App Claims to Blunt Intel's Compiler Edge on AMD Machines

[Xaled]

Why do some people take such news so personally?

 

[Static~Charge]

A ominously named app

Ominous? If you find the name "Intel Compiler Patcher" to be ominous, then "Streaming SIMD Extensions" must give you the willies!

The idea (suspicion rather), is that apps developed with Intel C++ compilers give modern AMD CPUs a performance disadvantage.

The idea is to see if you get a performance boost without causing instability. That would confirm or deny any optimization bias against non-Intel processors. It falls under the heading of "scientific observation".

We don't have an AMD machine at hand to put our benches ourselves, and so we invite AMD CPU users from our community to post their results by using this "patcher" at their own risk.

I recommend making a backup copy of the file prior to patching it. That will make it much easier to restore the file to its original state.

 

[dorsetknob]

I recommend making a backup copy of the file prior to patching it. That will make it much easier to restore the file to its original state.

don't forget the before and after benchmarks

 

[Bansaku]

" We don't have an AMD machine at hand to put our benches ourselves, and so we invite AMD CPU users from our community to post their results by using this "patcher" at their own risk. "

What? Not a single machine? That's, a fail!!

 

[Basard]

Isn't this software years old?

http://www.amdzone.com/phpbb3/viewtopic.php?f=532&t=138849

Holy crap, amdzone still exists? Or was it .org back in the 90s?
edit... Oh, it's just the forums, huh... I see...

 

[cadaveca]

" We don't have an AMD machine at hand to put our benches ourselves, and so we invite AMD CPU users from our community to post their results by using this "patcher" at their own risk. "

What? Not a single machine? That's, a fail!!

and to add, how can you not have an AMD cpu based rig at hand, you guys are suppose to be pros.

Talk to AMD as to why they don't send us hardware for review purposes. It's not like I haven't asked.

 

[TheoneandonlyMrK]

That's a shame as it is that this underhanded compiler shit goes on but ,I don't think Hsa or any derivative of Amd's new compiler software is optimised for intel or its gpus so what goes around comes around and all that.

 

[Mussels]

That's a shame as it is that this underhanded compiler shit goes on but ,I don't think Hsa or any derivative of Amd's new compiler software is optimised for intel or its gpus so what goes around comes around and all that.

i just figured out your name is 'one and only mr k' i've been reading it as 'andon lymrk' for way too many years.


today i learned...

 

[OneMoar]

yea sure lets patch in architecture specific compiler optimizations into a already built binary for use on unsupported hardware
what could possibly go wrong
never mind the fact that intel already removed the so called "cripple amd flag" YEARS AGO
or that fact that there are some pretty major implementation differences of SiMID extensions between AMD And Intel
go home bta you are drunk ....

 

[Solaris17]

i just figured out your name is 'one and only mr k' i've been reading it as 'andon lymrk' for way too many years.


today i learned...

iv always read it the one random link

I actually decided to break it up because of this post. wow you learn something new everyday.

 

[R-T-B]

This again is nothing new. This information has existed since I got into PC's like 10+ years ago no idea why anyone is surprised...

I agree. This is not news.

 

[Yorgos]

Talk to AMD as to why they don't send us hardware for review purposes. It's not like I haven't asked.

I bet those less popular tech sites that buy the h/w themselves from all vendors are doing it all wrong.
I've seen more whining here than reviews.

It's hard to believe that a tech site does not even have a phenom ii at least laying around to check this or even a 4 year old bulldozer to check the sse instructions that icc was turning to legacy x86 code when the executable was running on an AMD cpu.

To get this right, do you review only what you get for free?

 

[Yorgos]

I agree. This is not news.

It could be news only if this came back again.
It appeared a couple of times, iirc, in the past.
The bad thing is that the only proper compiler in windows is icc, for those that don't want to mess around cygwin or mingw ecosystem and use gcc, and the next useful compiler is the Microsoft one that sucks at everything beyond the basic.

For the sake of it, I might run some benchmarks. I have several xeon workstations, a devil canyon, a kaveri and a few mobile parts laying around(my gear might have more variety than tpu) and I'll be reporting back some comparisons on gcc and icc.
Let's hope I'll find some spare time to do it.

 

[R-T-B]

It could be news only if this came back again.
It appeared a couple of times, iirc, in the past.
The bad thing is that the only proper compiler in windows is icc, for those that don't want to mess around cygwin or mingw ecosystem and use gcc, and the next useful compiler is the Microsoft one that sucks at everything beyond the basic.

For the sake of it, I might run some benchmarks. I have several xeon workstations, a devil canyon, a kaveri and a few mobile parts laying around(my gear might have more variety than tpu) and I'll be reporting back some comparisons on gcc and icc.
Let's hope I'll find some spare time to do it.

Came back again? It's always been, AFAIK. And as a developer, I have never found Microsoft's compiler "sucky," but I've never done a proper analysis either.

Do run some benchmarks. I'd be interested.

