xServe - awaiting G5 urgently
Sorry for the alarmist headline, but I can't help think that Apple really needs to get a multi-way G5 - which was referred to last week as highly optimised for SMP - into its catalogue as soon as possible.
Why?
Because I just read this article, and nearly died laughing.
The idea that a 4-way Madison-based system will have a raw list cost of nearly $10K just for the processors alone (based on the midrange option) surely has to be an encouragement to Apple to develop systems that can do the same kind of work for a greatly reduced cost.
Reading the sidebar, the following things scream at me.
Now most of us already knew this, but this is a major win for 970 and the whole backwards code compatibility issue. Also not just compatible, but native performance.
I seem to remember Programmer killing my somewhat tortuous "Apple should buy SGI" thread from last year by chucking this information at me.
With the scalable performance of 970 (and hopefully 980 in the future), Apple should go after SGI's market with a vengence, kill it and then pick over the bones.
Build a mid-range server (4 and 8-way) that satisfies the needs of the scientific and CGI community and allow it to be clustered at high speed.
Reading the whole article is a hoot; firstly, the presence of ever-faster Xeons may actually prevent Itanium from developing a critical mass at the desired speed, and towards the end of the article, there's an acknowledgement that the lack of 32-bit compatibility requiring a comprehensive and expensive relicensing programme is an advantage of the Power architecture, which by implication extends to 970 and its successors.
Until last week, I would never have believed that Apple could ship a PC with workstation-level performance for $3000, and that the resulting beast would undercut a Dell box substantially on a like-for-like basis.
Now, Apple should take that damascene conversion to provding value for money and apply it to the server market. Could they ship a 8-way box that kicks the hell out of a Madison-based competitor for less money? I think so - $30,000 before you start sticking disks in and before you start paying MS its blood money means $40,000 to get a working system.
Come on Apple, show us something that'll make even the most jaded IT guy think twice.
Why?
Because I just read this article, and nearly died laughing.
The idea that a 4-way Madison-based system will have a raw list cost of nearly $10K just for the processors alone (based on the midrange option) surely has to be an encouragement to Apple to develop systems that can do the same kind of work for a greatly reduced cost.
Reading the sidebar, the following things scream at me.
Quote:
Runs 64-bit software that is incompatible with 32-bit software used by other Intel chips.
Runs 64-bit software that is incompatible with 32-bit software used by other Intel chips.
Now most of us already knew this, but this is a major win for 970 and the whole backwards code compatibility issue. Also not just compatible, but native performance.
Quote:
Server makers adopting the chip: ....SGI
Server makers adopting the chip: ....SGI
I seem to remember Programmer killing my somewhat tortuous "Apple should buy SGI" thread from last year by chucking this information at me.
With the scalable performance of 970 (and hopefully 980 in the future), Apple should go after SGI's market with a vengence, kill it and then pick over the bones.
Build a mid-range server (4 and 8-way) that satisfies the needs of the scientific and CGI community and allow it to be clustered at high speed.
Reading the whole article is a hoot; firstly, the presence of ever-faster Xeons may actually prevent Itanium from developing a critical mass at the desired speed, and towards the end of the article, there's an acknowledgement that the lack of 32-bit compatibility requiring a comprehensive and expensive relicensing programme is an advantage of the Power architecture, which by implication extends to 970 and its successors.
Until last week, I would never have believed that Apple could ship a PC with workstation-level performance for $3000, and that the resulting beast would undercut a Dell box substantially on a like-for-like basis.
Now, Apple should take that damascene conversion to provding value for money and apply it to the server market. Could they ship a 8-way box that kicks the hell out of a Madison-based competitor for less money? I think so - $30,000 before you start sticking disks in and before you start paying MS its blood money means $40,000 to get a working system.
Come on Apple, show us something that'll make even the most jaded IT guy think twice.
Comments
Originally posted by Mark- Card Carrying FanaticRealist
Because I just read this article, and nearly died laughing.
The idea that a 4-way Madison-based system will have a raw list cost of nearly $10K just for the processors alone (based on the midrange option) surely has to be an encouragement to Apple to develop systems that can do the same kind of work for a greatly reduced cost.
