Why would you need liquid cooling to add more than 4 GPUs to a PC? Do it the same as Mac Pro; add PCIe expansion chassis, or eGPU via Thunderbolt.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
MP 7,1 GPU rendering on the new Mac Pro. News, updates, chat.
- Thread starter vel0city
- Start date
- Sort by reaction score
I didn't realise eGPU over thunderbolt were well supported on PC - I thought that was a Mac special thing. I haven't heard on forums about that being done for GPU rendering on PC, but on Mac it seems to be how people are doing GPU rendering on say iMacPro. Does PC eGPU for GPU rendering work fine?
I think Apple bet high on metal to f##k-off from prooietary API as CUDA which would mean having to mandatory include nVidia GPU, given the pace (and high success) nVidia has, it could mean every game, graphics or gpucompute developer paying tribute to nVidia, and all other API being eclipsed and considered second class choice for developers (and users).
ok that's right, they had legit concerns about.
what happened?
Vulkan initially backed by Apple not just arrived late to the party, also delivered s conceptually inferior API to CUDA, khronos group definitely is not successful or well motivated.
Given OpenGL/opencl and Vulkan actually don't rival CUDA Apple decides to take it on own, developing metal.
Metal it's a success in Apple ecosystem for gaming 3d and some gpucompute/ml.
But metal has it's own strategic and fatal failures:
It's an closed API, even while apple uses llvm to compile to .msl it's not documented, you have to do some reverse engineering if you want to compile directly to .msl, it's definitely does not help, and rises walls to support open STDs (as an native opencl/hip/SyCL compiler targeting metal .msl) and made mandatory to use Xcode to develop GPGPU, it's has as drawback developers having to code twice (at least) their solutions to address a market which represent 10-15% of it's possible revenue, few developers has economical motivation to do that, so it's unlikely an new GPU accelerated applications or game to be first on the apple ecosystem unless it comes from a developers already having a functional team for Apple developments.
But then comes the devilish machine learning, while most framework relies on Opensource and openstds few key openstds do not work or are deprecated or just don't have access to GPGPU on macOS as opencl required for Tensorfow in absence of CUDA, two years ago apple promised to have Tensorfow in metal in a year, it didn't happen yet, those doing ML targeting Apple ecosystem are using client (mac)-cloud(Linux/CUDA) approach to be productive and have it's inference done, this approach is often followed for those doing natural language processing w/o regard the platform they target as its inference use to be too long and heavy for workstations and ussualy is done at inference farms, but other applications as image detection or manipulation don't require nothing you can't put in a workstation, these development/research cant be done in a Mac, even the Mac pro, while apple offers CoreML there is no ml/DL papers using it (or at least are as common as a white whale).
While the purpose behind metal is valid, metal has no chance to withstand against CUDA or SyCL/Vulkan or oneAPI in the long term, unless apple open it more and even re-open again macOS to 3rd party API as CUDA, oneAPI, and Vulkan, so macOS users are not doomed to be second class citizen on the new world of GPU accelerated applications (3D, VR, AR, ML/DL/AI ) and stay passive waiting for a developer to finally release a Mac or iOS version of it's window/Linux/Android successful titles.
ok that's right, they had legit concerns about.
what happened?
Vulkan initially backed by Apple not just arrived late to the party, also delivered s conceptually inferior API to CUDA, khronos group definitely is not successful or well motivated.
Given OpenGL/opencl and Vulkan actually don't rival CUDA Apple decides to take it on own, developing metal.
Metal it's a success in Apple ecosystem for gaming 3d and some gpucompute/ml.
But metal has it's own strategic and fatal failures:
It's an closed API, even while apple uses llvm to compile to .msl it's not documented, you have to do some reverse engineering if you want to compile directly to .msl, it's definitely does not help, and rises walls to support open STDs (as an native opencl/hip/SyCL compiler targeting metal .msl) and made mandatory to use Xcode to develop GPGPU, it's has as drawback developers having to code twice (at least) their solutions to address a market which represent 10-15% of it's possible revenue, few developers has economical motivation to do that, so it's unlikely an new GPU accelerated applications or game to be first on the apple ecosystem unless it comes from a developers already having a functional team for Apple developments.
