PDA

View Full Version : NASA Benchmarks G5


Catfish_Man
Jul 2, 2003, 02:01 PM
http://members.cox.net/craig.hunter/g5/

I'm fairly impressed. The G5 whoops arse on the multithreaded benchmarks and does respectably on the single threaded one (especially given the state of FORTRAN on the Mac).

KentuckyApple
Jul 2, 2003, 02:10 PM
Very interesting. Am I correct in thinking that the G5 benchmarks are based on the use of only one processor? If so, that machine would have really spanked the P4!

Pete_Hoover
Jul 2, 2003, 02:31 PM
You can't deny the benchmarks when NASA does them. I'd say that is a pretty reliable source. :)

Catfish_Man
Jul 2, 2003, 02:36 PM
Originally posted by KentuckyApple
Very interesting. Am I correct in thinking that the G5 benchmarks are based on the use of only one processor? If so, that machine would have really spanked the P4!

Indeed. Optimized for the G4 and running on only one processor.

Also, this benchmark has a working set of 1MB. That means two things:

1) it doesn't stress the system bus (so it doesn't use the G5's biggest strong point)
2) it will go much faster on the next gen P4 (P5?) and G3, because they have 1MB L2 caches.

iEric
Jul 2, 2003, 02:48 PM
Wow thats pretty cool.

I never knew NASA did benchmarks for computers.
(Dont they use PCs there?)

Catfish_Man
Jul 2, 2003, 02:54 PM
Originally posted by iEric
Wow thats pretty cool.

I never knew NASA did benchmarks for computers.
(Dont they use PCs there?)

Ummmm... yeah... they also use multiple CRAYs, a cluster or two of Macs, etc... 'PCs' as a group (including Macs) are rather inadequate for NASA's needs unless networked in very large groups. This benchmark is testing the performance of the G5 for some of the simulation code they use. They did a similar one on the G4, discovering (surprise, surprise) that it sucked badly at non-vector code, and blew everything out of the water at vector code.

szark
Jul 2, 2003, 02:56 PM
G4 (dual 1Ghz Xserve)
hw.machine = Power Macintosh
hw.model = RackMac1,1

I think I like the name "RackMac" better than Xserve. ;)

Ensoniq
Jul 2, 2003, 03:08 PM
I can see it now...

The NASA benchmarking will be criticised because they did not re-write their code to use SSE2 which "would have kicked Altivec's ass" I'm sure some will claim. :)

It's very interesting to see that the G4 and the P4 are pretty much identical in performance per clock. So yes...the P4 has twice the MHz now (P4 at 3 GHz, G4 at 1.42 GHz) so if you're not using AltiVec, the G4 may not be the best choice.

But the G5 definitely has caught up, if not completely surpassed the CURRENTLY SHIPPING competition. And if we really get to 3 GHz over the next year, Apple has nothing to be ashamed of anymore.

Abstract
Jul 2, 2003, 03:13 PM
An Evaluation of PowerMac G5 Systems for Computational Fluid Dynamics Applications

Oh, computational fluid dynamics. Of course. For that type of computation, the G5 will definitely be superior to the P4.

Okay, I don't know what I'm talking about. Kill me now. :(

PS: NASA.....should they be trusted? ;)

Catfish_Man
Jul 2, 2003, 03:17 PM
In response to the 2 posts above:

1) SSE/SSE2 is mentioned. They turned off the related compiler options because it LOWERED performance to have it on (???)

2) Yes, fluid dynamics is something the G5 should be good at. Lots of floating point math. The only better thing could be if it had a lot of memory traffic (although the P4's bus is almost as fast as the G5's).

destroyboredom
Jul 2, 2003, 05:56 PM
If you look at additional notes it says The g5 was running OS X 10.2.7?? Am i missing something? I though it only went to 10.2.6

ddtlm
Jul 2, 2003, 06:26 PM
Its worth noting that the dataset fit entirely into the G4's L3 cache but not into the G5's L2, yet the G5 still performed well.

yzedf
Jul 2, 2003, 07:24 PM
Considering the lack of NASA's ability to know the difference between meters and feet... :rolleyes:

Cubeboy
Jul 2, 2003, 08:39 PM
Hmm interesting, I do however wonder why they used 333Mhz PC2700 DDR SDRAM for the 2.66 GHz P4 system as it bottlenecks the Pentium 4's system bus and would probably lower performance by some amount over a RDRAM or Dual Channel DDR solution. This would also indicate that they were using a i845 motherboard which definitely isn't the best choice for performance in these types of applications. Which leads me to wonder about their estimations on the 3.2 GHz P4 as they were only assuming an linear performance increase due to clock rate, when the 3.2 GHz P4 utilizes a significantly faster FSB, superior motherboard (and this is compared to a i850e which itself performs much better on benchmarks than a i845 on a given system), faster ram+no memory bottlenecks, as well as some minor core revisions passed on from the 2.8/3.06 GHz Pentium 4s that on average improve performance about 4%.

