Suspected PayPal Hackers Seek Deal To Stay Out Of Prison
2 Men Arrested After Teen Was Run Over For His iPad
Intel forced to provide a compiler that isn't crippled for AMD processors
It's just one of the many sub-clauses of the settlement but an incredibly important caveat for software developers all over the world. If you run a program on two functionally-identical processors, one manufactured by Intel, the other by AMD, the program might run 47% faster on the Intel chip. Other than the open-source community's GNU GCC, there isn't really an alternative to the Intel compiler -- and for some applications the Intel compiler is the only option.
It's crazy but true! If the compiler detects that it's being run on a non-Intel chip, it produces slower, less-optimized code. You can see now why AMD -- and a huge number of software companies and programmers! -- was seeking such huge damages from Intel.
Let's hope the settlement in November means we'll see an end to the anticompetitive play, and a new compiler!
Comments
18
Subscribe to commentsJimScarJan 4th 2010 10:13AM
So all the benchmarks showing everything since Core2Duo stomping on Phenoms & Phemom II's has been a sham? It will be interesting to see some retesting with re-compiled benchmark software, though I'd be surprised if many companies bother to re-release existing products with updates.
Sebastian AnthonyJan 4th 2010 11:35AM
Yeah, basically. I can't believe benchmarks might be inaccurate, purely because the BENCHMARK (via the source code) is faulty, not the processor itself.
Crazy shit, really.
AlexLJan 4th 2010 11:10AM
I think this article may be overstating the market share and importance of the Intel compiler. GCC is the default compiler for development on Linux and OS X, and development on Windows is very often done with Microsoft's Visual Studio and thus uses the Visual C++ compiler, and interplatform development is usually not done with C/C++, so where does this leave the Intel compiler? Of course I'm just guessing here, I'd like to see some actual numbers about what kind of market share the Intel compiler has.
Sebastian AnthonyJan 4th 2010 11:37AM
Yeah, I looked for some numbers but couldn't find any.
The Microsoft compiler is definitely used a lot -- I don't think GNU GCC really figures into the 'grand scale' of things, considering relative market share of OSS and Windows-based systems.
Various sources seem to suggest that the Intel compiler is THE best for Intel chips. You can see how a developer might use the Intel compiler without knowing that they're crippling AMD users.
Chris StottJan 4th 2010 12:57PM
Personally, I don't like this precedent.
A platform provider should be able to have a compiler that has optimizations for their platform. Supporting other chips would add complexity to the Intel Compiler and take away from its primary purpose - being the fastest compiler for Intel chips.
There's nothing stopping AMD producing their own compiler.
Sebastian AnthonyJan 4th 2010 12:58PM
No, indeed, and I don't think anyone is saying that -- this is more the fact that Intel intentionally cripple non-Intel chips without telling anyone.
The linked article suggests that very few people actually know about this.
Of course AMD could make their own compiler, but if Intel make a good compiler... why bother? Reinventing the wheel...!
MyriaJan 4th 2010 3:34PM
"Of course AMD could make their own compiler, but if Intel make a good compiler... why bother? Reinventing the wheel...!"
So basically AMD should be able to leach off of Intel's IP?
There is a depressing amount of this sort of idiocy in the computer market. It's Intel's compiler, it should be their right to do with it as they wish. Instead we get bizarre IP socialism that allows smaller companies to leech off of direct competitors.
Why bother, indeed, when it's cheaper to get a court to make your competitor do your work for you?
JBJan 4th 2010 5:57PM
The whole point behind anti-monopoly law is that it hurts the CONSUMER (that's you by the way).
From what I'm reading, no one is asking Intel to write extra code for optimizations. They already WROTE extra code that deliberately chooses a slower code path for AMD processors. All they would have to do is take that out.
