PC Buying Guide 2013 - Q4

[RiGOD]

^^Nice one mate. I've read many posts in various forums saying that the BD architecture is not a failure actually, but its Windows which cannot utilise the full potential of BD's. And once it is optimised the BD's will run like anything they say. Could you shed some light on this?

BTW in many sites BenQ G2220HDAL is available for 7.5-7.7k, the price of HDL has gone beyond 8k.

 

[Cilus]

There are other things too. Most of the Benchmark suits as well as the software developers use Intel's compilers to build their applications for Machine Level language. The machine language conversion of a high level coding (done in C++, Java or other high level language) needs to be very much optimized for running in different processors. They can be executed in several ways and it is the duty of the compiler to pick up the most optimized one depending upon the instructions support in current CPU.
Now it has been observed that Intel Compiler just detects whether its a Non- Intel or Intel processor. If it is a Non-Intel one (read AMD) it simply picks up the unoptimized code path without using the advanced instructions like SSE4/3, FMA4 etc to reduce the execution speed.

The 1st hands on proof came with Via Nano X2 processors, a cheaper alternative of Atom X2 processors. ViA Nano normally performs around 10% better than the dual core Atom and it uses the cloned CPU interface of Atom to make it compatible with Intel Atom motherboards. When the CPU-ID of this VIA processor has been modded with the CPU-ID of a Atom X2, the performance enhancement jumped to a whopping 30%+ over the same Atom processor. It clearly indicates that software compilers based on Intel doesn't actually check the true Instructions support by the Non-Intel CPU, it just checks whether it is a Non-Intel one and picks up the unoptimized code path. Detail level check for the capabilities of the Processor is only checked if it has a CPU-ID of Intel Processor.

 

[rajan1311]

dude thats too small a cabinet for a 60k system...I mean, it needs to look 60k

There are other things too. Most of the Benchmark suits as well as the software developers use Intel's compilers to build their applications for Machine Level language. The machine language conversion of a high level coding (done in C++, Java or other high level language) needs to be very much optimized for running in different processors. They can be executed in several ways and it is the duty of the compiler to pick up the most optimized one depending upon the instructions support in current CPU.
Now it has been observed that Intel Compiler just detects whether its a Non- Intel or Intel processor. If it is a Non-Intel one (read AMD) it simply picks up the unoptimized code path without using the advanced instructions like SSE4/3, FMA4 etc to reduce the execution speed.

The 1st hands on proof came with Via Nano X2 processors, a cheaper alternative of Atom X2 processors. ViA Nano normally performs around 10% better than the dual core Atom and it uses the cloned CPU interface of Atom to make it compatible with Intel Atom motherboards. When the CPU-ID of this VIA processor has been modded with the CPU-ID of a Atom X2, the performance enhancement jumped to a whopping 30%+ over the same Atom processor. It clearly indicates that software compilers based on Intel doesn't actually check the true Instructions support by the Non-Intel CPU, it just checks whether it is a Non-Intel one and picks up the unoptimized code path. Detail level check for the capabilities of the Processor is only checked if it has a CPU-ID of Intel Processor.

please do post the link, would be a nice read

 

[Skud]

Here you go:-

VIA Nano CPUID Tricks

 

[rajan1311]

Here you go:-

VIA Nano CPUID Tricks

thanks buddy...Hope AMD have some tool to change CPUID, because this stuff is awesome!It works better than overclocking for free performance

 

[Cilus]

Thanks Skud for the Link. Rep+ for you.
Actually Bulldozer also performs better in GNU based Linux compilers where FX-8150's performance is comparable with 2600K. Check some reviews with Compiler like Open64, GCC etc. Here BD enjoys two things:
1. Linux has better scheduling logic and doesn't rom the thread scheduling problem like Windows 7.
2. Use of the Open Compilers extensively so that any CPU can have its max optimization code path.

 

[Skud]

You are welcome buddy.

 

[RiGOD]

Spoiler

There are other things too. Most of the Benchmark suits as well as the software developers use Intel's compilers to build their applications for Machine Level language. The machine language conversion of a high level coding (done in C++, Java or other high level language) needs to be very much optimized for running in different processors. They can be executed in several ways and it is the duty of the compiler to pick up the most optimized one depending upon the instructions support in current CPU.
Now it has been observed that Intel Compiler just detects whether its a Non- Intel or Intel processor. If it is a Non-Intel one (read AMD) it simply picks up the unoptimized code path without using the advanced instructions like SSE4/3, FMA4 etc to reduce the execution speed.

