"Mac Pro Label" on the new Mac Pro 2013?

thekev said:

Solutions not directly supported by Apple. They had to guide them toward something.

When was the last time they even updated that?

I wonder how it compares with the new ones. For most server duties the configuration wouldn't be very helpful. The dual gpu thing would be useful where gpu computation is used. If we're talking about a data center that just slicese these up into VMs, that makes zero sense.

Click to expand...

That's why I said such a cluster (of any considerable size) doesn't make any sense - like, at all. It's also why I said so much depends on the base price.

More than obviously anyone even slightly informed as to what's commercially available can customize nodes for very little money. Whether they're interested in multiple GPUs or CPU cores setting up or having set up, custom boxes can come in at around a half to one third the cost of what I guess Apple will ask for the MP6,1. This might not be a tremendous difference in small clusters of 8 or less but with 240 or even a quarter that many it just makes no sense at all.

One would have to be a complete idiotic buffoon unable to even add or subtract in order to consider clusters of MacPro boxes for cluster computing; rendering or whatever. Unless one is getting special deals from Apple (like Pixar might) such configurations are as I already said, ludicrous! How this discussion has even continued this far defies rational thinking.

On the other hand if Apple pulls a rabbit out of hat (about as likely as flying butt-monkeys IMO) and offers 12 fast cores and/or 2 pro grade GPUs for less than or about the same as other solutions then I could be wrong. And they have done this in the past with the 2006 machines - coming in at just slightly less than one could piece together a DIY of identical specification. The trouble with accepting this as a predictor is the current state of the marketplace. They would essentially have to offer a 12 core 3.5GHz at under $2.5k to make any sense to cluster customers. And if they intend on dual "workstation grade" GPUs that's not going to be possible.

I dunno the prices of systems capable of supporting 4, 6, or 8 GPUs but I'd be willing to bet the same thing there. More than likely the MP6,1 can NOT be priced competitively for either customer - and given the proprietary nature of the new design there won't be any end-around system mods to equalize the differences either. For individual workstation solutions the MP6,1 makes sense. For small clusters of 2 to 8 machines it MAY also make sense (doubtful really but some people may not mind losing $2k to $16K). But for sizable clustering no one on the planet is stupid enough to toss out an extra hundred-thousand dollars or two just to have the Apple logo on - and that's what they would be doing.

Tesselator said:

On the other hand if Apple pulls a rabbit out of hat (about as likely as flying butt-monkeys IMO) and offers 12 fast cores and/or 2 pro grade GPUs for less than or about the same as other solutions then I could be wrong. And they have done this in the past with the 2006 machines - coming in at just slightly less than one could piece together a DIY of identical specification. The trouble with accepting this as a predictor is the current state of the marketplace. They would essentially have to offer a 12 core 3.5GHz at under $2.5k to make any sense to cluster customers. And if they intend on dual "workstation grade" GPUs that's not going to be possible.

I dunno the prices of systems capable of supporting 4, 6, or 8 GPUs but I'd be willing to bet the same thing there. More than likely the MP6,1 can NOT be priced competitively for either customer - and given the proprietary nature of the new design there won't be any end-around system mods to equalize the differences either. For individual workstation solutions the MP6,1 makes sense. For small clusters of 2 to 8 machines it MAY also make sense (doubtful really but some people may not mind losing $2k to $16K). But for sizable clustering no one on the planet is stupid enough to toss out an extra hundred-thousand dollars or two just to have the Apple logo on - and that's what they would be doing.

Click to expand...

I'm quite aware of that. It should be recognizable by the simple fact that shops with both mac pros and render farms do not build render farms from Macs. It wasn't just the 1,1. The 3,1 was also extremely competitive. 2009-2012 the lower models have basically offered a box that could be customized further. I know you mentioned cheap external boxes. I never had much luck with low end eSATA port multipliers. Some chipsets could be really flaky. Perhaps usb3 is better given the lack of host card requirements. There's no good way to attach SAS/mini SAS das though.

