64 bit CPU. A true technical review please

[Vee]

Originally posted by: Artanis
And I still believe 4GB RAM is far away from being a standard, even in the today's workstations offerts, not talking about consumer market...

That (physical ram) is not completely relevant.

 

[cquark]

Originally posted by: Artanis
you can succesfully use /3 GB switch to enable 3GB RAM for appl. and only 1GB for the OS. And I still believe 4GB RAM is far away from being a standard, even in the today's workstations offerts, not talking about consumer market...

At Intel, we were buying HP workstations with 16GB of RAM about 5 years ago.

Workstations have been 64-bit for a long time, but whether you need more than the 2-3GB max depends on your application. Designing current microprocessors and manipulating large images are tasks that have needed that capability for a number of years already.

 

[Artanis]

OK, that's happening at Intel, I was talking about the most of us, consumers...Not many people, or A64 owners design new CPU-s

The majority 32 bits apl. which doesn't override Windows settings through their own memory manager, like Maya or Photoshop, can have IMAGE_FILE_LARGE_ADDRESS_AWARE flag activated (using imagecfg.exe utiliy for a 32bit executable - wich is a formality), for using 3GB of phisical (or virtual RAM). BTW, in Linux32 i heard 3GB limit for aplications is activated by default...
Not to even talk about posibility of rewriting 32bit apl. for PAE, for using RAM over 4GB barrier in 32bit OS.
Of course, this is not a elegant or desired solution and can't last forever (but is a solution)...

And Vee, if the quantity of phisical RAM is not relevant (while many of today's workstations are still sold with 1-2 GB RAM by default), why should we even care NOW about the 32 bit memory limitations ?
If next year a super game for instance, will be released, wich will require a 64bit OS, and a minimum 2GB of phisical RAM installed, do you believe it would be a succes? I sincerely doubt it would, considering the RAM prices, and why not, the big latencyes of 1GB modules...

 

[Vee]

Originally posted by: Artanis
The majority 32 bits apl. which doesn't override Windows settings through their own memory manager, like Maya or Photoshop, can have IMAGE_FILE_LARGE_ADDRESS_AWARE flag activated (using imagecfg.exe utiliy for a 32bit executable - wich is a formality), for using 3GB of phisical (or virtual RAM). BTW, in Linux32 i heard 3GB limit for aplications is activated by default...

Well, - bandaid.

Not to even talk about posibility of rewriting 32bit apl. for PAE, for using RAM over 4GB barrier in 32bit OS.
Of course, this is not a elegant or desired solution and can't last forever (but is a solution)...

Yes, I agree. - This is very undesirable. The idea is to port software to 32-bit segmented, poor performance format. A task many magnitudes more difficult, expensive, bugprone, and timeconsuming, than porting to 64-bit. This just to offer the short term option of buying a 32-bit only CPU instead of a 64-bit.
Kinda like: "It's winter, we're out of wood and freezing. Cutting more wood is not necessary. Let's set fire to the forest instead. Should keep us warm a while."

And Vee, if the quantity of phisical RAM is not relevant (while many of today's workstations are still sold with 1-2 GB RAM by default), why should we even care NOW about the 32 bit memory limitations ?
If next year a super game for instance, will be released, wich will require a 64bit OS, and a minimum 2GB of phisical RAM installed, do you believe it would be a succes? I sincerely doubt it would, considering the RAM prices, and why not, the big latencyes of 1GB modules...

Yes, I believe that 64-bit games will be successes next year. And the hardcore gamers that drive the market buzz, and the sales, are already on 64-bit, or will be shortly. But I do believe that there will be 32-bit editions. But they will be crippled in some way. Many gamers will not accept less than full content. This is perfectly evident by the sales of highend videocards.

Yes, the performance of large installements of DDR, is problematic. That's why we're going DDR2 and DDR3 gradually.
You misunderstand me somewhat. I said "not completely relevant".
But my point is this: You can run out of virtual address space on a 1GB machine as well. When you get that out of memory message, you've run into the 32-bit barrier. Installing more physical ram won't help you. More ram may help performance, but I don't think there's terrible much point with more than 2GB ram in a PC workstation, as long as we're still on 32-bit.

