OS X 10.8.3 beta supports AMD Radeon 7000 drivers, hinting at Apple's new Mac Pro - Page 2

Originally Posted by Marvin 

This bugs me. What happens to the workstation crowd? Are we to be stuck with 3 RAM slots? Even the iMac has more. The Nehalem/Westmere models already omit two of their maximum possible slots… 

I totally buy only one double-wide PCIe (16x 3.0) slot with the rest of the expansion expected to be Thunderbolt (and then six of those), though. Even on the full size model. Drop the internal PCIe, both optical drives, and trim out some of the excesses with the case build (because, hey, times have changed since 2003. The design itself is fine. Great, even. Why make a silent computer less so? Why make spectacular cooling less so?

The thing is, every time someone talks about a redesign, they're taking away all the features that make it a pro machine.   A new case would be nice (especially if it were lighter) but the iToy crowd just talks about how "antiquated" the mac pro is, and all they want to do to is take away functionality.  I know optical drives are over, but RAM?  PCI-E?  Multiple HDDs?  Proper air flow and heat dissipation?  

Originally Posted by Marvin 
It depends on who exactly we are talking about.

Your mockup appears more like an xMac than something designed to serve the needs of the Mac Pro's market right now.

If the true market is, as you say, those that push it to its limit, cutting its limit in half doesn't seem like the best idea.

I'm just… I'm losing it. You're right; it's four now, eight in the dual-chip, so did that mean Nehalem/Westmere can have 6 per processor normally? I guess all that I remember is they shorted us on the maximum number of slots. But that's probably up for grabs now, too… 

Oh hellzzzz nooo!  I don't want my Mac Pro to look like an Octopus full of cables. Going that route means one cable to the PCI-X/GPU box, another for the optical drive, another for external storage plus all the other wires for the displays and peripherals. I prefer to have everything internal if possible.

I wouldn't mind one optical drive on top and the second slot for a card reader like this: http://www.sonnettech.com/product/qio.html as an additional option, but removing the optical drive bay would be a bad idea!

In the case of the Mac Pro BIGGER is better.

As for gartner, they are full of baloney.

Quote:

Originally Posted by Marvin 


They might have gone the route of putting a single Xeon in the iMac but the model number MP60 listed in the Bootcamp plist suggests that won't be the case. It also provides good evidence that it won't have an optical drive, which indicates a redesign because that 5.25" unit takes up a lot of space. They could move the hard drives into that space and just lower the height or maybe have a different shaped power supply but I think they should go beyond that.
We get the same comments cycling round every time there's a Mac Pro reference such as putting in more PCI slots and an i7 because it's cheaper.
The entry Mac Pro uses a $294 processor, which costs exactly the same as an i7. The 1000W power supply and motherboard will be more expensive but we know that they upped the Mac Pro price by $300 after the first Mac Pro model so they obviously weren't selling enough to justify the margins they had. The lower the volume, the higher the price.
The parts that go into the Mac Pro can be bought for about $1200-1400 so they are easily at 40% profit margins.
When it comes to Thunderbolt, there's still the issue of how they get it to work with a dedicated GPU. The spec they are required to follow in order to call it Thunderbolt is that it has PCI and displayport on the same connection, no compromise. So they either have PCI slots and no Thunderbolt or they don't have Thunderbolt because if you put in another GPU, it can't know how to route the graphics out the TB ports. If you put in a non-standard GPU, it breaks the TB spec.
There's also the issue about the machine having 40 PCI lanes. If you have 4 slots, you can't give them all 16 lanes and if you max out the lanes on the slots, there's nothing to allocate to Thunderbolt. I think it's very much an either/or situation.
When you consider that the Mac Pro slots only have a 300W power allocation, you can only have multiple low-power GPUs or a single high-end one. The simpler option is the single high-end one.
Once you've decided on the GPU, Thunderbolt can take care of expansion. It would be better if Apple managed to get the 20Gbps Falcon Ridge controller though. This prevents the scenario where Macbook Pro/Air//iMac/Mini professionals are buying Thunderbolt peripherals and Mac Pro professionals are buying PCI cards. They all buy the same peripherals.
The single GPU would still be upgradeable but only from Apple as it has to work with Thunderbolt.
As far as the CPU goes, they can stick with allowing 2 CPUs but Ivy Bridge will bring 10-core chips, maybe 12. These will be expensive chips.
Right now, the highest-end MP uses 2x $1440 CPUs = $2880 but the performance is only about 20% faster than the $1885 single CPU 8-core E5-2687W. The equivalent Ivy Bridge chip will likely be 20% faster so they could offer the same performance as the current $6200 Mac Pro for:
$2499 - $294 + $1885 = ~$3999
While they could still offer a faster dual processor model at $6200, if few people are buying those, the better option would be to offer the best value to the highest volume of customers.
The entry model could do with a 6-core CPU and then have an 8-core in between.
By taking out the optical, the PCI slots and 2nd CPU, they can cut the power consumption down so the PSU can drop to 500-600W.
If they can fit this into a Cube, that would be great but I think they'd struggle with that. They can at least manage the following size as it's just a reworking of what they have already:

If they can put in functionality to allow zero-config connections over Thunderbolt, even better. You could buy as many $3999 models and just plug a TB cable between 1 and 2 then 2 and 3 etc.
Sure the complaints will come in about not being able to access PCI cards but for high-end tasks, wouldn't you rather spend $3999 on another MP and run any task natively on a dedicated 10-core Xeon than spend $4750 on a Red Rocket PCI card that only does one thing? There's always the backup of having an external PCI box anyway.
If they can figure out how to make PCI slots and Thunderbolt work together in a Xeon box, all the better I suppose but they still need to allocate 40 lanes between them so they won't have more than 4 slots.
Ultimately, just like FCPX they have to design this box for the next 10 years, not for the last 10 years and make it appeal to the widest Mac Pro audience. If leaving the design largely unchanged and leaving out Thunderbolt accomplishes this, so be it but I don't think it does. I think the USP should be performance-per-dollar, not expansion - make it more than twice as fast as the iMac for less than twice the price.
Remember what the original Macintosh said:
http://www.youtube.com/watch?feature=player_detailpage&v=2B-XwPjn9YY#t=224s
These big, heavy workstation form factors are becoming unnecessary for workstation use just like the mainframes. Same for servers. One day, so few people will buy them that they will be dropped:
http://www.gartner.com/it/page.jsp?id=2079015

Originally Posted by wizard69 
By the way guys, those AMD Radeons are OpenCL power houses delivering some impressive performance specs. By the time this machine ships there might even be lower power variants. The point is if this is coupled with just one of Intels coming many core chips we could see some rather impressive performance from a one socket machine. It is just another reason to not focus on the past when thinking about what is right for the future.

Just hope one of them works with my Early 2009…

Originally Posted by wizard69 
This whining about RAM is bothersome. First people don't seem to grasp that RAM is or will be performance limited by sitting in slots in the future. I'm talking very near future too, high performance RAM will have to be soldered on the motherboard. Further other technologies like multi chip 3D Ram means that that RAM will take up very little space on the motherboard. It will be rather easy to put 64GBs into a Mini in the not to distant future soldered right on the motherboard. In the end the discussions about RAM are just like the discussions about case size, to much focus on the past! People need to be open to the idea that motherboards will look dramatically different in the future.

Imagine how much controversy there will be when the Mac Pro comes with soldered RAM like half of their laptop lineup right now…

And price? I can see the threads now. 😱

Quote:

Originally Posted by Marvin 


If they can put in functionality to allow zero-config connections over Thunderbolt, even better. You could buy as many $3999 models and just plug a TB cable between 1 and 2 then 2 and 3 etc.
Sure the complaints will come in about not being able to access PCI cards but for high-end tasks, wouldn't you rather spend $3999 on another MP and run any task natively on a dedicated 10-core Xeon than spend $4750 on a Red Rocket PCI card that only does one thing? There's always the backup of having an external PCI box anyway.
If they can figure out how to make PCI slots and Thunderbolt work together in a Xeon box, all the better I suppose but they still need to allocate 40 lanes between them so they won't have more than 4 slots.
 

I'm tired today. I really wanted to reference highlander after you said something about people dying of old age (old movie), but I've got nothing other than the obvious joke about consolidating the lineup. Distributed computing has been around for a while, but I don't see this as a good option at a smaller scale like a 1-5 man shop. It might be better in a larger shop where some machines are under-utilized, but gpgpu computation is likely the future for this stuff. OpenCL and CUDA just need to support a wider range of functions. I'd suggest that in a couple years many of these customers will be better served by multiple gpu cards than an ever increasing number of standard X86 Xeons from a performance per dollar standpoint. The other problem is that starting at $4000 limits your market. Right now the $2500 option and the 12 core really cater to different markets. Trying to homogenize them at the $4000 mark means the imac has to pick up a lot of slack at the low end of that. Regarding thunderbolt and PCI slots, the problem was the lack of integrated graphics. It had nothing to do with PCI slots. The chip depended on specific logic board placement, and the certification requirements made integrated graphics the way to go. Right now we lack that on Xeons outside of the E3s. If that changes, you could have integrated graphics + PCI slots containing other cards.

As much as I enjoy the links and references, you miss many things. The last point is on single vs dual package. If you're starved for bandwidth, you can go with dual cpu packages for this reason. If it's a minor bottleneck, they can do what they've done up to this point. The current machine is slightly over-subscribed, and Apple has done this with many things. Some of the imacs have done this with firewire using a single bridgeport, so plugging in FW400 +800 meant they'd both run as 400. If the integrated graphics thing isn't so much of an issue, you can lose the lower x4 slot in favor of thunderbolt. Allocate 4 lanes to that with a x8 and x16 slot. It's not really ideal, but you could run gpus off both if they aren't oversized or constrained thermally. That basically leaves you with 12 lanes to cover the rest of the ports. It's cutting it pretty close, but it could work.

If Apple goes the route of just trying to chain X86 cores in boxes together, I see it as a product line with little remaining life. They need significant gains and performance per dollar to retain relevance. 20% faster means work gets done a bit quicker, but relying partially on gpu computation can actually alter workflows and enable things that weren't possible or feasible on the older hardware. It was actually one of your links that showed how much differentiation was present between a mobile gpu and a desktop gpu in the $500-1000 range  <200W.

Quote:

Originally Posted by Marvin 



Yes, it's like the FCPX face-lift. It doesn't matter much to people who ignore that but it broadens its appeal.
 

I forgot to reference this point. If you read over the most common complaints separated out from the general noise, it came down to lack of licensing options for the old thing after such a significant point of departure, and concerns that certain must have features were missing. Multi-cam support was the most commonly listed gripe, and that's one I can understand. Its name should be self explanatory. Apple remedied it with an update later, but the people who migrated were most likely already disappointed with Apple. My point remains that an increase in X86 cores may not be the way forward for much longer. I'd also suggest you'll retain some performance segment much longer. As we've seen outside of that, the point of focus has trended heavily in favor of mobile form factors with lower power requirements. If the power isn't being used, they'll optimize in a different direction.

Quote:

Originally Posted by jragosta 


And this is based on what? What makes you more of an expert in computer design than Apple?

You never used a thin Mac for anything other than net surfing, right?