I don't know that I really expect a price increase here. Similar pricing seems likely, but if it's a $2500-3k box that is outperformed slightly by the imac due to low end firepro cards based on older chips, there will be nothing to make that configuration appear desirable. The older one is quiet, handles additional types of storage, and can accept after market gpus. Taking that away may strain volume if they don't build significantly more value into the lower models.

thekev said:

I'm quite aware of that. It should be recognizable by the simple fact that shops with both mac pros and render farms do not build render farms from Macs. It wasn't just the 1,1. The 3,1 was also extremely competitive. 2009-2012 the lower models have basically offered a box that could be customized further. I know you mentioned cheap external boxes. I never had much luck with low end eSATA port multipliers. Some chipsets could be really flaky. Perhaps usb3 is better given the lack of host card requirements. There's no good way to attach SAS/mini SAS das though.

I don't know that I really expect a price increase here. Similar pricing seems likely, but if it's a $2500-3k box that is outperformed slightly by the imac due to low end firepro cards based on older chips, there will be nothing to make that configuration appear desirable. The older one is quiet, handles additional types of storage, and can accept after market gpus. Taking that away may strain volume if they don't build significantly more value into the lower models.

Click to expand...

I'm confused. Am I in the correct thread here?

Render nodes either GPU or CPU don't need much storage. 1TB would be overkill and a half.

For CPU nodes a headless system with lots of RAM and a fast nic is what's needed. Two or three hundred gigs is typically plenty and headless means no GPU is needed at all. RAM amount is determined by max project size.

For GPU rendering all that's needed is maybe 1x PCIe v3 slots and enough power for the number of cards you're running. Likely even a dual core CPU running at 3Ghz would be enough and system RAM probably isn't that critical. Likely also storage of 1TB would be way-enough and again a fast nic gets well used. I don't know that much about GPU rendering and I may be wrong about the amount of RAM.

Either way SATA and USB revisions aren't important and storage amounts have little impact for these purposes. Typically when running a cluster configuration there's a file server, a render controller, and the cluster nodes. The file server and the render controller can be the same machine in many cases. The render controller sends the jobs to the cluster nodes, which are instructed to save their results onto the file server (typically via ethernet). The Render Controller typically doesn't need much horsepower which is why it's often running on the fileserver itself.

The MacPro6,1 as shown at WWDC does not make a good fileserver. It does not make a good cluster node. It does make a decent workstation with which to edit, process, and otherwise manipulate the data which gets stored onto the fileserver - which the MP6,1 (typically) would get over ethernet.

Tesselator said:

The MacPro6,1 as shown at WWDC does not make a good fileserver. It does not make a good cluster node. It does make a decent workstation with which to edit, process, and otherwise manipulate the data which gets stored on the fileserver - which the MP6,1 (typically) would get over ethernet.

Click to expand...

You are reading too far into my words. I claimed that shops that employ both render farms and mac pros use the mac pros only as workstations, as they are not effective in cost/functionality for building a render farm. As you mention, they would make terrible nodes. The other part was addressing them as workstations, not in the context of nodes in a render farm. The main thing it has is computational power. I'm interested in what it offers sub $3k for the main thing, as I will still have to add additional hardware to that.

thekev said:

You are reading too far into my words. I claimed that shops that employ both render farms and mac pros use the mac pros only as workstations, as they are not effective in cost/functionality for building a render farm. As you mention, they would make terrible nodes. The other part was addressing them as workstations, not in the context of nodes in a render farm. The main thing it has is computational power. I'm interested in what it offers sub $3k for the main thing, as I will still have to add additional hardware to that.

Click to expand...

Ya, I dunno. I'm kinda scratching my head on that Q. People are saying that the new 12-core Intel chip is going to cost $2k or more. And they are also saying that the GPUs will be somewhere between $3k and $500 each. So with no RAM, no PSU, no SSD, no TB2 controllers, and no Case, we're already well past the $3k mark.

If Apple follows their current price|performance markup ratio and doesn't get any special Intel or AMD deals then these systems could very well start at around $5k for the base system. Specs are subject to change but the orator at WWDC did say that the MP6,1 "comes standard with dual workstation grade GPUs" so I'm guessing they don't currently intend to ship a single GPU configured model.

