New barefeats article proves G4s sometimes faster, sometimes slower than AMD/Pentiums

[javajedi]

Originally posted by ddtlm
Anyway I've had my fun here for now. I think it is settled that the G4 does poorly at this particular float test. I've done everything I can think of and gone though all sorts of variations of the loop trying to increase the IPC but I could never make significant headway on either the PC or the Mac.

That said, this test is essentialy a test where we do 400000000 double precision square roots which we don't even store and nothing else. There are no memory access, only very predictable branches. I have radically changed the loop and compiler flags and essentially nothing besides the sqrt() makes any difference.

I do not regard this test as important in the overall picture. It does not illustrate anything important to anyone, unless someone sits around doing square roots all day.

I might also add that designing a meaningful benchmark is very hard. I think SPEC is about as good as it gets, and yes the G4 looses in floats there too.

I'm in the process of figuring out vMathLib. I'm a Java guy, so all this Altivec stuff looks totally foreign to me

Never the less, once I get it working, I'll share the results with you folks.

Also: If anyone here wants me to try something, G3 vs G4, or whatever, aside from the square root and integer mult, let me know. I'd actually like to make full featured cocoa app full of test suites.

 

[jefhatfield]

impressive

ya guys lost me way back there, too

hey, do they use aluminum bats in the majors sometimes

 

[ddtlm]

Just passing through... an interesting test would be finding the determinants of large matricies of floats and ints. And I mean finding them by the straightforward stupid computation method, none of the simplification stuff.

Reasons:

1) Too large for all data to be in registers but easily small enough to fit in L1.
2) Takes a long time for surprisingly small matricies (20x20 is a huge number of calculations).
3) Stresses multiples and adds.
4) No massive-yet-trivial compiler simplifications, even for int.
5) The result has meaning.

 

[MacCoaster]

Overclocked my Athlon 1.4GHz Tbird to 1.522GHz, benchmark results (under C#, which was fastest) did an average of 7390 milliseconds (7.39 seconds). w00t!

Gotta love overclocking.

 

[springscansing]

My first post on macrumors! Wahoo!

This is actually my first post. Yay! Been a machead forever (using a IIgs when I was 4).

ANYWAY, regarding various posts about PCs encoding mp3s faster than macs. I am an audio engineer, and I must say the encoding algorithm is MUCH better sounding in iTunes than in Winamp, and I assume most of you are using iTunes in your comparisons. Different programs encode at vastly different rates. For example, I don't know if you recall an application called Soundjam and another called Audiocatalyst. Soundjam encoded 2.4x faster, but sounded like total junk.

Now.. I'm not part of the "MACS IS FASTR" group, because sadly, they aren't... I just wanted to point out the mp3 encoding tests weren't fair.

- Springs

 

[MacCoaster]

Re: My first post on macrumors! Wahoo!

Originally posted by springscansing
Different programs encode at vastly different rates. For example, I don't know if you recall an application called Soundjam and another called Audiocatalyst. Soundjam encoded 2.4x faster, but sounded like total junk.

Hmm? Have you tried to encode them at the same rate, same song, whatever--and documented the results. Would be cool to know.

 

[macwannabe]

Saying that the 2.8GHz P4 is no good because it is based on 25 year old architecture is nonsense as far as I'm concerned.

Can I take it then that you don't think that any of the cars on the market at the moment are worth having or have been improved at all on the grounds that they are based on an 80 year old design? "I don't think that BMW is any good as it is based on a Ford model T", hmmmmmmmm dodgy logic methinks.

 

[javajedi]

Originally posted by macwannabe
Saying that the 2.8GHz P4 is no good because it is based on 25 year old architecture is nonsense as far as I'm concerned.

Can I take it then that you don't think that any of the cars on the market at the moment are worth having or have been improved at all on the grounds that they are based on an 80 year old design? "I don't think that BMW is any good as it is based on a Ford model T", hmmmmmmmm dodgy logic methinks.

You are absolutely 110% correct. We've allready dismissed BackToTheMac's outlandish fallacies though

I think he gets the picture now....

