The Old Joel on Software Forum: Part 3 (of 5)

Athlon 64

Winner, loser, or also ran?

Dennis Forbes
Thursday, December 18, 2003

IMHO
Off to an early lead. Now Intel realized Itanic was a folly. I think Intel will catch up easily.

Mike
Thursday, December 18, 2003

All comments point to winner, even just running 32-bit apps. Just need to get a version of XP that supports it in 64-bit mode... nice thing about .NET is that the apps will transparently support 64-bit CPUs in 64-bit mode. You ask me, that's one of the major reasons they're pushing it.

Brad Wilson (dotnetguy.techieswithcats.com)
Thursday, December 18, 2003

I ask this after seeing a local computing warehouse selling the Athlon 64 3000 for $300 CDN, and a "has every feature imaginable" Asus K8V is only $190 CDN. At those prices I'm tempted to build a system around one just for playing with the technology. It does definitely look to be a great leap.

Dennis Forbes
Thursday, December 18, 2003

It's the Athlon's future. The Athlon XP is at end of life. Faster speeds will only be produced with the 64-bit chip.

Be aware that there are a couple flavors of 64-bit AMD chip. The Athlon FX (consumer) and Opteron (server) use the Clawhammer process and a PGA 940 socket. The very first Athlon 64s (3000+ and 3200+) use an older manufacturing process, and are placed into a Socket 754 socket.

So, if you drop money into an Athlon 64 board today, presumably with a Socket 754 and a 3000+ or 3200+, it's a dead end as near as I can tell (but a hell of a lot cheaper than an Athlon FX solution).

I have no idea why they decided to do that, but it's kind of obnoxious.

(And, of course, the Athlon FX is more expensive than an Opteron right now... more expensive than two Opterons, actually...)

Brad Wilson (dotnetguy.techieswithcats.com)
Thursday, December 18, 2003

Hrmm...you definitely have me thinking, and I would like more info on this. Mind you I generally upgrade motherboard and CPU at the same time, or relegate the whole PC off to some other task, and I haven't actually done a chip swap on a board for years (because there's always some new board technology that makes a new board a good idea....whether it's a faster variant of AGP, SATA, EIDE super ultra dma, faster RAM, etc).

However what I have seen thus far is that the 64 FX chips are actually the end-of-line chips, as the high end chips are switching to Socket 939 sometime in 2004. For the base 64 versions supposedly the socket is staying the same. The difference, of course, is dual memory channels on the former, and a single memory channel on the latter.

If you know any good sites that detail these, please point me in their direction as that whole field definitely seems hazy right now.

Dennis Forbes
Thursday, December 18, 2003

Well they certainly seem to be selling reasonably well at the moment ( http://www.theinquirer.net/?article=13249 ).

A cynic writes
Friday, December 19, 2003

I have used AMD Athlon XP, Duron and K6 and they have been very stable processors. Rock-solid, absolutely zero problems.

Also, AMD has a policy of having lower prices than the equivalent Intel processor.

Because of the price difference, for the same amount of money, I can build an Intel based computer, or an AMD based computer which is significantly more powerful.

At the same power (if we compare equivalent Intel and AMD CPUs), the AMD is often 30% less expensive.

So - when I will built my next computer, an AMD 64-bit processor will probably be the choice.

AMD makes very solid products and doesn't overprice them like Intel does.

MX
Friday, December 19, 2003

I love AMD processors.

On number 6 at the moment. No complaints.

#7 too will be an AMD

Tapiwa
Friday, December 19, 2003

AMD's 64-bit architecture typifies the "Worse is Better" philosophy. I predict Intel's painfully elegant IA-64 will be steamrolled by AMD's extend-the-registers-again approach.

There is just too much investment in x86 development tools, and too large an installed base of x86 software to ignore. Apple has gotten away with major architecture shifts (680x0->PPC->OSX) but very slowly and with a very captive market. You can't do that in the Wintel world.

Despite the futurism of the IA-64 architecture, I haven't seen any benchmarks that convince me it is fast enough to justify the shift from x86 (and certainly not at prices of $3,000+ for the CPU only!).

I'm not saying Intel is out of the running. They'll come out with an extended x86 eventually. Itanium will be remembered in computer history as one of the mosy expensive catastrophes ever.

Dan Maas
Friday, December 19, 2003

AMD is out way ahead on this one. The only question is if there actually is a race or not.

Stephen Jones
Friday, December 19, 2003

I agree, the itanium is up there with the Edsel.

In related matters, how does Transmeta end up with a chip called the Efficeon. What in the name of Zeus whatchamacallit were they thinking.

Tapiwa
Friday, December 19, 2003

"IMHO... Off to an early lead. Now Intel realized Itanic was a folly. I think Intel will catch up easily. "
----------------------------------

Could Intel design and release a chip that outperforms the Athlon64/Opteron? Sure. They could even brute-force it and throw buttloads of transistors at it, wind up with a huge die, and sell them at a loss just to crush AMD. It's rumored they already have an x86-64 chip ready to go, as a contingency plan.

I have series doubts that they'll release a chip like that anytime soon. They've sunk hundreds of millions if not billions into Itanium. With a ton of money invested, and pride/credibility at stake after pushing it so heavily, how quickly will they be willing to abandon that? If they're willing at all?