As an example, even if an AMD processor supports SSE3, the complied code will give the AMD processor the slower SSE2 or even plain SSE version of the program.
gamechldJan 4th 2010 5:12PM
Then why stop at crippling? They should just have the compiler not work on AMD chips at all. Probably because they don't want people to blatantly see a software issue on Intel side, they would rather it appear to be a hardware issue on AMD side. You can't side with Intel on this because its not like they took a defiant stand against IP sharing. They simply tried some sneaky code writing to make their chips look faster in certain situations. There's a difference between withholding help to a competitor and active sabotage. This was the latter in my opinion.
dmorris68Jan 4th 2010 6:12PM
As noted, this article drastically overstates the market share and significance of Intel's compiler. While they do have a reputation as a good compiler toolchain, I've been a C/C++ developer for over 20 years, and have rarely used it or seen it used. Microsoft, Borland, and Watcom compilers account for probably 99% of all x86 code I've ever done on Microsoft x86 platforms, although lately I've been trying to use GCC whenever possible. And in the vast majority of commercial Microsoft x86 C/C++ software where I've paid attention to the compiler used, Microsoft's compiler was by far the most common. So I'm thinking this particular point hasn't impacted AMD all that much.
Sebastian AnthonyJan 4th 2010 6:15PM
Well, fair enough, and I defer to your greater knowledge on the market-share fact!
If it wasn't an issue, it wouldn't have been brought up in the anti-trust case! I don't suppose we'll ever know what the actual percentage of Intel's compiler usage is tho'.
minibarJan 4th 2010 6:41PM
Most corporate marketing skews perceptions of competitive analysis to favor themselves and in this case the Intel compiler serves as marketing propaganda. Further, it's widely expected that dominant market players control market share by controlling the manner in which the products are measured because it happens across so many markets where there is a dominant market leader. Hard to believe this rises to the level of civil liability. AMD should spend more money developing better silicon, but I suppose suing your competition is one way of gaining competitive advantage.
dmorris68Jan 5th 2010 12:25AM
I agree that the compiler is likely a non-issue, and in response to Sebastian's reply to my earlier post, I expect it was just one of several examples AMD could point to as underhanded/anti-competitive practice on Intel's part, which is why it was included in the list.
However from what I recall reading about this case, I believe AMD had a valid argument. AMD initially developed (and I believe, patented) the AMD64 extensions for 64-bit processors, which Intel adopted and basically absconded with. This was a case of Intel using AMD IP without agreed-upon credit or compensation. So while I do frown on frivolous litigation, especially around software patents, there has to be some protection for the little guy who innovates from the big bullies (Intels, Microsofts, etc.) of the market. To wit, Microsoft's recent slap-down over their stealing of XML format patents in Office, that they now have to remove under court order. Microsoft is famous for such tactics, since their pockets are generally deep enough to pay off anybody who fights back, assuming the little guy can survive a protracted litigation battle. Intel's pockets are similarly deep, although I have to expect a 1.25 BILLION dollar settlement had to sting even them!
AstralmindJan 5th 2010 6:15AM
I feel tricked by Intel, i had the feeling they where the best, but when you put stunts like that you can acutely see they are afraid of AMD. And here's a question to think on: If you are so good you say why cheat ?
AlexJan 6th 2010 10:39AM
A lot of you are missing the point. Intel's compiler is NOT free, not by a long shot. Developers *pay* for it, and it still cripples apps compiled with it for AMD processors.
All of you howling that AMD is benefiting from Intel's IP are missing the point - *developers* are paying for Intel's IP, as a software service. Their software service, which again is paid for, is expected to function properly - and it isn't on AMD (and other) processors.
Now do you see why it isn't just AMD, but Intel's customers that are angry too? They pay an arm and a leg per station to use the ICC.
AerowsJan 6th 2010 10:49AM
@ Chris Stott
"A platform provider should be able to have a compiler that has optimizations for their platform. Supporting other chips would add complexity to the Intel Compiler and take away from its primary purpose - being the fastest compiler for Intel chips."
That would be well and good, if they weren't charging customers a fortune to use it and selling it as a software service. They are in the software business with the compiler, not the hardware business. If it was given away, that would be one thing - it isn't. Intel is selling their software and service IP, then turning around and misrepresenting it.
Intel's own customers, those who bought the compiler, are angry about it, too.
anuJan 12th 2010 10:15AM
this article is grossly misleading... it says:
- Intel's being forced to fix the issue: it was fixed back in 2007 in version 10
- it caused a 47% performance hit: the Intel compiler disabled some SSE instructions on non Intel cpu's resulting in a maximum 10% performance hit on Intel compiled apps, oh and it's their software
- it affects a large percentage of the software market: as another commenter pointed out Intel probably has
anuJan 12th 2010 10:40AM
- it affects a large percentage of the software market: as another commenter pointed out Intel probably has