 

[uuuaaaaaa]

It could be news only if this came back again.
It appeared a couple of times, iirc, in the past.
The bad thing is that the only proper compiler in windows is icc, for those that don't want to mess around cygwin or mingw ecosystem and use gcc, and the next useful compiler is the Microsoft one that sucks at everything beyond the basic.

For the sake of it, I might run some benchmarks. I have several xeon workstations, a devil canyon, a kaveri and a few mobile parts laying around(my gear might have more variety than tpu) and I'll be reporting back some comparisons on gcc and icc.
Let's hope I'll find some spare time to do it.

I hope that you find some spare time good Sir!

 

[Basard]

Well, I may as well try it out.... see what happens.... I don't have much for programs to benchmark though.... Any suggestions for this four-day weekend?

edit: Meh... I scanned and it found a bunch of old installer files for CCC, which I don't use anymore. Found nvcuda.dll and some stuff for flash....

I've got this weird problem with videos skipping in firefox when i put my mouse over videos, I'm pretty sure it has to do with Adblock or Noscript. Because it works fine on an a bare version of firefox. Gonna see if it miraculously fixes that when I patch everything and my computer doesn't explode.

Otherwise, wish me luck! haha.

 

[GoldenX]

Cinebench 11,5 stock: 4.61
Cinebench 11,5 patched: 4.65

Cinebench 10 stock: 16316
Cinebench 10 patched: 16570

 

[cdawall]

" We don't have an AMD machine at hand to put our benches ourselves, and so we invite AMD CPU users from our community to post their results by using this "patcher" at their own risk. "

What? Not a single machine? That's, a fail!!

I'm not testing it because at the end of the day it really doesn't matter lol. I am posting from an ancient athlon II X3 as we speak and its more than powerful enough to browse the web and play even most AAA title games.

 

[truth teller]

[off-topic]
so my comment went down in flames? no warning nor notification of removal nor anything?
yeah, thats fine, whats a warning anyway. after all its your site you can share whatever "content" you damn well please.

now i want to play some c&c, and i hope im the only one...
[off-topic]

 

[cadaveca]

To get this right, do you review only what you get for free?

Nope. However I do have an eternal Non-Disclosure contract with AMD that prevents me from doing reviews on anything other than what they have OK'd. I gave my word and signed a legally binding document, and I plan to live up to it.

I've seen more whining here than reviews.

Whining takes less time and effort than a review does? Blame the hookers and blow.

 

[GoldenX]

Nope. However I do have an eternal Non-Disclosure contract with AMD that prevents me from doing reviews on anything other than what they have OK'd. I gave my word and signed a legally binding document, and I plan to live up to it.

That explains the low amount of low/mid range AMD products reviews.

 

[Yorgos]

Nope. However I do have an eternal Non-Disclosure contract with AMD that prevents me from doing reviews on anything other than what they have OK'd. I gave my word and signed a legally binding document, and I plan to live up to it.

I was in the same situation with several DELL, Mellanox and Intel systems and technologies(actually that cpu+fpga thing from intel was introduce to us 2,5 years ago) that were introduced to us inside dark rooms when I was working at CERN, but it didn't forbid me to buy a CPU from the shelf and test it.

 

[cadaveca]

I was in the same situation with several DELL, Mellanox and Intel systems and technologies(actually that cpu+fpga thing from intel was introduce to us 2,5 years ago) that were introduced to us inside dark rooms when I was working at CERN, but it didn't forbid me to buy a CPU from the shelf and test it.

Meh. Not all agreements are the same. Signed many NDAs over the years and no two have been alike. I value my position in this community and won't do anything to jeopardize it.

But then I'm left with a question for you:

So... umm... why are you complaining and not doing the testing, then? Like, who's the pot and who's the kettle? Do the testing, write a review and post in in our forum with results that anyone can repeat. An informational news post for our AMD users doesn't predispose us to have an obligation to review it. I'd do it, buy the hardware, no problem. I simply do not have time for it. Doing reviews is my hobby, not my job, and I only have so much time for my hobby, unfortunately. I have other reasons not to as well, this is but two. If they aren't good enough for ya, oh well. Sorry to disappoint.

 

[TheGuruStud]

Cinebench 11,5 stock: 4.61
Cinebench 11,5 patched: 4.65

Cinebench 10 stock: 16316
Cinebench 10 patched: 16570

Cinebench would be the worst test. Not only is it not a real, but it's written explicitly for intel only (very fine tuned for intel µarch, there's lots of hands in intel's cookie jar). Plus, IIRC, there was a cinebench scandal years ago about this and has since stopped the practice. There's nothing to be gained, now.

 

[Pill Monster]

Cinebench would be the worst test. Not only is it not a real, but it's written explicitly for intel only (very fine tuned for intel µarch, there's lots of hands in intel's cookie jar).

That's why Cinebench is a perfect test to run on AMD. I suspect this is meant to work like BDC which unblocks x87on Bulldozer based CPU's (though I don't know how this Intel Compiler software would do it).

I've tested a patched Phenom II and Piledriver on Super Pi, Cinebench and Games like Skyrim which are compiled with x87, and as in the Cinebench results posted earlier the effect was negligible.......