The New Z170 Motherboards

[Scali]

People don't "have" to do anything. Apple could afford to switch architectures because of their limited market share and the types of people who use Macs. (Not big business, and not people who are concerned with cost.) That would never work with PCs. Intel can only sell what people buy, and in the business world, evolution beats revolution every time.

Sure it would work.
It's very simple.
Developers buy IA64 systems, and recompile their applications to IA64. This is quite trivial (and if you don't, you lose business to competitors that do).
Now, consumers can:
1) Stick with their x86 hardware and software forever, and no longer get any updates or performance improvements
2) Upgrade to IA64

The choice is very simple.
Big business is a bad argument in the first place, because they are/were the ones that used non-x86 machines in the first place.

"Evolution beats revolution" is a fallacy: the argument here is that evolution is stopped dead. So revolution is your only option.

IA64/EPIC is much more difficult to optimize for than x86; this is a well-established fact.

No it isn't, and I already explained why. You provide no arguments, so this is just a fallacy: appeal to the people.

Intel themselves struggled to write an optimizing compiler.

Perhaps you missed Tor's post, where he said that the IA64 compiler generated considerably faster code than the x86 compiler did.

In the server market, Itanium 2 competed directly with Opteron.

No, back in those days, x86 was still kids' stuff. As already said, IA64 competed against high-end CPU architectures. x86 had not closed the performance gap with (post-)RISC CPUs yet, in those days.

If you wanted a 64-bit server, those were your two choices.

No, as said, there were many more choices.

And of course when running 32-bit legacy code

But why would you?
x86 servers were mainly popular as cheap *nix-based web/db servers (BSD/Linux, Apache, MySQL etc). Most of that stuff is open source anyway, and was already compiled to 64-bit (both x86 and IA64).

The choice for businesses was a no-brainer

Exactly, the people who chose AMD64, had no brains. They made the stupid decision to lock themselves into a horrible 64-bit architecture with virtually no performance advantages over 32-bit, while there were various superior options on the market. All because of "Look! Our shitty outdated legacy stuff benchmarks higher on these CPUs, and we totally forgot to see what new 64-bit software would perform like! Legacy software is all we'll ever need!"

I also don't buy the multicore argument. Itaniums ran hot as hell, and heat is a bigger problem than die size.

It's all about transistor-count. Itanium cores had a lower transistor-count than x86 cores, simple fact. So, with everything else equal, the Itanium cores would not run as hot as the x86 ones.

 

[Scali]

How about the history of PC computing since the IBM 5150? Backwards compatibility has always been king, and progress has always been through evolution.

How about this is nonsense?
There have been various computers introduced after the 5150, which managed to cut out a nice bit of marketshare for themselves. Apple's Mac being the most obvious example.
Commodore 64 is another machine that sold way more than any PC.

The PC 'won' in the end because the performance kept improving, and it had a large library of software available.
No reason why IA64 couldn't win on the same merits.

Not being complete idiots, they realized what the market wanted and delivered.

It made good business sense, because the market is stupid, as already pointed out. They ask for the poor short-term solution, and AMD was willing to provide it.
We were making a technical argument however, not a business argument.
For technology it was a huge set-back.

 

[Tor]

How about the history of PC computing since the IBM 5150? Backwards compatibility has always been king, and progress has always been through evolution.

AMD didn't "block" anything. Intel was free to do whatever they wanted. Not being complete idiots, they realized what the market wanted and delivered.

I hope you realize how little sense that argument makes.. to spell it out: After AMD came out with AMD64, Intel no longer had any choice in the matter. Evolution ran into a wall.

An apples to oranges comparison. Two cores vs one in the absolute best case scenario for parallelism. And regardless, my comparison was AMD64 to IA64.

Who talks about cores? One IA64 core was faster than one high-end x86 core. And AMD64 isn't faster than x86 except for in the <5% range - 64-bit isn't faster than 32-bit, everything else being the same (x86 vs amd64 isn't exactly the same, unlike MIPS or SPARC or PPC, so amd64 is a tiny bit more efficient than x86). And, as you mentioned floating point elsewhere - yes, we got that ia64 server to test our heavy FFT application, but that's not what I used it for. I tested everything else, which was *not* about floating point, and it performed very well indeed.