Well, don't forget the more expensive copy of windows. Don't forget that you will have to find appropriate software. 32 bit software just can't be used on such a system. These boxes are for the few people who have the apps that they needed compiled for the Itanium- very few people.
Now most of us already knew this, but this is a major win for 970 and the whole backwards code compatibility issue. Also not just compatible, but native performance.
It is a major win for the Opteron as well.
I seem to remember Programmer killing my somewhat tortuous "Apple should buy SGI" thread from last year by chucking this information at me. With the scalable performance of 970 (and hopefully 980 in the future), Apple should go after SGI's market with a vengence, kill it and then pick over the bones.
SGI doesn't really have many bones left to pick. They are mostly dead.
Build a mid-range server (4 and 8-way) that satisfies the needs of the scientific and CGI community and allow it to be clustered at high speed.
Good idea. Apple would need to recoup the development costs for the Mobo. Survey says that people would be willing to pay $4500 for a 4 way machine with pretty basic specs. Add enough RAM and HD space and you are easily up to $7000 (still cheap!). 8 way would be less of an option because it is too much of a dedicated server machine- there is no option for using the Mobo on desktops and so it would have to be significantly more expensive so that it could recoup the R&D costs.
Reading the whole article is a hoot; firstly, the presence of ever-faster Xeons may actually prevent Itanium from developing a critical mass at the desired speed, and towards the end of the article, there's an acknowledgement that the lack of 32-bit compatibility requiring a comprehensive and expensive relicensing programme is an advantage of the Power architecture, which by implication extends to 970 and its successors.
Yes, Intel's worst cometition is Intel. It is hard for itanium to compete with Xeon in the low end server area. Itanium's only advantage is that it can address more RAM and do 64 bit math. Not very good advantages. Worse still, Opteron is here and this makes Intel's life difficult.
Until last week, I would never have believed that Apple could ship a PC with workstation-level performance for $3000, and that the resulting beast would undercut a Dell box substantially on a like-for-like basis.
Now, Apple should take that damascene conversion to provding value for money and apply it to the server market. Could they ship a 8-way box that kicks the hell out of a Madison-based competitor for less money? I think so - $30,000 before you start sticking disks in and before you start paying MS its blood money means $40,000 to get a working system.
Yes, I think that Apple could ship an 8 way, but they don't need to. Right now, a 4 way G5 would compete well with an 8 way Xeon box at floating point math, but it would absolutely crush it when it comes to the performance/price ratio. I think Apple should do it and leverage the mobo to make an uber pro desktop machine.
Then again, I'm just giddy over the G5 finally being here and I want to see a machine that no desktop PC can get near.
The first thing Apple should do is strengthen its current 1U-high server for its target audience: creative professionals (audio/video/3-D, etc.), scientific users, and education.
Many educational users are probably well served by the current G4 models -- a G4 7457 update should be more than enough for most educational market needs, and could be offered at an affordable price.
Creative professionals and scientific users, on the other hand, really need the number-crunching power of the new G5s. I don't know, however, if anything more than a dual G5 is practical or badly needed right now. The practical problems are heat dissipation, price, and market interest. If Apple can first establish a solid reputation with affordable, powerful single and dual G5 systems in something like the current 1U chassis, then it might be time to get into a bigger, beefier chassis with quad processors.
Without having established its credentials in the server market first, however, I think coming out right away with a quad (or higher) G5 system would be a poor gamble for Apple to make.
One word: fans.
Four words: Big f*cking huge fans.
The xServe is *noisy*... you don't really care about noise in a server environment (trust me, the HVAC systems are so loud in most of them you can't hear yourself think), so they can pull out all the stops and have the fans move massive amounts of air across the mobo.
Not in a PB.
Won't IBM be competing in exactly the same market, using exactly the same chips? That would give IBM all the same advantages as Apple against Intel.
But I would bet that most IT gnomes would by an IBM over an Apple any day.
Am I missing something?
Originally posted by Transcendental Octothorpe
[Edit: I'm talking about 4 and 8 way systems, as Mark initially recommended.]
Won't IBM be competing in exactly the same market, using exactly the same chips? That would give IBM all the same advantages as Apple against Intel.
But I would bet that most IT gnomes would by an IBM over an Apple any day.
