The AMD Execution Thread [2007 - 2017]

[hoho]

A little overview of Spec benchmarks.
They consist of lots of small application benchmarks. Regular SpecFP/SpecInt scores are found by running one instance of every benchmark in sequence. Basically that shows single threaded performance*. Spec_rate scores are found by running several copies of the benchmarks in parallel, usually as many as there are CPU cores. In SpecFP_rate score is usually limited by availiable memory bandwidth, in SpecINT_rate memory bandwidth is not as important.

*) With certain compiler swiches it is possible to automatically parallelize code to run on multiple cores so one has to be careful when comparing the scores.

Reason why Barcelona has showed excelent SpecFP_rate scores is that it has a lot more memory bandwidth availiable compared to Intel (roughly 2.5-3x more on dual socket motherboard). If one looks at the non-rate scores then Intel leads in integer and matches in floating point.



Btw, newer AMD CPUs support SSE3 (some random page showing it). Though it does seem like AMD will not have SSE4 in their CPUs. Also if AMD has SSE5, probably the biggest update since the introduction of SIMD instructions, and Intel doesn't then I wonder how well it will be accepted. Also I wonder what will Intel have in place of this, early information about Larrabee and Gesher seem to indicate that both have multi operand instructions just as SSE5 does.

 

[_xxx_]

Since when did Nvidia have to license technology from 3dfx to compete with them?

They were sued by 3dfx for stealing their tech (multi-texturing or whatever the point was) and that case was buried due to them buying 3dfx. So effectively, they should have licensed that and some other stuff but they never did.

 

[dizietsma]

A little overview of Spec benchmarks.
They consist of lots of small application benchmarks. Regular SpecFP/SpecInt scores are found by running one instance of every benchmark in sequence. Basically that shows single threaded performance*. Spec_rate scores are found by running several copies of the benchmarks in parallel, usually as many as there are CPU cores. In SpecFP_rate score is usually limited by availiable memory bandwidth, in SpecINT_rate memory bandwidth is not as important.

*) With certain compiler swiches it is possible to automatically parallelize code to run on multiple cores so one has to be careful when comparing the scores.

Reason why Barcelona has showed excelent SpecFP_rate scores is that it has a lot more memory bandwidth availiable compared to Intel (roughly 2.5-3x more on dual socket motherboard). If one looks at the non-rate scores then Intel leads in integer and matches in floating point.



Btw, newer AMD CPUs support SSE3 (some random page showing it). Though it does seem like AMD will not have SSE4 in their CPUs. Also if AMD has SSE5, probably the biggest update since the introduction of SIMD instructions, and Intel doesn't then I wonder how well it will be accepted. Also I wonder what will Intel have in place of this, early information about Larrabee and Gesher seem to indicate that both have multi operand instructions just as SSE5 does.

Many thanks for taking the time to write that peice in an understandable and fairly short form. Yes, it makes sense from how you describe it how they can be so far ahead in those benches.

 

[3dilettante]

Just one point here. AMD is no so much restricted by Intel architecture. They were the first to implement x86-64 in 2000, and so it was Intel had to follow AMD´s standards. Check this out:

http://www.amd.com/us-en/Corporate/VirtualPressRoom/0,,51_104_543_552~715,00.html

The big reason was that Microsoft really wanted a 64-bit x86 variant so it could push Windows into a market that was completely closed to it.
Intel balked because that would cannibalize its new IA64 architecture, and AMD managed to slip in.
By the time Intel changed its stance, Microsoft refused to support another x86-64.

Unless SSE5 has a "Increase OS Revenue" instruction ahead of Intel, any further AMD extensions will not receive the same support.


Back on the financial front:

AMD's manufacturing shortfall is directly impacting the major gains its made with large OEMs.
The big reason they took so long to sell AMD systems (besides pressure from Intel) was the fear AMD would be unable to supply them.
AMD's regression on this front is not healthy for its relationship with those OEMs AMD has gone out of its way to screw the channel for.

 

[ShaidarHaran]

They were sued by 3dfx for stealing their tech (multi-texturing or whatever the point was) and that case was buried due to them buying 3dfx. So effectively, they should have licensed that and some other stuff but they never did.

