Xserve G5-why bother????

[jeff.macaddict]

I was looking at the specs of the Xserve G5, and it supposedly has a peak performance of 9 Gflops.

If I'm not terribly mistaken, the G5 Tower can achieve 80 Gflops.

So why would you want to build a cluster of Xserves? I think Virginia tech would have had to buy something like 9800 Xserve computers.

[kettle]

has to be a typo, G4's were doing that.

[jeff.macaddict]

Look at this! AH HA! I am correct.

http://www.apple.com/xserve/

HERE is what is THERE.

[kettle]

Is Double Precision Floating Point any different than Floating Point?

I hope so.

[CrackedButter]

Originally posted by jeff.macaddict
I was looking at the specs of the Xserve G5, and it supposedly has a peak performance of 9 Gflops.

If I'm not terribly mistaken, the G5 Tower can achieve 80 Gflops.

So why would you want to build a cluster of Xserves? I think Virginia tech would have had to buy something like 9800 Xserve computers.

Are you going to buy an Xserve? How does this deriding of "the" favourite company help you or us?

You sound like a PC user who is happy with his P4 extreme, or something.

[edesignuk]

Originally posted by CrackedButter
Are you going to buy an Xserve? How does this deriding of "the" favourite company help you or us?

You sound like a PC user who is happy with his P4 extreme, or something.
It's actually a good point, and a resonable question http://upload.edesignuk.net/uploaded_data/smilies/smilie_shakeheadfrown.gif There's nothing PC troll like about this at all.

[Mantat]

Double precision allow you to make precise computation when you are for exemple multiplying a big number by a very small one.

So I guess that the G5 is better at the sort of thing than other servers.

And the 80gFlops is a typo obviously....

[Tim Flynn]

I see the G5 XServe compute servers and what Virginia tech is doing as Apple's run and the supercomputer market. As well as other markets.

[jnasato]

Yah, Virginia Tech's cluster is ~10Teraflops=~10,000Gigaflops.
10,000Gigaflops/80Gigaflops=125 PowerMacs; not quite 1100.

[hugemullens]

I want an 80 gigaflop powermac, that'd be awesome.

[Adman_Rob]

From the Apple site:

The 1U (1.75-inch) Xserve comes with your choice of single or dual 2.0GHz G5 processors running at speeds of up to 30 gigaflops, a 1.0GHz frontside bus per processor for up to 8GBps throughput, two full-length 64-bit, PCI-X slots for up to 533MB/s throughput, and up to 8GB of 400MHz RAM with Error Correction Code (ECC). ...


http://www.apple.com/xserve/specs.html

[Rower_CPU]

I'm not seeing any info on the PowerMac pages to indicate how many gigaflops it gets, but if you divide the VT cluster's Tflops by 1100 you get about 9Glops.

Before you crazy cluster performance nuts jump all voer me, yes, I know that performance doesn't scale 1:1 like that, but it's at least an indicator.

[Tim Flynn]

The volume (in^3) of the G5 XServe is about 1/3 the volume of a Power Mac G5.

So Virginia Tech or others following can put 3 times the number of processors in the same space as Virginia tech super cluster. Or looking at it the other way, for similar performance, 1/3 the space.
Note: very simple assumptions :)

[bousozoku]

It might be good to remember that, since the G4, Apple has been telling us about Gigaflops from AltiVec, which only does single precision. They haven't even talked about the floating point unit and I would also be embarrased to do so. The latest G3s have better floating point capabilities.

Apple is talking about the floating point units in the PPC970 and how fast they are but in cases where a certain level of precision is not required, they could get more out of the AltiVec unit in sheer volume.

[jeff.macaddict]

If the G5 doesn't do 80 gflops, which I know it does, then what does it do. Your answer better be higher than 18.3 -- Thats how many the G4 systems crank out.

[jeff.macaddict]

The volume (in^3) of the G5 XServe is about 1/3 the volume of a Power Mac G5.

So Virginia Tech or others following can put 3 times the number of processors in the same space as Virginia tech super cluster. Or looking at it the other way, for similar performance, 1/3 the space.
Note: very simple assumptions


Umm, just remember that three times the computers is:

3X electricity
3X heat
3X noise
3X maintinence
3X repair costs
3X weight (almost)

I think clustering G5 towers is a much better idea than clustering xserve cluster nodes.

By the way, in case you heard otherwise, I HATE windows, and linux, but do like PCs, (personal computers) as long as they are made by Apple, run Mac OS, and bear the stylish Apple logo and are connected to a gorgeous Apple display.:D

[johnnowak]

Stop making up numbers you nucklehead! 80 gigaflops... lol

[johnnowak]

The dual 1.42 G4 was like 9 gigaflops MAX. The dual G5 is about 14 I think.

80... christ.

[Rower_CPU]

jeff-

Your 3x the costs, etc. statement makes little sense in light of the fact that they will also be getting somewhere near 3x the performance.

If you can fit 1100 Xserves in the same space as 366 PowerMacs, the Xserve is the better option.

[jeff.macaddict]

I am have the G4 tower pamphlet (sp?) in front of me, and it says right here that the G4 tower, the 1.25 GHz one, does 18.3 Gigaflops. I also read this in MacAddict magazine, in the back. It was an advertisement for, I think MacMall. 18.3 gigaflops, clear as day. No typo. And my dual 1Ghz does 15 Gflops.:D

[OutThere]

Ok. Stop. This is nuts. Everyone has a different number and everyone is throwing it around.

All these numbers are the results of different tests. If one were to run the same test on a stock Xserve G5 and a Stock PowerMac G5 you could get an idea of what their performance is next to eachother. Finding random numbers and taking them out of context doesn't work. Each test determines the strengths in each category, so it says that "Doing x task, the Xserve can put out y gigaflops" now, in a different task the Powermac would put out some other number.

I think that each computer will perform pretty much alike, and a good way to compare them will be to compare some G5 Xbench scores with the G5 Xserve Xbench scores. Just wait and see.

[jeff.macaddict]

I am seriously, sweat to god, seriously, when I get about $30,000 in my savings account, I am going to spend about $20,000 on a top of the line PM cluster. I fold!

I will get seven computers, and one will be the controller. The other six will number crunch. The main one will be an $8,000 on with maxed RAM, top of line GPU, etc, and hooked up to a Cinema Display.:D

I am serious. Really seriously serious.

[Frohickey]

If you can put 3x the number of XServes into the space of a G5 Tower, then, what is going to happen is NOT that you will have 3x the electricity/heat/etc.

What you will have is 1/3rd the space.