That may also be some kind of guide to when we need to be migrated to 64-bit. 2GB standard is somewhere we need to go 64-bit. Let's see...

 

[Artanis]

Originally posted by: Vee
Yes, I believe that 64-bit games will be successes next year. And the hardcore gamers that drive the market buzz, and the sales, are already on 64-bit, or will be shortly. But I do believe that there will be 32-bit editions. But they will be crippled in some way. Many gamers will not accept less than full content. This is perfectly evident by the sales of highend videocards.

Let me say what I believe: If Intel doesn't come up shortly with a viable and a suficient production of 64 bits desktop CPU's, AMD wouldn't have the force to dictate software's 64 bits migration on it's own powers. And that's what's makes me sad, because the Intel's logo "we don't need 64bits CPUs for desktops" is still remnant in my head, on Intel's loyal hardware analisys websites, and even in the head of developers. They won't make any move till the 'sleepy giant' will make some relevant moves...
But as you said, let's wait and see

P.S. please forgive my language, I'm not a native, I'm just trying

 

[imported_Gatak]

I think AMD has managed to get thge 64bit into mainstream quite well considering everything. Many vendors are switching to AMD 64 and sales are very high. Some games already do have extra benefits for the AMD 64 platform which are not available for 32bit systems. Of course this is a marketing trick, but it shows they are determined. I think Intel have a difficult time as they can't seem to produce new generation CPUs just yet. Also their future CPUs will support AMD's x86_64 which is also a vitory for amd and the 64bit platform.

 

[Artanis]

Now, for AMD it seems to be a victory, but for 64bits users maybe a 'clean' 64bits technology (without x86 reminiscence) would have been preferable...

 

[imported_Gatak]

Yes, the old backwards compatible problem still exists. Maybe this is the way we need to go? Intel tried with Itanium which failed. The IBM PowerPC platform also exist at 64bit, so did Alpha (are they still existing?)... None have really been very successful in giving users 64bit. Apple is perhaps closest with their G5 and OSX Tiger?

 

[Artanis]

Nor IBM or Alpha had enough marketshare in CPUs to drive things toward 64bits platform. Itanium is way too expensive and Apple's G5 lacks performance. Today engineers from Alpha work with Intel and from IBM with AMD, as I know...Let's see the results, but as I said, the consumer market doesn't seems to NEED 64bits right now...

 

[cquark]

Originally posted by: Artanis
Nor IBM or Alpha had enough marketshare in CPUs to drive things toward 64bits platform.

While UltraSPARC still dominates the workstation market, it's also a 64-bit processor. PA RISC makes up another chunk, and they're also 64-bit. The workstation market has been mostly 64 bit for a decade, but you're right that it'll take a few more years before the PC market is dominated by 64 bit machines.

OK, that's happening at Intel, I was talking about the most of us, consumers.

You were talking about workstations, which aren't what most consumers have, but which are what engineers at large companies have. I also pointed out that Intel had workstations with 4 times as much memory that you mentioned 5 years ago. A lot more workstations have 4gb today than had 16gb in 1999.

Today engineers from Alpha work with Intel and from IBM with AMD

Actually, AMD hired most of the Alpha design teams from DEC to produce the Athlon. You can definitely see their influence in the design. Intel did get some of DEC's ARM development, which led to the XScale processors.

Not to even talk about posibility of rewriting 32bit apl. for PAE, for using RAM over 4GB barrier in 32bit OS.
Of course, this is not a elegant or desired solution and can't last forever (but is a solution)...

Why do that when AMD is offering 64-bit processors today as a much cleaner solution?

These transitions need to happen before most consumers need the address space. If CPU manufacturers waited until most application sets required 2GB, it would be too late. If the average consumer PC today is within a factor of 2-8 of the most common limit, then it's definitely time for CPU manufacturers to be making 64-bit CPUs for the higher end users.

 

[Artanis]

Why do that when AMD is offering 64-bit processors today as a much cleaner solution?