Since no decision was ever reached in that case all you can do is speculate. I really hate to come down on the side of NV over 3dfx but in the eyes of the law they're clean. You'll also note that NV counter-sued 3dfx for patent infringement, so saying that one party sued another doesn't really mean anything without a decision.

 

[INKster]

The right to use the x86 ISA, extensions, and IP related to x86 is part of a cross-patent agreement between Intel and AMD.
Both companies are allowed to use each other's instructions, but Intel is still the owner of x86. AMD making payment to Intel has been part of previous agreements.

There are a number of restrictions AMD is subject to for it to have an x86 license.

For example:
There are limits to what percentage of its production can be made by an outside foundry.
There are limits to how large a company AMD can buy.
It is forbidden for AMD to undergo a change in control (buyout, etc).
It is forbidden for AMD to go bankrupt.

Any violation of these rules terminates AMD's right to use x86.
This is why hopes AMD can restructure if it goes bankrupt, or it gets bought out by some company or consortium, are mostly pointless.

Unless the buyer has full rights to Intel's current and future ISA and its extensions, AMD's x86 design and infrastructure is worthless.

That's like saying to someone that he is forbidden to die, or he'll have no more Dunkin' Donuts after he's dead.
Does it make any sense ?

Also, i thought the terms of the agreement between AMD and Intel were not in the public domain...

 

[3dilettante]

Better wording would be it would be forbidden to restructure financially and still have the right to make x86.

The terms for termination of the current agreement are available, though many parts of the full agreement are confidential.

Here's the text, though exact numbers are often confidential.

Section 6 deals with termination.

http://contracts.corporate.findlaw.com/agreements/amd/intel.license.2001.01.01.html

 

[MTd2]

Better wording would be it would be forbidden to restructure financially and still have the right to make x86.

No, that's not it. It doesn't cover the right to x86. The original is expired, and the license to the x86-64 belongs to AMD, as you confirmed right above. You told me that.

What is important there is the linceses to the original SSE1,2 and 3-. I will say again, SSE 1 and 2 expires in 2009 and 2010. So, any company will be able to do what AMD does now, in relation to those protected IPS, except for the instructions in SSE3.

If anycompany aquires AMD in 2010, they will be able to produce Fusion normaly, excluding the instructions in SS5, thar are also contained in SSE3 and SSE4.

 

[3dilettante]

I'll go with the idea AMD can continue to produce models based on older instantiations of x86.

6.3. Survival. The provisions of Sections 1, 2, 4.5, 5, 6.3, 7 and 9 will
--------
survive any termination or expiration of this Agreement.

The above seems to support this.
It also may mean Intel has right to produce x86-64 as well, since it has rights to it under the current agreement, and no liability can be assigned to products made prior to the expiration.


For the future:
The license, in any instantiation, includes the assumed right to produce Intel-compatible x86 when Intel makes new extensions.

Even if AMD should get away with producing chips running the older SSE variants, termination of the agreement will mean that it cannot produce chips compatible with later Intel extensions.

Given Intel's rate of new architecture introductions, that would leave AMD perhaps a year.
Unless AMD can get some kind of critical mass to force Intel to renew or software developers to take notice, it will erode out of the market.

Intel's position as dominant player means it can afford to lose AMD compatibility.

 

[MTd2]

Intel's position as dominant player means it can afford to lose AMD compatibility.

I had other thought when I read this. In 2000, Microsoft was aware that a superior position would be troublesome for AMD, and thus it could be that it could be crushed, so they favored AMD. Maybe this could be a coincidence, but with Vista right now being so bloated, maybe they can give AMD a hand again...

 

[MTd2]

IBM Servers Use Upcoming Intel 45nm Xeon 5400 Processors

http://www.x86watch.com/news/ibm-blade-forty-five-nm-xeon-144.html

It sucks to be DAAMIT...

 

[3dilettante]

What could Microsoft do?

If it favors AMD over Intel, when x86 fractures it will put its OS against the majority of the PC market, which is going to want to stick with the dominant platform.

I don't see what Microsoft has to gain favoring AMD's extensions.
It needed x86-64 for high-end systems that wanted the address space.

An FMADD or new data type isn't going to change the bottom line when it comes to the OS market.