Am I missing something?
Besides, it might be a better deal for Apple to leave the really heavy iron to IBM, if they can get IBM to offer OS X as an option for such servers.
Originally posted by Rhumgod
Doesn't anyone else think that if they can build an xServe mobo built on the G5, they can build a PowerBook mobo based on it too? They are nearly the same type of enclosure to cool, and I guess with the massive amount of fans, heat sinks in the new Power Mac that it will be a while before we see them in a small enclosure like the 1U xServe and the PowerBook? Someone, please tell me differently, I am dying for a new PowerBook, just waiting for the iminent next release.
The XServe mobo essentially is the desktop mobo. They are virtually the same.
A powerbook mobo needs to be much smaller, consume rless power, dissipate heat, etc. There is a big difference between a desktop mobo and a laptop mobo.
XServes can have LOUD fans to deal with heat, but laptops can't (or at least, they should not).
Originally posted by shetline
Besides, it might be a better deal for Apple to leave the really heavy iron to IBM, if they can get IBM to offer OS X as an option for such servers.
Not going to happen. IBM has Linux for its entry servers and AIX for its high end servers. Why would they need need OS X? More likely (but still not going to happen) would be Apple making OS X run on some mid range IBM hardware.
Originally posted by Transcendental Octothorpe
[Edit: I'm talking about 4 and 8 way systems, as Mark initially recommended.]
Won't IBM be competing in exactly the same market, using exactly the same chips? That would give IBM all the same advantages as Apple against Intel.
But I would bet that most IT gnomes would by an IBM over an Apple any day.
Am I missing something?
To be honest, you're probably not way off the beam except Apple has the shiny beautiful OS X and IBM, er, doesn't.
The beauty of Apple shipping a 4-way is that starts to fill out the server line and the boxes start to make sense for 200-500 seat companies to run line-of-business applications (i.e those applications that are typically fuelled by the frontline RDBMS systems like Oracle and Sybase), as well as going after render farm and all the standard scientific/visualisation stuff.
The 8-way market is, as has been pointed out, more of a risk. However, the appeal here is to go after the line-of-business apps for the corporate market, those companies for whom Sun would traditionally be the box of choice and for whom Itanium 2 and 64-bit Windows is going to be an attractive prospect.
Apple now has the building blocks to provide not just binary compatibility across the software range, but also consistent 64-bit silicon across everything from the server to the laptop.
For the developer community, either ISV or internal MIS types, that has to be an attractive proposition for reducing development time and cost, as well as reducing TCO across the lifecycle.
It also means that an ISV application can exist all the way from a single user configuration running in a faux client-server mode to an enterprise solution running on a cluster of 8-ways.
IBM with the best will in the world cannot deliver on that promise: 970 is not a desktop/laptop proposition for them, it conflicts too much with their PC business. Therefore they cannot leverage the binary and silicon compatibility through the whole of an installed infrastructure.
I think anything more than 8-way would be dick-swinging to be honest, and better addressed through creating some sort of NUMA architecture using HT or Infiniband.
Seriously this is the way for Apple to start to claw back some brownie points in the corporate world: OS X is slowly maturing and is infinitely preferable to Linux, as well as being more cost effective than Windows, especially as was pointed out earlier in its 64-bit edition.
4-way by 2004, 8-way by 2005; there's an ambition for you.
Originally posted by Yevgeny
The XServe mobo essentially is the desktop mobo. They are virtually the same.
A powerbook mobo needs to be much smaller, consume rless power, dissipate heat, etc. There is a big difference between a desktop mobo and a laptop mobo.
XServes can have LOUD fans to deal with heat, but laptops can't (or at least, they should not).
I was well aware of the loudness that would be acceptable in an xServe. I guess I was referring to a lower MHz processor (1.0, 1.2GHz G5). This is the internal xServe:
There are no external fans other than on the power supply, and the two internals are mainly for the (up to) 4 hard drives. You certainly wouldn't need that in a PowerBook. One HD one CPU and one fan. Anyone think that could happen inside a PowerBook soon? I sure do. Mobo designing with the additional HT and various other components could be the one thing holding up a G5 PowerBook. Seeing as how a 1.2GHz PPC970 consumes 19w according to IBM and the 1.0GHz G4 7455 consumes 20w according to Motorola, it doesn't seem like CPU heat is the main issue.
