Stereo Tool

I think the problem lies elsewhere - it lies in the CPU architecture itself - you can't simply optimize the code for all kinds of CPUs You just can't do it. And it's not really a problem, but this is what I call research and development . Core 2 architecture is a totally different beast from Sandy Bridge and its successors. Hence, more instructions per cycle can be executed, and this results from CPU optimizations itself, but also implementations of new instruction set that can be used.

Having tested various versions of ST on multiple Intel and AMD platforms, I'd say Hans is optimizing for Intel. You can see it for yourself when comparing ST performance on Intel/AMD with synthethic benchmark results, such as passmark. The two can do equally well in passmark, while ST will work significantly better on Intel. That's for sure. As an example, Athlon II 250 dual core (3.0 GHz, 2M, 4 Gb DDR3 1333, Win7Prox64SP1, Passmark score 1.7k) performed only marginally better than C2D E6400 (2.13 GHz, 2M, 3Gb DDR2 800, Win7Prox64Sp1, Passmark score 1.3k). The latter on the other hand performed much better than E2220 (2.40 GHz, 1M, Passmark score 1.3k) on the same system config. That's why I asked how much ST depends on CPU cache, it looks like it uses cache a lot.

The same happens when older and newer generations of Intel CPUs are compared.

Well, I asked the new compiler to optimize for an older CPU... So it should not perform worse (except if the compiler generates worse code than the version from 2007 did - which is actually very likely because I've had a performance degradation with every new compiler that I tried, that's why I was still using the one from 2007).

What I'm going to do: I'm going to compile it again, with the old compiler but with the new libraries. That should at least answer the question if the degradation comes from the compiler or from the libraries. Should be ready in about 1.5 hours, I'll compile the Winamp plugin version again so we'll have 3 different versions to compare.

Slawomir,
I propose Hans to take some middle class AMD like those you mentioned . I call it middle class bcause it`s double core , not too slow in speed and so on . Then to compile optimized versions for AMD cpus , so we can compare it to the other versions. It`s not a bad idea but looks like Hans don`t have AMD at all..

That sounds like a good idea. Let's see what Hans thinks about it. Currently at home I have a spare rig powered by Athlon II x4 620. It is exactly the same core architecture as dual Athlon II, except for the cache. Clock speed is 2.6 GHz, however only 2M cache (independent 4x512). Phenom II would be way better (it has L3 cache).

I believe that somewhere in the Raspberry Pi thread Hans mentioned why he's optimizing for Intel in particular.
And it kinda makes sense for one reason. Intel CPUs have generally higher IPC than AMD CPUs (Since Core 2 and Athlon II/Phenom II era up till now nothing changed in that matter). When you need to heavily exploit two threads, that's exactly what you are looking for - as fast single core performance as you can get. And this is where Intel shines. Another advantage would be significantly lower power consumption.

I have a general suggestion for everyone: when comparing ST performance on your system with others, please indicate which OS it is based on (also whether 32 or 64 bit).

I have Athlon II X2 240e 2.8GHz - K10
and Athlon 4850e - 2.5 GHz - K9
There is double CPU load difference between these two - The K10 is about 20-23% . The K9 is around 50%
But they have only 300Mhz difference in clock speed. I don`t think that the Intel`s analog to 240e is much faster in calculations. This myth is not valid for years ago..

Equivalent to 250e (in synthethic benchmark) would be E7200, which is clocked slower. Unfortunately I don't have that one. But you might be surprised - it is 45 nm Wolfdale core with 3M cache shared between the two cores (unlike Athlons II) and supports SSE 4.1 - that actually might make a difference over Conroe C2D series 65 nm (which has only SSSE 3 support and no SSE 4.1).

Please check out "Why choose Intel over AMD" thread, since I don't want to double myself on AMD vs Intel. And just to clarify - I'm not an avid fan of Intel or AMD But it is always nice to see AMD users here on this forum, dominated by Intel users

And I really think we drifted away from the main topic of this thread, which is ST 7.51 BETA. So, DJ-DOGGY, I will discuss the AMD-related stuff with you via PM. And I will do it with pleasure.

Hans, it is really nice to see development for AVX support on newer CPUs!

I`m not dominated . You take wrong CPUs to "compare". What is the idea to compare a cpu that NOT support 4.1 SSE with CPU that support 4.1 ? This is not a comparision ! Give me an Intel that NOT support 4.1 and supports only things that ARE supported by the 250e , undestand ? Two CPUs must have even the same architecture ( and frequency ) - let`s say 45nm or 32 nm .. or whatever you want. This IS a comparision for me , not the ones that are in internet - they pick whatever they want to compare without any EQUAL parameters!


See this one !!! What is the stupid idea of compare these CPUs ?This only shows that AMD production of 2012 will BLOW OUT this Intel !
What is the idea of comparing CPU with 2 years difference in production ?
And the points ARE similar !! 6.1 vs 6.0 ? What do you say for this one ? Even that they have different nm technology - one is 32 nm the other is 45 nm
http://cpuboss.com/cpus/Intel-Core-i3-2 ... II-X2-250e

OLD AMD operates even to the "NEW" Intel ? don`t make me smile please, i know that old ison 3 Ghz , but the new is 2.2 . I already show you the difference about 300Mhz . Look previous post

Actually you've made a good point on that, but you won't find one. And I will tell you why - Athlon II 240e supports SSE 4a instruction set, which is not found throughout Intel CPUs and is not fully consistent with 4.1, but it is there Take into consideration also differences between system bus, which would be HT link vs FSB. The closer one without SSE 4.1 (at least passmark-wise) would be C2D E5400 2.7 GHz 2M, 800FSB, right? But I also don't have it. No need to be aggressive.

And I wish it would be like you show i3-2328M vs Athlon II x2 250e, but it is not. You don't own i3 2328M, do you? If you had, you would've known that. Also note, that IPC is way higher for this i3, in comparison with Athlon, as it matches it in synthethic tests.

Regards,
Slawomir

P.S. Maybe Hans would comment on SSE 4.1 vs SSE 4a support, I'm curious.

Like i said look at this one and tell me

http://cpuboss.com/cpus/Intel-Core-i3-2 ... II-X2-250e

If we put AMD product of 2012 like this intel is .. what will happen to the Intel .. which works like this OLD AMD ? Scores are identical , but the AMD is 2 years behind the I3 ? So tell me what will happen when we put 2012 AMD to compare with this intel ? I think it will blow out the I3 ......That is why i tell yopu to compare SAME products of different developers... This paper shows that they compare some UNcomparable things .... that operates (for my surprise) THE SAME .

Yes And I do agree with you, at least to some extent. But this is true only for synthethic tests. ST runs and behaves different and to me looks like Intel optimized. Also, you will not find same products of different developers - it's like comparing oranges to apples.

Can anyone help on this topic, Bojcha maybe?