Benchmark G4 vs Intel vs AMD

[gekko513]

I'm a "high-performance-programmer" and recently I decided to benchmark some of the computers I have available using one commonly used critical subroutine for my line of work. Here are the results measured in time to solve the problem:

Apple Powerbook G4 @ 1GHz : 4,17s
Dell Notebook Pentium M @ 1,6GHz : 4,43s
Intel Xeon @ 2GHz : 3,51s
AMD Opteron @ 2GHz : 2,22s

Compared to the G4 this means
188% Opteron@2GHz
118% Xeon@2GHz
100% G4@1GHz
94% PentiumM@1,6GHz

Pentium M is Intel's mobile processor and it's currently at max freq 1,7GHz.

For these kind of problems the G4 is actually not that "old and slow". I just wish I had a G5 to compare to

 

[Dont Hurt Me]

I would say this is a best case scenerio for G4, with a FSB that is stuck at a max of 1.3gbps once that is hit there is no way to feed it more data.
A lot of us here have said a 1.25 G4 is about equal to a 2.0 P4 under ideal conditions. your numbers about show that.
take a dual G4 machine and those 2 cpu's are still fighting over a single FSB that is limited. Bring on the G5.

 

[gekko513]

Originally posted by Dont Hurt Me
I would say this is a best case scenerio for G4, with a FSB that is stuck at a max of 1.3gbps once that is hit there is no way to feed it more data.
A lot of us here have said a 1.25 G4 is about equal to a 2.0 P4 under ideal conditions. your numbers about show that.
take a dual G4 machine and those 2 cpu's are still fighting over a single FSB that is limited. Bring on the G5.

Yes, I think you're right. This is a best case scenario for the G4. The problem is not memory intensive at all, it is mostly computationally intensive. It is also very data-dependent which will lead to pipeline bubbles for processors with long pipelines. (And by data-dependent i mean that instructions often require the result from recent instructions)

 

[Mord]

those are damn good results for a laptop compared to two desktops one being 64 bit both being twice it's clock rate i heard that those centrino laptops beat the powerbooks speed wise.

 

[AMDMACMAN]

The G4 would truely be blown away in a better benchmarking set up. Same ram, hard drive, and what not.

The pentium M, and opteron are the most advanced x86 processors on the market. They should be compared to a g5 not a g4.

 

[varmit]

post the routine

If all possible, post the routine and how to run it. Then people can post what they got on there G5 Macs. That is if you are allowed to post it up. Make a new thread pointing to this one, and also pointing to where we can get the code or program, or what ever we need to replicate it, then people can post what they get. Also, we can post what are PCs can do too.

 

[abhishekit]

Originally posted by AMDMACMAN
The G4 would truely be blown away in a better benchmarking set up. Same ram, hard drive, and what not.

The pentium M, and opteron are the most advanced x86 processors on the market. They should be compared to a g5 not a g4.

opteron...ok.. ..but pentium m is no big deal..it can easily be compared to g4..for g5 intel needs an improved prescott for a respectable loss....

 

[Mav451]

Originally posted by Hector
those are damn good results for a laptop compared to two desktops one being 64 bit both being twice it's clock rate i heard that those centrino laptops beat the powerbooks speed wise.

I seriously doubt that those benches were run in 64-bit mode. Unless you are running beta Windows 64-bit Edition, then it is running in its "native" 32-bit mode. 64-bit has nothing to do with the results at this point. It does however have a lot to do with its lower latencies from its on-die memory controller.

 

[gekko513]

Originally posted by Mav451
I seriously doubt that those benches were run in 64-bit mode. Unless you are running beta Windows 64-bit Edition, then it is running in its "native" 32-bit mode. 64-bit has nothing to do with the results at this point. It does however have a lot to do with its lower latencies from its on-die memory controller.

The Opteron test was run on RedHat Enterprise Linux 3 for AMD64. (So yes it was in 64-bit mode). On the other hand it was a 32 bit program, so it doesn't matter. I could however run a similar 64 bit problem on these same computers if anyone is interested.

 

[gekko513]

Re: post the routine

Originally posted by varmit
If all possible, post the routine and how to run it. Then people can post what they got on there G5 Macs. That is if you are allowed to post it up. Make a new thread pointing to this one, and also pointing to where we can get the code or program, or what ever we need to replicate it, then people can post what they get. Also, we can post what are PCs can do too.

I think I better not, the routine is borderline company confidential. However I'm currently trying to convince my boss to inherit my AMD internet computer, so I can get a G5 Wish me luck

 

[Mav451]

haha oops Gekko. My mistake. Why did i forget about all the other 64 bit OSes out there...

 

[gekko513]

And the results are in

... for the 64-bit version. I've rewritten the program to solve a 64-bit problem, and here are the somewhat surprising results:

Opteron@2GHz: 2,24s
Xeon@2GHz: 11,78s
Pentium4-M@1,6GHz: 16,92s
G4@1GHz: 6,18s

Comparing speed to the G4 again (higher better):

276% Opteron@2GHz
100% G4@1GHz
52% Xeon@2GHz
37% Pentium4-M@1,6GHz

The G4 does even better compared to the Intel competition for 64 bit problems. Why? My educated guess: Because the machinecode to calculate a 64 bit value using 2 or more 32 bit calculations is even more data-dependent than the original problem, causing even larger bubbles in the Intel processors' long pipelines.

The Opteron processor is even faster. This is no surprise considering it's a 64 bit processor and this is a "64-bit problem".

 

[DrBoar]

http://n0cgi.distributed.net/speed/
show in RC-72 the 1.4 GHz G4 to be 3 times as fast as a 3 GHz P4 and twise the speed of any AMD. However that does not stop the P4 and AMDs to spank the G4 in real applications.

So while it is interesting to compare how different CPUs in a set of calulations those results are specific for those calculation not an indication of general performance

 

[gekko513]

Originally posted by DrBoar
http://n0cgi.distributed.net/speed/
show in RC-72 the 1.4 GHz G4 to be 3 times as fast as a 3 GHz P4 and twise the speed of any AMD. However that does not stop the P4 and AMDs to spank the G4 in real applications.

So while it is interesting to compare how different CPUs in a set of calulations those results are specific for those calculation not an indication of general performance

I agree. Testing specific calculations doesn't tell the whole truth about the processors' general performance. It does, however, say something about the processors' potential for tasks in "family" with said calculation.

Real applications does not only include Photoshop, zipping, mpeg-encoding and Unreal Tournament 2006FGX. Real applications does also include diverse scientific simulations and problem solving. Different CPU's can perform very different depending on application in this setting.