Gilbo

Intel's Caved.

Recommended Posts

Not much else to say except maybe wow. A demo of the technology is supposed to be given at IDF, according to CNET. I think this is above Inquirer rumour status now.

I can't believe that it will be compatible with AMD's implementation of 64-bit extensions despite the fact that the article seems to suggest that. I would be astounded if that were the case. It's about time Intel began to retreat on the EPIC front. It should be clear by now that processors need to be designed to run software, not the other way around.

Overall, I have to admit I never thought Intel would go this route. I guess AMD's year may not be as easy as I, at least, expected it to be.

Do well.

Jonathan Guilbault.

Share this post


Link to post
Share on other sites

Last I heard M$ kicked Intel in the balls saying, no we won't create ANOTHER Windoze version for a proprietary version of an Intel CPU. It's either AMD64 instructions or nothing!

However I am tired right now and I could have dreamed the entire thing up.

Share this post


Link to post
Share on other sites
I can't believe that it will be compatible with AMD's implementation of 64-bit extensions despite the fact that the article seems to suggest that.  I would be astounded if that were the case. 

Maybe AMD swapped x86-64 for SSE2/SSE3 or something like that.

It would be great for users (but highly unlikely) if both chips used the same socket as well...

Share this post


Link to post
Share on other sites

Intel and AMD have had a cross licence for years (on instruction sets) this is why the Athlon and Pentium chips are still fairly compatable after all these years... Intel wouldnt have to grovel for any licence from AMD...

Intel will apparently be showing off a 64 bit chip early in Feb.... So perhaps Yamhill is alive and well.

Share this post


Link to post
Share on other sites

I knew that AMD had a pretty liberal agreement that they've been able to use to hold Intel's feet the the fire for compatibility, but does it work both ways?

Share this post


Link to post
Share on other sites

i don't believe the big deal is 64bit itself. it's a good gimmick for making sales, but as shown with other 64bit platforms like sparc64 the gains are usually not there for typical users - infact there are drawbacks. the athlon64 has powerhouse speeds, but i do not believe this is not due to its 64bit nature.

i think what this means to amd, to intel, and to us is that amd gets to fight on a level playing field. intel was going to shut them out with epic/itanium and lock in the corporate customer base on that level. they failed, so now they have to go heads up until they get another opportunity to separate. it should be a while.

Share this post


Link to post
Share on other sites

Microsoft already ported Windows and SQL to Itanium and AMD64, so they may be reluctant to do a port for a third 64-bit architecture, although I doubt this is the reason.

Opteron has twice the registers and other optimizations which cause most of the increase in performance, although 64-bit addressing dramatically improves performance for large DB or memory hungry apps.

I, too am surprised that it will be compatible with AMD's extensions, but customers would go nuts if it wasn't. All that means is the same number of registers and support for the extra instructions, etc.

Share this post


Link to post
Share on other sites

Unbelievable news. AMD has been blowing in the wind for close to a year with the lack of a 64-bit version of Windows to support their new flagship CPU. Now, to coincide with the launch of WinXP 64, Intel is going to release their own x86-64 CPU (presumably 100% compatible with AMD's offering).

On top of that, the Intel CPU will probably feature:-

1. Hyperthreading

2. SSE3 aka the Prescott extensions

3. double FPU performance (compared to Opteron)

4. large L3 cache

5. support for unbuffered, non-ECC memory

6. support for true 64-bit virtual mem. addressing (not 48-bit virtual) -- if only for purely a marketing reason.

The only thing missing would be the integrated memory controller.

The above could all be pure fantasy and speculation, just like all the talk of a future dual-core Opteron...

Share this post


Link to post
Share on other sites

The real master stroke will be that it will probably be "compatible" but still different. In other words I imagine that they will probably add a bunch of extentions so that over time, if they can get some developers to incorperate them into applications, then they can make the Athlon64 be seen as the "somewhat incompatible" 64 bit cpu.

Share this post


Link to post
Share on other sites

it makes sense for both players to have compatible CPUs since that will increase the market. With too many consumer 64-bit processors around, people will be reluctant to adopt it, which will result in lower returns for both companies.

Share this post


Link to post
Share on other sites
The sad part is that Intel probably squeezed MS to prevent them from from releasing the 64-bit port of Windows for AMD's 64-bit implementation.

No it that is isn't finished. Intel wasn't involved in that discussion.

Share this post


Link to post
Share on other sites
i don't believe the big deal is 64bit itself.  it's a good gimmick for making sales, but as shown with other 64bit platforms like sparc64 the gains are usually not there for typical users - infact there are drawbacks.  the athlon64 has powerhouse speeds, but i do not believe this is not due to its 64bit nature.

i think what this means to amd, to intel, and to us is that amd gets to fight on a level playing field.  intel was going to shut them out with epic/itanium and lock in the corporate customer base on that level.  they failed, so now they have to go heads up until they get another opportunity to separate.  it should be a while.

