Toggle sidebar Toggle sidebar
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

Xserve FLOPS

Z

Zyniker

macrumors 6502
Original poster
Not exactly certain where to put a hardware-related thread about the Xserve...so, I'll just put it here. 😉

At any rate, does anyone know how many FLOPS the new Nehalem-based servers push? I'm in the process of designing a high-end server cluster and I'd very much like to be able to use Xserves; however, I cannot seem to find any good performance data (I'm looking for numbers here). Any help will be much appreciated. 🙂
 
If you have the physical space, get Mac Pros instead.

They are faster, cheaper, look better and thye can act as servers.
 
ZunePod said:
If you have the physical space, get Mac Pros instead.

They are faster, cheaper, look better and thye can act as servers.
Click to expand...

Space is the major consideration here. Rack-mountable is a must (yes, you can mount Mac Pros; no, I'd rather not). To give you an idea of what I'm designing: think with a minimum of 200 servers in mind. Yes, that's the minimum.
 
For compute nodes I would go for the nehelam xservers.

Have Apple send you a seed machine to test and run the Livermore kernel benchmarks and the stream + flops benchmarks.

I have the source code and I compiled then via icc and gcc. You would be amazed at the performance increase vs GCC when using the intel compilers and vector code.

200 nodes, sounds like a heavy setup. Is this for validating W88/W87 physics- packages?

What is amazing is the power of these desktops, they now beat the Y-MP systems that 15 years ago were running in the millions. A YMP-EL in the early 90's which had a fraction of the power of 1 core cost starting in the 6 digits.
 
hugodrax said:
For compute nodes I would go for the nehelam xservers.

Have Apple send you a seed machine to test and run the Livermore kernel benchmarks and the stream + flops benchmarks.

I have the source code and I compiled then via icc and gcc. You would be amazed at the performance increase vs GCC when using the intel compilers and vector code.

200 nodes, sounds like a heavy setup. Is this for validating W88/W87 physics- packages?

What is amazing is the power of these desktops, they now beat the Y-MP systems that 15 years ago were running in the millions. A YMP-EL in the early 90's which had a fraction of the power of 1 core cost starting in the 6 digits.
Click to expand...

While I won't be the one executing the actual code on this cluster (I'm simply in charge of designing and purchasing), I do know that the primary function will be biological sciences (protein folding, drug analysis, et cetera). Realistically, I imagine this cluster will be called upon to run numerous different programs. (Business forecasts, at the very least.)
Your suggestion of a seed machine for testing/benchmarking sounds like my best option. Thanks for the info. 🙂
 
Zyniker said:
While I won't be the one executing the actual code on this cluster (I'm simply in charge of designing and purchasing), I do know that the primary function will be biological sciences (protein folding, drug analysis, et cetera). Realistically, I imagine this cluster will be called upon to run numerous different programs. (Business forecasts, at the very least.)
Your suggestion of a seed machine for testing/benchmarking sounds like my best option. Thanks for the info. 🙂
Click to expand...

Your welcome. And Apple will send you one, especially if you are considering 200 nodes. Trust me. Apple sent me in 2006 a Mac Pro seed machine. At that time I was using a Dell Precision and a Sun workstaion. I was able to use it for 30 days and at the end it was hard to pry that Mac away from me, I told them to let me use it a little longer while I wait for the Mac Pros to arrive.

Anyhow I switched 🙂 The Mac replaced all the precisions and sun workstations. 1 workstation vs 2.

Vendors will always be willing to send you eval equipment, they need the business.
 
A good place to look at scientific benchmarks is the fluent site. Fluent is a popular CFD code, so it might not be relevant to your particular application but may be better than trying to parse the forum site where many of the discussions are based around video editing and photoshop, neither of which have much relevance. In most of the benchmarks, the Nehalems substantially outpaced the Harpertowns. I put an order this week for 24 Dell R610s (1U, Dual Socket Nehalem, 2.66 GHz, 12 GB of 1333MHz mem). Nodes run about $3500. People are down on Dell but we have around 150 servers that have been running for 4 years now without any issues beyond a hard drive or two. Personally I would prefer to run xserves, but you you can't touch Dellmart's price performance. -G


click on benchmarks --> new

http://www.fluent.com/software/fluent/fl5bench/
 
FLOPS isn't really a very good indication of performance. It's pretty close to meaningless, actually. Just look at the type of Nehalem-based Xeon they're using to get a good idea of how they perform.
 
Zyniker said:
While I won't be the one executing the actual code on this cluster (I'm simply in charge of designing and purchasing), I do know that the primary function will be biological sciences (protein folding, drug analysis, et cetera). Realistically, I imagine this cluster will be called upon to run numerous different programs. (Business forecasts, at the very least.)
Your suggestion of a seed machine for testing/benchmarking sounds like my best option. Thanks for the info. 🙂
Click to expand...

I don't know if you are able to say, but I was just curious where this type of hardware is going. Nothing specific, more like is it a university research center, pharmaceutical, etc. Just curious
 
PowerBookRelic said:
I don't know if you are able to say, but I was just curious where this type of hardware is going. Nothing specific, more like is it a university research center, pharmaceutical, etc. Just curious
Click to expand...

Private research firm. Does a bit of everything. At the moment, biological sciences are the bulk of the work. 😉
 
Zyniker said:
Space is the major consideration here. Rack-mountable is a must (yes, you can mount Mac Pros; no, I'd rather not). To give you an idea of what I'm designing: think with a minimum of 200 servers in mind. Yes, that's the minimum.
Click to expand...

200? Is this a joke?

Can I ask: is this a Linux-based server. Because if it is, I can think of FAR cheaper and better things.

If it wasn't a necessity to be rack mounted, you could have used PS3s.
 
You must log in or register to reply here.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Top Bottom
Back