AMD focuses efforts on Palomino core

eviljolly writes: "ZDnet's Gamespot reports about the new AMD Palomino core which will be released at 1.5ghz. They also mention something about AMD's first 64bit processor called the ClawHammer which will come out in early 2002 at 2ghz"

Comments

Re:No, it's VERY necessary.

[peter]

Note that execution speed on an IA64 is extremely dependent on the compiler. To get acceptable, let alone good, performance, you need to recompile for each implementation of the architecture, since different implementation could have different instruction issue capabilities. (I think this is true, but I haven't checked recently. It might be the case that the IA64 arch specifies enough about the issue capabilities to make all implementations be the same. If so, that would be a severe limit for future implementations. Either way, compiling for IA64 is not something that you do just once.

This is more serious than running i486 optimized code on a PIII or Athlon, since those CPUs do dynamic out-of-order execution to get good performance when running pretty much any code. (You can get better performance by optimizing specifically for the target CPU, of course). You still get decent performance running i586 programs on an i686, for example. This probably won't be the case with IA64 (and I'm not talking about their hardware IA32 mode, I'm talking about their EPIC design which exposes so much CPU detail that recompiles will become almost necessary between major CPU upgrades. Fortunately with a CPU this fast under the hood, I wouldn't building all the Debian packages I want from source, as long as it was automated :)

The point of this post is that migrating users over to IA64 won't really work, except for people with enough sense to use software they can compile themselves. Unless I'm missing something about IA64 that would allow future IA64 CPUs to get good performance on old IA64 code, commercial software vendors would have to have a plethera of different versions of their software. I guess they could put all the binaries on the same CD, and select which to install at install time, or even use shared libraries so the choice happens at run time.
#define X(x,y) x##y

ack! upgrades...

[MoNsTeR]

I'll need a new motherboard for this, won't I? :(

I swear, "upgradability" is a myth. Anytime a processor worthy of upgrading to comes out, it either has a different interface or voltage, requiring new motherboards for all. I've never been able to put two consecutive chips in the same board.

sigh....

Re:Which AMD ~= which Intel?

[luge]

It's hard to say exactly, since different chips tend to do better at different things. However, if you mean raw math, a prof in my LUG (who happens to be a beowulf big shot) did some benchmarks that more or less show that for large vector math athlon 750s are more or less equivalent to a PIII 933. Obviously, this has little direct bearing on you if you aren't doing particle physics but it would tend to suggest that (at worst) you can assume that Athlons and PIIIs at the same clock speed perform the same. Unfortunately, most of the performance data he put on the web is not up right now because of some legal hassles with AMD but the data is quite interesting. If he is able to repost it (he should be able to soon) I'll try to find an excuse to post it in another thread, or you can email me for some more details.

Palomino

[PD]

You mean there's a PONY in there? Cool!

Re:startrekmail??

[PD]

It forwards to my real e-mail address for free. I only use startrekmail.com on the internet. Sort of the same thing as using the name "King Africa" even though you're a 9 year old girl.

Re:this is stupid

[wik]

Take a look at itanium. Sure, it'll run slowly with 32-bit code, but not businesses simply can't recompile everything for the new processor. In many cases, they probably don't even have the source code to do so (e.g. if they use 3rd party libraries from a party that ended a long time ago). If you are suggesting that companies can do this port, it's non-trivial. If you have ever gotten sloppy in C and casted a pointer as an integer, then your code will eventually bomb on IA64, because ints are still 32-bit, but pointers are 64-bit (at least on Win64). That's just one of many gotchyas. Ouch.

Re:this is stupid

[um... Lucas]

I think it's morally wrong to advocate a solution that leaves users in a position of losing their data, and running inferior programs to boot. StarOffice is no substitute for MS office, except to the most diehard individual, for instance. Yes it has the same checklist of features, but it runs slow as molasses and crashes CONSTANTLY (under both Windows and Linux... i uninstalled it from Windows and kept it under linux to use in a pinch, until finally settling on VMWare.

Until free software can do everything commercial software can, including reading all of the file formats flawlessly, no amount of "freedom" can substitute for the loss of one's investments in time creating their documents and money spent on programs that won't work upon such a switch.

Re:this is stupid

[um... Lucas]

Backwards compatibility is why x86 is the leading architecture. Backwards compatibility is why DOS & Windows are still the leading platforms for developers. Backwards compatibiltiy is how the PowerPC got it's foothold into the market, and is what Transmeta's trying to do to get their processors accepted by the market place (though in both cases, backwards compatibitliy was achieved through the use of software rather than actual silicon).