 

[MacCoaster]

Originally posted by javajedi


You are absolutely 110% correct. We've allready dismissed BackToTheMac's outlandish fallacies though

I think he gets the picture now....

Yup. Proven technology. I sure hope he gets the picture.

 

[Sherman]

Yes, you are correct

Another rumor I've heard going around is one of Intels Pentium 5.

We all know about the amazing 4.7Ghz P4 but it's actually not that much faster than a 2.5Ghz P4 because they added so many steps to get up to that clock speed.

In the P5 they tried to stop this troubling trend and found out, they could only get a 1.3Ghz P5, pretty much equal to the G4, without all those extra steps.

Amusing...

 

[ddtlm]

Sherman:

Hmm, not sure where you got that rumor, but it reeks of uninformed "macz rulez!" PC bashing. They did not lengthen the pipeline to get the 4.7ghz P4. The P5, according to conventional wisdom, is the 90nm P4 sporting SSE3, not some totally new chip.

they could only get a 1.3Ghz P5, pretty much equal to the G4, without all those extra steps

Load of crap. Plain and simple. You know there are Pentium 3's available for sale at 1.4ghz, don't you? And lets not even contemplate for fast Athlons are clocking without the P4's super-long pipeline.

 

[springscansing]

Re: Re: My first post on macrumors! Wahoo!

Originally posted by MacCoaster

Hmm? Have you tried to encode them at the same rate, same song, whatever--and documented the results. Would be cool to know.

Yes I have actually. iTunes IS slow, but it's the best. There was an article in MacAddict a few years ago comparing the speeds and quality of different mp3 encoders at the same bitrates.

 

[javajedi]

ddtlm,

I have my theory as to why java took the lead over C in the sqrt example. There is quite a common misconception about Java that it's always slow, and there is a reason for it. Back in the early days prior to 1.2, it wasn't uncommon to see something like we did here run 10,20, or even 30 times slower then C. VM's today (1.4 /w hotspot) are much smarter than they were years ago. IMO, Hotspot makes the conventional "just in time compilers" look like a thing of the past.

Anyways, when you really think about it, Java really has an extra card up it's sleeve. Sure we tell GCC we want max optimizations, (03, etc), but GCC is limited to compile-time optimization. I think since java has adaptive runtime optimizations, specifically hotspot, the runtime optimization is what really makes the difference.

The reason why it's called "HotSpot", is literally because it looks for "hot spots" by profiling on the fly at runtime. Pretty cool, huh? Your first adaptive optimizations kick in second time the loop is ran. Not to mention the conventional JIT optimizations... code will natively compile and so you eliminate the costly overhead of bytecode translations.

Lastly, I am going to do the matrix operation you spoke about, I have to finish up some course work, so I may not get to it tonight, but as soon as I can devote some time to it, I will.

 

[nixd2001]

Originally posted by javajedi
ddtlm,

I have my theory as to why java took the lead over C in the sqrt example....

It might be worth finding C and Java language lawyers as well. ISTR that their treatment of IEEE FP values is different in subtle areas. I can tell you from past experience that these subtle areas are often what hammers performance. I'm talking about treatment for NaNs and that sort of thing. So this may be relevant?

 

[javajedi]

yeah, that's certianly possible. I'm not sure if that is or is not the case, but wouldn't be suprised if it is. I'll find out.

 

[ddtlm]

javajedi:

Lastly, I am going to do the matrix operation you spoke about, I have to finish up some course work, so I may not get to it tonight, but as soon as I can devote some time to it, I will.

Good to see the topic lives on. I thought about doing it yesterday but couldn't decide how I wanted to. I think it should be nonrecursive but honestly I haven't even decided how it can be reasonably done.

Anyways, when you really think about it, Java really has an extra card up it's sleeve. Sure we tell GCC we want max optimizations, (03, etc), but GCC is limited to compile-time optimization. I think since java has adaptive runtime optimizations, specifically hotspot, the runtime optimization is what really makes the difference.

JIT compilers are a mystery to me. I might add that they do exist at least a little for other languages too, read something somewhere about HP using them on their mega-servers for compiled apps. Can't remember details but it was said to help.