1- Pick some music that has some dBA level. Either heavy metal, punk, or Aqua's Barbie Girl will work fine (http://en.wikipedia.org/wiki/Barbie_Girl)

2- Use big, full, earphones, and listen to such music at high volume (but take care of your eardrums, please).

3- Start "The Sims 3" in your brand-new ultra-thin Mac (if you pretend you're not a gamer, you can say you run a render job in Maya if you prefer).

4- While "The Sims 3" starts a new game, notice the subtle details in Aqua's Barbie Girl.

5- No, the wind effect in the song's drum base isn't a new mix of the song: It's your ultra-thin Mac hurrying the fans for the CPU not melting.

6- You can now turn off the music, quit "The Sims 3", and continue using your ultra-thin Mac for net surfing. Yes, it was designed for that.

*GASP*
This (hopefully) means I can shove a Radeon 7XXX series card in my 2012 MacPro and reap all the nerdy delights a high powered graphics card brings!

I have the 6-core Uni Processor model and that CPU is on par performance wise with the current 6-core E5-XXXX Sandy Bridge processors (albeit running at a higher clock frequency), so its by no means slow. A 7XXX series graphics card would be the cherry on top.

Quote:

Originally Posted by rob53 

Case size is important. Just ask all those enterprise technicians who have to move the old Mac Pros around. For me, a Mac Pro case needs to be just large enough to hold a motherboard with several CPUs, plenty of RAM, and a few specialized PCIe cards. Standard I/O ports don't really take up much room. As for optical drives, I wonder if not having them would be better, letting me attach my specialized and constantly changing drives to whichever Mac Pro I want instead of having to settle for older internal drives. The amount of space for internal disk/SSD drives is something that I'd have to think about. There needs to be enough disk room for individual users but for groups of users, having SSD boot drives and external shared drives might be a better option. This would mean a much smaller case design. Think about a stack of Mac mini. Add better CPUs with heat sinks (tall-mini, size of two minis) and something on the order of four tall-minis would only be a foot tall. Redesign to get rid of the optical drives (two quad- or six-core CPUs could fit in the bottom two or three tall-minis) and you could put disks and I/O cards in the top tall-mini space to get four to six CPUs with disks and I/O cards for a nice new Mac Pro. All in a much smaller design and (maybe) with redundant power supplies. Actually, Apple could have fun and build the new Mac Pro as a modular computer: redundant power module, dual CPU module, disk module, I/O module. Just stack them to plug everything together. Start with a single CPU module but allow 2-4 (or more) to be stacked. Same with disk and I/O modules. This ends up looking like a blade server and could provide the same functionality (maybe even as a reborn OS X Server).

Why the hell would you do that? Why would you have it all external modules when its already modular inside a single case? You musn't know much about professional hardware because those Xeon CPUs, even the bleeding edge chips, needs a lot of room, big fans and larger heatsincs to keep cool. I have yet to see a small (1U/2U) rack server with more than two CPU sockets and I have never seen a tower with more than two CPU sockets as small as the Mac Pro. People that require that much horse power should turn to computer clusters. You would a LOT more space than the foot print of a Mini (even one 3x as tall) for even a single high power Xeon chip.

Then you have the issue of cost; connecting all those external modules together would require very expensive, high bandwidth, enterprise super/cluster computer grade fibre optic cable. Making drives external is just... dumb... Why would you take your disks out of your system, away from the SATA or SCSI backplane and connect them externally? Thunderbolt is far to expensive and internal disks offer direct connection to the disk controller without going through the PCIe first as with thunderbolt. Its also more secure being inside a locked case, its better for cooling (pro systems and servers have fans for the hard disk bays) and external units would end up being gigantic in order to furnish a fan and the interface ports.

You're suggesting all of this to make the system smaller, when in realistically all you're doing it introducing more parts to potentially fail, dangerouse interface cables, racking up the price and actually make the system unfeasibly large.

Quote:

Originally Posted by ecs 

Finally, an interesting rumor!!! (I was quite bored of iOS rumors)

 

I'd like to have a new Mac Pro with as few mechanical parts as possible (no HD, fan-less PSU... and if the whole computer can be designed with just one fan, better than two).

Even the lowest power desktop PCs have fans in the PSU. Those things get HOT. Its not a wall-wart or power brick connecting to a battery in a laptop, its a big brick of a thing that has to power several high performance devices.

Also, one fan? Why would that be a good idea in a high performance system? High performance servers and workstations are designed to stay below at least 60C. The Mac Pro has a fan on the GPU, fan in the PSU, fan for the HDDs, a fan for the daughter card holding the processor and ram, a fan inside the processer heatsinc itself and an exhaust fan. It needs all those fans - you can't just chop them down to one and expect it to function. The only single fan I can see dropped in the near future is the HDD fan once HDDs become obsolete in everything except specialist systems requiring large amounts of high-density storage.

Its comments like the above two that really infuriate me because they are completely unrealistic with little or no thought towards reality and the laws of physics.

Quote:

 

(text cut out because WALL-O-TEXT!)

What you have proposed here is quite possibly one of the dumbest things ever for a professional workstation machine.

You've removed the second processor, crippling the machine in the high-end market, you've removed the PCIe slots, meaning it can no longer be upgraded with extra expansions boards, you're removed a drive bay leaving only three (making almost all RAID configurations useless if three drives are employed) and you've removed the top of the machine above the drive bays that held both the ODDs and the PSU, meaning no space left for extras such as a card reader. Your choice of putting the PSU behind the processor and ram daughter board means that cooling has been compromised as there is now no exhaust fan at the back of the machine for said daughter board (with the heat from the PSU also affecting the airflow around the CPU). Smaller fans would have to be employed for the middle of the tower running at a higher RPM, causing lots of excess noise and their is less space around the CPU heatsinc for it to "breath", causing other components around it to heat up from being in such close proximity, needing either lower performance parts or faster, louder fans.