The AMD64 architecture does not suffer those drawbacks to the same degree. Because an int is defined as 32-bit unless a 64-bit number is explicitely requested, 64-bit software will not be much larger than 32-bit software. Only pointers will double in size, which AMD has predicted will increase code size by about 10%, depending on the program.

Share this post


Link to post
Share on other sites
i don't believe the big deal is 64bit itself.  it's a good gimmick for making sales, but as shown with other 64bit platforms like sparc64 the gains are usually not there for typical users - infact there are drawbacks.  the athlon64 has powerhouse speeds, but i do not believe this is not due to its 64bit nature.

i think what this means to amd, to intel, and to us is that amd gets to fight on a level playing field.  intel was going to shut them out with epic/itanium and lock in the corporate customer base on that level.  they failed, so now they have to go heads up until they get another opportunity to separate.  it should be a while.

The AMD64 architecture does not suffer those drawbacks to the same degree. Because an int is defined as 32-bit unless a 64-bit number is explicitely requested, 64-bit software will not be much larger than 32-bit software. Only pointers will double in size, which AMD has predicted will increase code size by about 10%, depending on the program.

They have also said the 10% program size increase will be offset by the registers, and that perf in 32bit mode and 64bit mode will be almost the same on typical apps.

Apps that it hurts could be run in 32bit mode, but the os will still be 64bit.

AMD's 64 bit perf is entirely the good perf of the Opteron in 32bit and 64bit modes. I would perfer Intel to wait a long time to release an AMD64 compatible system.

MS is not interested in porting windows to a forth architecture and even if it does notice the fast timing of the AMD64 version of windows.

Share this post


Link to post
Share on other sites
i don't believe the big deal is 64bit itself.  it's a good gimmick for making sales, but as shown with other 64bit platforms like sparc64 the gains are usually not there for typical users - infact there are drawbacks.  the athlon64 has powerhouse speeds, but i do not believe this is not due to its 64bit nature.

i think what this means to amd, to intel, and to us is that amd gets to fight on a level playing field.  intel was going to shut them out with epic/itanium and lock in the corporate customer base on that level.  they failed, so now they have to go heads up until they get another opportunity to separate.  it should be a while.

The AMD64 architecture does not suffer those drawbacks to the same degree. Because an int is defined as 32-bit unless a 64-bit number is explicitely requested, 64-bit software will not be much larger than 32-bit software. Only pointers will double in size, which AMD has predicted will increase code size by about 10%, depending on the program.

Isn't the definition of data types the area of a compiler designer, and not the area of the CPU designer?

Also, code may depend on being able to store pointers in ints and the other way around. And most of the time, ints are as large as the native registers.

Share this post


Link to post
Share on other sites
i don't believe the big deal is 64bit itself.  it's a good gimmick for making sales, but as shown with other 64bit platforms like sparc64 the gains are usually not there for typical users - infact there are drawbacks.  the athlon64 has powerhouse speeds, but i do not believe this is not due to its 64bit nature.

Well, it won't take too long before the 2 gbyte physical or virtual address space becomes a bottleneck.

One gbyte RAM isn't uncommon at this moment and more will soon be common I think.

Share this post


Link to post
Share on other sites

It is important to remember that AMD's 64-bit implementation offers significant advantages in the areas where performance is still very relevant. Particularly in any type of compression algorithm, for example media encoding gets a massive boost (here's a 64-bit version of lame vs a 32-bit version). I, personally, can't wait to see a AMD64 optimized Xvid codec. Also many server applications appear to recieve a significant boost(here's apache), although not nearly as remarkable as that observed with compression algorithms.

There were also demonstrations of encryption algorithms improving ~4x, which perhaps someone else can find. This was in higher end encryption. Before you scoff, understand that this makes sense if you consider a Xeon or P4 needs to use every register it has got to store a single 256-bit number while the Opteron or Athlon 64 can store 4 such numbers in its registers! The ratio of 4:1 stands for lower strength encryption like 128 bit keys too. This work is becoming very important these days.

jpalmer,

3. double FPU performance (compared to Opteron)

Ya, that'll be the day :blink: Oh ya, especially note the multi cpu comparisons to appreciate the difference in environments where this stuff is taken very seriously.