Who wants to "upgrade" their system only to find their applications and more importantly, their documents inaccessible? Yes, you could use linux and open source products so that you wouldn't have any disruption, but 95% of the world (at least) don't have that luxury.

Which AMD ~= which Intel?

[magic]

I was shopping for a new computer and wanted to get an AMD chip because of the lower cost. I couldn't tell what the relative power of the AMD and Intel chips are, however.

The MHz is only a useful comparison between two of the same processor model (e.g. PIII 500Mhz vs PIII 800MHz). A fast PIII outperforms a slow P4-- and I have no idea where AMD fits in at all.

Does anyone know what the relative ranking of chips is?

-m

Re:Which AMD ~= which Intel?

[JoeGee]

For price/performance right now I feel Thunderbird leads. If you want bleeding edge performance *I* would recommend a 1.33 GHz T-bird system, 256 MB of PC-2100 DDR SDRAM, and a GeForce 2 video card.

The P-4 1.5 GHz may come *close* to the performance of the 1.33 GHz Thunderbird in a few areas, and may exceed its performance in Quake 3, but for those dubious "achievments" you'll pay an extra $500 or more, money that you could use to upgrade your Thunderbird system to a nice 21" monitor, or a flat panel display.

If you are using real world applications, need floating point power for 3-D, and want your machine to be powerful across a wide variety of potential applications then the Thunderbird is the chip for you.

If you want sheer gigahertz bragging rights, do not use any office applications, don't care that Quake 3 is the only game you can play that you can boast to friends has a fantastic frame rate, don't care that the current motherboard will not support the next generation of Pentium 4's, and use Flask DiVX ;-) to compress your DVDs onto a single CD then by all means go out and grab a P-4. :)

Re:this is stupid

[WNight]

The problem with that is that Microsoft and other closed-source companies will just tack on a new file format, make it the default in the next gen, and then people will claim the open-source product is crap because it can't read the new format.

What is really needed is proper Word97 support, or some other older but still usable format. The new formats don't support anything that 99.8% of people will ever use, they're just there to keep other applications from being compatible.

And this isn't a open-source whine. The closed-source companies are trying to proprietize to stop any competitors. WordPerfect support MS formats, so MS invents new ones. WordPerfect developers (back before it was owner by Corel) admitted to such, and they said they knew that Microsoft was doing the same for the same reason. In fact, things like the Halloween document prove it.

Athlon vs PIII/P4 GHZ comparisons

[SpinyNorman]

The PIII and Athlon are roughly MHz comparable when running with SDRAM, except of course that the PIII tops out at 1GHz while the fastest Athlon is currently 1.33GHz.

If you're using DDR RAM, then Athlon beats PIII at any given MHz rating because it's FSB is twice as fast as that of the PIII and DDR takes advantage of that.

Things get more difficult with the Athlon vs P4 comparison, but the benchmarks out so far seem to show that the 1.5 GHz P4 is about as fast as a 1-1.2GHz Athlon. When both are configured for max. speed with PC800 RDRAM for the P4 and PC2100 DDR RAM for the Athlon, then a 1.33GHz Athlon blows the 1.5GHz P4 away.

In theory the P4 with PC800 has far higher memory bandwidth than Athlon even with PC2100 DRR, and this should give it an edge in some benchmarks, but in reality it only shows in synthetic memory benchmarks and Athlon's more balanced architecture proves unbeatable. P4 *without* PC800 RDRAM is severly crippled, and not worth considering.

Re:my stupidity.

[el_chicano]

gee, i sure love trolling. don't you?

jackass.

and for the record, i get linux running fine. i just use a real os for my server needs, thanks.

Nice troll! Putting a troll in a post denying that you are a troll is an inspired bit of trolling. Score: +5.
--
You think being a MIB is all voodoo mind control? You should see the paperwork!

Re:Correction

[Kingpin]

Perhaps you should as well :)

Re:this is stupid

[roca]

> The fast clock is the real selling feature of this
> chip.

Read my post again, from the top. That's pretty much my point.

Note, BTW, that people have spotted Hammer #defines in Microsoft header files. Microsoft are obviously considering support for the chip. They certainly need to have that option, because they don't want a wave of people turning to Linux/*BSD to run cheap 64-bit servers on AMD's hardware.

Re:this is stupid

[roca]

You are correct.