But then comes the devilish machine learning, while most framework relies on Opensource and openstds few key openstds do not work or are deprecated or just don't have access to GPGPU on macOS as opencl required for Tensorfow in absence of CUDA, two years ago apple promised to have Tensorfow in metal in a year, it didn't happen yet, those doing ML targeting Apple ecosystem are using client (mac)-cloud(Linux/CUDA) approach to be productive and have it's inference done, this approach is often followed for those doing natural language processing w/o regard the platform they target as its inference use to be too long and heavy for workstations and ussualy is done at inference farms, but other applications as image detection or manipulation don't require nothing you can't put in a workstation, these development/research cant be done in a Mac, even the Mac pro, while apple offers CoreML there is no ml/DL papers using it (or at least are as common as a white whale).
While the purpose behind metal is valid, metal has no chance to withstand against CUDA or SyCL/Vulkan or oneAPI in the long term, unless apple open it more and even re-open again macOS to 3rd party API as CUDA, oneAPI, and Vulkan, so macOS users are not doomed to be second class citizen on the new world of GPU accelerated applications (3D, VR, AR, ML/DL/AI ) and stay passive waiting for a developer to finally release a Mac or iOS version of it's window/Linux/Android successful titles.
Last edited:
Vulkan is only few years around. How long is CUDA development?
Vulkan is open source inititative, and development of OSS usually takes of in a factorial manner. Look at where Linux is right now!
Vulkan is not there yet. But knowing OSS and how it works, we can expect that Vulkan will be the "OneAPI to rule them all"
.It just needs one breakthrough in some way.
I was enthusiast about Vulkan, now I'm disappointed, Vulkan maybe an foss but is developed by an group paid by the industry, it won't happened spontaneously as Linux, you can't apologize it slow pace on being fossVulkan is only few years around. How long is CUDA development?
Vulkan is open source inititative, and development of OSS usually takes of in a factorial manner. Look at where Linux is right now!
Vulkan is not there yet. But knowing OSS and how it works, we can expect that Vulkan will be the "OneAPI to rule them all"
It just needs one breakthrough in some way.
It is exactly the same development process, and monetization as GNOME development, or Blender's.
Yes, the development of Vulkan has payed developers. Yes, they are payed by the industry: AMD, Nvidia, Intel, Valve, etc. But it is still collective job, not one company overpowering others.
It is new API, which is also the reason why it hasn't got traction in GameDev. Lack of experience with it puts Devs off of it. That doesn't change the fact that it is the best API of them all.
Give it time. Its inveitable, with development and maturity of Linux, ("cough", consoles, "cough") that Vulkan will get traction.
Errrr. That is not what happened. Maybe in some alternative universe, but not this one.
Apple really didn't back Vulkan specifically all that much at all. Khronos was having lots of trouble getting to a "what's next" solution for OpenGL. That standards committee was having even more hiccups than OpenCL had getting going. In part, because looking to break with OpenGL so had additional overhead of the faction who don't want major changes not being happy.
"OpenGL Next" was never meant to be a primary competitor to CUDA. It was definitely intended to couple to OpenCL in some explicit way, but it is modern "replacement" for OpenGL ( and OpenGL sharder language as a side effect ).
Microsoft ( Direct3D 12) , AMD ( Mantle ) , and Apple ( Metal ) all were riffing off the same baseline research that was pointing toward a solution that was more aimed at graphics libraries (game engines , os foundation graphics ) than at apps were had higher concurrency and closer to the "metal" API. ( more work but more control ). To kickstart "OpenGL next" from being somewhat stuck, AMD contributed Mantle as something that was concrete and working that Khronos could modify to turn into a standard.