Dobby.
Originally posted by dobby
2U and 2PSU as well. We opted for 10 SunFire V240's over 10 Xserves as we needed fault tolerance. The Xserve was perfect for the job (especially price) but we needed that extra PSU.
Dobby.
And that's a good point!
The xServe is a nice first effort, in terms of price and packaging, but it cannot be taken seriously in a real line-of-business or 24x7 function until it has concepts like redundant PSUs or bullet-proof remote management.
And these concepts also applies to the RAID product.
It's not enough for Apple to say: "We use xServe in our business, and these concepts are not an issue". If you're a successful company with an IT setup that works, you're not going to swap things like the Remote Insight functionality of a Proliant or the N+1 engineering of an IBM x440 for a box that can't do the same things.
I'm holding out for 3U, 4-way, dual PSU, dual gigabit, with a big hole where a SDLTxxx or an AIT-x would fit.
Also, I wouldn't be surprised if we saw an acknowledgement of storage over IP (SoIP) at some point in the near future.
The search for perfection shall continue though and I'm sure the next Xservers will have more variety and have some really BIG servers with hardcore features.
I wonder how many Xserves Apple is selling as servers? And what they're being used for?
Originally posted by Rhumgod
There are no external fans other than on the power supply, and the two internals are mainly for the (up to) 4 hard drives.
Nope.
One blower (the bottom one in your pic) goes straight over the top of the dual CPUs.
The other blower goes over PCI/AGP slot above in the event that you put a hot card in there (GPU, etc).
The HDs need cooling, but not that much. The real problem is that there is little air intake/output room so you have so shove air through in a massive way. Plus, if you've got a Dell dual Xeon straight above and below the box (as is the case with mine) then you can't depend on any conductive/convective cooling like you can with a laptop.
This is the real reason why a G5 Xserve is a challenge. The 4 HD bays obstruct airflow so the blowers are going to have be unbelievably loud to cool 2 95W 970s. The Xserve cluster should be much easier to pull off for this reason.
Originally posted by Aquatic
I wonder how many Xserves Apple is selling as servers? And what they're being used for?
I have one. Dual 1GHz w 2 120GB drives. Nice machine.
It's being used as a workgroup server. 12 users boot off of it, about 1GB of databases being served off of there. Website, a few other little things.
It's a bit more than I needed but I wanted a server that would last a while. We don't need the CPU as much as we needed the storage. I have no complaints about the speed or the software aside from the shitty RAID set rebuilding. Hoping that's fixed in 10.3. I have a few problems with Workgroup Mgr, but I suspect it's an operator problem rather than a software problem. Either way, it could stand some improvement.
It's been running for 10 months and I've never had an OS X service go down. I've not had to reboot it except for OS updates. I'm very happy that I got it.
I say we'll see an Xmas Xserve
Originally posted by g::masta
surely Apple will wait for Panther Server to be rock solid before loading anything remotely hot into an Xserve? 10.3 Server HAS to be solid solid solid before they can risk releasing a G5-based Xserve AND new software at the same time.
I say we'll see an Xmas Xserve
Why? Apple released the original Xserve under 10.1, which was far from rock solid. I actually held off buying mine until they shipped preloaded with 10.2, but units went out for some time with 10.1.
I'd be fine with them shipping 10.2.7. Now, you might not want to rely on that combo, but I don't see why they shouldn't ship them.
Originally posted by Mark- Card Carrying FanaticRealist
And that's a good point!
The xServe is a nice first effort, in terms of price and packaging, but it cannot be taken seriously in a real line-of-business or 24x7 function until it has concepts like redundant PSUs or bullet-proof remote management.
Cluster them and you have full redundancy
We'll see G5 Xserves before the end of the year, and before the Powerbooks. But remember that Apple isn't going to step on IBM's toes by building 8 way Xserves. It would be silly for Apple to do that anyways, because in the big iron market service contracts are as important as the hardware, and why would Apple want to go head to with IBM, the creator of Neo, Apple's savior?
I'd rather see Apple focus on the low end server market, and supply IBM with software for the high end market, in the form of OS X Server.