The number one goal is to run 32-bit x86 applications fast. On top of that AMD adds a 64-bit mode that cleans up the instruction set somewhat, adds another 8 general purpose registers and of course 64-bit instructions and addressing, all designed to be fairly easy to target since it's just a variant of the x86 instruction set. The 64-bit mode is there for high-end servers or any OS that wants to use it --- the extra registers and IP-relative addressing should make it faster than regular x86 mode if you can recompile your code. But since the 64-bit and 32-bit code runs on the same core, it doesn't matter too much if people don't use the 64-bit stuff initially.

If AMD ships the Hammer processors according to schedule then Intel is going to be in a very bad position. IA64 performance simply *sucks* and it always will; static scheduling just does not work, and the smartest compilers in the world can't get around that. Intel gambled and lost on that one. Consequently IA64 will keep slipping and probably never ship in volume. There's just no market for a very expensive, very slow, incompatible CPU.

However, because of the investement and credibility, Intel can't just abandon the IA64. It'll be dragging them down for years. Meanwhile the best they can do is to keep revving the P4, which is already slower than the Athlon, and marketing it as hard as they can. That'll keep them going for a while but customers who want actual performance or a 64-bit architecture, not to mention value for money, will increasingly go to AMD.

The only bright spot for Intel is the rumoured SMT capabilities of the future P4 rev, Foster. That could give them a boost; we'll have to wait and see.

Sun and Intel

[Big Jojo]

Was Intel targetting Sun when they released the first 32-bit PC space processors? No. Incorrect. Sun was running on Moto 68K at the time. The i386 series was very clearly an attempt to catch up to Motorola.

Curiously enough, the Sun386i was Sun's first non-Motorola platform. Yep, it ran SunOS 4.0 more usably than the Motorola boxes. Of course Sun killed that product line since it competed too effectively with SPARC boxes like the Sun4/110 ... and they had just decided to get into the CPU business. Know anyone who managed to get a pre-release Sun486i? Collector's item. Even though it was Intel CPUs that got Sun into some of its current major (non-Scientific) markets ... customers bought the Intel name more than the Sun name at that time.

Re:Disagreements with your last paragraph.

[0xA]

I doubt this. Intel is not aiming for Sun's market. Anybody who buys Sun hardware does so becuase it's insanely reliable, very scalable, and usually better suited to specific tasks (serving, crunching, etc). x86 simply does not match the benefits of using Sun hardware. Moving the PC space into a 64-bit architecture is the next fundamental step forward in faster home/business computing. Support? Was Intel targetting Sun when they released the first 32-bit PC space processors? No. The market wanted faster computers that could handle more memory as demand from applications increased. We're simply doing this again.

This market is exactly what the Itanium is aimed at. If it wasn't do you really think that HP would be working with the Itanium rather than the PA/RISC? HP's PA/RISC stuff is in exactly the same market as Sun's SPARC offerings.

Biggest issue for AMD

[dgb2n]

The most critical issue for AMD is not CPU speed but power consumption. In addition to the increased speed, that is the real promise of the Palamino core.

The newly released Thunderbird chips at 1.3 GHz are great chips but I wonder if AMD can really increase their portion of the server market if they don't introduce cooler chips that work well in slim rack mounted configurations. Until then I will continue to use my Athlon 500 that is running at 700 but sounds like a jet engine with all the cooling ;-)

ClawHammer?

[n3rd]

Sounds like some subliminal messaging from AMD as to what to use on your Intel processors.

AMD's 64-bit architecture

[Animats]

AMD has disclosed what the program-level architecture looks like. It's an reasonable extension of the x86 architecture to 64 bits. The chips will run in 32-bit mode, and support the ability to run 32-bit programs under a 64-bit OS. Some of the less-used features of the x86 are gone in 64-bit mode; most notably, segmented memory management. There are few surprises in this chip. gcc has been ported; it was an easy job.

This is the machine Intel probably should have made, rather than the Inanium.

Re:AMD's 64-bit architecture

[Animats]

The basic problem with the Inanium is that it requires near-omniscience from the compiler to make VLIW-type explicit parallelism work. Dozens of architectures requiring explicit parallelism in the code have been developed over the years, generally as "supercomputers". None ever got any significant market share.

It will be amusing if, say, nVidia puts an IA-32 core on their graphics chip and starts taking over the low end with a one-chip computer.

Re:AMD's 64-bit architecture

[jeti]

You mean, "Intel should have made some half-assed extension to a painfully obsolete ISA rather than have the balls to invent a radically new and innovative ISA that might actually be able to exploit the rediculously large number of transistors that will be on future chips for something other than cache."

Actually, when I first heard about IA-64 (the new ISA for Itanium) I thought it was about time.

...It was probably around four years ago, and Itanium is still not finished...

