, then click on System Summary to get the Family, Model, and Stepping of your K6. [It beats the hell out of unscrewing the case, prying off the heat sink, and reading it directly from the processor.] Anyway, it looks like we have [and these are PDF documents, by the way]

Family 5, Model 7 == K6
http://www.amd.com/us-en/assets/content_type/white _papers_and_tech_docs/21641.pdf

Family 5, Model 8 == K6-2
http://www.amd.com/us-en/assets/content_type/white _papers_and_tech_docs/21641.pdf

Family 5, Model 9 == K6-III
http://www.amd.com/us-en/assets/content_type/white _papers_and_tech_docs/22473.pdf

Anyway, thanks again!

Here's what I've discovered: If you're on a Windows platform, invoke Start | Run | winmsd.exe, then click on System Summary to get the Family, Model, and Stepping of your K6. [It beats the hell out of unscrewing the case, prying off the heat sink, and reading it directly from the processor.] Anyway, it looks like we have [and these are PDF documents, by the way]

Family 5, Model 7 == K6
http://www.amd.com/us-en/assets/content_type/white _papers_and_tech_docs/21641.pdf

Family 5, Model 8 == K6-2
http://www.amd.com/us-en/assets/content_type/white _papers_and_tech_docs/21641.pdf

Family 5, Model 9 == K6-III
http://www.amd.com/us-en/assets/content_type/white _papers_and_tech_docs/22473.pdf

Anyway, thanks again!

btw, I've done some digging, and found this document, scroll down to page 33 & 37-39; which I wasn't aware that software could also enable/disable the LME register. /* but then again, I don't know how/what they define 'software' to really mean/be. App & OS, OS not App, App not OS?...*/

/* thing that stikes me most, if the OS wanted to execute something, it got to have the CPU in it's 32-bit mode /* as M$ XP is 32-bit compatible only...*/, meaning that the 64-bit software first has to 'active long mode', execute it's instuction(s), 'deactivate long mode', give back control to OS, OS executes it's instuctions in 32-bit mode.... App then restarts it's cycle again by 'activating...' ... *you get my drift..* /* and is this possible, to begin with? */

... and is this 'activaing / deactivating' any/much efficient in the long run? *anyway, moving along..*
*/

and scroll down/up to page 2: "Defaults can be overridden in most modes using an instruction prefix or system control bit" - page 2 - table 1. "Operating Modes" /* but what's "most modes"? */

But, then again, does their 'proprietary' software do this 'instruction prefixing or system control biting'? is it possible for them to switch to '64-bit addressing' from inside 'legacy mode'?

Yes the Mobile Athlon 64 has clock throttling. It will modify clock and power draw in respect to both heat and if the processor is idle or not.

PowerNow

Its something to note that while many chips can have 64 bit pointers, the chip does not necessarily support 64 address lines. For example, from the Athlon 64 FX Datasheet found here, we know that the Athlon 64 FX has 40 physical address lines, Granted, that's still a Terrabyte of physical address space, but, its nowhere near the numbers you quote.

Mind you, the originaly 68000 was like this, with only 24 physical address lines, as were the 80486SLC's with only 24 physical address lines, despite being 32bit internally. Oh, and I believe MIPS arches have 30 address lines because they do not support non word-aligned read/writes, but that may or may not be true.

Oh, another thing, the Athlon 64 does clock in 64 or 128 bits per read/write cycle, so even if it uses the physical address lines for the high bits (most likely) its still not the full 64 bit address space.

I wouldn't call it _smoking_ but the truth about the Opteron is commoditization of the 64-bit space. Wasn't that supposed to be the impetus behind Itanium in the first place?

In lots of 1000:
Athlon 64 FX-51: $733
Athlon 64 3400+: $417
Athlon 64 3200+: $278

So retail using your $400/3200+ as mark-up ratio, that should be something like $1040, $600 and $400, respectively.

Also, on the P4/Athlon war I haven't checked lately since I'm happy with the XP2000+ I have, but at the price range I've been at AMD has come out on top for my last three processors (Duron 700, Athlon 1200 and the above mentioned XP2000+). Maybe the P4C is different, right now I really don't care though :)

Kjella

If you haven't seen any documentation it's only because you haven't looked hard enough!

There have been a number of tests comparing Opteron and Athlon64 processors in 32-bit and 64-bit mode under Linux and even a few that use Windows (beta version of WinXP). Here are a few links:

First off, some SPEC CINT2000 numbers: 32-bit OS + 32-bit apps, 64-bit OS + 32-bit apps and 64-bit OS + 64-bit apps. Unfortunately there are no similar CFP2000 numbers since GCC Fortran isn't up to the task, so you end up with lots of different variables making it nearly impossible to compare.