All you've done is cripple the machine and made it useless in the high-end market. If these are the omissions you'd make to the Mac Pro then you are clearly not its target market.

Machines of this type from any manufacturer have all the features you ommited because they are required by the high-end user. These machiens need to run cool, perform mind-bending calculations quickly, be rated for 24/7 operation with pin point precision and accuracy and have 99.9% uptime at minimum. A consumer PC (home, office, gaming etc.) are none of these things. The Mac Pro and other such workstations should not be treated and thought of as a consumer desktop computer.

Quote:

Originally Posted by benanderson89 

[...]

Even the lowest power desktop PCs have fans in the PSU. Those things get HOT. Its not a wall-wart or power brick connecting to a battery in a laptop, its a big brick of a thing that has to power several high performance devices.

 

Also, one fan? Why would that be a good idea in a high performance system? High performance servers and workstations are designed to stay below at least 60C. The Mac Pro has a fan on the GPU, fan in the PSU, fan for the HDDs, a fan for the daughter card holding the processor and ram, a fan inside the processer heatsinc itself and an exhaust fan. It needs all those fans - you can't just chop them down to one and expect it to function. [...]

Yes, one fan, because I don't really need a Mac Pro elephant. What I actually want is an iMac without the display and with proper cooling. Put the most complete iMac configuration into a moderately sized cube, and you won't need more than a big high-quality silent fan to keep it cool at intense CPU/GPU work.

Problem is that Apple, with the only exception of the Mac Pro, doesn't seem interested in cool (I mean cool thermal-wise) machines, maybe because if they run hot, they last less, and you buy machines more frequently. Put it together with beautiful aesthetics, and all factors are put there to increase sales.

This forces you to consider the Mac Pro even when you don't need one. But I don't want an elephant, that's why I'm asking for just a simple machine that can be used for intense CPU/GPU, with as little mechanical parts as possible, while keeping the chips cool. It's not hard to achieve that, although I don't see Apple doing it, for the reasons above.

Quote:

Originally Posted by ecs 

 

Yes, one fan, because I don't really need a Mac Pro elephant. What I actually want is an iMac without the display and with proper cooling. Put the most complete iMac configuration into a moderately sized cube, and you won't need more than a big high-quality silent fan to keep it cool at intense CPU/GPU work.

 

Trying to engineer in favor of smaller sizes can actually drive up costs. Right now what most people fail to realize is that they're already leveraging a portion of the xmac crowd to keep mac pro sales at an acceptable level. The xmac machine people want aligns quite well with the specs of the base mac pro. Adding in a few drive bays and PCI options has little impact on pricing. They're using high markups for a line with little growth and low volume relative to their other products. I realize you dislike this, but that is how they've chosen to address this market up to this point. Windows is also showing terrible growth on desktops, so I can't see any big changes coming there.

Quote:

Originally Posted by PhilBoogie 


Lion only sees 96GB (don't know if ML will). Strangely, Windows uses all 128GB under Bootcamp.
OWC 16GB Memory Modules for 2009/2010 Mac Pro — 48GB / 96GB in Mac Pro


Photographers disagree. Read this piece on a person using Photoshop to the max, so to speak. Easily needing 80GB, wishing for more.
I really enjoy this thread. Don't have time to discuss everyones' take on the subject, but I do have one thing to say:
 

Most photographers will never use that much. You'd have to be stitching some enormous files with a lot of layers. Adobe does recommend 64 GB for optimal performance with After Effects on a 16 core machine. They basically allocate 2GB per logical core. I don't think the ever increasing core count will keep going, but dual package models still provide twice the bandwidth. Whether Apple sells them is another matter. This represents a small portion of their total users. I'd expect Windows to represent a significantly larger chunk of this. The thing is it doesn't matter for most of these applications. When you're in them, they're 90% the same. People usually just continue with whatever operating system they already use unless hardware or software requirements force a change.

Quote:

Originally Posted by ecs 

 

Yes, one fan, because I don't really need a Mac Pro elephant. What I actually want is an iMac without the display and with proper cooling. Put the most complete iMac configuration into a moderately sized cube, and you won't need more than a big high-quality silent fan to keep it cool at intense CPU/GPU work.

 

Problem is that Apple, with the only exception of the Mac Pro, doesn't seem interested in cool (I mean cool thermal-wise) machines, maybe because if they run hot, they last less, and you buy machines more frequently. Put it together with beautiful aesthetics, and all factors are put there to increase sales.

 

This forces you to consider the Mac Pro even when you don't need one. But I don't want an elephant, that's why I'm asking for just a simple machine that can be used for intense CPU/GPU, with as little mechanical parts as possible, while keeping the chips cool. It's not hard to achieve that, although I don't see Apple doing it, for the reasons above.

Machines running hot does affect reliability - but Apple machines are consistantly some of the most reliable on the market; so much for your "they make them hot so they break" theory. The problem is that you don't seem to understand the requirements of Professional IT equipment: Apple clearly state that the MacPro is a workstation and not a consumer oriented desktop computer, therefore meaning it will cator to workstation users. Workstation users want power, versitility, accuracy and reliability: The way to make the Mac Pro powerful is to use high performance parts - high performance equipment runs hot, so large cooling systems are needed. The system needs to be versitile and ready for any situation, hence the massive expansion capabilities of the Mac Pro right down to removable daughter boards. They need to be accurate and reliable, again pointing towards the performance and cooling but can also mean the ruggedness of the system, its components, chasis and motherboard.