It's not a technical issue, it's a management/business decision issue which would basically amount to them admitting that Itanium was a huge cock-up.

John Rose
Friday, December 19, 2003

"AMD's 64-bit architecture typifies the "Worse is Better" philosophy. I predict Intel's painfully elegant IA-64 will be steamrolled by AMD's extend-the-registers-again approach."
--------------------------
Well, nobody really likes the x86 instruction set, that's for sure. But on the other hand, like you noted, the tools for dealing with it are extremely mature. In a very real sense the "problem" of x86 has been "solved". When you look at it that way, 64-bit extensions to a well-understood instruction with tons of stable tools built around it don't look so bad.

On the other hand, the EPIC architecture of Itanium, which apparently very advanced, is supposedly a real beast to deal with as well. It's a beast for precisely the opposite reasons as x86, being *advanced* instead of *antiquated*, but it's a beast nonetheless and doesn't have the benefit of decades of understanding at the hands of the IT world.

Now, I'm only going by what I've read here as far as the Itanium is concerned... I've no direct experience whatsoever with it. I would love to hear from people who have worked with it...

John Rose
Friday, December 19, 2003

"AMD makes very solid products and doesn't overprice them like Intel does"

A more accurate sentence may be "Intel doesn't underprice like AMD does" considering that AMD is losing money.

That said AMD makes some great products and they have positioned the Athlon 64 well. It will gain market share due to its superior 32-bit performance even while the 64 bit market is still small

Mike McNertney
Friday, December 19, 2003

Dan Maas: "I predict Intel's painfully elegant IA-64 will be steamrolled by AMD"

This is a common misconception!

Just because Intel finally moved from the legacy x86 instruction set doesn't mean the new one is "elegant". Have you tried to program it? I can imagine it being a product of endless meetings, the fusion of CPU designers from different companies etc., as it the "new" architecture is just as old and legacy-ridden as the old one.

I haven't programmed the chip myself, but there are some opinions about it that I've heard from everyone who has done some bare-steel hacking on it. The interrupt system is a great example. It is one of the ugliest things from the legacy x86 architecture. And they got rid of it now, but only to replace it with something just as hairy and legacy-ridden although I can't remember the details.

Instruction sets are irrelevant to everyone except OS kernel programmers. The optimal characteristics vary over time depending on programming style and manufacturing technology. Right now we're almost around full circle to full CISC style instructions as they use the cache much better. Linus Torvals have posted some interesting insights on this from time to time on the lkml.

Jonas B.
Saturday, December 20, 2003

I certainly agree - look at Itanium binaries, they are 2-3x the size of equivalent x86 versions. But RAM and cache cost the same for everyone.

It will be very interesting to see how much further Intel and its partners take IA-64 before they wake up to the fact that it's DOA. Presumably many executives at those companies have a lot to lose when it finally comes crashing down :). (already you see Intel positioning Itanium as a product for high-end servers only, not the mass-market that AMD is targeting)

The one thing that has always marred AMD's CPUs is the sub-standard quality of the chipsets and motherboards needed to run them. When you buy an Intel motherboard, you're pretty much guaranteed to get a rock-solid product. But most AMD-compatible motherboards are made in cost-cutting fashion by fly-by-night companies that care only about the home PC/gamer market, with its associated lower level of quality control.

(this is the reason all of my recent purchases have been Intel systems, even though I think AMD CPUs are better - I don't want mysterious crashes, I want it to work right out of the box...)

Dan Maas
Saturday, December 20, 2003

"It will be very interesting to see how much further Intel and its partners take IA-64 before they wake up to the fact that it's DOA."

The Alpha chip was arguably the best shot Windows ever had at leaving the x86 architecture. Not only was it a superb chip in its own right, with growth opportunities, but it was well supported by the x86-to-Alpha just in time cross compiler that even made it usable to run x86 binaries on. Give or take, it was available for more than half a decade (I think perhaps 7 years).

It was a dismal failure.

Intel may be willing to stick with it 10 years, but I doubt they'll stick with it longer than that. And, like the Alpha, the IA64 will see its place in a few installs, but never be overwhelmingly popular; certainly, it has almost not shot of becoming the mainstream desktop CPU. Ever.

AMD did the right thing. They said: "We're moving to 64-bits, but you can still run 32-bit programs and 32-bit OSes if you want. On the 32-bit stuff, you'll get performance that's equal to, or better, than our old 32-bit line CPUs. And when you need that terrabyte of RAM, we'll be here for you."

Very wise, I think.

Brad Wilson (dotnetguy.techieswithcats.com)
Sunday, December 21, 2003

I agree Alpha was probably a much nicer architecture than IA64 ever will be. But it's dead now, and Sparc64 is heading the same way. We're left with PPC, x86-64 and IA64. Intel needs an marketplace miracle if the latter one is the one to prevail.

Jonas B.
Monday, December 22, 2003

This is an old thread now, but Dan, I think the new opteron chips have memory controllers on board. I think the thirdparty chip set problem will be much of problem in the future.

christopher baus (www.baus.net)
Wednesday, December 24, 2003