I'm just hoping that they have assumed internal GPU/PSU/SSD/MotherBoard fabrication and are getting "special" deals from AMD and Intel - all to the affect of being able to deliver a base unit under $2.5K - or at least close to that mark. They certainly did streamline the assembly process - here's hoping they are extending that kind of streamlining to all aspects of production where costs can be cut lower.

Tesselator said:

It's only good for a few specific apps tho right?

Click to expand...

I've only used it with the FCPX suite, but according to Wikipedia it works for any app exposed via the command line.

It's not meant for things like scientific apps, but plenty of third party apps fill that role on the Mac.

Tesselator said:

That's why I said such a cluster (of any considerable size) doesn't make any sense - like, at all.

Click to expand...

That's a bit of an over-exageration.

It would make a lot of sense for clusters that use GPGPUs, which is almost most popular these days than CPU based clusters. GPUs work great for clustered rendering. H.264 encoding...

Actually, what sort of tasks would these machines be BAD at? Clustered compiling is about all I can think of, but support for that got dropped long ago on the Mac anyway...

goMac said:

I've only used it with the FCPX suite, but according to Wikipedia it works for any app exposed via the command line.

It's not meant for things like scientific apps, but plenty of third party apps fill that role on the Mac.

Click to expand...

Oh cool.

That's a bit of an over-exageration.

It would make a lot of sense for clusters that use GPGPUs, which is almost most popular these days than CPU based clusters. GPUs work great for clustered rendering. H.264 encoding...

Actually, what sort of tasks would these machines be BAD at? Clustered compiling is about all I can think of, but support for that got dropped long ago on the Mac anyway...

Click to expand...

Naw, no exaggeration. If you're going to do GPU computing you can do double the performance at nearly half the price on a non-mac system - if I guess the MP6,1 price right. So no, it's not an exaggeration to say the MP6,1 won't make a good GPU cluster node. Additionally other systems can accommodate 4 or more NVidia cards which have CUDA as well as OpenCL. This of course isn't even mentioning that other systems can select any grade of card for such GPU computing - so that the node systems can be tuned to the best price performance.

Once again GPU render nodes do not need fast expensive CPUs in them making the MP6,1 an even worse choice! You might love this new Mac and I wouldn't blame you but be realistic concerning what it's good for.

And BTW, GPU rendering/computing is in it's infancy. Almost no applications can make use of it yet. This may change and is changing even as we speak but currently about 80 or 90 percent of all distributed rendering systems rely on CPU not GPU.

Tesselator said:

Ya, I dunno. I'm kinda scratching my head on that Q. People are saying that the new 12-core Intel chip is going to cost $2k or more. And they are also saying that the GPUs will be somewhere between $3k and $500 each. So with no RAM, no PSU, no SSD, no TB2 controllers, and no Case, we're already well past the $3k mark.

Click to expand...

Yeah. You know to me the design seems more oriented with the Xeon 1600s cpus. When I saw single chip 12 core, it occurred to me as being a move just to maintain parity in core count and cpu benchmarks with the current model. Look at past gpus. They've used Radeons before. They're sometimes more or less ram with a bit more markup. OSX requires some amount of separate driver development either way, and AMD makes up a fraction of their sales from Firepros in general. I suspect those cards are 7970s or something similar with larger framebuffers. A 7970 Mac edition is $600 retail. I could see such a card at over $1000 each. I do not think Apple is going to shoehorn in $3000 cards. I don't even think those are going to stay at $3000 much longer given that both NVidia and AMD often reduce pricing on longer cycles. If you work downward from the "up to" model it may not be as expensive as it looks. It would still be a higher cost with 2 gpus, but I suspect the equivalent of 2x 5770 at the lower end, however that works out from AMD. I have no idea what 3x TB2 controllers will cost.

Tesselator said:

If Apple follows their current price|performance markup ratio and doesn't get any special Intel or AMD deals then these systems could very well start at around $5k for the base system. Specs are subject to change but the orator at WWDC did say that the MP6,1 "comes standard with dual workstation grade GPUs" so I'm guessing they don't currently intend to ship a single GPU configured model.

Click to expand...