There is also this areticle at Ace's Hardware, and this little bit on Anandtech. Other tests exist.

Long story short, 64-bit support on the Opteron can and often does improve performance, even on apps that don't require lots of memory or use 64-bit integers. The extra registers help.

As for compilers, Microsoft plans on supporting AMD64 in their Visual.net 2003 compiler (beta versions are available now) and GCC supports the instruction set now. That makes up the compilers used for the vast majority of applications. As you mention, PGC is also doing a compiler, and it seems that Sun will support AMD64 with their compiler as well. Here is the AMD64 Developer Resource Kit page which lists all sorts of software with support for the AMD64 instruction set.

This linked article is an excellent roundup of the ongoing battle between AMD and Intel. It holds a lot of insight for people who have not been following the news closely.

However, it has to be pointed out that he missed several important incidents:

- AMD alliance with SUN: news article

-AMDs deal with Tippet studios: We built some prototype desktop workstations powered by AMD Athlon(TM) MP processors. We had tried systems powered by a competitor's processors, and they worked fairly well. However, we absolutely preferred the performance of the AMD Athlon(TM) processor. A good part of the advantage comes from the performance of AMD's floating point engine, which is very important to compute-intensive operations such as rendering.

-Intels new challenge in process technology with a cheap strained silicon process, finally unveiled at the iedm. AMD, this will be a touch one: IEDM article

• contains the "magic sequence"
• is contained in a valid ethernet frame
• is address to the target's MAC- or a multicast address (including broadcast).

Because of the handshake You cannot use TCP, but any UDP or ICMP (ping!) packet including the magic would do it. It has to pass the firewall (if any). The dest address could be

• unicast if the last router has a static arp entry for the dest
• broadcast if the last router forwards broadcast packets
• multicast if You have a multicast routing path from You into the last subnet.

Read AMD whitepaper and a howto.

/graf0z.

On the other hand, NetBSD has had amd64 support since 2001.

So NetBSD supports it before AMD even -officially- released it? The specs and emulator(s) have been around for a while, but I seriosuly doubt you could even get AMD's 64bit processors before this year (or possibly maybe even last year, but still, 2002 > 2001).

Although, there could be a blurring of lines on what 'support' means. I look at 'support' as being able to be sure it's stable, and problems are minimal (or non-existant). This is basically what defines (as well as a few other factors) a tier-one platform on FreeBSD.

Check the release dates for the Athlon 64 and Opteron.

On the other hand, NetBSD has had amd64 support since 2001.

So NetBSD supports it before AMD even -officially- released it? The specs and emulator(s) have been around for a while, but I seriosuly doubt you could even get AMD's 64bit processors before this year (or possibly maybe even last year, but still, 2002 > 2001).

Although, there could be a blurring of lines on what 'support' means. I look at 'support' as being able to be sure it's stable, and problems are minimal (or non-existant). This is basically what defines (as well as a few other factors) a tier-one platform on FreeBSD.

Check the release dates for the Athlon 64 and Opteron.

Well, some of us ARE working on that very thing. (In all honesty, there's another distribution that I know of that we're working with, but since it's not public knowlege, I can't point people that way- YET.)

Shortly, with the efforts of companies like ourselves, SCI and hopefully others, that is going to change.

(Oh, and on a different note, HOW did you get the ability to have links without the [foo.com] after them? :-)

from the horse's mouth:

Q: What markets are you targeting for the AMD Athlon 64 processor?

A: PC users are looking for industry-leading performance on today?s 32-bit applications with capability for the 64-bit future. A growing number of end users need to access large amounts data and/or physical memory, including gamers and people doing advanced content creation work with photos, video and other digital media. AMD is targeting these tech-savvy consumers and businesses who want outstanding 32-bit performance for today with 64-bit capabilities that will enable them to run tomorrow?s advanced operating systems and applications.

Quote (comparing Athlon 64 FX vs. Opteron, FX is a Athlon64 tweaked for gamers):

"Additionally, the processors differ in that the AMD Opteron processor features three HyperTransport links, compared to the one HyperTransport link of the AMD Athlon FX processor. They are also tested to different electrical specifications."

Quote (comparing Athlon 64 FX vs. Opteron, FX is a Athlon64 tweaked for gamers):

"Additionally, the processors differ in that the AMD Opteron processor features three HyperTransport links, compared to the one HyperTransport link of the AMD Athlon FX processor. They are also tested to different electrical specifications."