As for 64-bit platforms, I ran the Itanium server as a build- and test machine in a farm which included 64-bit MIPS, 64-bit Alpha, and two more. So Opteron and ia64 weren't the only 64-bit platforms.

I'll say it again: The ia64 that Intel could produce with the same-generation fab tech as used for x86, performed much better than the x86.

 

[Plasma]

Sure it would work.
It's very simple.
Developers buy IA64 systems, and recompile their applications to IA64. This is quite trivial (and if you don't, you lose business to competitors that do).
Now, consumers can:
1) Stick with their x86 hardware and software forever, and no longer get any updates or performance improvements
2) Upgrade to IA64

The choice is very simple.
Big business is a bad argument in the first place, because they are/were the ones that used non-x86 machines in the first place.

"Evolution beats revolution" is a fallacy: the argument here is that evolution is stopped dead. So revolution is your only option.

You have no understanding of business, time, and profit.

No it isn't, and I already explained why. You provide no arguments, so this is just a fallacy: appeal to the people.

What part of memory latency is non-deterministic is unclear to you?

Perhaps you missed Tor's post, where he said that the IA64 compiler generated considerably faster code than the x86 compiler did.

I seem to have misinterpreted his post. However, my argument was never that IA64 was slower than x86. It's that for the same money, an Opteron would outperform an Itanium.

Exactly, the people who chose AMD64, had no brains. They made the stupid decision to lock themselves into a horrible 64-bit architecture with virtually no performance advantages over 32-bit, while there were various superior options on the market. All because of "Look! Our shitty outdated legacy stuff benchmarks higher on these CPUs, and we totally forgot to see what new 64-bit software would perform like! Legacy software is all we'll ever need!"

Again...you have no clue how actual businesses operate. "Drop everything and start over because something new and shiny is available" is not something you do.

It's all about transistor-count. Itanium cores had a lower transistor-count than x86 cores, simple fact. So, with everything else equal, the Itanium cores would not run as hot as the x86 ones.

Power consumption and heat dissipation are not directly related to transistor count. So everything else is not equal.

 

[Plasma]

How about this is nonsense?
There have been various computers introduced after the 5150, which managed to cut out a nice bit of marketshare for themselves. Apple's Mac being the most obvious example.
Commodore 64 is another machine that sold way more than any PC.

The PC 'won' in the end because the performance kept improving, and it had a large library of software available.
No reason why IA64 couldn't win on the same merits.

The large library of software was possible due to backwards compatibility. So no, IA64 couldn't win on the same merits, because its x86 performance was miserable.

It made good business sense, because the market is stupid, as already pointed out. They ask for the poor short-term solution, and AMD was willing to provide it.
We were making a technical argument however, not a business argument.
For technology it was a huge set-back.

I disagree. It AMD64 lowered the cost of 64-bit PC computing for the consumer.

 

[Plasma]

I hope you realize how little sense that argument makes.. to spell it out: After AMD came out with AMD64, Intel no longer had any choice in the matter. Evolution ran into a wall.

AMD64 was the evolution. Intel certainly had a choice; they could have kept pushing IA64. Nobody forced them to implement AMD64.

 

[Scali]

The large library of software was possible due to backwards compatibility.

Apparently you didn't get my point that millions of people moved FROM C64 TO PC.
And there were also people who moved from PC to Mac.

I disagree. It AMD64 lowered the cost of 64-bit PC computing for the consumer.

64-bit computing was (and for the most part still is) irrelevant to the consumer.
The cost of 64-bit computing would have come down eventually anyway. Look at ARM for example. There's no immediate need for 64-bit support, yet more and more ARM SoCs for phones/tablets are 64-bit at little or no extra cost.

 

[Scali]

You have no understanding of business, time, and profit.

Great arguments! Oh wait, there aren't any.

What part of memory latency is non-deterministic is unclear to you?

Memory latency is not an Itanium-specific issue. In fact, modern Itanium implementations implement out-of-order execution to mitigate this issue in a very similar way to how x86 does it.
So this is in no way an architectural issue. I can understand that those with a lack of understanding (or some kind of agenda) may think otherwise.

It's that for the same money, an Opteron would outperform an Itanium.

Itanium was never about bargain-basement offers. It is a high-end solution where reliability is more important. In that it still outclasses Xeons and Opterons, which is a reason why it is still preferred over x86, in professional environments.
Having said that, Itanium offered performance at a level that Opterons could not attain. Cost is not always the most important factor.