There are $5k+ workstations sold today, but I don't think the volume would be enough to keep a company the size of Apple interested in such a product line. Its fanboys have been calling for product cancellation for a long time, but really that happens with whatever line of rumors. It always amused me when people tried to call the demise of a line.

Tesselator said:

I'm just hoping that they have assumed internal GPU/PSU/SSD/MotherBoard fabrication and are getting "special" deals from AMD and Intel - all to the affect of being able to deliver a base unit under $2.5K - or at least close to that mark. They certainly did streamline the assembly process - here's hoping they are extending that kind of streamlining to all aspects of production where costs can be cut lower.

Click to expand...

There was some silliness about how being made in the US would drive up costs. I figured it might be that they kept it closer to home to make oversight easier on a smaller operation. I'm not sure if they are using a dedicated factory. I would guess they know how many units they have to ship for a viable line, and I wish I could offer more reference on that. It's going to be expensive overall no matter what. If you're starting from nothing, display costs and any storage enclosures have to be accounted for. My own experiences with certain generic ones haven't been as great as yours with the lower end solutions. If you're throwing $2k+ into storage, you can buy better cards and things. My own strategy for anyone cost sensitive would be to minimize the number of extra hardware layers. Not having to deal with drivers that wake the computer up or cause hanging on restarts is generally favorable.

Tesselator said:

Naw, no exaggeration. If you're going to do GPU computing you can do double the performance at nearly half the price on a non-mac system - if I guess the MP6,1 price right. So no, it's not an exaggeration to say the MP6,1 won't make a good GPU cluster node. Additionally other systems can accommodate 4 or more NVidia cards which have CUDA as well as OpenCL. This of course isn't even mentioning that other systems can select any grade of card for such GPU computing - so that the node systems can be tuned to the best price performance.

Click to expand...

Hasn't this always been true?

Not to mention, does it even matter?

I think a lot of the use case of the Mac Pro is that you're farming out the extremely heavy work to a server farm. Apple doesn't care if that server farm is running Mac or Windows or Linux or Amiga as long as you're running that Mac Pro at your desk. If you're working in CS suite and need to do a big encoding job, no reason you can't just farm that out to a Windows blade cluster sitting in a closet in the building somewhere. Instead of dropping another $2000 on a second CPU at everyone's desks. just spend $10k-$20k on a Windows or Linux farm and have the Macs at the desks farm out intensive jobs to the farms.

But, if your workflow is very Apple based, there also isn't much of a reason you couldn't get a cluster of either Mac Pros or Mac Minis for Qmaster.

thekev said:

Yeah. You know to me the design seems more oriented with the Xeon 1600s cpus.

There are $5k+ workstations sold today,

There was some silliness about how being made in the US would drive up costs.

Not having to deal with drivers that wake the computer up or cause hanging on restarts is generally favorable.

Click to expand...

Yeah, that could all be true. I hope you're right about the CPU and GPU on the base model. It still wouldn't make the MP6,1 an even half way decent cluster node but it might make the price of admission for us Workstation upgradders more appropriate - that would be nice.

goMac said:

Hasn't this always been true?

Click to expand...

No.

does it even matter?

Click to expand...

Well, that's what the thread is about, so... yes I guess it does.

I think a lot of the use case of the Mac Pro is that you're farming out the extremely heavy work to a server farm. Apple doesn't care if that server farm is running Mac or Windows or Linux or Amiga as long as you're running that Mac Pro at your desk. If you're working in CS suite and need to do a big encoding job, no reason you can't just farm that out to a Windows blade cluster sitting in a closet in the building somewhere. Instead of dropping another $2000 on a second CPU at everyone's desks. just spend $10k-$20k on a Windows or Linux farm and have the Macs at the desks farm out intensive jobs to the farms.

Click to expand...

That's basically what I have been saying here, so you won't get any arguments from me on that.

But, if your workflow is very Apple based, there also isn't much of a reason you couldn't get a cluster of either Mac Pros or Mac Minis for Qmaster.

Click to expand...