Vulkan didn't hit initial release until 2016. Mantle was "out" in 2014. Direct3D 12 (with Mantle notions) in 2015 (announced in 2014). Metal 2014 ( on macOS 2015 ). Apple was pounding the drums on Metal well before Vulkan finished up. Apple hasn't done much at all to help Vulkan along. ( they haven't gone to much measures to slow it down either.... at least so far. Just more heads down trying to move Metal forward as quickly as they could. )
A billion and growing iOS+macOS+tvOS devices says otherwise. Metal is on more devices than Windows at this point. And Microsoft has their own 3D graphics API that is surviving just fine.
There are some sub areas that Metal won't penetrate but the notion that it doesn't have a large enough ecosystem to survive and fund platform ports to is a bit more than myopic. Yes, Metal isn't about completely maximizing Mac Pro sales. But that isn't the role the Mac Pro plays in the ecosystem.
Apple should put more effort into facilitating SYCL (and SPIRV ) . They should do needlessly hostile things to Vulkan. Microsoft has their own stack but they also left folks tack on other stuff also the the baseline stuff.
Hey Dec, usually your replies deserves big respect, even when we disagree (as most your prior reply except about SyCL), but admit it, a billion and growing...? Actually doesn't seems something written by you but from a hardcore fanboy, the numbers "accounted" by Tim cook needs a lot of salt...
Fwiw Android Android account a conservative 2.5 billion active users and windows 1.4 billion (licensed, pirated unknown).
Android grew faster at India and China where iOS devices are White Wales.
I read days ago there are about 15 million active professional software developers, a special group where Macs used to be the dreamed workstation, gives 1/4 or less from three will but a Mac and most likely a MBP 16 or iMac or mini, unlikely to buy the imMP, which IMHO won't be seen very often outside video studios and narco's rooms.
Users hanging onto their iOS devices for more years amounts to growing. As long as the devices retired into complete disuse don't outnumber numbers bought then it is growing to grow. What you are doing a misdirection on is the number of new devices sold. That has slowed. However, that doesn't make a big impact if users don't dump the old ones at a fast then new rate. Really think that folks are throwing way 80M iphones year? Hand me downs , used sales , etc. sure but turned off and never used again?
Another market where there is pragmatically little to no CUDA and doesn't really make any 'end of the world" story about that ecosystem going forward.
iPhone SE 2 ( won't be a SE but will have lower price point). Apple was a bit slow in turning the ship but not as slow as for the Mac Pro sub-sub-subsegment.
Again MBP 16" is going to do more to quell that rebellion (with esc key and revised keyboard ) than the MP will. The bulk of those folks were already on MBP (or Mini or some non Mac Pro variant ) already.
Interesting assessment on the Vega VII card. I hope this "loop whole" will remain open for the cost-conscious MP 7.1 buyer.
A standard MP 7.1 with the lowest speck 580X could be upgraded with less expensive Vega VII card(s), maybe slightly used cards to get a stronger machine to (at least) beat a standard iMac Pro 2017 from a GPU point of view.
Call me a snob, but if I was spending $5999+ on a macpro, I wouldn’t be putting a ‘used’ secondhand gpu in it....... 😜
snob! It all depends on cost of Vega II cards I think.
You can get 2nd hand Radeon VII for £500. Two of those would be good for GPU rendering (and noisy)
If Vega II costs £2000 for similar performance doesn't seem to make sense to me, but yeah if Vega II costs £1000 it would be worth it for the silent running.
I seem to be the only person who thinks its a good idea to put Radeon VIIs in a Mac Pro though.
(swap £ to $ for conversion)
Do you really think people holding a 6yr old phone still purchases apps? I didn't think so, so big 90+% of the apps sales goes to devices with less than 3yr old, where low quality in Android jungle forces new devices purchases, indeed having programmers more customers investing in their new communicators.
Crazy idea:
Get a MacPro7,1 with the Radeon 580X for macOS operations, install a much more powerful Nvidia GPU in a different slot for Windows/Bootcamp usage. Use a KVM switch or something to swap between the cards after switching OS.
I haven't seen much documentation on if having GPUs from competing brands in the same system will gel well, though. I'm heavily invested in macOS for creative content production but likewise in Nvidia for gaming. So... kinda at a crossroads here. Thanks, Nvidia and Apple....