Because the 64bits software or even OS seems not to move as AMD hardware does

 

[cquark]

Originally posted by: Artanis

Why do that when AMD is offering 64-bit processors today as a much cleaner solution?

Because the 64bits software or even OS seems not to move as AMD hardware does

That's a good point. Changes like the 32-bit to 64-bit migration are always difficult, because no one can write the software without the new hardware and no one will buy the hardware if there's no software. However, AMD64 offers a nice upgrade path from 32-bit to 64-bit systems. While it's not as clean as a pure 64-bit solution like most of the alternatives there, I think x86 compatibility is a necessity for consumer systems. The only alternative I see is doing a dual core design, with a 32-bit x86 CPU and a modern 64-bit CPU as the other core, which is likely too expensive/difficult to deal with to be popular.

 

[Artanis]

I don't know how difficult it is, because I'm not a programmer, but it seems to me is unwanted by the Wintel corp. According to the old Intel's roadmaps, we should all have by now 64bits computing, and XP/2003 server x64 edition would allready have been released...

 

[cquark]

Originally posted by: Artanis
I don't know how difficult it is, because I'm not a programmer, but it seems to me is unwanted by the Wintel corp. According to the old Intel's roadmaps, we should all have by now 64bits computing, and XP/2003 server x64 edition would allready have been released...

Intel expected the 64-bit Itanium to replace x86, but it's not the first time they made that mistake. Remember when Intel's 32-bit i432 was going to replace the x86 processors of its day, but the backwards compatible 386 succeeded instead?

 

[imported_Gatak]

That Dell is "considering" AMD is a big step. Dell has been notoriously a Intel-only vendor. Dell would not say they are considering unless they see that their customers are demanding the AMD platform. http://www.techspot.com/story16286.html

 

[imported_devnull]

Unfortunately the computing business is no different than any other business... And like a business what it is meant to do is to create money.

That being said you should try to visualize what is offered not only from an engineering point of view but from a marketing point as well.

Do we need 64bits? Well, a lot, REALLY lot may be written here..... and most of it has already been spoken.

Yet one may visualize it this way. When AMD introduced the first Athlon family it competed with PentiumIII. It presented comparable performance and it scaled (in terms of MHz) quite comparably with Intel opponent. That gave from a marketing point a -let's say- tide and AMD started to gain market share.

But a great problem emerged : thermal problems of the Athlon processors. In my country there is a saying : it is better to loose your eye than your reputation. AMD is still haunted by the problems of this era.

The first P4 processors were not good enough, giving Intel bad time. Before the core update, Intel relied on the bad reputation of AMD to hit sales (it also tried to burry the PIII processor as soon as possible!). By the time AMD introduced the Athlon XP line with heat dissipation problems solved, Intel managed to beat AMD (again from a marketing point of view) with two key elements : very high clock speeds and hyper-threading. AMD's campain of "efficient computing" could not convince that many people, because high MHz are very easier to sell. Also the whole campain for Hyper-Threading made quite an impact (more than it deserved anyway) and made it very recognizable.

So what AMD had was a very good processor that it could not market. AMD engineers saw that the performance of their opponents was better so the started working on K8 to beat Intel. But it would surely not be enough to build a better processor if it had not something to REALLY make an impact from a marketing standpoint.

That is where 64bit comes... You don't need to think much, just look all the hype it has raised in just this thread at the forum... People who don't understand what 64bit is, think immediately that 64bit is better than 32bit!!! Now, you may realize that AMD marketing department did a VERY VERY clever thing!!!!!!!!

The story does not stop here... AMD had virtually no presence at the server/workstation market. So they needed a strong point to make an impact there. 64bit was the right thing to do. Give people backward compatibility with no penalty + 64bit capabilities + comparably cheap prices + very good scalability for multiprocessor configurations = TOTAL SUCCESS!!!!!

It was also very very clever of them to introduce the Opteron (AMD64 for servers) first. This gave the ignorants an impression of a very serious chip and of a high end performer, while giving the ability to AMD to explore the manifacturing process abilities of their new chips. Server demand is much smaller than mass desktop so it was fairly easier to correspond to the demand and to finetune their production before going to the masses!

