New Mac Pro

[Neil Dearman]

so a base model 12 core will cost 7 Grand? I hope this is not true the base model 12 core cost around 4/5ish I believe. So sad will buy one just means will have to whore myself out for a couple of months. i wonder if you can buy 3rd party Ram and install if yourself?

 

[Bruce Allen]

Prices are not too bad, really. The 12-core CPU is the one option that seems out of line. Although, this just makes the decision to go for the 8-core that much easier. I was on the fence trying to decide between 12 and 8 core, but I think the $1500 additional to go from 8 to 12 just makes it easier to stay with 8, especially since the base internal clock on the 8 is 300MHz faster.

Agree heartily.

PCIe SSD pricing is better than I expected. GPUs are what they are and Fire GL D600's are no slouch GPUs and pricey in the regular PC world too.

But would you actually buy two FireGL cards with only 3GB VRAM each for a PC when you have much better options?

Personally, for me, I have difficulty with the logic that "the Mac Pro GPU's not a bad deal because the PC version is expensive too" because there's no way in hell I'd buy those cards for a PC.

Also... you'd be able to sell that PC FireGL card afterwards. Not sure what the aftermarket demand for FireGL daughterboards for Mac Pro will be if and when they give you the possibility to upgrade.

RAM pricing is right where I expected it to be. 4 x 16GB 1866MHz RDIMMs should be right about $1K. So about 60% more than what it *should* be for the upgrade, but their pricing is right in line with HP, Dell, other workstation vendors so I wasn't expecting a deal here. Or anything different. In a couple months' time when this RAM is more plentiful and available from OWC and Crucial, I would probably buy the third-party RAM. For now, I'm planning to just order it from Apple so I can get it here and everything is ready to go.

So, I think for me and most R3D related workflow tasks or general workstation use. The 8-core is going to be the sweet spot. With 1TB PCIe storage, dual D600's and 64GB for me.

I would recommend the 12-core to people who have ample room in their budget and who will be spending most of their time in apps that can make use of that many cores regularly. For a system primarily used for Resolve or 3D apps like Maya where the extra cores can be readily utilized, the 12-core could be their ideal system. I will be running Resolve and Maya myself, but I run lots of different apps and the apps I use the most are not going to benefit from 12 cores over 8, especially when most will use fewer than 8 cores / 16 threads and could benefit from the faster clock speed of the 8-core CPU. The 6-core also has an advantage there as well, but I think 6 cores and 12 threads is cutting it a bit tight with some current softwares like Adobe CC.

Interesting - so you're actually going for one, Jeff?

Personally I'm still feeling that for the work we do, the GPUs are too underpowered (not enough compute, plus 3GB VRAM per GPU is cutting it really close) and a nVidia Maxwell setup with enough VRAM will make everything else look a little silly by next year.

What I saw at SIGGRAPH with V-Ray RT on nVidia GPUs blew me away. And I'm sure that nVidia (who own Mental Ray) will likely have a compelling GPU-rendering answer to V-Ray RT for VFX / film / motion graphics pipelines soon too. Which will not be available to Mac Pro users who have AMD.

So I'm basically looking at getting in 2014 (eg less than 12 months away):

Cheap PC option: 8-core Haswell-E (Extreme Edition, not Xeon) with 3x nVidia Maxwell GPUs with 8GB VRAM each plus 64GB system RAM.
or
Apple option: Cylindrical Mac Pro v2: Haswell Xeon (8-16 cores?) plus 2x nVidia Maxwell GPUs (hopefully) with 8GB VRAM each (hopefully)... if they go AMD no chance.
or
Workstation PC: dual Haswell Xeon (16-32 cores), with 4-5 nVidia Maxwell GPUs with 8GB+ VRAM each plus 128GB system RAM.

For my own work, I'll probably go the Cheap PC route and take the extra $$$$ I save and put that into cloud rendering credits.

But if Apple don't make stupid GPU decisions, I'll take a look at version 2 of the Mac Pro. Also, it'll have HDMI 2.0 built in, which should make things a lot nicer.

This version just seems to be too much "early adopter tax" to me. It will do 4K. But it possibly won't do it nearly as well as a 3x Maxwell GPU system could for less money in less than 12 months.

Anyway, I love reading your insights so am fascinated with what you think!