The Mac Pro is no smaller than any other professional workstation - infact its quite tiny in comparison with others in the same market segmant (ProAm, Medium and Enterprise). Dell's new T5600 workstation is toated as having a "Compact Chasis":

Its 41cm high, 17cm wide and 47cm deep.

The MacPro is around 40cm high (sans the large handles), 20cm wide and 47cm deep.

The Mac Pro is 3cm wider, but it has 4 hard disk bays vs Dell's two. The MacPro case is on par with what other manufacturers are calling "Compact" - cases for this class of machine can get much, much larger and much, much heavier. Your typical gaming computer has a larger case than the MacPro.

If you think its easy to keep high-performance chips cool then you really need to take another look at the heatsincs on the market and those installed in other professional workstations; they are gigantic for a reason. To get a desktop computer to stay at the heat levels gained by the MacPro, you better be prepaired to shove one of these on your motherboard + high performance fans:

Lets not forget that a single fan will not be enough to also cool a dedicated graphics card as as well the power supply unit. Even the cheapest desktop computer with a modest GPU card has a minimum of three fans.

Quote:

Originally Posted by hmm 

Most photographers will never use that much. You'd have to be stitching some enormous files with a lot of layers. Adobe does recommend 64 GB for optimal performance with After Effects on a 16 core machine. They basically allocate 2GB per logical core. I don't think the ever increasing core count will keep going, but dual package models still provide twice the bandwidth. Whether Apple sells them is another matter. This represents a small portion of their total users. I'd expect Windows to represent a significantly larger chunk of this. The thing is it doesn't matter for most of these applications. When you're in them, they're 90% the same. People usually just continue with whatever operating system they already use unless hardware or software requirements force a change.

Given the sheer pixel density of modern cameras, its very possible that a photographer could use that much RAM in a heartbeat. I draw colour comics in photoshop, and I have no trouble reaching the 32GB+ mark of Real Memory on my MacPro after several solid hours toiling away over a wacom tablet. If I was doing professional grade work with CMYK and/or print ready that would easilly top 64GB+. A CMYK or print ready file (300DPi+, 16-bit colour or better and/or very high resolution) can be hundreds of megabytes in size, some even a gigabyte, and this is just the file on disk!

Quote:

Originally Posted by benanderson89 

 

Quote:

Given the sheer pixel density of modern cameras, its very possible that a photographer could use that much RAM in a heartbeat. I draw colour comics in photoshop, and I have no trouble reaching the 32GB+ mark of Real Memory on my MacPro after several solid hours toiling away over a wacom tablet. If I was doing professional grade work with CMYK and/or print ready that would easilly top 64GB+. A CMYK or print ready file (300DPi+, 16-bit colour or better and/or very high resolution) can be hundreds of megabytes in size, some even a gigabyte, and this is just the file on disk!

Scratch disks always built up over a number of hours. With adequate ram you're just storing the information directly in memory. Anyway I've dealt with CMYK files in the past. I'm not sure why you'd feel the need to retain 16 bpc there. CMYK spaces tend to be pretty locked down, so you shouldn't run into banding issues. Hundreds of megabytes in size is nothing. You can deal with 2GB (on disk) files comfortably on modern hardware. It was possible a decade ago. It's just you wouldn't have used 16bpc modes. You didn't have cpu intensive things like smart objects. You didn't assemble large spherical hdri imagery directly in PS (it didn't even support the radiance format). Digidlloyd talks things up a bit at times, but he does provide file sizes for reference and lists the actions applied. When you look at a 16 bpc 15-20k image and run a set of intense tasks, it allows you to really see stratification within the lineup, and it is nice being able to deal with things in real time without lag. It's just silly to suggest it's otherwise unworkable. I mentioned that Adobe suggests 48GB of ram for a 12 core machine or 64 for a 16 core if you want to retain maximum performance, especially during rendering. After Effects is the most memory intensive application they publish. Photoshop isn't as bad, even in CMYK or indexed color and 10k files.

Quote:

Originally Posted by Marvin 


He didn't quite max out 64GB but a few extra layers might do it. Working with multiple 16bpc 22MPixel images and saving a 24GB PSD file isn't respresentative of a widespread need though - he even said it was for comparisons so presumably loading a whole load of images in as layers to see the differences. Photoshop isn't meant for that. But one processor support 6 slots anyway so if they went this route, Apple could support 96GB. Photoshop shouldn't use that much RAM for doing this. Either they need to figure out how to keep layers compressed in RAM or using an intelligent proxy system.

 

You can't save a 24GB PSD file. I don't know how much of that bulk is layers or if its compressed, but PS has a limit on size. Once you go over 2GB on disk, you have to save in .PSB (large document format) anyway. My primary use these days would be for stitching spherical hdr images. You could do a lot of these things all the way back to the G4/G5 days, but it was a lot slower. At that time 16 bpc was also uncommon in PS. It's overrated anyway. 32 bpc is useful if you need to work with linear data.

I was going to avoid that topic, but photoshop can run in a lot of ways. If you load it with ram and set its memory allocation high, it will use it. People still worked on huge files for movie posters and things prior to 64 bit photoshop. It's just that today it's practical to let it cache more data to ram rather than scratch disks if the resources are available. It's not the leanest application out there, but memory prices are low. For anyone dealing with large files, I'd just say max ram and if necessary, add an ssd after that.

