Sir Lawrence Olivier contemplating the new Mac Pro
EDIT FEB. 28, 2014 - Updated to OSX 10.9.2 a couple days ago, no issues to report. See the link below for the update guide and also to get CUDA and OpenCL working properly if you have a Nvidia GK110B GPU. To get my GTX780 up and running took a bit of work.
EDIT FEB. 21, 2014 - Updated to Mavericks 10.9.1 following MacMan’s guide found here:
Just do exactly as he says as usual and all will be well. Everything works as it did in 10.8.5.
For anyone using Assimilate Scratch or Scratch Lab, support has given the go ahead for Mavericks. I have been running the v8 beta for a few days and am loving it but have yet to give it a full workout.
Now for the exciting part and I quote from support - “Note that LAB V8 now uses multi-GPUs to debayer RED material.”
That is for MX media, not Dragon yet but that is in the works as well. Will have to test my GPUs vs Red Rocket one of these days. Also from Assimilate support regarding some v8 questions - “Scratch is 1 GPU only. When processing wrappers (MXF / QT), GPU does the color but encapsulation remains a CPU task” Lots of great news with the v8 release overall.
EDIT JAN. 28, 2014 - Updated with R3D playback performance below
A lot of people in our industry are wrestling with inevitable hardware purchase options given that the 2010 Mac Pro design is EOL. What makes this year’s decisions particularly difficult is that if one wants to invest in Apple to manage demanding workstation tasks and use their existing PCIe hardware, the 2013 Mac Pro requires an external Thunderbolt expansion infrastructure if it is to directly replace last year’s model.
Most concerning for me is that despite the incredibly innovative new design, Thunderbolt expansion solutions for multiple PCIe cards unfortunately mandates potentially significant PCIe lane sharing that was formerly a manageable issue for power users on the previous Mac Pro platform. Apple is apparently not concerned with this impending headache for many of us though:
"It’s our most expandable Mac yet." Really? How so? With the 2009 / 2010 design, the 2nd PCIe x16 slot allows for an external PCIe expansion chassis to be connected to the Mac Pro with a bidirectional data pipeline of 8000MB/s including overhead. (I’m assuming you occupied the first x16 slot with your OSX GPU.) Not to be forgotten are the two PCIe 2.0 x4 slots offering 2000MB/s bandwidth each. That’s 24 PCIe lanes total available for expansion.
With the 2013 model, this is what you get for PCIe card expansion:
12 PCIe lanes total. Each one of the 3 Mac Pro Thunderbolt 2 Falcon Ridge controllers connects to the PCH / CPU via a Gen2 PCIe x4 bus, period. In theory, the fastest you can expect out of each TB2 controller or any single TB2 port will be approximately 2000MB/s given an approximate 20% overhead penalty. (20Gbps = 2500MB/s.) Old Mac Pro fastest expansion option - 8GB/s vs new Mac Pro fastest expansion option - 2GB/s. Bummer.
Or is it? To be fair, 2GB/s per controller is fast enough for my needs most of the time. My 24TB RAID10 (8 x Hitachi Ultrastar 3TB SAS drives - 12TB usable) and Atto R680 would easily run full speed on one TB2 controller provided nothing else is connected. I say that because unless TB2 has reduced latency over TB1, I have found that IOPS / read / write speeds take a nosedive when a TB controller has to manage more than 1 working device at once. Multiplexing multiple devices’ data into one PCIe x4 bus takes time and resources that do affect individual device performance. With 2 controllers remaining, the 2nd Falcon Ridge controller would be saturated with my Red Rocket as x8 Gen1 = x4 Gen2 bandwidth. That leaves 1 TB2 controller to take care of my Atto H608 HBA card for LTO-5 and eSATA shuttle drives, Sonnet Qio E3 for SxS cards, Myricom 10GbE fibre card (for Phantom docks), Decklink 3D Extreme SDI card for Resolve, Aja Kona 3G for Scratch Lab, FW800 PCIe card, and Highpoint Rocketraid 2314 for Redmag readers.
Obviously it’s unlikely all of those are goIng to run at full speed sharing the same TB2 controller but in reality I probably wouldn’t configure the cards that way anyway. I’d test thoroughly to find out the best way to share the available bandwidth between the 3 controllers. Until then I can’t say what the actual speed penalty would be using Thunderbolt 2 expansion vs. x16 PCIe expansion, but what I do know from experience is when you’ve got a half million 5K frames in the render queue and multiple delivery formats to churn out ASAP, every little bit of performance helps.