Bruce Allen
www.boacinema.com

 

[Bob Gundu]

With 1TB PCIe storage, dual D600's and 64GB for me.

You mean Dual D700's right Jeff?

 

[DJ Meyer]

Workstation PC: dual Haswell Xeon (16-32 cores), with 4-5 nVidia Maxwell GPUs with 8GB+ VRAM each

I doubt you'll see 2P Haswell Xeons in 2014. Or if so, not till laaaate 2014.

 

[Brian Merlen]

For Avid MC7, Redcine, CS5 Ppro, what is the sweet spot core wise? I dont want to over buy and I need to be realistic about my needs. I might begin transitioning into learning Resolve, assuming I could get a decent experience without spending over 5k or so. I was thinking of going for a fully kitted rMBP but I been putting it off waiting to weigh in on tower BTO pricing. I heard a lot of good stuff about rMBP and Red workflows, but I imagine this tower is going to blow the heck out of it and a 4 core tower is about the same price as a fully kitted rMBP which is why I am considering springing for a tower at all...

 

[Bruce Allen]

I doubt you'll see 2P Haswell Xeons in 2014. Or if so, not till laaaate 2014.

Well, we are almost in eeeaarrly 2014...

http://vr-zone.com/articles/haswell...ntel-server-cpu-gravy-train-speeds/66841.html

Sounds like Haswell-E and 2P Xeons by end of 2014, yes!

Bruce Allen
www.boacinema.com

 

[PatrickFaith]

My MacBook Pro retina is doing great with nuke 8, I think people will be surprised how much less they need gpu with deep workflows, and how they won't need massive render farms for deep renders.

 

[Frank Weeks]

Do you really think that at this point Apple is incapable of developing/marketing consumer gadgets and professional apps/hardware at the same time?

That might have been true a couple years ago, but now that Apple is one of the largest corporations on the planet and is sitting on a pile of cash that would make Smaug blush I'm willing to bet they have enough organizational capacity to pull it off. From a PR perspective, even if the "pro" side only made marginal profits, it would make sense to make a super cool computer just to burnish the brand. I believe the auto makers call it the "halo" effect. Just sayin'

Cheers - #19

This is what I believe. I hope tomorrow. I'm going to drop a load on the future. Following Jeff's lead. I doubt I'll be alone.

 

[Jeff Kilgroe]

OK, I'm calling shenanigans on the MacRumors pricing reveal..

I came home tonight to find email in my box from my business sales rep who said he'd have pricing today for me. Let's just say that what he sent me and what is on the MacRumors site are two different things. I'm not going to post what I was sent. For one thing, I'm supposed to hold it in confidentiality. And I also have a couple questions that need answering or at least some degree of clarification.

@Bob Gundu -- Yes, I meant dual D700's

@Bruce Allen -- I agree with a lot of what you're saying, but then again for the tasks you're talking about, I would be building PC's. Here we're talking about a Mac Pro. Yes, I'm buying a Mac Pro and it will be used to do things that it will handle well. For tasks it won't handle as well, there are other tools for those jobs. I need to remind people from time to time that I own and use several computers. As do most any content creation businesses and professionals. I'm planning to start with one 8-core Mac Pro. Probably not a 12-core at this time due to pricing and Intel not really offering a compelling option for 12-core CPU speeds. I will most likely buy a couple more Mac Pro systems after they have been around and we get an update of some sort. While Haswell core Xeons are still a ways off, there are other models to the E5-V2 Ivy Bridge Xeon lineup coming over the next few months. Including faster 12-core options.

FWIW, the Haswell Xeons are on the roadmap for Q2 2015, or I should say that's what they're roughly penciled-in for. Given the track record of Xeon releases over the past few years, I wouldn't hold my breath or try to plan for them. nVidia is also behind schedule on GPUs. Not sure what's going to happen here, but they can't even get all their announced Kepler products to market. Who knows where Maxwell and other upcoming GPU lines are headed. IMO, I think they need to do away with their schizophrenic product catalog and consolidate several of their product lines into a more common base. Then they need to actually make more than a minimal effort to properly support OpenCL and find a way to better incorporate OpenCL and CUDA so the industry can move on to better things. The industry at large is shifting to OpenCL and CUDA is only attracting certain highly-specialized tools as we move forward. The CUDA-optimized V-Ray is nice, but with the shifting and posturing going on in the GPU industry right now, I'm not so sure I'd want to standardize on a proprietary platform. Either as a developer or an end user