One may that Apple has acted quite similarly with the introduction of their 64bit machines...

 

[borealiss]

There are other advantages. If you are using WinXP SP2 and have a x86 64bit system with the noexecute option enabled, then you are taking advantage of the 64 bit features of the cpu. This is because of an extra page table level that exists, and is only possible on systems that have 64 bit technology.

 

[Gamingphreek]

That is not true, the Intel processors without EMT64 enabled can do this in hardware.

As for purely 64bit... we cant that would mean completely ditching the x86 micro-architecture. Everything that we run today would have to be completely rewritten. It would be like trying to run an X86 windows based program on an Itanium 2 or a MAC... you cant do it its not the same architecture.

-Kevin

 

[borealiss]

yeah, you're right. i thought this was a feature exclusive to the EMT64 enabled processors.

 

[flamingspinach]

Wow. This thread just changed my mind about the AMD vs. Intel issue, I think - I was on the Intel side for at least 4 years. Thanks! O_O

My question is: we've heard a lot about AMD's Athlon64 (which is the same as K8?), but what is Intel's plan for a 64-bit architecture? Will the two be compatible? Will there be some sort of a split of the IBM-compatible (commonly referred to as "PC") standard? And if so, would AMD's 64-bit or Intel's 64-bit be the one to go with?

I don't plan on upgrading my computer for a loooong time (once I get it in a couple of months) so I want to be sure I'm not missing anything.

-fs

 

[Varun]

K8 is the Athlon 64, just to clarify that for you. The K7 was the Athlon (and Athlon XP).

Intel is implementing x86-64, but with a couple differences I think. It SHOULD be compatible completely, but I guess that remains to be seen. Intel is releasing the 64 bit extensions on the Xeon first, as they are still saying that the desktop doesn't need 64 bits.

Hopefully the Intel implementation will perform to the standard that AMD has set down for x86-64. I really can not see why they would not have all of the same commands, other than to have the Beta vs VHS war.

 

[flamingspinach]

Ah, so as Intel has the 80x86, AMD has the Kx? Cool... I guess Intel has already stopped using the x86 naming convention though, since the P4 must have been the equivalent of the "80986"...

Anyway, yeah I guess that does make sense. Intel is in a fragile enough position, probably doesn't want to go off on its own trying to lead the market away from the already-established Athlon64, it would be suicidal... cool I think I'll get a K8 then. Thanks!

-fs

 

[Vee]

Or maybe Pentium/PentiumMMX = '586, PentiumPro/PentiumII/PentiumIII/PentiumIIIe = '686, Pentium4Willamette&Northwood = '786, Pentium4Prescott = '886.

I think Intel has some "P" numbers, not "Pentium", that denotes the basic core design. Pentium was P5. Rather than call P6 "Hexium", Intel decided to use the market value of the name "Pentium", so P6 are PentiumPro to PentiumIIIe. Willamette was P7. I don't know if Prescott is the P8, but it's sufficiently different from Willamette/Northwood to be so, I think. The reason for not calling it Pentium5, could be its disappointing performance.

 

[flamingspinach]

O_O Are Pentium Pro through Pentium IIIe all the same architecture!? I didn't know that...

Intel really should have called the Prescott something different - if not Pentium 5, at least "Pentium 4 something"... now consumers will be confused...

-fs

 

[Vee]

Originally posted by: flamingspinach
O_O Are Pentium Pro through Pentium IIIe all the same architecture!? I didn't know that...

Basically, yes. Details like slot or socket, onchip L2 cache or in package L2 cache, differed. PIII added the SSE extension. PIIIe went to smaller but fullspeed cache.

And of course clockspeed increased all the time. And I think that may be the reason for the various changes to the cache. To get the best value from the clockspeed.

Intel really should have called the Prescott something different - if not Pentium 5, at least "Pentium 4 something"... now consumers will be confused...

In a way they are. They are called P4E and P4A (no HT and 533FSB) (and maybe also P4F eventually,- 64-bit support). There is one single Northwood P4 that is also "A" to distinguish it from a Willamette, but it's old and much slower clock, so there's no risk of mixup. Northwoods are P4, P4B and P4C.