Then there’s durability. PCIe cards are fragile unquestionably but mounted properly inside a PCIe slot, there’s really no reason for them to get physically damaged except maybe from electrical issues. All I know is that it’s less mess and more robust for me to take a mac pro or hack to set than it is to take a thunderbolt Mac and multiple expansion chassis plus additional thunderbolt peripherals. By the time you get all the AC cables plugged in and all the thunderbolt daisy chain cables connected, you’ve got many potentially disastrous connections in that scenario. With a tower, all of the PCIe peripherals are bus or PSU powered and there is only 1 AC cable to worry about if you’re not running a PCIe expansion chassis. If you are you then have 2 more cables to worry about. No big deal. Plus the footprint is actually smaller with a tower and internal peripherals than an equivalently equipped Thunderbolt setup.
There is one last reason I’m not a huge TB fan. I have experienced rare but painful TB peripheral crashes especially when daisy chaining devices. A restart is usually required. I don’t have this problem on Mac Pro or hackintosh setups with the same cards. Not much more to say about that. I get it though, not many of us need this much expansion bandwidth and Apple will undoubtedly expand it’s Mac Pro market by making it this small and cool looking. Too bad their wasn’t yet another version that would cater to Apple’s top end power users made by Apple.
Assuming that your expansion needs will be met with Thunderbolt 2 and you’re ready to pull out your wallet, you might want to take a look at the new Mac Pro graphics vs. other options one last time before committing.
Truly the 2013 MP is well equipped in the graphics department with dual AMD D700s available but it’s still frustrating being unable to take advantage of additional PCIe GPUs for applications like Davinci Resolve and Adobe Creative Cloud. CUDA acceleration is not supported on the available Mac Pro graphics options but that will become less of a concern as the months roll by and OpenCL becomes better utilized in pro media apps.
Still, as good as the dual FirePro GPUs are, they aren’t quite as impressive as dual Titans or GTX780s, at least on paper. For example, the EVGA ACX GTX780 Superclocked Edition tops out at 4.3 teraflops vs. the mac pro D700 at 3.5 teraflops in single precision performance. Here’s a few more specs:
NOTE: I am not 100% sure of their compatibility with Mavericks as I am still running 10.8.5 on all of my Macs.
EVGA GTX780 ACX Superclocked Edition:
NOTE: GTX 780 ACX SC OpenCL not working on 10.8.5, OpenGL / Cuda is working properly..
GTX 780 ACX SC Cuda-Z score:
(Sorry about the lame pics - I did this in a hurry…)
EVGA GTX Titan:
NOTE: I have not personally tested the Titan but rumour is some people have both OpenCL and OpenGL working properly in Mavericks.
ASUS GeForce GTX 770 DirectCU II OC:
Graphics Engine: NVIDIA GeForce GTX 770
Bus Standard: PCI Express 3.0
Video Memory: GDDR5 2GB
CUDA Cores / Shading Units: 1536
Memory bandwidth: 224GB/s
SMX Count: 8
Pixel Rate: 33.9 GPixel/s
Texture Rate: 135 GTexel/s
Floating-point performance: 3.25 TFLOPS
Two of these should perform similarly to the dual D500 Mac Pro option considering the nearly identical specs.
NOTE: OpenCL and OpenGL / Cuda working 100% on this GTX 770 card running 10.8.5 and OSX default Nvidia drivers.
GTX 770 Cuda-Z score:
GTX 770 Luxmark OpenCL score:
Heaven Benchmark 4.0 - Extreme settings but Tessellation not working in 10.8.5 to my knowledge:
Added Jan. 28, 2014:
R3D playback clip: Epic 5120 x 2700 Redcode 8:1 23.976
CPU - i7 3970x Overclocked to 4.4GHz
RAM - G-Skill 64GB 1866MHz
GPU - 1 x Asus GTX770 2GB
Storage - 12TB RAID10 - ATTO R608 & Hitachi Ultrastar 3TB SAS x8
Redcine-X Pro version 22.1.31200
1/2 debayer - 23.976 FPS - no dropped frames
Full debayer - drops frames - stalls briefly every 65 frames
Davinci Resolve 10.0.2 (switching between the OpenCL / OpenGL GPU setting didn’t change results)
1/4 good - 23.976 FPS
1/2 good - 21 FPS
1/2 premium - 10 FPS
Full premium - 5 FPS
2013 Mac Pro GPU options:
Click on this link for a great Nvidia GPU comparison chart found at guru3d.com
I am damn impressed though by the D700 specs considering the tiny size of the new tower design. In real-world use there will probably be little meaningful difference between dual Titans and dual D700s according to these specs but of course this is all guessing for the time being.
Moving Awkwardly Forward:
It was with all this in mind that I reluctantly committed to another hackintosh build. Before I get into that though I must give a huge thanks to Tonymacx86 and the amazing contributors on his website and of course Apple. I mention Apple because so far, they have been turning a blind eye to the Hackintosh community. Without them both, hackintoshes would not be an option.