I'm not sure there's a good way to implement more than a single thunderbolt chip, and each chip supports 2 ports according to everything I can find. In the past Apple has limited the number of "specialty" ports. They never had more than 3 firewire ports. This assumes we'll see thunderbolt right now. It could skip a generation, as it would have little impact on the overall shape of the case. The best way to implement it remains integrated graphics, which you wouldn't have on the E5 you mentioned. You're still stuck on engineering it to be smaller. At a $4000 price point, I don't see how that could possibly matter unless you're going the lenovo C20 route and trying to make more fit in a server rack. It's actually more expensive than some of the others. When you engineer specifically for size, it costs money. The cost of materials to build a large aluminum case is paltry compared to the costs of trying to make something as compact as possible while retaining performance parts. I think you just like to imagine this stuff. You told me you like new/innovative solutions before. I get that, but I think your abstract concepts are misaligned with the priorities of such a machine. I'm also skeptical of the Radeon drivers when the rest of the lineup has moved to NVidia for now. The ability to leverage OpenCL and CUDA is important when it comes to reaching the widest market possible. If you're offering a fixed graphics solution that is limited to AMD, you've limited your market yet again.

You knew this was nonsense logic when you typed it. I'm really puzzled by this. At $4000 the price is a large determining factor in your potential market. You're looking for users with either bleeding edge requirements or complex needs, and this solution does basically nothing to drive anything forward. If you're looking for a design that must last for a number of years, gpu computation must be a part of the core design, not something allocated to third party generic boxes and a limited set of cards with hit or miss OSX support. Apple really needs the widest market possible. If they're limiting it to a high price point, the worst thing they could do would be to drive away anyone that can afford it. I think that kind of pricing + artificially imposed limitations would be enough to finally kill the line due to negative growth.

I should add that most of the people who want a smaller case assume that this would directly lower the price. It's a false assumption in terms of product positioning. Apple could lower the price if they wanted to, and they would sell more. Using thick aluminum doesn't contribute more than a few dollars to the material cost, and any initial setup costs should have been covered long ago given the age of the outer shell design.

Quote:

Originally Posted by wizard69 

…you call yourself a professional but can't see the importance of these standards so it makes me wonder just what you are profession wise.

I am a recording engineer. In my post, I made several references to audio recording, and included a link to PCI audio hardware made by Mark of the Unicorn. I stand by what I said: USB 3.0 and Thunderbolt are fine, but that is not how we—nor anyone else in the industry—gets multiple channels of 192kHz audio onto a hard drive. We have a frightfully large investment in PCI hardware because that's how this works.

Quote:

Originally Posted by Marvin 


For 4 ports, I reckon they'd get away with a 4-lane 20Gbps per lane controller. This would let you run 2 displays and have 2x 20Gbps ports free. If they can implement two controllers or have 8 lanes, that would be the preferred setup.
I wonder if the Thunderbolt setup has been the major cause of the current state of the Mac Pro because there was really no other reason to avoid Sandy Bridge.
They can obviously avoid using it in the Mac Pro but they've put so much effort into pushing it as a standard. Leaving it out of their most expensive machine doesn't seem like a sensible idea.
The Mac Pro is $6200 at the top-end just now. The new one would be $4000 with a single 10/12-core CPU. It would start at $2499 with a 6-core.
When you say it's not driving anything forward, what exactly does leaving the machine unchanged do? You're not getting any more GPUs in there with a 300W power limit on the slots. Just put one high-end GPU in and computation is covered.

I misinterpreted part of your post. I thought you meant just offer a $4000 model, and I'm not sure that would bring in enough volume to be considered sustainable for Apple. It could be completely different for a smaller vendor. Apple has somehow supported dual gpus in their cto options before, although they weren't as power hungry. If they see this as a growing use case, it would make sense to design a box that could adequately cool a couple of them without tacking additional coolers directly to the card. Using 2 might require power cables. As you mentioned the PCI lanes have a 300W limit. This is something that is currently better accommodated by some of the PC vendors. They make significant design updates more often, and more of them have been going this route with the current generation of machines. It makes sense to look at opportunities for growth if they're going to keep making this line. Growth requires real improvements across the line. Right now the higher end mac pros have greatly outpaced the lower ones, which have remained somewhat stagnant.

I forgot to include this before. Ivy Bridge E5s don't make it any easier to implement thunderbolt. What have you read that suggested otherwise? It doesn't integrate thunderbolt at a native level. It doesn't provide integrated graphics. Consider that they may not have devoted resources to the project, especially after repeated delays from intel. There was likely no mac pro team to work on it. While I still see perfectly valid use cases for such a machine, I thought they'd either update or cancel it. As for thunderbolt, I don't think Apple cares beyond their own peripherals. It was initially placed where they had a mini displayport connection. In either case you could plug in a display, and they built out from that. It allowed them to offer integration that wasn't possible on the notebooks and imacs, which obviously make up the bulk of mac sales.

If the pricing strategy remains somewhat similar, you'll still attract most of the same people. It needs to attract some kind of growth, especially in driving faster repurchasing cycles. People buying these probably update every 2-4 years. Buying the new one with 15% faster X86 cores doesn't do much. It usually means that anything bound by machine time gets done a bit faster, which doesn't describe the overall for most of these users. The other circumstance is that the machine you already own is limiting so any amount of additional speed is welcome. I've been interested in things like CUDA because they open up newer workflows. After Effects probably would not have implemented a raytracer if they were still bound to X86 cores. I agree that the $2500 model should have shifted to a hex to maintain some kind of value. I don't agree with the idea of starting with a form factor and determining what will fit within it when building a workstation as opposed to determining what should be included and designing around that.

Quote:

Originally Posted by PhilBoogie 


There are quite a few people posting this very request; a smaller MP, an in-between iMac and MP. Knowing Apple, they never seize to amaze people. You might get your wish, but I doubt it. Because:
iMac starts at $1299 Mac Pro starts at $2499. Say they want to release an mid-Mac, if you will. I think that price will need to be in between there, $1899.

