Rumor Wrapup - Rackmounts and Beyond

[mc68k]

Originally posted by Zenith


I believe the PCI bus is 64-bit, 33 Mhz. That means 266 MBps bandwith.

Here are the specs from Apple:

I/O connectivity
Two full-length 64-bit, 66MHz PCI slots (lower slot filled with PCI graphics card in standard configurations); supports 3.3V 32-bit or 64-bit PCI cards running at 33MHz or 66MHz

One half-length 32-bit PCI/AGP combo slot with one of the following:
— Secondary Gigabit Ethernet card in standard configurations

— AGP 4X graphics card (build-to-order option)

So barring the combo slot, you would have 64-bit, 66MHz PCI slots which would effectively have a theoretical peak of 528MB/s bandwith. But that's just for full-length PCI cards, which are only things like Ultra SCSI controllers, Gigabit cards, capture cards, etc.

 

[cryptochrome]

Did anyone hear about using these servers for scientific and media computing? While the storage is good the processing capabilities leave a bit to be desired (Two 1Ghz G4s? They had that already). It seems to me that the current G4 architecture is holding things up a bit.

With high-speed multiprocessor G5s, and a new mobo architecture based on hypertransport (both of which have been predicted for a long time) you have some serious processing power in one unit. But, that's neither here nor there yet. The real potential could be in a version of OS X for regular macs (and one for the server) that supports one-preference-panel-configured processing and storage sharing over a wide variety of methods (including internet, lans, wans, and firewire). A no-hassle distributed computing solution for the masses. It sounds like Rendezvous could help lead to that.

 

[Catfish_Man]

The G4...

Originally posted by cryptochrome
Did anyone hear about using these servers for scientific and media computing? While the storage is good the processing capabilities leave a bit to be desired (Two 1Ghz G4s? They had that already). It seems to me that the current G4 architecture is holding things up a bit.

With high-speed multiprocessor G5s, and a new mobo architecture based on hypertransport (both of which have been predicted for a long time) you have some serious processing power in one unit. But, that's neither here nor there yet. The real potential could be in a version of OS X for regular macs (and one for the server) that supports one-preference-panel-configured processing and storage sharing over a wide variety of methods (including internet, lans, wans, and firewire). A no-hassle distributed computing solution for the masses. It sounds like Rendezvous could help lead to that.

...is not the problem. The problem is the MPX bus. Dual 1GHz G4s is PLENTY of power for a 1U, but they're nowhere close to their potential without full DDR support (MPX+, RIO, or integrated DDR controller).

 

[ftaok]

Re: The G4...

Originally posted by Catfish_Man


...is not the problem. The problem is the MPX bus. Dual 1GHz G4s is PLENTY of power for a 1U, but they're nowhere close to their potential without full DDR support (MPX+, RIO, or integrated DDR controller).

So is the lack of MPX+ a Motorola issue or an Apple one? I'm under the impression that Apple makes the motherboards. Would the G4 need major tweaking to get it to work on a MPZ+ bus, or will we have to wait for the G5 (or how about the 7470)?

 

[jacka55]

How about thin clients?

I was thinking. Now Apple has these rack mounted servers, there'll also be a raid array supporting terabytes of storage, and then couple that with Mac OS X NetBoot, there would be a real potential for a veritable corporate thin-client network workhorse. Apple could release a sub-sub $1000 thin-client (Maybe even a G3: Keyword=Inexpensive), without a hard drive of course, and BAM! What more could a office-type corporation want? Many of them are still haven't been able to break free from the good ol' UNIX terminals. It just hit me, that with the new network hardware, everything's there for a corporate network...except the the thin-client.

 

[Wry Cooter]

Re: How about thin clients?

Originally posted by jacka55
I was thinking. Now Apple has these rack mounted servers, there'll also be a raid array supporting terabytes of storage, and then couple that with Mac OS X NetBoot, there would be a real potential for a veritable corporate thin-client network workhorse. Apple could release a sub-sub $1000 thin-client (Maybe even a G3: Keyword=Inexpensive), without a hard drive of course, and BAM! What more could a office-type corporation want? Many of them are still haven't been able to break free from the good ol' UNIX terminals. It just hit me, that with the new network hardware, everything's there for a corporate network...except the the thin-client.

Its certainly set up better for that now that the server is here. Even educational markets might be highly interested in this for several reasons, cost being one of them. Not sure how it might rely on changes in the current infrastructure of software licensing. And there is always some perceived need for some local storage. Previous resistance to thin client paradigms still exist.

 

[Onyxx]

the bus

the bus is 133. see here if you don't believe me. http://www.apple.com/xserve/specs.html

Did anyone take a look at the powerdraw of these little beauties? 125 watts! now this may not strike most as significant, but compare the 125 watt draw to say dell's 1u server (either the pitiful celeron model or the p4 version) and you halve your energy bill (at least halve) for the year. The servers will pay for themselves before long and since maintainence is, well, "mac-like" (aka "virtually non existant") it makes it even better.

As to the comment about scientific/media computing, these things are perfect. Many scietific apps allow you to run the process over a number of computers. This is called clustering and many universites and scientific institutions have been using this method for years. Now consider the power draw of your average tower at about 350 watts (remember you want a lot of towers so power consumption can get rather high) and the 125 - 200 watt draw of these servers there is a big advantage. And since you could fit almost 4 servers into the same space that 1 tower occupies, these servers make clustering even more feasible.

As for media, cg companies do the same thing. The way it works is that the artist creates a scene/animation/etc. and sends it out to the servers to render while they work on something else. Many 3d apps let you cluster using plugins or an extension of that program. One great example is cinema 4d and cinema 4d net. Combine the multithreading/processing capabilites of C4d net with the STANDARD gig-e and a gig-e hub, and you have instant renderfarm.

sun servers can cost a heck of a lot of money and they draw a heck of a lot of power too. Well now you can get a "machine" with 90 processors, up to 21.6 tb's of storage, and 90 gigs of ram for under 250k!! take a look at the sun line and show me a server that will match that for a render farm. And when you want to get rid of the sever and get a new g6 system (he he), instead of having to hide a hulking 7 foot high "refrigerator" in the company basement, you can split the thing up into the 45 individual servers and see who wants an "outdated" machine.

the possibilites are endless. Keep it up apple.