IvanV

Here's something you probably don't know, but really should - especially if you're a programmer, and especially especially if you're using Intel's compiler. It's a fact that's not widely known, but Intel's compiler deliberately and knowingly cripples performance for non-Intel (AMD/VIA) processors.
____________
Source: OSNews

__________________
If anyone has Portal 2 and hasn't played the co-op and wants to do me a favour, let me know (PM me or whatever).

deBelly

Not very surprising, unfortunately. Is that compiler used in many important applications?

IvanV

Someone else will be able to provide a better answer, but I'm guessing lots. As long as their CPU is later used to run the application (and they do hold a vast majority of the market), few other compilers are capable of optimizing the code so well and they also provide many extremely well written, and therefore popular, specialized libraries compiled with that compiler.

Sihastru

Well it is made by Intel and they target Intel CPUs. It's like blaming nVidia for not making CUDA compatible with ATi cards. Anyway the author is clearly seeking it's 15 min of fame.

Intel makes more then one compiler, let's assume he talks about the C++ compiler. The way to "optimize" compiled binaries is to use instructions from the extended set (like MMX, SSE and others not so much known in the marketing channels).

It's obvious that an AMD CPU will not be fully compliant to the SSE4.2 extended instruction set (AMD only has the 4.1a extended instruction set implemented in their newest CPUs). The same can be said for an Intel CPU as not beeing compatible with the 3DNow! instruction set.

If you want an analogy is like having AA implementations on videocards. nVidia has MSAA, ATi has MSAA, but nVidia also has QxAA. nVidia goes up to 16x, ATi goes up to 32x on enhanced modes. These methods are similar, but not quite the same, and one card will be more efficient at doing 8x MSAA then another card (nVidia vs. ATi). And you can't ask an nVidia card to do 32x "ATi" AA and you cannot ask an ATi card to do the 16x "nVidia" QxAA.

For two CPU's to behave completely the same all it takes is identical CPUs. We all know Intel does faster Integer computations. Both CPUs are x86 compatible, but one is better at something then the other is. So wouldn't you find it normal for a company to fully take advantage of their products' strengths?

Intel's biggest strength now is Hyper Threading a.k.a. SMT. I bet their compilers spit out SMT optimized code like there's no tomorrow. AMD doesn't have Hyper Threading at this time, but they will in the future. When they will, the Intel Compilers might use them efficiently too.

This is getting ridiculous... Next I will hear about Intel not letting AMD use the color blue...

It's a free market, AMD can make it's own compilers. There's a reason they are called Intel compilers.

IvanV

It's not about that at all, if the compiler only detected the feature set and made multiple execution paths based on that, that would be fine (and desirable), but it detects the brand and provides non-Intel CPUs with code that sucks (there have been experiments with VIA Nano, since VIA, unlike AMD and Intel, does not lock manufacturer ID so it's possible to forge it, and when given Intel code it performed up to 50% better in some cases).

As for the comparison with CUDA technology, that does not hold water either. CUDA is Nvidia's proprietary technology which they decided not to license to ATI. On the other hand, Intel and AMD have cross licensing agreements and they share technologies. This is just an abuse of a good position in the compiler market in order to gain an unfair advantage in the CPU market. It's like going to a Shell petrol station and getting diluted petrol poured if non-Shell oil is detected in your engine.

__________________
If anyone has Portal 2 and hasn't played the co-op and wants to do me a favour, let me know (PM me or whatever).

OldBuzzard

You have that reversed. They target NON-Intel CPUs.

Arestechnica proved that when they tested the Intel ATOM against the Via NANO. The Via CPU allows you to change it's CPUID, and they played with that feature.

Low-end grudge match: Nano vs. Atom

What's interesting is that they got three difference results. One for when the CPUID was set to VIA, another that was a bit faster when they changed it to AMD, and they a heck of a lot faster when they set it to Intel.

Yeah, no shenanigans there, right?

__________________
When looking for a reason as to why things go wrong, never rule out sheer STUPIDITY

Sihastru

I still don't see why Intel should put their time, money and man-power into something that should equally benefit them to the same extent it benefits the competition.

Again, it is the Intel Compiler, not the AMD-VIA-Intel-Tylera-ARM Co-Developed for the good of all Mankind Compiler.

IvanV

You have, either because of ignorance or deliberate misunderstanding, confused a few things. It's "Intel" because it's made by Intel, just like there is Microsoft compiler (which should only work on MS CPUs, right?). Like all compilers, it has a target architecture which is called the x86 architecture (ARM that you mention doesn't belong there and wtf is Tylera?!).

Intel hasn't even bothered to tell those who licensed it (and payed for it - they sell it and the compiler is a product of it's own) something like: "This compiler is only for our CPUs, if you use a program compiled with it on a non-Intel CPU, we will make sure that your execution speed sucks!". It's pretty much a definition of anti-competitive behaviour - if you go to our competition we shall punish you.