Looks like Crucial is now taking orders for Mac Pro late 2013 RAM. For 64GB it's around $900 give or take, depending on if you have to pay tax. I'm debating on this one, but with what I've found with the pricing above, I'm going to wait for the official Apple info to show up. And like I said, the quite I received to day leaves me with questions. Not so much questions about what I was sent, but what I was not sent. IMO, it looks like the info is starting to show up in their systems and they're feeding pieces to some of their customers. I probably have questions because they don't have all the answers yet.

 

[Frank Weeks]

8 core, dual D700s 6GB, 1 TB flash storage, 32 GB RAM. This is where the sweet spot lies (I hope). And I'll need a 3-slot pci extension board. (Red rocket, Black Magic (for divimci) and E-sata extension card)
If I'm misguided here Jeff, let me know.

 

[Dave_S]

Does new mac pro allows pcie expansion for gpu via thunderbolt?

 

[Brian Merlen]

I wonder what you weren't sent...love the cryptic messages Jeff! U got us on the edge of our seats...I am betting right now the GPUs cost wayyyy more then what macrumors says...

 

[Brian Merlen]

Does new mac pro allows pcie expansion for gpu via thunderbolt?

as of now, no (it does not exist) :smiley:

BUT

even if it did exist, it still does not allow it cuz of interpolarity issues... hopefully one day they allow it, but TB2 still will limit the through put even if it did...

 

[Jeff Kilgroe]

Does new mac pro allows pcie expansion for gpu via thunderbolt?

This is an OSX problem, not a hardware problem. So in a nutshell, it works just fine on the new Mac Pro if you're running Windows. GPU expansion via Thunderbolt does not currently work with OSX or Linux. Whether you're running common PC hardware or Apple hardware.

Apple knows people want this, but it's going to require a major change to how the graphics driver interface is incorporated into the OS. There were some guys out there who made an OS patch/ hack to allow for Quadro 4000 cards to be installed in a Thunderbolt expander, but it didn't work all that well and the performance was terrible.

I wonder what you weren't sent...love the cryptic messages Jeff! U got us on the edge of our seats...I am betting right now the GPUs cost wayyyy more then what macrumors says...

Heh heh.
Been on the phone this morning. The CTO options are not all available in the product database yet. So we were spoon-fed the few morsels they had. There is no pricing yet on the dual D700 GPU upgrade and a few other things. I also wish that people would stop with the gloom and doom over the D700 and saying that they each only have 3GB of VRAM. Not true. It clearly says in multiple places on the Apple site and Mac Pro pages that *EACH* D700 GPU will have 6GB DDR5 VRAM, that means 12GB total VRAM. I've been ignoring it when people keep saying it, but just thought I'd lay that out there.

I'm a bit concerned about pricing on the D700. They're spec-for-spec identical to the AMD FirePro W9000, just in the new GMX form-factor. The W9000 cards are $3300 each and direct competitors to the Quadro K6000. No way the upgrade to dual D700's is only $600. If it truly is, then Apple has worked some serious magic here.

 

[Rudy Lehfeldt-Ehlinger]

Ive been looking at the configuration prices that macrumors have been posting and while they all look like what I was expecting ram has always been cheaper to not buy from apple...

Does anyone have a reason why I shouldnt get the base ram and buy 4x of these http://www.crucial.com/store/partspecs.aspx?IMODULE=CT16G3ERSDD4186D and save my self $800?

 

[Bruce Allen]

@Bruce Allen -- I agree with a lot of what you're saying, but then again for the tasks you're talking about, I would be building PC's. Here we're talking about a Mac Pro.

Sorry, I wasn't very clear... at the place I work, we use over a dozen 12-core and 8-core Mac Pro Mac Pros right now for Maya, After Effects, etc.

For a couple reasons:
1. Mental Ray renders faster on Linux and Mac than Windows
2. More freelancers who can design well are more comfortable with Mac OS X than Windows
3. Our Avids are all on Mac OS X and it's nice to have parity there - eg highlight files in the Finder and say "grab the RED file!"
4. We have projects that go on for years and wouldn't want to swap network shares, templates, render farm setup etc.