I would camp outside an Apple store and pay $1500 for a mid sized Mac that had an i7 processor, some expansion (2 hard drives plus room for an optical drive for folks like me that still sues it) and didn't have a built in monitor.

The Mac Pro is simply overkill in size, price and processor horsepower.

But the mini and the iMac are too restricted. Built in monitor? No way. Not for me. Mini with no expansion? No way. Both are desktop (stationary where a little bit of size and weight isn't that big of a deal) but have been shrunk so much that you can't even get an optical drive in one.

$1500 mid sized, some expansion, sans built in monitor Mac and I'm all over it.

Quote:

Originally Posted by MacTac 

I would camp outside an Apple store and pay $1500 for a mid sized Mac that had an i7 processor, some expansion (2 hard drives plus room for an optical drive for folks like me that still sues it) and didn't have a built in monitor.

The Mac Pro is simply overkill in size, price and processor horsepower.

But the mini and the iMac are too restricted. Built in monitor? No way. Not for me. Mini with no expansion? No way. Both are desktop (stationary where a little bit of size and weight isn't that big of a deal) but have been shrunk so much that you can't even get an optical drive in one.

 

$1500 mid sized, some expansion, sans built in monitor Mac and I'm all over it.

The time to implement such a machine would have been a little after they moved to intel to grab some of that market. It's been in a growth slump, which tends to be a bad thing for new products, especially when it comes to a company the size of Apple. I'm not discounting the advantages of the form factor. I just don't expect to see this. The idea of the mac pro becoming an xmac and falling to this price level is really unlikely. The people who criticize the mac pro as being overkill often ignore that the base model is very xmac like in its hardware choices. The hardware used there isn't that expensive, and sharing a backplane adds negligible costs. In fact the daughter board design was likely a cost cutting measure to prevent having to use more expensive dual package parts in the single version they split off as of 2009. Really it wouldn't be much cheaper to build the xmac than it would the bottom mac pro. The performance would be pretty similar. People just misunderstand the cost of that model. It's there because Apple wanted it there, and they use it to maintain minimum volume for the rest of the line. If you don't believe me, look up the cpu costs at the launch of the $800 quad mini and the mac pro at the last "refresh". They drop a $300 cpu option into the $2500 mac pro, and they aren't paying the costs of a full dual package board due to the daughterboard design. As to thick aluminum and extra drive bays, they don't add much. It's cheaper aluminum, but it's thick. Things like drive bays and other random features are also  commonly found in xmac like machines on newegg.
 

What all of you really want is the base mac pro in a smaller case for $1000 less, and this would require a change in philosophy for the machine rather than a reduction in costs.

Quote:

Originally Posted by Marvin 


So you mean they have to get round Intel's slow upgrade cycle somehow like by allowing you to buy multiple machines and easily hook them together to get guaranteed performance scaling with both CPU and GPU. Yeah, that's a good idea. I wonder how they'd connect them together to allow you to control a CPU and GPU in a plug and play manner though.

This isn't quite how I see it. I've mentioned several things such as performance per dollar, potential range of use cases, and what upgrades actually enable. These are aimed at users with either demanding workloads or atypical requirements. I think we can agree on that much. They're not really designed for distributed computing as a primary use. You could try that, and it might work if you were looking to harvest extra cycles from under utilized machines in a larger shop driven by mac pros. It's also true that as they become more powerful, they could leverage some workloads that were previously dedicated to clusters, assuming those workloads hit flat growth and workstations have now caught up. I was merely pointing out that the most cost effective improvement to a wide range of workloads is currently tied to computation allocated to the gpu. The comparisons are generally a CUDA card to a 12 core mac pro, and even with the testing biases, it presents a strong performance per dollar. Take a look at NVidia's propaganda. What's really interesting is how well it works at various levels. Adobe certainly didn't optimize in favor of cpu computation, but the results are still really impressive. Looking at where things will go moving forward, I see that as a better thing to address at a core level. The gains are likely to be much stronger, and it would be something to drive growth in the mac pro as the software matures. If you're just looking to build a server farm, there are cheaper methods. Do you see Apple as a company that would research ways to daisy chain mac pros at a plug and play level if they really were recently considering its cancellation? This isn't the kind of thing they've pursued since the Xserve. Given their professed interest in OpenCL and recent migration back to NVidia, I assumed a strong focus on GPGPU to align better with their current path.

You're leaving out the potential for a late Sandy Bridge E. If you look at Westmere, they still used some nehalem options. It could be the same thing here. It may be mixed either way. If they're waiting for Ivy, these drivers will likely never be used. The AMD 8XXX will be out by the time we have Ivy Bridge E5s. NVidia seems like a better option anyway. I think leaving the Mac Pro as the only computer without CUDA options is just a terrible move for its future when it needs to soak up as much volume as possible. No more mac pro makes less sense to me unless they were undecided. They could have cancelled it when they announced new products or sunset it like they did with the Xserve. I only see a redesign if they think they can capture new customers with it. This is the whole point of what I've mentioned. Even in workstations, not everything is a $10k/seat configuration. They've been going kind of cheap on X86 cores with just a quad in the $2500 model. If that is the continued direction, they need something else to prop up its value.

I agree. Pushing it back to next year likely means that no one was allocated to work on it.

I think the last two lines there are more your imagination, although it does make these discussions more interesting.

Quote:

Originally Posted by Marvin 


But you have to be suggesting that if they offer better performance per dollar with multiple GPUs that they include more double-wide slots and increase the power limit of the slots. They'd have to double that power allocation. Bigger PSU, more slots, possibly a bigger case on top of the standard prices. You would be able to get a quad-core with 2 GTX 780s, which would be more cost-effective for certain tasks than a dual processor machine but still pricey and not that appealling.