Power consumption and heat dissipation are not directly related to transistor count. So everything else is not equal.

Well no, but if you want to go there, clockspeed tends to have an effect on it as well, and Itaniums generally run at considerably lower clockspeeds (because as stated, they are far more efficient, so have much higher IPC). So yes, some factors are actually in favour of Itanium.

 

[krebizfan]

IA-64 was most effective for a limited problem set; most software worked poorly on it. Instead of being twice as fast running x86 code and 10 times as fast running IA-64 specific code, Itanium was only slightly faster than the best x86 chips from both Intel and AMD. A challenge to sell a much more expensive CPU that doesn't offer performance benefits. Energy efficiency also played a role as many servers switched to slow power saving CPUs instead of powerhouse CPUs thanks to the California energy crisis.

Alpha was a more traditional 64-bit design and it couldn't keep enough of a performance advantage to make up for the much higher per unit development costs relative to X86. IBM had to pay Global Foundries to take over Power production. CPUs have followed the same consolidation process as the rest of the industry; it just took longer.

 

[Scali]

x86 isn't efficient. The fact that billions of dollars have been poured into it, gave them a position that nobody could compete with anymore. The elephant in the room is that it's Intel who has been pouring all these resources into x86. What they can do with x86, they can do with a more efficient architecture, only better (note that they only allocated about 1/10th of their resources to Itanium, the rest to x86).
Look at AMD to see what the second-largest CPU company can do with x86... and it's downright ugly.

 

[Plasma]

Compared to the PC, C64 was a toy. Millions of people also moved from NES to SNES; it means nothing.

Most companies don't have the source code to all the software they use. There is no "just recompile everything". Switching to Itanium meant a big performance hit on legacy software, or paying lots of money to upgrade to native IA64 versions.

64-bit computing was quite relevant for the consumer when we hit the 4 GB memory barrier.

 

[krebizfan]

A new CPU can cost a billion dollars to implement. Unless one can sell many millions of units, the cost will block adoption. I expect that the future will be extensions to existing instruction sets and gradually only the extensions will be used with a small amount of X86/X64 or ARM legacy instructions to bootstrap into the new instruction set.

 

[Trixter]

I am surprised there is any comparison of IA64 to X86-64 because they're completely different architectures. IA64's innovation was the very large register sets designed for flexibility and parallelism, not that the base word size was 64-bit. Any comparisons trying to equate the two should probably stop.

You know what else should stop? A long flamewar about the history of failed enterprise-grade architectures in a thread where a guy is asking advice about what modern motherboard he should buy.

 

[Scali]

64-bit computing was quite relevant for the consumer when we hit the 4 GB memory barrier.

Which was more than a decade after the whole IA64 vs AMD64-scenario.
As I already said, being 64-bit was not very relevant to AMD64's success. Being able to run legacy 32-bit x86 code fast was (as in: winning benchmarks in consumer-oriented reviews).

 

[Scali]

A new CPU can cost a billion dollars to implement. Unless one can sell many millions of units, the cost will block adoption. I expect that the future will be extensions to existing instruction sets and gradually only the extensions will be used with a small amount of X86/X64 or ARM legacy instructions to bootstrap into the new instruction set.

Either that, or intermediate bytecode. GPU development has been much faster than CPU development over the years. One of the main reasons is that applications don't run code directly on the hardware, but use APIs to compile the code to native format before it is run.
This means that there doesn't have to be any backward-compatibility at the hardware level, only in the drivers. Which means designers can start from a clean slate with every generation.
The same could happen with CPUs (and it has, for the most part, with Java/Android, .NET, HTML5/JS and such), but we're not going to take the 'hump' of "but my legacy code benchmarks 0.005% slower on this new and superior architecture!" if hardware design continues to focus on legacy performance.
The reason why Intel stopped x86 development is exactly to get over that hump with IA64.

 

[Scali]

I am surprised there is any comparison of IA64 to X86-64 because they're completely different architectures. IA64's innovation was the very large register sets designed for flexibility and parallelism, not that the base word size was 64-bit. Any comparisons trying to equate the two should probably stop.