Get a MacPro7,1 with the Radeon 580X for macOS operations, install a much more powerful Nvidia GPU in a different slot for Windows/Bootcamp usage. Use a KVM switch or something to swap between the cards after switching OS.
I haven't seen much documentation on if having GPUs from competing brands in the same system will gel well, though. I'm heavily invested in macOS for creative content production but likewise in Nvidia for gaming. So... kinda at a crossroads here. Thanks, Nvidia and Apple....
Why to spend 6000$ on a single basic Mac? Instead spend that on a PC clone with Threadripper and an iMac, interact with the PC thru vnc; simpler shot on ground
When you get more than 3 GPUs, you don't use a tower form factor anymore, you go to a rackmount server. For example, the Dell PowerEdge R740xd supports 6 GPUs in a dual-socket 2U chassis.
Dual socket is necessary because it gives you twice the PCIe lanes and twice the memory bandwidth.
Yes. I feel the same as you, Octane has the special something to its renders, but the price you pay is crash crash crash. I've been working on a personal project on my PC (my 4.1 doesn't reliably turn on anymore, so I build a cheap PC for the meantime), and C4D has crashed so much. Don't think this is windows making it happen, think its just octane crashing C4D all of the time.
If they somehow do manage to make Octane stable on the Mac it would be incredible, but I'm not expecting it tbh, think that is just the life of Octane. As you say Redshift is mega stable.
As we haven't heard much from the devs, my guess is we won't have metal Octane or Redshift GPU renderers by the end of the year like they said. I'd love to be proven wrong though.
Big Redshift update coming next week, the material system is using the native C4D node editor and native C4D noises are finally supported. The devs posted this image to Twitter but since deleted it. I saved it from a C4D Slack group. Maybe we will get Metal news with this release?
OTOY also announced Octane 2020 this week which looks awesome - their CEO said that "it was just too much to fit Octane X into everything else happening with today’s releases, but it’s coming up soon." He also said "Octane X needs 10.15 and a Metal GPU" so we can run this on our cheesegraters with a Metal GPU.
@shuto - sorry but I can't find the forum post about Apple deving Redshift for Metal, they must have deleted it.
I'm really excited about 3D on the Mac again. Good times ahead!
[automerge]1575108538[/automerge]
I don't think it's Windows causing the stability issues either, everyone complains about Octane but you rarely hear anything about crashing on any of the other big renderers. A lot of studios use Octane for style frames and lookdev, but all the final outputs are done in more stable renderers like Redshift, or if you really need to scale it up - Arnold. Even though Octane's output is aesthetically amazing it can be a nightmare to clear noise and fireflies and you can get horrible blotchiness from GI. I love it for stills but don't have the patience to babysit it through animation, as much as I'd love to. Whenever I see an animation or mograph that has been fully rendered in Octane it just has a certain look that looks beautiful. Really hoping for massive stability improvements in Octane X.
OTOY also announced Octane 2020 this week which looks awesome - their CEO said that "it was just too much to fit Octane X into everything else happening with today’s releases, but it’s coming up soon." He also said "Octane X needs 10.15 and a Metal GPU" so we can run this on our cheesegraters with a Metal GPU.
@shuto - sorry but I can't find the forum post about Apple deving Redshift for Metal, they must have deleted it.
I'm really excited about 3D on the Mac again. Good times ahead!
[automerge]1575108538[/automerge]
I don't think it's Windows causing the stability issues either, everyone complains about Octane but you rarely hear anything about crashing on any of the other big renderers. A lot of studios use Octane for style frames and lookdev, but all the final outputs are done in more stable renderers like Redshift, or if you really need to scale it up - Arnold. Even though Octane's output is aesthetically amazing it can be a nightmare to clear noise and fireflies and you can get horrible blotchiness from GI. I love it for stills but don't have the patience to babysit it through animation, as much as I'd love to. Whenever I see an animation or mograph that has been fully rendered in Octane it just has a certain look that looks beautiful. Really hoping for massive stability improvements in Octane X.