That's just in the theatrical graphics department. Add in broadcast, editorial, finishing etc and we have a LOT of Mac Pros.

In the past most of what we do has been CPU-bound. And just as GPU rendering is about to make sense, Apple chose an unsuitable GPU. Plus they limited us to only 2 GPUs (overkill for many tasks but not quite enough for heavy GPU tasks).

I agree that in an ideal world, OpenCL is nice. But if you compare actual running apps and even what is on the horizon... it's difficult to deny that nVidia is way ahead of AMD and everyone else and will likely stay that way for the next few years at least. Giving Apple some cards badged as FirePros at a killer discount so that Mac Pro has AMD inside might be a good way to fight back a little but I don't really think it's enough. In the end, the performance needs to be there.

But as you say... for what we're doing right now, PC seems the better choice. So I think you see our dilemma since we are currently Mac Pro users!

I also wish that people would stop with the gloom and doom over the D700 and saying that they each only have 3GB of VRAM. Not true. It clearly says in multiple places on the Apple site and Mac Pro pages that *EACH* D700 GPU will have 6GB DDR5 VRAM, that means 12GB total VRAM. I've been ignoring it when people keep saying it, but just thought I'd lay that out there.

Well that's good at least! Missed that.

I'm a bit concerned about pricing on the D700. They're spec-for-spec identical to the AMD FirePro W9000, just in the new GMX form-factor. The W9000 cards are $3300 each and direct competitors to the Quadro K6000.

Well... just because AMD priced it to compete with a K6000 / K5000 doesn't mean it actually competes with a K6000!

In fact, a $1000 Titan or even a cheap 780 card could work much better in many cases.

While Haswell core Xeons are still a ways off, there are other models to the E5-V2 Ivy Bridge Xeon lineup coming over the next few months. Including faster 12-core options.

FWIW, the Haswell Xeons are on the roadmap for Q2 2015, or I should say that's what they're roughly penciled-in for. Given the track record of Xeon releases over the past few years, I wouldn't hold my breath or try to plan for them. nVidia is also behind schedule on GPUs. Not sure what's going to happen here, but they can't even get all their announced Kepler products to market. Who knows where Maxwell and other upcoming GPU lines are headed.

Thanks for the good info! Yeah, I'm just going to wait and see.

IMO, I think they need to do away with their schizophrenic product catalog and consolidate several of their product lines into a more common base.

Oh, totally!

Then they need to actually make more than a minimal effort to properly support OpenCL and find a way to better incorporate OpenCL and CUDA so the industry can move on to better things. The industry at large is shifting to OpenCL and CUDA is only attracting certain highly-specialized tools as we move forward.

Well, if you look at 3D rendering, we have:

V-Ray: originally OpenCL, now focused on CUDA because AMD OpenCL sucks currently
Mental Ray: owned by nVidia, next gen stuff will be CUDA
Arnold: CPU
RenderMan: CPU

Plus, the commercial-grade GPU challengers (Octane, Arion, FurryBall, etc) either run faster on nVidia or are nVidia-only.

The CUDA-optimized V-Ray is nice, but with the shifting and posturing going on in the GPU industry right now, I'm not so sure I'd want to standardize on a proprietary platform. Either as a developer or an end user

I agree. I hate the shifting, posturing, etc.

Although if you're talking about proprietary platforms... I don't care as much whether the developer uses OpenCL or CUDA or Mantle - as long as they use the fastest tech available and have an engine that can shift to another tech if it becomes faster.

For example: V-Ray is kinda the opposite of proprietary. The same shaders work in both CPU and GPU. And Vlado has tried to do OpenCL in the past and will try in the future assuming AMD can actually make it work one day.

I think that much of the shifting and posturing would stop if AMD actually built a fully-working OpenCL solution that competed with nVidia's existing CUDA stuff!