I think only one person is trying to equate the two. The rest seems to understand they are different products, aimed at different markets, and that 64-bit in itself indeed wasn't the key point. It's just that if you design a new CPU architecture in the late 80s or beyond, you factor in 64-bit from the get-go.
The one thing about 64-bit is that eventually x86 would run into the limits of its address space, and IA64 would be the only way forward in an Intel world.
So given the timing, it would have provided Intel with an extra reason to get people to migrate to the new architecture. It will be a long time until we run into the limits of 64-bit addressing, so a new chance will not likely present itself anytime soon.

 

[Scali]

By the way, IA64 wasn't Intel's first attempt to move away from x86. They tried it before in the 32-bit era with a RISC-architecture named i960. They had an earlier failure with an overly complex 32-bit architecture named APX 432.

 

[Plasma]

If there is no comparison between IA64 and AMD64, then you can't make the argument that AMD64 killed IA64. So which is it?

AMD64 came out in 2000, Athlon 64 came out in 2003. Much less than a decade. Windows XP 64-bit was available in 2005. By 2007, I had 16 GB of RAM on my CAD PC.

My point is, if you wanted the ability to run native x86 code on a 64-bit CPU, those were your choices. Yes, IA64 was initially aimed at the enterprise market, but Intel hoped it would eventually make inroads with consumer PCs. AMD64 "won" because it was cheaper at the same performance level and ran legacy code faster. What you consider a "huge setback for technology", I consider a win for consumers. Which is what the last half of this thread has been: how competition benefits consumers.

 

[Scali]

If there is no comparison between IA64 and AMD64, then you can't make the argument that AMD64 killed IA64. So which is it?

Sure you can.
It seems you have problems thinking ahead and seeing the bigger picture.
There is a difference between instruction set architectures and the products built on these architectures.
IA64 was launched as high-end and would trickle down into the mainstream over time.
AMD64 was mainstream, and was trying to claw its way up to the high-end.
Because AMD64 was there, IA64's road to mainstream was blocked.

So, indeed, you cannot compare between the actual products built around these architectures, because they served different markets.
This however has nothing to do with the fact that ISAs can enter new markets over time (just look at x86, which started out as a cheap solution for home/personal computers, and is now in virtually every market, from embedded to HPC).

AMD64 came out in 2000, Athlon 64 came out in 2003. Much less than a decade. Windows XP 64-bit was available in 2005.

XP 64-bit gained virtually no marketshare whatsoever, and even with Vista and Windows 7, the 32-bit versions were still quite popular.
Just because the OS was there is no indication whatsoever that the mainstream user was running into the address space limits.

By 2007, I had 16 GB of RAM on my CAD PC.

Funny. We were talking about mainstream computing. CAD is done on workstations. Guess what? People were doing CAD on 64-bit workstations long before AMD64 arrived. You make it sound like we needed AMD64 for this, which is completely wrong obviously.
I can't shake the feeling that you try to make yourself sound like a professional, but you've actually never used anything but x86, and base everything else on hearsay, and the distorted/rewritten history by the x86/AMD fanboys.
Especially the way you completely ignored what Tor said (who, like me, has hands-on experience with actual non-x86 machines, including high-end 64-bit ones, and who isn't just some user, but actually developed code on them).

My point is, if you wanted the ability to run native x86 code on a 64-bit CPU, those were your choices.

In which case you are completely missing the merits of the IA64 architecture.

What you consider a "huge setback for technology", I consider a win for consumers. Which is what the last half of this thread has been: how competition benefits consumers.

Yes, and you still don't get the point. Consumers would be much better off today if we had taken a different route. We'd have more performance and lower power consumption, because we would no longer have to bake in all the legacy x86 crud into every CPU. Our codebase would also be more portable, rather than hardcoded to x86. So we would once again be free to choose the best hardware for our software, rather than be stuck with x86 by default, no matter how good or bad it is.
So it was a shallow, short-term win, but a loss in the long term.

 

[Plasma]

IA64 was a commercial failure even in high-end markets, long before AMD64. You can't blame that on AMD.

You are missing my point. Because of AMD64 I could do CAD on a "regular" Windows XP box with 16GB of RAM. I no longer needed a "workstation". Maybe your mother doesn't need more than 4GB, but it was a welcome capability for PC enthusiasts and business users.

Nothing is based on hearsay; I lived and worked through the history. You don't want to turn this into a pissing contest.

I really don't know what else to say to you. Most of the free world understands business and economics. You are entitled to your opinion that we'd be better off with monopolies, but I think you're delusional.