Benchmark: Dual G5 vs Athlon 64 vs Itanium 2 vs Pentium 4 vs Power 4 vs Power 3

[Eug]

Originally posted by: DAPUNISHER
Which A64 did they use Eug? I don't keep up with Apple, but it seems to me@2.5ghz they will make a few waves no?

A64 2.0 GHz was used according to the summary. A single G5 2.5 GHz will be very fast, but I don't think it will be faster than an Opteron 2.4 GHz or Xeon 3.4 GHz on average. It will likely be competitive though. IBM may be able to get the 3.0 out by fall though, as Prescott/Athlon speeds ramp up.

Originally posted by: Pariah
Wow, a dual 2 GHz G5 beats up on a 2.6GHz P4 and 2 GHz A64 in number crunching, who'd have thunk it? A dual Opteron 248 or dual 3.2GHz Xeon would have put up a much better showing. Heck, a single FX-51 or P4EE 3.2GHz would have looked a lot better.

If I read this right, the performance of a single 2 GHz G5 beats up on the 2.6 GHz P4 and A64 2.0. (ie. The dual G5 is running 3X as fast as the P4 2.6 for some of the tests, and over twice as fast as the A64 2.0.) But I'll let the others comment since I don't understand the benches.

 

[HokieESM]

Originally posted by: Sunner

Is Intel's Fortran compiler better than ICC?
I've heard many people talking about the shortcomings of ICC in other areas than performance.

Intel's Fortran Compiler is excellent. It is THE defacto Fortran compiler for x86 machines--and BY FAR the fastest for Intel chips themselves. I've used a lot of the different x86 compilers for my research (for all three of the different FE methods I use), and IFC is splendid. But I do adhere to formatted code, and most of my code uses very few libraries (its mostly +-*/).

I've never heard complaints about ICC for speed itself--but it is VERY funny about libraries and "messy" code. Supposedly it is absolutely great if you code in ANSI C.

I'm certainly no expert on the architecture or compiler design--I've just had to "use" a lot of different computers to do my work. So my comments are based on "real-world" benchmarking experience with some finite-element codes--although, they do have different "road blocks"--one is very cache-dependent (loves Itaniums), another is very CPU floating-point performance limited (likes Athlons a lot), and one is very memory bandwidth dependent (loves a P4EE, let me tell ya... between the cache, the good memory bandwidth and the Intel compiler optmizations). I've also run some on VT's "supercomputer" (when it actually was not crashing)... which was a dual G5. The G5 itself wasn't as impressive as the xlf95 compiler--actually the G5 ran like crap with g77--Apple scores a big + for working with IBM on this one.

 

[Pariah]

If I read this right, the performance of a single 2 GHz G5 beats up on the 2.6 GHz P4 and A64 2.0.

Well, if you had read the background information on the tests you would have seen this:

"A graphical representation of the important data is in this chart. The machines tested are

Athlon64: at 2GHz clockspeed, running SuSe Linux 9.0, compiled with ifc 7
G4_667_PB: Apple PowerBook G4 at 667MHz, 1GB RAM, running OSX 10.3 Server, compiled with IBM xlf95 compiler
dG5: Apple dual G5 2GHz, 2.5GB RAM, running OSX 10.3 Server, compiled with IBM xlf95 compiler
Itanium2_1.3GHz: Altix 32 CPU Itanium2 at 1.3GHz, running Linux, compiled with Intel Fortran compiler v8
P4_2.53GHz: Intel P4 at 2.53GHz, running FreeBSD, compiled with Intel Fortran v7 (FreeBSD native executable)
P4_2.6GHz: Intel P4 at 2.6GHz, running Linux 2.6.1, compiled with Intel Fortran v8
PWR3: IBM SP nighthawk-II Power3 System, AIX 5.1, compiled with xlf95
PWR4: IBM Regatta Power4 System, AIX 5.1, compiled with xlf95 "

Notice the "d" in front of the G5 on all the benchmarks? My assumption is that stands for dual.

 

[zephyrprime]

Good point, but he's hand coding Altivec (since xlf doesn't auto-vectorize).

BTW, I know nothing about coding, but I'm just pointing a potential argument.

I do know about programming and your point is valid.

There are so many problems with the bench results here. Nobody should really give them any consideration.
And nobody seems to have mentioned yet that the Power4 is a dual core processor.

 

[zephyrprime]

--but it is VERY funny about libraries and "messy" code. Supposedly it is absolutely great if you code in ANSI C.

Yeah, I've run into some weird problems with icc compatibility also. I had this program where it wouldn't compile until I just rearranged some of the lines of code. I didn't really change them, I just rearranged them slightly. icc is the buggiest compile I've ever used and I've used quite a few. To their credit though, each new version is a little less buggy.

I've never had a problem with their speed either. When they first came out, they were the champs of speed. But with VS 2003, MS was able to match them for speed which is something I never expected out of MS. (but now VS 2004 & ICC 8 are out and I don't know exactly how well they perform.)

 

[Eug]

Originally posted by: Pariah

If I read this right, the performance of a single 2 GHz G5 beats up on the 2.6 GHz P4 and A64 2.0.

Well, if you had read the background information on the tests you would have seen this:

"A graphical representation of the important data is in this chart. The machines tested are

Athlon64: at 2GHz clockspeed, running SuSe Linux 9.0, compiled with ifc 7
G4_667_PB: Apple PowerBook G4 at 667MHz, 1GB RAM, running OSX 10.3 Server, compiled with IBM xlf95 compiler
dG5: Apple dual G5 2GHz

Notice the "d" in front of the G5 on all the benchmarks? My assumption is that stands for dual.

Of course. However, my point was that the scores of the G5 in some of the tests in GL's link were over twice as fast as the Athlon64, which means that a single G5 would still be faster. I will quote myself:

The dual G5 is running 3X as fast as the P4 2.6 for some of the tests, and over twice as fast as the A64 2.0.

 

[Pariah]

Now upgrade the A64 to an FX and add 400MHz, while taking out one of the G5's and what do you think the benchmarks will look like? Quite a pounding by the FX in most of the tests. The dual G5 only did really well in 2 out of 3 of the Lorentz tests, it was pretty much already even or slower against the lower end single A64 in the rest except one aspect of the Solar model.

 

[Eug]

Originally posted by: Pariah
Now upgrade the A64 to an FX and add 400MHz, while taking out one of the G5's and what do you think the benchmarks will look like?

Yes, a 2.4 GHz Opteron or whatever is very fast. In fact I had already so in the thread. I will quote myself again (referencing the 2.5 GHz G5 970FX rumoured to come out soon):

A64 2.0 GHz was used according to the summary. A single G5 2.5 GHz will be very fast, but I don't think it will be faster than an Opteron 2.4 GHz or Xeon 3.4 GHz on average. It will likely be competitive though. IBM may be able to get the 3.0 out by fall though, as Prescott/Athlon speeds ramp up.

 

[n0cmonkey]

I think the tests are fairly valid. Yes, the compilers are different and handle the code to varying degrees. Is this an artificial benchmark, or benchmarks of something he is running in the real world? If it is something he is running in the real world, and just decided to benchmark the various arches, good for him. It shows us how they compare for the work he is doing.

Most of the benches probably wouldn't match up to what most of us do with our computers anyway, so basing a computer purchase off of them would be foolish.