No one is asking them to make a compiler that will optimize for AMD CPUs, as I said, compilers have a target architecture and all CPUs which share an architecture can execute the same code, so if they want to optimize for their CPUs, that's fine, just let there be one execution path, even if it's Intel optimized.

__________________
If anyone has Portal 2 and hasn't played the co-op and wants to do me a favour, let me know (PM me or whatever).

BlueMak

It's one thing to benefit your own products and a very different thing to impede the performance of another company's product.

It's like giving your son a Ferrari, and giving someone's else's kid a Fiat 128 with tan oil instead of proper gas/oil. Just giving a Fiat 128 would be fine.

__________________

The people who are regarded as moral luminaries are those who forego ordinary pleasures themselves and find compensation in interfering with the pleasures of others(Bertrand Russell)"You go into Afghanistan, you got guys who slap women around for five years because they didn't wear a veil,You know, guys like that ain't got no manhood left anyway. So it's a hell of a lot of fun to shoot them." - Lt. Gen. James N. Mattis
This is slavery, not to speak one's thought. [Euripides-The Phoenician Women (c.411-409 B.C.)] http://www.macedonia.info/FALLACIESANDFACTS.htm Sic semper tyrannis.

OldBuzzard

This is the key. What they are doing is DE-OPTIMIZING for CPUs other than Intel.

The original code allows processors that have the capability, to use things like SSE2 and SSE3 (which by the way are LICENSED to AMD), then when an AMD CPU is detected it forces code that won't allow the AMD CPU to use the SSE2 and SSE3 optimizations, thus crippling their performance.

It appears that a lot of this code has found it's way into the various benchmarks being used as well.

It kind of makes you wonder. Are the current Intel processors are really all that much better than the AMDs, or are they "better/faster" because of "smoke and mirrors" in the benchmarks?

__________________
When looking for a reason as to why things go wrong, never rule out sheer STUPIDITY

comp_ali

This matter is like some nvidia's game "TWIMTBP" that have issues with Ati cards. The game was programmed when detect non-nvidia gpu it will disable some features "And I am not speaking of Physx"

Cow_160483

Shady dealings by Intel I had no idea ... makes one wonder how much better their CPU's are compared to AMD's

__________________

Decode!
Muhahahahah!

BlueMak

It is very interesting indeed.

__________________

The people who are regarded as moral luminaries are those who forego ordinary pleasures themselves and find compensation in interfering with the pleasures of others(Bertrand Russell)"You go into Afghanistan, you got guys who slap women around for five years because they didn't wear a veil,You know, guys like that ain't got no manhood left anyway. So it's a hell of a lot of fun to shoot them." - Lt. Gen. James N. Mattis
This is slavery, not to speak one's thought. [Euripides-The Phoenician Women (c.411-409 B.C.)] http://www.macedonia.info/FALLACIESANDFACTS.htm Sic semper tyrannis.

Sihastru

I am still standing strong defending my opinion. I know it's easy to point fingers at someone that did wrong in the past, but it doesn't mean they do wrong every time.

May I try another analogy, the first one wasn't that good...

Let's say you buy an external USB Western Digital drive ( = Intel CPU). This drive comes with Western Digital branded automatic backup software ( = Intel Compiler). Let's say that you also buy an USB Seagate drive ( = AMD CPU) and another similar drive from Samsung ( = VIA CPU).

Now you use the Western Digital Drive and the Western Digital backup software and perform a backup of your entire system in 30 min dead. Now you want to try the Western Digital backup solution with your Seagate and Samsung external drives.

Let's assume Western Digital does not lock it's backup software to be used only with external Western Digital drives and they say that the software will work in a "compatible mode" with any other USB drive on the market ( = x86_32/x86_64, Intel says the ICC is compatible with other CPUs), as opposed to an "optimized mode" when it is coupled with an WD branded drive (ICC produces optimized code when an IntelGenuine branded CPU is found)...

You try the software and it works with your Seagate drive (AMD) but it does the backup in about 4 hours, a staggering 8x slower then with the WD drive.

Next you try it with the Samsung drive (VIA) and it still takes 4 hours. But the Samsung drive can be "hacked" and you can make it look to the system as a Western Digital drive (VIA = IntelGenuine Processor). You try the Western Digital backup software again and it takes only 30 minutes this time. OMG!

Are you entitled in this situation to call Western Digital and complain about the slow speed their backup software has when it's not working with a Western Digital branded external drive when Western Digital specifically stated the software will only work in "compatible" mode with other drives?

Translation: Are you really expecting Intel to support any other brand of CPU? Are you really expecting Intel to test each individual CPU from the competition? As I said before, the CPUs are compatible, not identical.

Your only real argument is that Intel should not test for the CPU brand, but instead test for the CPU feature set. I agree about that, but this means Intel has to test each CPU on the market from any current or future manufacturer (as I stated before, an AMD CPU feature set is not identical to an Intel CPU feature set, and even if the feature set is identical, the implementations might not be, resulting in different performance numbers), but since Intel is the one that does all the work, it's not our place to tell them how to do their job.