No, this is where it becomes necessary to alter the workflow. And it's so easily done. IF your workflow needs a large cluster, and IF the cluster nodes need to be OS X, and IF that means spending an extra $100K or more then no one sane would consider maintaining that workflow. For almost every pro app that would require OS X nodes there are better more capable App solutions on other platforms. And the pro apps that don't require OS X nodes don't even enter into this discussion - so we're right back to the same statements: MP6,1 is suited as a workstation or perhaps as nodes in a small cluster but it's absolutely NOT suited for large clusters as outlined in the OP of this thread.

Tesselator said:

Yeah, that could all be true. I hope you're right about the CPU and GPU on the base model. It still wouldn't make the MP6,1 an even half way decent cluster node but it might make the price of admission for us Workstation upgradders more appropriate - that would be nice.

Click to expand...

I'm speculating like everyone else here. The 4,1 and 5,1 are priced really aggressively, so I suspect they have some leverage in pricing there. As you've noticed with Apple in the past they like to consolidate changes into big updates, so I'm not sure their current pricing strategy will carry over. Adding up a $2000 cpu + 2 x $3000 gpus and whatever else with Apple's typical markup would probably not make for a workstation that many people would purchase. It's not that interesting to me to see "up to" specs. It's interesting to see something I would actually purchase as opposed to just post about on here. I don't mean on day one, although it would amuse me if people lined up to buy mac pros.

thekev said:

It's not that interesting to me to see "up to" specs. It's interesting to see something I would actually purchase as opposed to just post about on here.

Click to expand...

Yup. I guess we'll have this info round about November or December of this year. Till then all we got is the WWDC info. I could see Apple releasing some more preliminaries in September or there abouts tho.

mwhities said:

This is interesting.



http://www.macstadium.com/blog/new-mac-pro-hosting-and-colocation-will-be-here-soon/

Click to expand...

I'm just wondering why you need web hosting with dual GPUs. Does Apache use OpenCL all of a sudden?

(no seriously, why?)

slughead said:

I'm just wondering why you need web hosting with dual GPUs. Does Apache use OpenCL all of a sudden?

(no seriously, why?)

Click to expand...

My same thoughts. I guess it's for the power users? No idea. My little 09 mac mini worked just fine for everything I needed.

slughead said:

I'm just wondering why you need web hosting with dual GPUs. Does Apache use OpenCL all of a sudden?

Click to expand...

Depends upon what kind of web services hosting. All web services don't necessarily pass through HTTP and/or require Apache.

It is a colocation shop too. Can do....

Metered hosted XCode terminal services.

Batch job computational services.

Dynamic custom rendered images.


Multiple racks of that? Probably not.

It does tap into pretty well their whole value add SAN network back end services. The GPU overhead may just turn into "noise" in the costs over a couple of years of service. It really depends if the GPUs goose the prices higher than the $2,499 point or not.

The good thing about the modern AMD FirePros is that the go to sleep pretty well if not using them. That just gives the CPU all the more thermal management headroom to work and doesn't incur hosting costs. ( would be true if in a standard slot or not but that's were the space consumed comes in. Not three Fans/heat sinks... just one. )

slughead said:

I'm just wondering why you need web hosting with dual GPUs. Does Apache use OpenCL all of a sudden?

(no seriously, why?)

Click to expand...

If you're doing audio or video processing it would be helpful, but not for just normal HTTP hosting.

Tesselator said:

Ya, I dunno. I'm kinda scratching my head on that Q. People are saying that the new 12-core Intel chip is going to cost $2k or more. And they are also saying that the GPUs will be somewhere between $3k and $500 each. So with no RAM, no PSU, no SSD, no TB2 controllers, and no Case, we're already well past the $3k mark.

If Apple follows their current price|performance markup ratio and doesn't get any special Intel or AMD deals then these systems could very well start at around $5k for the base system. Specs are subject to change but the orator at WWDC did say that the MP6,1 "comes standard with dual workstation grade GPUs" so I'm guessing they don't currently intend to ship a single GPU configured model.

I'm just hoping that they have assumed internal GPU/PSU/SSD/MotherBoard fabrication and are getting "special" deals from AMD and Intel - all to the affect of being able to deliver a base unit under $2.5K - or at least close to that mark. They certainly did streamline the assembly process - here's hoping they are extending that kind of streamlining to all aspects of production where costs can be cut lower.