Also, nobody has any experience of using VLIW (or EPIC, whatever) style ISA on a general purpose processor before, so that will probably cause problems in the beginning. (Ok, Transmeta CPUs are VLIW, but you won't run them natively, and so far their performance haven't been stellar either.)

So just fixing up some of the worst problems of IA-32 ISA might actually be a good idea (another way would have been to design another 64bit ISA and just make better work than Intel has been doing). AA-64 ISA gives us twice the number of general purpose registers (which are twice as big as the old ones) and gives us register based FPU insted that horrible stack based monster we have suffered two decades. (Well, it was probably a good idean when FPU was it's own CPU but Intel should have at least given us a register based alternative to access it when it designed Pentium...)

The bottom line: IA-64 looks good in paper, but after the difficulties in implementation it doesn't look so good anymore. The AA-64 doesn't look that good in paper but doesn't seem to suffer same kinds of problems as IA-64 chips. However, neither is yet officially released (and AA-64 won't for a long time) so it will take some time untill we see which is actually better.

Re:No, it's VERY necessary.

[rgmoore]

Intel's strategy for their upcoming CPU is to not make it backwards compatable. This is a *serious* flaw in their plans and I don't think we're going to see them succeed with this venture. There won't be apps for it! Users would bitch and complain and only a few vendors will actually properly update their apps.

Actually, Intel's plan makes perfect sense. They're trying to market their chip first to the kinds of users who are willing to rewrite/recompile their software to take advantage of the new instruction set. Those people are the ones running Open Source code like Linux and Apache or homebuilt applications for special purpose applications. Note, for instance, that RedHat is already working on an Itanium version of their distro. That market will then give vendors like Microsoft a reason to develop versions of their software for the Itanium. You can be that if Linux + Apache + Perl on Itanium turns out to be a successful web serving environment that MS is going to want to produce a Windows + IIS + ASP competitor. Once the hard part of porting Windows is done, MS is going to want to move their other apps to the new architecture, too, and the whole market will move over. Intel is looking at it taking 5 years or so to move completely from x86 to IA64, but they've actually figured out a way of breaking backward compatibility and not dying for it. Of course they're also keeping an alternative around by keeping their x86 development, so if things blow up on them they'll still have chips to sell.

Skip the middle man...here's the map

[AdmiralNanook]

http://www.amd.com/news/virtualpress/roadmap.html

Re:No, it's VERY necessary.

[n0ano]

Intel's strategy for their upcoming CPU is to not make it backwards compatable. This is a *serious* flaw in their plans and I don't think we're going to see them succeed with this venture.

If I have to respond to this kind of comment one more time I will SCREAM. Intel is, repeat is, retaining backwards compatibility with IA32 application code on their IA64 processor. Read the literature that is out there right now before you make this completely erroneous assertion.

I have an IA64 machine in my office running Linux right now. I run Quake 3 Arena on it. I guarantee there is not an IA64 version of Q3A, I'm running IA32 Q3A binaries, I run IA32 StarOffice binaries, I run IA32 Oracle binaries. I don't know of any IA32 Linux application that doesn't run. If fact, if you know of one, please let me know, one of my jobs in life is to make sure that IA32 applications run on IA64 Linux.

--
Don Dugger
VA Linux Systems

Correction

[jrcamp]

The article states:

Further down the road for AMD's desktop plans come Thoroughbred and ClawHammer. Both are on the same early 2002 schedule AMD announced last fall.

CmdrTaco should do a "make sense" check before he posts stories. It's well past early 2001.

Re:No, it's VERY necessary.

[MrBogus]

The guy is snotty, but he has a proper point in his reply. AMD releasing an 64-bit CPU with NO commercial operating system compatibility it totally fucking braindead. And as for Open Source OSes, if that's all you got, wtf hang onto the i386 ISA?

The best AMD can hope with their 64-bit consumer CPU is some optimized Windows video drivers to improve people's Quake scores. Not quite the same market as Sparc or Itaninum, so it's pointless to compare AMD-64 to any real server/workstation chip.

Just like how the i386 brought 10 years of "extenders" instead of real 32-bit OSes, I bet a good number of these 64-bit chips WILL be nicely chunking 80-fucking-86 real mode code in some consumer's Windows ME machine. Yes - this sort of backward compatibility rules, if you like the idea that 20 year old asm code is necessary to run your computer. All Fanboys rally around in support!

Not to mention that your post is incorrect at many points: IA64 does have (slow) IA32 compatibility, and besides it's primarily aimed for a market which is currently buying Sun Sparcs and the like and doesn't give a shit about IA32 compatibility. Furthermore, there's nothing about NT's DOS emulation which makes the OS unstable or slow - it's all in userspace.