Actually workstations don't always run the hottest gpus. This is mostly unique to Apple, but even Apple has allowed 2 in the 150W range on several prior mac pros. A lot of workstation gpus are clocked lower, but they can perform significantly better under certain circumstances. It varies to a degree, which is why a lot of people used GTX580s for CUDA processing. Supposedly the Teslas held up better with double precision math, and obviously you have more ram available there. I expect integrated graphics may eventually show up on some E5 variants. The E5 equivalent of Haswell is the earliest I'd expect that if they go that route. Broadwell is more likely, as I don't expect them to go up in core counts indefinitely. So far that strategy hasn't worked perfectly. A lot of these algorithms weren't written to be split among so many threads, and algorithm development tends to be largely academic. If you ever read Siggraph articles, you'll find a lot of Phd dissertations referenced among them.

Anyway I think you're a bit imaginative in the way you envision these things. In 10 years a lot of things that exist as primary computing devices today may operate more like slim clients. It's just too early to gauge with 100% accuracy. People claim the cloud will take over in every regard, yet it's useless without the proper infrastructure and back end programming.

I could see distributed computing in a larger shop if it could be worked out as a way to harvest extra cpu cycles from workstations that are not used at the time or under- utilized. Apple lacks any kind of efficient infrastructure for this though, and they've been moving away from this direction. In the G5 era, Xgrid was supported. There was a reason for the choice in some of these clusters, although I can't remember what it was. That video doesn't really go much into the logic behind their hardware choices.

I don't necessarily view this the same way. The mini still aligned well with Apple's future goals. In the case of the mac pro, they had plenty of time to redesign for Sandy Bridge E if they wanted to do this. Even when they released Westmere, Sandy Bridge E was already looking pretty far out. At that time Sandy Bridge E was scheduled for the first quarter with E5s late in the third quarter. The opportunity was there. It's more likely that they simply did not allocate any kind of team to such a project, then later decided they didn't want to cancel it. They have pushed out two redesigns this year, so it could happen. I don't agree with your list of priorities though. The historical price points of this line dictate its markets.

You're ignoring what I mentioned before. Apple may not have a starter option within Ivy Bridge E. Look at the mac pro now. The lowest option is nehalem from 2009. The low end from westmere would have been the 2.4, and it was still above their price target, and possibly not compatible with a single package board. The other thing I would mention is that Intel's release dates are not necessarily accurate. Sandy Bridge E5s officially launched in early March. Most oems weren't shipping until early July. Typically the supercomputer vendors get first pick in terms of contract purchases. Q3 could easily mean machines shipping in December by Intel's math. Apple likely has access to more information on this matter, which is why I've stated the possibility of a late Sandy Bridge E still exists. I'm not denying that is an incredibly screwed up release cycle.

What they wanted to ditch was their consumer desktop market. The margins on their Z series workstations are quite high, and they already built the Z1 as an AIO design. They priced it pretty aggressively. Well outfitted it comes out around $5k. I think it's higher if you go with the Dreamcolor display, which is necessary if you're coming from NEC/Eizo. HP's markups on upgrades have always been extremely high. If I bought one, I'd probably wait for a configuration I like to show up as one of the standardized configurations because of this.

Regarding the server market, it's not all a trend to ARM.

http://www.wired.com/wiredenterprise/2011/11/server-world-bermuda-triangle/

http://www.wired.com/wiredenterprise/2012/09/29853/

Note the two links. The first is regarding Facebook and others directly negotiating server purchases with ODMs. The second relates to companies like Google building their own servers.

If anything X86 has grown in the server market. I don't expect them to simply ignore ARM. Viewing this as a foregone conclusion without even examining what ARM gives up in favor of power efficiency is really illogical. The two designs remain far enough apart to where I'm not sure how it will end.

Quote:

Originally Posted by Marvin 


It would all be Ivy Bridge, none of this selling old CPUs on the low-end. It's about performance-per-dollar. They would design it and price it in a way that they could use the latest architecture in the whole lineup. The latest 6-cores would be around $500-600. They can absord the extra $300 just from the margins but also from the redesign (no optical, smaller PSU etc).

We aren't in disagreement here about what they could do. It just contradicts their previous behavior. I'm not sure what their strategy is at the moment. Their current pattern of behavior is inch up pricing and cut costs. That's pretty much a product death spiral. It could be that they foresee the imac cannibalizing enough in the future to phase this out in another generation or two. I'm not sure though. I don't disagree with you that they could eat the price increase. It's just that this contradicts their prior behavior. Intel cut the price of 6 core cpus long ago. They only repriced it at WWDC to $3000. I'd expect that price point to most likely carry over to Sandy or Ivy. 

We discussed the possibility that it was initially slated for cancellation. If this was the case and they had no one working on it, it would make sense. I thought Ivy Bridge only went to 10 cores? Anyway I've never seen workstation boards morph on in between generations like that. The chipsets right now are the same ones they'll have available then. You get less mileage from them when skipping the first generation. This means they'd have to fabricate a new board for Haswell E5s. I wonder if we'll see them before 2015. Going with a single package 10 core system would offer some explanation, but I'm still not seeing it. My best guess is low priority + change of plans. At this point they can't be making much off the line. You really think an Ivy machine will be ready by WWDC? Even with delayed shipping dates, that seems unlikely to me given the disparity between Intel's "launch dates" and actual shipping dates from oems.

I'm still curious when Apple will support OpenCL 1.2. If only I could get Snow Leopard with OpenCL 1.2. It's much leaner than Lion and ML. The one problem is the lack of ability to leverage newer things.