Click to expand...

These are my concerns about the thing as well. The economy of scale on this thing is just not going to be good enough, I think--basically I think it's going to cost significantly more than the sum of its off-the-shelf component parts to build. That leads me to think that maybe they'll sell at least the base units at a loss or close to the margin for a few months. Even then, it seems like it's going to be painful.

goMac said:

If you're doing audio or video processing it would be helpful, but not for just normal HTTP hosting.

Click to expand...

There for encrypting all of the 40960-bit SSL traffic. Take that NSA.

Dr. Stealth said:

This is right from Macstadium... They've been doing this for 14 years with Mac Minis and are expanding. Makes sense to me.

"For Datacenter operators like us, the costs of operating are significantly driven by the cost of Datacenter floor space (square footage), the cost of power, and the cost of Air Conditioning. Apple products have always adopted themselves to greatly reducing these costs. At MacStadium, we are are able to facilitate 240 Mac mini’s per 8 square feet with a total power draw of less than 30AMPS @ 110VAC – that is about 1/6th the space and power utilized by traditional 1U Rack Servers. The reduced operating costs are then passed on to our subscribers in the form of hosting and collocation services at a fraction of the price typically associated with such platforms by other dedicated server hosting companies."

Click to expand...

Do people that care about space and power savings per server use 1U rack servers anymore?
Wouldn't they be more likely to use something like this?

http://h20584.www2.hp.com/hpgt/guides/select?lang=en&cc=us&prodTypeId=3709945&prodSeriesId=3707371

32 server nodes in 10 rack units, or if you prefer, up to 256 processor sockets/1024 processor cores in a standard 40U/42U rack.

These days you might use the BL460 Gen 8 with 2 x Xeon E5-2670 (8 cores), with 16 blades per enclosure and the same number of cores per rack.

Plus with the HP Bladesystem (or the IBM equivalent), we are talking about a standard rack which is about 600mm wide (outer dimensions). Plus you can add ethernet and fc switches in the back of the blade enclosure which help with the cable management.

While in that picture of the Mac Pro rack, we see 9 x Mac Pro in each row across the rack and each Mac Pro is about 6.8 inch or 168mm diameter (without cables).
http://www.macstadium.com/images/landing/mac-pro_datacenter_rack.jpg

So these are not standard 600mm racks, in fact they look about three times the width of a standard 600mm rack... and a lot taller than standard 42U racks (which are about 2000mm tall). So while it looks good to get that many servers per rack, it's not so impressive when you have a super sized rack.

Looking at some of the Mac Mini racks, these seem to be extra wide as well.
http://www.macstadium.com/images/datacenter/rack1c.jpg

Now there might be some valid reasons for putting Mac Mini and Mac Pro machines in racks, but I don't think there's any space savings going on.

Tesselator said:

Next I expect the same company to produce a server rack design comprising 256 iPhones.

Click to expand...

That's the funniest thing I've read all week.

handsome pete said:

Precisely. Though you would have thought they would do their homework on that front. Specs for the new chips weren't exactly top secret.

Click to expand...

Do you really want to trust your data center/hosting needs to a company that can't even get their basic marketing facts straight on a web site? Details matter! What else don't they get right?

I for one would never market 100% uptime. Once you've learned the difference between 99.9% and 99.99% -- you've got to know 100% is almost impossible to promise.

----------

slughead said:

i'm just wondering why you need web hosting with dual gpus. Does apache use opencl all of a sudden?

(no seriously, why?)

Click to expand...

+1

sorry to dig this up from the grave but.

It would be interesting to see if any more progress was done on this new rack.

Also apple needs to put the apple logo on the front of the mac pro!

ClassicII said:

sorry to dig this up from the grave but.

It would be interesting to see if any more progress was done on this new rack.

Click to expand...

Unless Apple shipped them a "beta-tester" unit, there isn't much they can do with this new rack.

Also apple needs to put the apple logo on the front of the mac pro!

Click to expand...

As long as Apple is the only vendor shipping a cylinder shaped workstation, the product itself says "Apple". Sticking their logo onto something for giggles doesn't really accomplish much.