The 1st hands on proof came with Via Nano X2 processors, a cheaper alternative of Atom X2 processors. ViA Nano normally performs around 10% better than the dual core Atom and it uses the cloned CPU interface of Atom to make it compatible with Intel Atom motherboards. When the CPU-ID of this VIA processor has been modded with the CPU-ID of a Atom X2, the performance enhancement jumped to a whopping 30%+ over the same Atom processor. It clearly indicates that software compilers based on Intel doesn't actually check the true Instructions support by the Non-Intel CPU, it just checks whether it is a Non-Intel one and picks up the unoptimized code path. Detail level check for the capabilities of the Processor is only checked if it has a CPU-ID of Intel Processor.

Now that's some shocking information. So if the most optimised path of the FX is chosen it'll perform close to SNB's right? So once that fix is done in Windows (atleast in 8) then the FX line up will have a future I guess.

 

[Cilus]

No man, Windows 8 will only fix the scheduling logic of the threads but how the instructions present under each thread will be executed will be still dependent upon the compiler used by the specific apps.
Sadly in Windows platform, Intel has invested around US 50 Million to develop several Compilers and let the developers use it for free. So it is very unlikely that commercial software developers will use any different compilers.

There are certain Open Compilers under GNU licencing like X264, Open64, 7Z etc which can be optimized to take the advantage of the advanced instruction set supports of any processors and apps developed with these will surely show some huge performance boost once the Thread issuing logic is fixed in Windows 8.

In Linux, you are having more options, I think Ico can throw some lights here.

 

[RiGOD]

^^That's sad news. So if the same conditions persists there isn't much to expect from the Piledriver series too.

 

[Cilus]

^^That's sad news. So if the same conditions persists there isn't much to expect from the Piledriver series too.

It's not like that. Even fixing the thread scheduling will increase performance of BD modules. Piledirver has better IPC performance and power management, can run cooler. AMD is also focuing to add the instruction sets which are Intel compiler compatible like FMA3. BD used to have FMA4 which is advanced than FMA3 but Intel suddenly dropped FMA4 from their Compiler specifications, added FMA3 support with Sandybridge-E and Ivybridge, just before the release of BD.

Hope, you are guessing the reason correctly,

 

[dfcols71]

if based on your reports intel and windows are ganging up to kickbut amd
why dont amd find some software developers who optimize the software to to
fully optimize the bulldozer range cpus

 

[Skud]

Not our reports buddy.

And AMD simply don't have the money, it's as simple as that.

Similar story in games also.

 

[rajan1311]

Not our reports buddy.

And AMD simply don't have the money, it's as simple as that.

Its not that they don't have money, just that Intel has 10x more

 

[Cilus]

Guys, I am thinking to replace Phenom II X6 1090T, available around 9K, by FX 8120, currently available at 9.4K at SMC. The reasons are:-

1. Gaming performance at 1080P is almost same.
2. FX has support for higher overclocking compared to Thuban.
3. The multi-threaded performance is better while keeping the single-threaded performance almost same.
4. Thuban has reached its EOL and there is nothing new it can offers apart from overclocking. All the other hand, FX 8120 is a relatively new architecture, promises the increase of performance with Windows 8, new Instructions like SSE4, FMA4, AVX support etc.

What do you guys think?

 

[RiGOD]

^^Sounds good to me. Its a new architecture compared to that of Phenom and has scope of improvement in performance in future. And yeah, the price drop too. Wise choice IMO.

 

[Cilus]

^^ Thanks for sharing your opinion. Here have a read of this article where the Author investigated thoroughly Intel's compiler and found out that it deliberately cripples AMD's performance. He was also able to create a patch to bypass the CPUID check in some areas:-

Swallowtail

 

[Omi]

^^^
But if you look at the difference between day to day performance or FPS, it will not be observable.
Your machine is already a beast, but if one looks at the EOL aspect, its not bad either.
Just that the %increase should justify the cost.
Things may be interesting after Windows 8 release (tough I doubt by how much)

8150 in Win8 Preview
*images.anandtech.com/reviews/cpu/amd/Bulldozer/Review/win8.jpg

Swallowtail

That was a very Good Read BTW.

 

[dfcols71]

well my point if amd have solid evidence they can outbench intell cpus with the compiler check discriminating against amd,they will defiantly sue intell for unfair practices.
as an after thought amd should develope their own compiler stuff.
I think the real problem is both intell and amd are focusing more on apu and mobile chips

 

[Cilus]

Omi, thanks for your concern but I'm not talking to get a FX for me, I was talking about the builds here where we're suggesting 1090T.

well my point if amd have solid evidence they can outbench intell cpus with the compiler check discriminating against amd,they will defiantly sue intell for unfair practices.
as an after thought amd should develope their own compiler stuff.
I think the real problem is both intell and amd are focusing more on apu and mobile chips

AMD as well as some other players in market have already sued Intel and there were several lawsuits going against Intel back in 2009. But most of 'em have been settled outside court and there are certain things you can't directly accuse. After money power of Intel is something.