Bruce Allen
www.boacinema.com

 

[paul schefz]

I'm a bit concerned about pricing on the D700. They're spec-for-spec identical to the AMD FirePro W9000, just in the new GMX form-factor. The W9000 cards are $3300 each and direct competitors to the Quadro K6000. No way the upgrade to dual D700's is only $600. If it truly is, then Apple has worked some serious magic here.

that was the bit of info that made me doubt the whole thing as well....even at 600 each it would be very very nice...
on the other hand i just can't see the pricing to be 4000 base+3000 for the 12 core + 3000 each for the D700....that would seriously turn people off and feed the whole apple tax thing....
so as always it will probably be somewhere in the middle....i won't believe anything until i see it up on the apple store.....

 

[Jeff Kilgroe]

I think that much of the shifting and posturing would stop if AMD actually built a fully-working OpenCL solution that competed with nVidia's existing CUDA stuff!

Funny thing is, that was 4 or 5 months ago. Different story now. OpenCL in its latest incarnation is kicking ass and the AMD and PowerVR chips are doing well to support it. Resolve on W8000 and W9000 GPUs is smokin'! While still technically inferior to CUDA, it's gaining traction because it's an open spec and has direct integration with OpenGL now. The only reason CUDA exists is because nVidia wanted to take the industry in this direction, but they had to create their own solution. It came by way of a few different techs like nVidia Cg, that evolved into CUDA. By the time OpenCL was a real spec on a drawing board, CUDA was already seeing real-world use.

I totally understand all your dilemmas, Bruce. We face many of the same issues here. Predominantly our renderers we use most are Mental Ray, Modo and Lightwave. FWIW, the Mental Ray team is also dabbling with OpenCL.

Not sure how you feel about it, since it's somewhat of an unknown factor at the moment, but the compactness of the Mac Pro and direct Thunderbolt integration are going to be huge. This is a computer designed for clustering. Apple even said that at WWDC, not in the keynote, but with their tech people talking. We can get the same CPU and GPU density out of the new Mac Pro systems as we can with dual Xeon + dual GPU 1U server "blades" and even with a single link, we have 20Gbps interconnect fabric. So, software dependent of course, if you find 1 x 8-core and dual GPUs to not be enough, set another Mac Pro next to your current system and link it at 20, 40, 60 or even 80 Gbps… Chew on that one for a moment.

I'm not trying to say the new Mac Pro is the end all, be all of workstation tech or anything like that. For many jobs, it's not the right tool. OTOH, I think it's greatly misunderstood. On its own, it can't replace a dual Xeon tower with more than two GPUs. So in this sense a bit of out of the box thinking needs to be applied. Likewise, when people start whining about the lack of cores or the lack of GPUs and then start referencing various PC solutions I have to wonder… If these issues are so important, why are we not seeing a greater prevalence of quad-CPU systems? I build one here and there and used a couple for a time for rendering. Works great, although expensive. We see lots of quad-CPU 32+ core systems in the high-end server world.

When it comes to tasks like rendering, that are easily distributed, I think I would much rather go for a distributed and nodalized approach of one good CPU and a couple pretty good GPUs per node. This makes a lot of sense. Fill in the render farm with add on Linux or Windows PCs that can stack 2 to 4 CPUs and/ or 4 to 6 GPUs in a single box.

While we can talk of clustering the new Mac Pro, what is really going on here is a play on desktop workstation setups. While this iteration may not have all that we could want or hope for, it's a step in the right direction, IMO.

FWIW, the chatter about the new Mac Pro must be getting pretty thick. I've noticed HP has really stepped up their workstation marketing the past couple days...

 

[Jeff Kilgroe]

Mac Pros are actually here. Resellers are taking delivery of the stock configs! Today! Authorized business sales resellers and business sales reps at Apple Stores can take orders for them starting today!

No official word on CTO pricing or the systems going online on the Apple Web store. Indicators seem to be saying tomorrow or thursday.


…Just thought y'all should know. MacRumors and similar news sites are failing on this. Sad, really sad.

 

[Blair S. Paulsen]

Some of Jeff's notes mimic my thinking on how to best exploit available resources. The real world performance of TB2, particularly when multiplexed/bonded will be critical to practical clustering and what I call ad hoc render farms.

I am very attracted to the potential flexibility of having a 6 pack of MacPro cylinders cranking out renders one week and then each off doing other tasks the next week. Its essentially an external version of parallelism when max crunch is needed, but with the advantage of units that can exist on their own as well.

Or maybe I'm just crazy...

Cheers - #19