Still, the G5 Powermac performs admirably and their is little doubt that with a good compiler, it can achieve parity or surpass the fastest Pentium 4 in single threaded Jet 3D.

Off topic but isn't Portland the company thats writing the compiler optimized for the Athlon 64/Opteron? It looks promising from preliminary benchmarks.

ddtlm
Jul 2, 2003, 08:44 PM
Cubeboy:

I don't pay a lot of attention to P4 chipsets, but I was under the impression that there was at least one supporting dual channel DDR-333 (well, there was a dual DDR-266 and a dual DDR-400 anyway).

MorganX
Jul 2, 2003, 09:16 PM
Originally posted by Ensoniq
But the G5 definitely has caught up, if not completely surpassed the CURRENTLY SHIPPING competition.

Hey come on now play fair. The G5 isn't "currently shipping" just yet itself. ;)

Cubeboy
Jul 3, 2003, 07:45 AM
Originally posted by ddtlm
Cubeboy:

I don't pay a lot of attention to P4 chipsets, but I was under the impression that there was at least one supporting dual channel DDR-333 (well, there was a dual DDR-266 and a dual DDR-400 anyway).

I assume that they used single channel DDR as any RAM thats Dual Channel would be listed something like DDR333 SDRAM, CL2 to signify that it's Dual Channel.

Jaykay
Jul 3, 2003, 08:06 AM
Originally posted by destroyboredom
If you look at additional notes it says The g5 was running OS X 10.2.7?? Am i missing something? I though it only went to 10.2.6

Yeah, thats smeagol isnt it? They need an optomised version of OSX for the G5 processor before 10.3.

solvs
Jul 5, 2003, 12:22 AM
Originally posted by Cubeboy
I assume that they used single channel DDR as any RAM thats Dual Channel would be listed something like DDR333 SDRAM, CL2 to signify that it's Dual Channel.

You do know that CL2 means Cass Latency 2.0 right? Dual Channel RAM is just called Dual Channel. Although there is Dual Channel RAM, some single channel RAM will work in Dual Channel motherboards. It's all about the Mobos.

MorganX quote:
-----------------------------------------------------
Originally posted by Ensoniq
But the G5 definitely has caught up, if not completely surpassed the CURRENTLY SHIPPING competition.
-----------------------------------------------------

Hey come on now play fair. The G5 isn't "currently shipping" just yet itself.

I think that's what he meant.

Cubeboy
Jul 5, 2003, 07:36 AM
Originally posted by solvs
You do know that CL2 means Cass Latency 2.0 right? Dual Channel RAM is just called Dual Channel. Although there is Dual Channel RAM, some single channel RAM will work in Dual Channel motherboards. It's all about the Mobos.

Ahhh, oops, my mistake, thats what happens when you pay attention to someone with as short of a memory span as myself. ;)

Sabenth
Jul 5, 2003, 07:49 AM
I am not one for taking bench tests but got to say that this looks impressive only one problem with it makes no sense to me ha ha ha..


THEN AGAIN CANT SEE A FLOATING POINT IF IT FLOATED BY ME LOL

the g5 is fast we know that its a great chip and its going to make a big diffrance to us all hey presto even floating point nutters will be happy

solvs
Jul 5, 2003, 09:09 AM
You know, after reading some of that Haxial.com guy's comments, I tend to see his point. Some of it IS smoke and mirrors. Not too long ago I was one of many complaining about Apple's price/performance/features. Surprised I wasn't called a troll.

However, some of his comments were kinda stupid. $2999 is only $1 cheaper than $3000. :rolleyes: 3.2GHz P4s JUST coming out - of course Apple didn't use them. And the fact that he seemed to take other benchmarks at face value, versus Apple's. Which he attacked, I think, mostly because Apple is very blatant about their Worlds First/Worlds Fastest. But this IS just marketing. Benchmarks mean very little.

In real world performance Apples may be faster. They may not be. They may be better. Or not. It doesn't really matter. If you play certain games or use certain programs, Wintel may be a better option. Sometimes Linux is... for certain things. For those who want to do audio/video/image editing, I feel Macs are superior.

I've used Windows to do this stuff, and it's been painful. Problems abound. Others may have had better luck, but many who've tried both would agree with me.

What's my point? I forget.

But benchmarks, shmenchmarks. The G5 is a good processor. Especially for what some people (in this case this NASA guy) do. And that's good because it's good to have choices and competition. Apple seems to be on the right track at least. Yeah Apple.

I love Macs... I tolerate PCs (barely).

SteveG4Cube
Jul 5, 2003, 09:50 AM
Is this the same benchmark test NASA used to fake the Apollo moon landing? :D

iJon
Jul 5, 2003, 09:59 AM
one of slashdot guys pointed this out. how come these are some guys homepage and not on nasa's site?

iJon

P-Worm
Jul 5, 2003, 10:58 AM
I'm not sure if NASA wants this as 'official.' To me, it just seems that a small group of guys got together and wanted to test the latest and greatest.

P-Worm