Since a CPU is better then another CPU at doing some random job, what is optimized for one might be very bad for the other. It worked ok for the VIA CPU, but that's one case, and you cannot generalize it to the point of drawing an incontestable conclusion. The only way to make sure you indeed optimize the compiler for all the CPUs is to test all CPUs. This means Intel spending money on making AMD/VIA CPUs run faster.

What AMD/VIA/Intel share is a license for x86, x86_64, MMX, SSEx. The license is not extended to the ICC. ICC is the property of Intel.

Judas

No it's not..

It's like Nvidia Purposely slapping extra code in their cuda to detect and PREVENT such working on ATI.

Just like Nvidia has and continues to Purposely DETECT and PREVENT Physx from running even on their own card IF an ATI card is present.

__________________

IvanV

This is where you keep getting it wrong, the CPUs don't come with the compiler, it's a completely separate commercial product and it's description says:
It promises "advanced optimization" and support for Intel and compatible (non-Intel) processors, not a word about the compiler being recommended for other CPUs any less than for genuine Intel! And due to cross licensing agreements between Intel and AMD I have already mentioned, their CPUs are architecturally compatible. And yes, AMD's CPUs may not support Intel's latest extensions to x86 instruction set at the moment, but neither do 90% of Intel's CPUs in the wild and they don't get stuck with crappy code.

You may continue to disagree, but this is what I and FTC have to say. I know it may seem that it's their product and they can do whatever they want with it, but there is extremely good reason why certain forms of market behaviour have been strictly forbidden in all of the civilized world for some 150 years or so. And Intel know that they broke the law or else they would have documented this feature (and by not documenting it and making everything look rosy, they also committed a fraud against anyone who licensed the said product).

__________________
If anyone has Portal 2 and hasn't played the co-op and wants to do me a favour, let me know (PM me or whatever).

OldBuzzard

And anyone that bought a system with an Intel Atom processor instead of one with a VIA processor based on the PCMark 2005 benchmarks.

It wouldn't surprise me if the same thing is happening in the other benchmarks as well.

Intel have always been weasels, and this is just one more indication of just how slimy the company is.

Sihastru, I'm done here, as it's obvious that you aren't capable of understanding business ethics/law, or you are just an Intel fanboi that will support them no matter what.

__________________
When looking for a reason as to why things go wrong, never rule out sheer STUPIDITY

Sihastru

@Judas: They don't block AMD specifically, they only optimize for Intel. They look for the "IntelGenuine" string, not for the "AMD Genuine" string. AMD is not singled out, they are just not included. It's a small difference concerning the end result, but it's a big difference when it comes to ethics/law.

@IvanV: Yes they say that on the website, but every website has disclaimers. So you cannot use the information about a product on the website to legally attack the company. For example, on the Intel site:

But, if you check the individual product documentation of some of the Compilers (the same documentation can be found in printed materials), you'll see that they don't claim optimizations for other processors then Intel and Intel platforms.

At any moment they can say that it was a simple case of editing error. Any website that has this kind of disclaimer (all of them really) can make this claim.

@OldBuzzard: Business ethics is something business men/women laugh at. Like 14 is the new 18 and so on. Law is difficult (to me) to understand (as is to them, the ones that wrote it), so I'm surely not able to defend my point in a court of law (if the law would be perfect, we wouldn't need a jury), but I can try on a forum. Hopefully next time I will not be accused of fanboism, when pretty much else fails.

@all. I'm not saying that what Intel does is commendable. What I'm saying is that it is not illegal. Ethics? Not really a business word.

ICCs have been behaving like this since forever. It's not something that was discovered last week. Because Intel is at a low point because of the antitrust problems, some random guy on the Interweb found it to be a good moment to have it's 15 minutes of fame.

It's the let's kick them while they are down attitude that I dislike. It's easy to put them in a bad light since they didn't left that place yet.

My last post in this thread unfortunately, I draw the line at fanboi.

BlueMak

If it can do better but the software recognizes that it is not an Intel CPU and thus doesn't allow it to work as fast as it could, then it is hampering and I don't know how you can see it in any other way.
If it would do that by overclocking the CPU, I could see how perhaps you would be correct, but it simply doesn't let it work at full speed. This is not giving a boost to Intel cpus, this is slowing down non Intel cpus.

__________________

The people who are regarded as moral luminaries are those who forego ordinary pleasures themselves and find compensation in interfering with the pleasures of others(Bertrand Russell)"You go into Afghanistan, you got guys who slap women around for five years because they didn't wear a veil,You know, guys like that ain't got no manhood left anyway. So it's a hell of a lot of fun to shoot them." - Lt. Gen. James N. Mattis
This is slavery, not to speak one's thought. [Euripides-The Phoenician Women (c.411-409 B.C.)] http://www.macedonia.info/FALLACIESANDFACTS.htm Sic semper tyrannis.

Judas

It was a crude example but a example of putting extra code in to specifically prevent or hinder the use of other hardware fully capable of doing the exact same job from actually being able to do it properly and as effectively as it should.

This is clearly rediculious

__________________