Last edited:
Redshift news from today:
"We’re still working (and making nice progress!) on this but I’m afraid we won’t manage to have a release within 2019. At this point it looks more like a Q1 2020 release but of course things could change fast (for the better, I mean ). There are lots of things that are already working fine but there are also a few roadblocks which we’re working diligently to overcome."
Sounds quite far from release and will probably launch in beta, like they did with Redshift 3 which has been in beta and not production ready for nearly a year.
"We’re still working (and making nice progress!) on this but I’m afraid we won’t manage to have a release within 2019. At this point it looks more like a Q1 2020 release but of course things could change fast (for the better, I mean ). There are lots of things that are already working fine but there are also a few roadblocks which we’re working diligently to overcome."
Sounds quite far from release and will probably launch in beta, like they did with Redshift 3 which has been in beta and not production ready for nearly a year.
I did read some recent posts by the OTOY CEO on Facebook about Octane X, but the amount of noise getting posted about the price of the Mac Pro has made the thread unbearable to read. But yes, it looks like a 2020 release for Octane X too as you said, which is fine, happy to wait for early reports and for the Macs to get into people's hands before deciding on my own BTO. It's not like my 5,1 and GTX1080ti has suddenly stopped working overnight.
Thanks for the update @vel0city
Yeah after you posted it I've been reading the facebook octane thread. The latest news I can see is "Preview is ready now, waiting on drivers for public release, stages will be a few months after"
So yeah early next year. Didn't think it would be easy to remake GPU renderers in Metal. Would love a teaser benchmark though! The only thing I have heard is Radeon VII about same speed as NVIDIA 2080, no sign of 2080ti speeds unfortunately.
Not sure if I should get VegaII now or later. Damn expensive. Also thinking a £700 Radeon VII as second compute GPU makes lots more sense than second VegaII at £2520, when apparently they are similar performance speed, but need benchmarks to prove it I guess.
Yeah after you posted it I've been reading the facebook octane thread. The latest news I can see is "Preview is ready now, waiting on drivers for public release, stages will be a few months after"
So yeah early next year. Didn't think it would be easy to remake GPU renderers in Metal. Would love a teaser benchmark though! The only thing I have heard is Radeon VII about same speed as NVIDIA 2080, no sign of 2080ti speeds unfortunately.
Not sure if I should get VegaII now or later. Damn expensive. Also thinking a £700 Radeon VII as second compute GPU makes lots more sense than second VegaII at £2520, when apparently they are similar performance speed, but need benchmarks to prove it I guess.
The first real test will be how MP7,1's are embraced by "industry" at conferences like NAB. If they're supporting and developing tools to support macOS Catalina+ and MP7,1 specifically, things should move fairly rapidly. If they're mostly waiting it out and still on TBv1 style products, it will be a longer road ahead.
I'm hearing AVID will not certify MP7,1 until late January 2020 at the earliest.
Ironically, despite the MP7,1 demos with HDX cards, ProTools still has this warning posted:
ALERT: Do not upgrade to macOS Catalina version 10.15
Current versions of Pro Tools are NOT COMPATIBLE with macOS Catalina (10.15). We are working closely with Apple in order to support macOS 10.15 Catalina as soon as possible. Until further notice, we recommend that you do not upgrade your Mac operating system for your Pro Tools machines. In the meantime we recommend that you continue to use macOS Mojave or other compatible macOS.
I'm hearing AVID will not certify MP7,1 until late January 2020 at the earliest.
Ironically, despite the MP7,1 demos with HDX cards, ProTools still has this warning posted:
ALERT: Do not upgrade to macOS Catalina version 10.15
Current versions of Pro Tools are NOT COMPATIBLE with macOS Catalina (10.15). We are working closely with Apple in order to support macOS 10.15 Catalina as soon as possible. Until further notice, we recommend that you do not upgrade your Mac operating system for your Pro Tools machines. In the meantime we recommend that you continue to use macOS Mojave or other compatible macOS.