This actually wasn’t a difficult decision once I started looking into it again despite having sworn that I was done building hacks. I’ve built 6 hackintoshes since Leopard and they are unquestionably a pain in the ass to perform updates on. Much less so than in the past but still not totally stress free. Once dialed in though stability has not generally been an issue for me. I rarely feel like an OSX crash is because of my core hack components - namely CPU, RAM, motherboard and platform specific kexts. Instead, they’re almost always caused by application bugs or peripheral hardware conflicts that likely would have also caused a problem on my real Macs. I say this confidently because as I’ve mentioned I regularly share hardware between my hacks, Mac Pro, and TB Macbook Pros. They all have issues at some point. I consider my hack builds stable when I get all graphic benchmarks performing at expected or better speeds and when the system can successfully run the Prime95 Torture Test (Blend mode) for 24hours while staying under 80 degrees Celcius on all CPU cores.
In the end my gear decisions have to be based on overall value plus meet a few basic needs. The 1st need is getting the job done right as quickly as possible within my available resources. Translated this means: faster transcodes = more sleep. Simple. 2nd to that is ensuring my business plan remains intact which is definitely getting harder to do each year. That could easily be a post on it’s own though. When I look at all available options as objectively as possible I have to admit that if it became impossible to build hackintoshes tomorrow it would make more financial sense for me to migrate to Windows than to invest in the new Mac Pro system. Prores encoding is the main technical obstacle I would face in that scenario and there are now economical solutions for that such as Cinec. For me, at least for now these needs are met best using PC parts running OSX.
One Last Hurdle:
If you are a more morally upstanding person than myself or part of a large organization there is at least one additional consideration you will be faced with if considering a hackintosh. That is, the legal and moral implication of installing OSX on PC hardware. No matter how you slice it, it is a violation of Apple’s OSX EULA to install OSX on non-Apple hardware. This is a huge debate and has been discussed many times in many languages but If you’re a large company, the Apple OSX EULA violation makes Hackintoshes a no go. No project manager or legal department would likely allow it. For small to medium sized businesses particularly in the VFX / edit house world, hackintoshes are not uncommon but generally kept out of the client’s line of sight. Although extremely unlikely, hopefully Apple will someday open up the OSX EULA to accommodate hackintosh installations from a legal standpoint.
Above: OSX reads 4.3GHz but I it is actually running at 4.4GHz. Haven’t looked into a fix yet.
Here is a list of my hackintosh parts that should closely match the performance of the new 3GHz 8 core E5-1680 Mac Pro model with 64GB RAM:
Intel i7 3970x CPU - $700 (eBay) Overclocked to 4.4GHz
64GB G-Skill 1866 RAM (2 x F3-1866C9Q-32GXM) - $660
Gigabyte X79S-UP5 Wifi - $355
Asus GTX770 OC GTX770-DC2OC-2GD5- $370
(my EVGA GTX780 ACX SC is collecting dust until the 10.8.5 GK110B OpenCL issue is resolved)
Silverstone 1200W PSU - $233
Intel RTS2011LC CPU cooler - $55
Coolermaster Elite 330 ATX case - $55 (already owned)
2 x Cooler Master Blade Master 120mm fan - $30 (already owned)
OSX drive - 1TB Hitachi Deskstar 7200RPM - $100 (already owned)
OSX Mountain Lion $20
Hackintosh install guides - tonymacx86.com - by donation
Total - $2558
As an aside I am not installing OSX Mavericks until I get the go ahead from Assimilate. I need Scratch Lab to be as stable as possible which means sticking with 10.8.5 for now.
The performance estimate I’m using as a reference for the 2013 Mac Pros can be found here on the Geekbench website.
My Geekbench 3 64bit score @ 4.4GHz 24/7 stable:
2013 Mac Pro Geekbench 64bit estimates:
There is a new Geekbench 64bit score for the new 12 core Mac Pro: 33066
For now I’m satisfied getting close to the 8 core model performance-wise relative to the cost. My 3970x build is about 30%-50% faster than my workhorse i7 970 / X58A-UD5 / 24GB RAM OSX system with a GTX660 for what it’s worth.
Lastly, here are some estimated performance / price comparisons for you to digest:
2013 8 core Mac Pro - $7700 - 5% faster than my build - $5142 more than my build
2013 12 core Mac Pro - $9700 - 30% faster than my build - $7142 more than my build
At least you get a significant speed bump for an extra $7000…
Anyway, I don’t know how many of you are in the same boat as I am but I hope this helps you find the right solution for your workstation needs whatever they may be.