On the subject of x86-64 Windows. I think that, no matter what Microsoft said, Intel could convince them to release a version for their implementation if it was different. I feel this is a likely source of the ridiculous delays the product has seen. I mean they had working beta's ages ago (at least for AMD). And I am fully aware of AMD and Intel's cross-licensing agreement and that there would be no barriers to prevent Intel from producing a compatible chip. I just can't see them admitting that so much of what they've been claiming for years now is rubbish, and, on top of that, adopting their rival's technology. We have all seen that Intel doesn't exactly deal with the market in a humble way. Maybe we're about to witness a change of heart though. I'll be shocked if that's they way of it though.

Do well.

Jonathan Guilbault.

Share this post


Link to post
Share on other sites

Here's an encryption related link. It notes a 1.5x to 3.0x benefit from going 64 bit relative to 32 bit. It's not the original benchmark I saw, but it's relevant.

Do well.

Jonathan Guilbault.

Share this post


Link to post
Share on other sites
It is important to remember that AMD's 64-bit implementation offers significant advantages in the areas where performance is still very relevant.  Particularly in any type of compression algorithm, for example media encoding gets a massive boost (here's a 64-bit version of lame vs a 32-bit version).  I, personally, can't wait to see a AMD64 optimized Xvid codec.  Also many server applications appear to recieve a significant boost(here's apache), although not nearly as remarkable as that observed with compression algorithms. 

There were also demonstrations of encryption algorithms improving ~4x, which perhaps someone else can find.  This was in higher end encryption.  Before you scoff, understand that this makes sense if you consider a Xeon or P4 needs to use every register it has got to store a single 256-bit number while the Opteron or Athlon 64 can store 4 such numbers in its registers!  The ratio of 4:1 stands for lower strength encryption like 128 bit keys too.  This work is becoming very important these days.

Most of the time, the key in secret-key encryption is only used once to initialize certain data structures. So it's not related to key-size.

I think 32-bit vs 64-bit doesn't matter as much either (because most algorithms are designed to work good on 32-bit systems, although many use 64-bit blocks), I think it's the larger number of registers.

Share this post


Link to post
Share on other sites
Here's an encryption related link.  It notes a 1.5x to 3.0x benefit from going 64 bit relative to 32 bit.  It's not the original benchmark I saw, but it's relevant.

Do well.

Jonathan Guilbault.

Ah, but that's more about public-key encryption I think. In that case, keys are usually between 512 and 2048 bits.

Share this post


Link to post
Share on other sites

Olaf van der Spek,

I think it's the larger number of registers.

The bit width of the registers matter as well. A 64-bit number needs two 32-bit registers. The K8 has twice as many registers which are all twice as wide (hence when dealing with big numbers it has 4x the useable register addresses). Also the actual arithmetic takes only one operation with a 64-bit ALU, although, as you say, the register space and the effect it has on loads and stores is likely the relevant part of the performance issue in this case.

Do well.

Jonathan.

Share this post


Link to post
Share on other sites

Some very big names in non-consumer markets have been working on AMD-64 versions of their software for quite a while now. They've been more than impressed with the speed compared to P4 Xeons.

Intel would be insane not to run with a compatible version.

Share this post


Link to post
Share on other sites
The sad part is that Intel probably squeezed MS to prevent them from from releasing the 64-bit port of Windows for AMD's 64-bit implementation.

As an AMD executive has been quoted as saying: Microsoft needs no help delaying a product...

I think this is smart for intel, because it recognizes that thier current 64 bit stratagy isnt working and are thus expanding thier 64 lines... (I dont think they'll drop Itanium for the desktop 64 bit chip)

Share this post


Link to post
Share on other sites
i don't believe the big deal is 64bit itself.  it's a good gimmick for making sales, but as shown with other 64bit platforms like sparc64 the gains are usually not there for typical users - infact there are drawbacks.  the athlon64 has powerhouse speeds, but i do not believe this is not due to its 64bit nature.

i think what this means to amd, to intel, and to us is that amd gets to fight on a level playing field.  intel was going to shut them out with epic/itanium and lock in the corporate customer base on that level.  they failed, so now they have to go heads up until they get another opportunity to separate.  it should be a while.

The AMD64 architecture does not suffer those drawbacks to the same degree. Because an int is defined as 32-bit unless a 64-bit number is explicitely requested, 64-bit software will not be much larger than 32-bit software. Only pointers will double in size, which AMD has predicted will increase code size by about 10%, depending on the program.

Isn't the definition of data types the area of a compiler designer, and not the area of the CPU designer?

Also, code may depend on being able to store pointers in ints and the other way around. And most of the time, ints are as large as the native registers.

No, AMD has defined the AMD64 ABI to define an "int" as a 32-bit number. It is possible that a compiler designer could do otherwise, but then that compiler would not follow the ABI.

If a program wants a 64-bit int, it can ask for a long long.

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now