Microsoft may be best served by speeding up the time when thing are back to one ISA for the sake of smoother development and compatibility, whether it means AMD survives or not.

 

[Fox5]

IBM Servers Use Upcoming Intel 45nm Xeon 5400 Processors

http://www.x86watch.com/news/ibm-blade-forty-five-nm-xeon-144.html

It sucks to be DAAMIT...

IBM servers have always favored intel, fwiw.

BTW, there seem to be so many interpretations of the x86 license agreement, it'd be nice to get an official position on it.

 

[MTd2]

If it favors AMD over Intel, when x86 fractures it will put its OS against the majority of the PC market, which is going to want to stick with the dominant platform.

I think the power of Microsof when it comes to the desktop OS is too big, but it's just a guessing for now. I guess they could really make a market shift when dealing with Intel and AMD.

BTW, there seem to be so many interpretations of the x86 license agreement, it'd be nice to get an official position on it.

Agreed. I will think who I may ask to.

 

[INKster]

If anycompany aquires AMD in 2010, they will be able to produce Fusion normaly, excluding the instructions in SS5, thar are also contained in SSE3 and SSE4.

You are incorrect.
SSE5 will not be a superset of SSE4, like the past generations were (SSE, SSE2, SSE3).
A significant amount of instructions from SSE4 will not be in SSE5, and vice-versa:

http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=3073&p=1


Intel has decided that it will not support SSE5, in the foreseeable future anyway.
That means that their compiler (which is dominant) will probably prevent most software developers from using AMD's SSE5, limiting its reach.
Much like things AMD decided to do earlier, like "MMX+", "3DNow" and derivatives, etc.

AMD64/x86-64 was a different matter, because it was a structural break that had to be done anyway, as the x86 32bit memory addressing got to the 4GB wall and PAE wasn't the best solution, obviously.

 

[3dcgi]

Intel has decided that it will not support SSE5, in the foreseeable future anyway.
That means that their compiler (which is dominant) will probably prevent most software developers from using AMD's SSE5, limiting its reach.
Much like things AMD decided to do earlier, like "MMX+", "3DNow" and derivatives, etc.

AMD64/x86-64 was a different matter, because it was a structural break that had to be done anyway, as the x86 32bit memory addressing got to the 4GB wall and PAE wasn't the best solution, obviously.

I don't know compiler market share and it would be interesting to see some stats, but I doubt Intel has dominant market share. When I looked a few years back it cost more that Visual Studio. I bet Microsoft's compiler has the dominant market share on Windows and gcc on Linux/Unix.

 

[_xxx_]

I don't know compiler market share and it would be interesting to see some stats, but I doubt Intel has dominant market share. When I looked a few years back it cost more that Visual Studio. I bet Microsoft's compiler has the dominant market share on Windows and gcc on Linux/Unix.

I'm quite sure many companies making professional SW are using the Intel compiler. Adobe for example.

 

[MTd2]

You are incorrect.
SSE5 will not be a superset of SSE4, like the past generations were (SSE, SSE2, SSE3).

Well, I was not quite incorrect... because i implicitly included the possibility of not having SSE4. The the number set of instructions contained only in SSE4 and SSE5 is 0, so out of 0 instructions, 0 are contained . What you posted is interesting and enlightening, i wonder how many are shared with SS3, though...It's interesting that AMD is trying to lose its depence on its main compatitor. But as FOX5 said above, it´s better to check the agreements with some secure sources.

 

[Albuquerque]

I think the power of Microsof when it comes to the desktop OS is too big, but it's just a guessing for now. I guess they could really make a market shift when dealing with Intel and AMD.

Yes, but Microsoft's bread-and-butter business is Software. And to sell the maximum amount of software possible, your software should run on as many platforms as possible.

They have no vested business interest in pushing one CPU hardware provider over another. And in fact, if push came to shove, they'd likely want to have less hardware options to support versus more -- that makes debugging far easier.

So the chances of Microsoft making any sort of wide sweeping change to support (or not support) a major CPU vendor is pretty low at this point. At least IMO...

 

[ban25]

I'm quite sure many companies making professional SW are using the Intel compiler. Adobe for example.

I doubt it's really all that big, to be honest. Most companies are happy with VS.