Re:Disagreements with your last paragraph.

[MrBogus]

To the end user in the market, this is the same as not working

The "end user" is some spare tire-wearing system administrator. The only IA64 applicaiton Microsoft plans to ship on launch is SQL Server. Somehow I think that some 150K DBA can figure out that the IA32 compatibility is for legacy object code, or maybe some little utilities (like WinZip, etc).

I doubt this. Intel is not aiming for Sun's market.

This incorrect statement just shows that you haven't been paying any attention to Intel's IA64 marketing.

x86 simply does not match the benefits of using Sun hardware.

And that's why IA64 is not based on x86, DUH.

Was Intel targetting Sun when they released the first 32-bit PC space processors? No.

Incorrect. Sun was running on Moto 68K at the time. The i386 series was very clearly an attempt to catch up to Motorola. Unfortunatly for Intel, IBM and Microsoft fucked up the OS support.

NT's old compatibility code is a serious cause of slow-downs an instabilities

What old compatibility code? Neither WOW or NTVDM even run unless you launch a 16-bit app. Try reading up on NT sometime. My guess is that your super reliable source doesn't even know the difference between Windows NT and Windows 9x/ME, and neither do you.

BTW, I'm actually hoping that I'm getting trolled here and that you are not as blockhead stupid as you seem to be.

Re:ARG! wait too long!

[atrowe]

A LOT of people still use their 32 bit software. I'd sure hate to get all new games and apps just so I could use my new $2000 computer.

My discrimination function

[2nd Post!]

I dunno if the other replies are of this vein, but in y experience there's two pretty reliable ways to do this kind of comparison shopping; either getting the fastest MHz at the price you are willing to pay (say $90, you can get a 500MHz Celeron, a 650MHz Duron, or a 550 MHz Athlon, I'd get the Duron), or you can get the fastest processor at the highest price; 1.12GHz Intel, 1.25GHz AMD, I'd get the AMD Athlon...

It makes sense, because the market uses $$$ as the optimal means of conveying the value of performance!

Geek dating!

I somehow doubt that...

[2nd Post!]

I am so very wrong.

The Intel P4s are something like $600 to the Athlon's $300. You acknowledge that yourself. But I think you misinterpret my words:
"get the fastest processor at the highest price"

I never meant for a buyer to spend the most money possible; that's silly and stupid. Rather take the most the can afford, say $170, and find all the CPUs at that price, and take the highest clocked CPU. This is perfectly valid for x86 CPUs because so much is weighted by clockspeeds, but across the various flavors of AMDs and Intels.

Using SharkyExtreme, $170 gives us an Athlon 950MHz, An Athlon T-bird 1.1GHz, and a P3 850MHz. Guess what? The T-bird wins.

I can see why you can interpret my statement to mean "Buy the most expensive CPU on the market", but with that kind of reasoning, one would buy, san, an SGI P3 Xeon or something!

Geek dating!

Disagreements with your last paragraph.

[Lethyos]

Not to mention that your post is incorrect at many points: IA64 does have (slow) IA32 compatibility,

For all intensive purposes though, can you really call this compatibility? Doesn't compatibility also encompass and denote a certain level of acceptable performance? I've read that 32-bit code running on IA64 is unusably slow. To the end user in the market, this is the same as not working ("something is obviously broken, my apps are slower than they were on my old machine").

and besides it's primarily aimed for a market which is currently buying Sun Sparcs and the like and doesn't give a shit about IA32 compatibility.

I doubt this. Intel is not aiming for Sun's market. Anybody who buys Sun hardware does so becuase it's insanely reliable, very scalable, and usually better suited to specific tasks (serving, crunching, etc). x86 simply does not match the benefits of using Sun hardware. Moving the PC space into a 64-bit architecture is the next fundamental step forward in faster home/business computing. Support? Was Intel targetting Sun when they released the first 32-bit PC space processors? No. The market wanted faster computers that could handle more memory as demand from applications increased. We're simply doing this again.

Furthermore, there's nothing about NT's DOS emulation which makes the OS unstable or slow - it's all in userspace.

I disagree. My old college buddy, who now works for MS, has had a lot of exposure to members of the Windows team. The general consensus is that NT's old compatibility code is a serious cause of slow-downs an instabilities - especially the DOS stuff, which they would LOVE to ditch entirely (because it would make NT more lucrative to those well versed with *nix). I can't really go into detail because OS's in general are not really my forte (but hey, lots of us regurgitate information passed to us by sources we consider to be trustworthy).