World Exclusive Preview of IWill's ZMAXdp

[DR]

Dual Opteron in small form factor works, and it works well. I've left off talking about the noise the system made while under heavy load until this conclusion, since it's the noise I maybe was most impressed with. With low-power Opterons, CPU temperature never rose over 48C under load, and with a BIOS SmartFan threshold of 58C, the fans never made more than a half-hearted whisper. The ZMAXdp, with those CPUs, is nearly silent.

It's disgustingly well engineered and it works incredibly well. Just think about the implications of all that performance in such a small box for a second or two. It stretches much further than being a nice toy for the well-heeled Photoshop addict.

http://www.hexus.net/content/reviews...lld19JRD04NzY=

[Steve]

Just imagine a beowulf cluster of those!

/slashdot

Great article Rys - my wallet is looking nervous.

[Gordy]

Looks like shuttle have some proper competition in the sff world at last

[turkster]

Yeah nice review .

That is one compact box of tricks. Pretty impresive .

[Rexticle]

Dimensions?

[Rys]

Like the units displayed in Akihabara, Japan, recently, the chassis colour of the preview unit is a deep red. In terms of dimensions, the main unit is roughly 12.5 inches deep, 8.25 inches wide, 7.5 inches tall, and as you'll see soon, there's an extra 1.5 inches of protruding PSU assembly at the rear.

Rys

[directhex]

Just imagine a beowulf cluster of those!

http://space-simulator.lanl.gov/photos/ss1.jpg ?

one presumes that's part of the idea, but proper cooling would be nightmarish

[Butuz]

As would proper managment what with no managbility features at all in a shuttle. That shuttle supercomputer was a total waste of money, you could have got more performance per square inch out of some dell rack mount servers, plus the servers email/txt you if one of their fans fail. Imagine working out which shuttle fan has failed in that lot?

Wouldnt be suprised if the guy/gal who had that bright idea gets sacked in the near future.

@ Iwill. Mmmmmmm i think i have found my next PC upgrade. Bye Bye SN45G, Y'hellow Dual opteron goodness!!!

Butuz

[directhex]

As would proper managment what with no managbility features at all in a shuttle. That shuttle supercomputer was a total waste of money, you could have got more performance per square inch out of some dell rack mount servers, plus the servers email/txt you if one of their fans fail. Imagine working out which shuttle fan has failed in that lot?

actually i disagree. what you lose with the commodity machines is snmp traps directly on the hardware, a software solution at the management node is pretty trivial - in many cases at work we've found it easier to use the consumer-level solutions to problems (such as basic ssh-based probing machines to get SMART data) than trying to work around snmp - and even then, if you're collating your syslogd logs on the head node, scanning & checking them... the money saved (or the extra computing power obtained) with a big-ass beowulf cluster compared to a more traditional rack hooked up by myrinet is huge. if it weren't viable, then there wouldn't be such a push in recent times towards beowulfs for things, and the shuttle simulator is just one of a number of high-profile machines (remember that big apple G5 one at Virginia Tech?)

the only real concern is over system support - i.e. if something fails, then are iwill gonna have someone out to fix it by the following day, the way you'd get with a sunfire or xseries? no - and that (coupled with earlier heat concerns) is the big problem.

[turkster]

one presumes that's part of the idea, but proper cooling would be nightmarish

Well it depends on how complex a cooling sytem your thinking of. Yeah if you're gonna have to plumb up the processors to some kind of external cooling system but the Liebert XD high density rack cooling solutions used at the Virginia Tech Xserver cluster would surely be able to cope with shelf of ZMAXdps. If you were building a large cluster with ZMAXdps you'd probably have to build your own racks to hold them and could integrate cooling sytems into the racks.

[directhex]

Well it depends on how complex a cooling sytem your thinking of. Yeah if you're gonna have to plumb up the processors to some kind of external cooling system but the Liebert XD high density rack cooling solutions used at the Virginia Tech Xserver cluster would surely be able to cope with shelf of ZMAXdps. If you were building a large cluster with ZMAXdps you'd probably have to build your own racks to hold them and could integrate cooling sytems into the racks.

it's not so much a question of having the cooling, it's a question of it cooling all the right places. the xserve is designed to sit in a rack environment (i.e. take in air-conditioned freezing air from the front, vent boiling hot air from the back), and the racks are designed to get rid of the hot air sharpish.

even with rack kit the heat can be an issue if the hot air isn't pulled out of the rack fast enough (and allowed to recirculate). for something like the zmaxdp, which doesn't have apple's gaping ventilation holes, you'd need to completely alter your rack cooling system - or just not bother

[Butuz]

Ive just re read that preview. Excellent article as always Rys.

That is my next PC purchase for sure.

Rys - you still got it? Can you run seti classic benchmark for me?

Butuz

[Apex]

Can i have one please bob !

/me starts to drool.....

Right now for my thoughts on certain things.

Firewire support, this is a big let down imho 1 port...is that it ?

The heatpipes do not have a shround on them thus imho this is a poor design.

Air will take the path of least ressitance so would it not make sense to force the air threw the heatpipes instead of allowing it to go threw the gap?



The sata external port, while a good idea, is poorly implemented.

Other then a few other niggles this does look like it might get purchessed over the amd64 rig i was going to get......time will tell

[DR]

The SATA rear ports are going to change, there is also going to be a black one.

Don't forget with thermals you need to get air throught the case.

IMHO it is better than the old Shuttle ICE system.

[spikegifted]

Nice little (p)review... I particularly like the Auto Gordian Knot benchmark - reliable multithreading realworld apps for benchmarking are difficult to find.

As a SMP freak, I think this boxen will eventually ended up in my collection of rigs, one way or another.

Just a curiousity, is there a particular reason why the Iwill box (nForce) was equipped with a pair of 244 and it is benchmarked against a pair of 250 on a different platform (AMD)? This doesn't really allow the platforms to be judge on a same footing...

[DR]

Because they are low power units, the price point of 244sLP are much higher than that of a normal 244.

Also it was for comparison - we will have it soon with 2 * 250s