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
In terms of raw TFLOPs they are about the same. 2080TI has tensor cores, which in some applications could dramatically improve performance. 32GB of memory would allow for larger scenes, models, etc to fit in memory.
Yeah this is a good question. Are the Vega II cards a good choice for GPU rendering? I think they have spec'd the cards with 32GB memory so it is good for 8K footage, not really for GPU rendering. 16GB memory is still really good for GPU rendering.
Before the Vega pricing came out I was really liking the idea of getting a Mac Pro to use for After Effects, Cinema 4D and GPU rendering with Octane and Redshift. I thought with 2 x Vega II duo cards the Mac Pro can have 4 gpus internally, which would be amazing expandability for GPU rendering. But now pricing is out it costs $11,200 to buy 2 x Vega II Duo cards as separate purchases. Ouch.
Its one thing not supporting nVidia cards, but then to only sell very expensive Vega II cards doesn't feel like they have build this for GPU rendering. If Apple really wanted to use AMD cards (instead of nVidia) but knew this would make things harder for their users, they could at least have made these at cost.
On the PC side, 4 x 2080ti hybrid cards cost $1350 x 4 = $5,400 (hybrid because you would be able to fit 4 on a PC motherboard. People are estimating Vega II performance will approximately match 2080 performance. So PC example would beat 4 x Vega II here at half the cost. Is it just a case that Macs are always twice the price!
I like the idea of Radeon VII in Mac Pro as this should be similar performance to Vega II, but much cheaper. But you would only be able to have two of these internally (plus more on x4 PCIe speed eGPU thunderbolt).
The question I'm asking myself is - has this this new Mac Pro been designed for Editors & Colourists rather than for 3D work?
So as much as I'd love it not be true, is a PC the only sane option for GPU 3D work, even if that PC is solely used for 3D work and everything else is done on a Mac?
Except there is a problem: The Vega 2s have Infinity Fabric, the 2080tis don’t.
I don’t think Nvidia even supports 4 way SLI anymore, but even if they did, SLI isn’t really supported by pro apps and isn’t as fast as Infinity Fabric.
So while each 2080 might be faster alone, they’ll be handicapped because they’re all talking to each other on a slow connection. 2 x Vega 2 Duos on a Mac Pro will probably be quite a bit faster than 4x 2080 Tis, not slower.
Could depend on the workflow. Would like to see benchmarks. But that’s my thought.
Good thought. I hope the infinity fabric saves the day, and means these are great for GPU rendering.
I'm a bit sceptical to be honest though, PCIe connection speed doesn't really matter that much for GPU rendering. See here...
https://www.pugetsystems.com/labs/a...ne-Scaling-in-OctaneRender-and-Redshift-1259/
But yeah that would be great if infinity fabric helps speed up GPU rendering. It sounds like Apple are helping with drivers for Octane, and working on Redshift, so it is in their interests to make it work as quick as possible.
This isn't necessarily true. While 4-way isn't supported in SLI. NVLink is now supported on 2-way which is leaps beyond SLI and is faster than infinity fabric to my knowledge. It also allows for pooled memory. So 2 – Titan RTXs look like one massive pool of 48GBs of memory.
@shuto - I think you're correct that the Vegas are more for editors. For our use (GPU rendering/3D) I think the Radeon Pro W5700X with 16GB are going to be more suitable. Worth holding out for prices and Octane/Redshift performance before diving in.
Still interested in seeing comparisons between W5700X and RX 5700 XT. Both have 40 compute units and nearly identical specs. Only major "difference" seems to be 16GB vs 8GB.
REALLY hope this means the drivers for RX 5700 XT will be improved in Catalina soon.
REALLY hope this means the drivers for RX 5700 XT will be improved in Catalina soon.
bah, i think apple doing the work for redshift now is showing what it is. apple needs some success stories to sell their pros and even more the image of still being pro. and it may be that they come up with some good results for their proprietary gpu modules. but this is not the future. it is a sidestory that will die off very soon because it is unlikely that these modules will ever get updated later on. all of this happened already with apple. no gpu options from apple, then (if lucky) maybe 1 update after 2-3 years, then things die off and the uncertain phase begins again where you dont know if apple will continue with the pros. redshift probably never would have done it themselves because there is just no userbase. the old macs are getting old and there will be just a very small number of people willing to pay 20k for a mac pro, while they can have a faster gpu system for a third of that price.
i personally would be surprised if the mac pro 2019 becomes a success. and specially for gpu rendering it is really not wise to go with a mac. all software is made for pc and cuda, first. the mac ports come later, are normally buggier, etc etc
[automerge]1576190525[/automerge]
that is exactly my thought if you like macos. have a pc for 5-8k and use it only for the 3d apps. keep an imac for photoshop, office, mail, small side tools, etc
i personally would be surprised if the mac pro 2019 becomes a success. and specially for gpu rendering it is really not wise to go with a mac. all software is made for pc and cuda, first. the mac ports come later, are normally buggier, etc etc
[automerge]1576190525[/automerge]
that is exactly my thought if you like macos. have a pc for 5-8k and use it only for the 3d apps. keep an imac for photoshop, office, mail, small side tools, etc
Last edited:
If anyone is interested in 3D stuff, I'd recommend watching this video showing real time ray tracing performance with Metal (start at the 46:00 mark). Very impressive:
https://developer.apple.com/videos/play/wwdc2019/613/
That's running on a MacBook with 4 eGPUs. Id imagine that the performance using a MacPro with 4gpus would be even more impressive.
https://developer.apple.com/videos/play/wwdc2019/613/
That's running on a MacBook with 4 eGPUs. Id imagine that the performance using a MacPro with 4gpus would be even more impressive.
My thinking exactly - they probably took an offer from Apple that cost them nothing. Unless the port is done in such a way that Redshift then can effortlessly update the thing without getting their hands dirty themselves, they'll just let it die off.
Remember the trashcan and the presentation of the Foundry's then new and much hyped application Mari on it?
If memory serves the software wasn't available for a good while on the platform after release of the machine - and when users got their hands on it finally it they discovered that it ran abysmal on the Dxxx GPU's and was a bit of a bugfest on the platform too. Last time I checked Mari is a full version number behind on macOS...
I ordered my 2019 Mac Pro, but I'm going to wait until redshift or octane are out before I get beefy GPUs to go with it.
After much deliberation I went with 16C / 96GB / 580X / 1TB / Nano XDR VESA + AppleCare.
My thinking for the time being is to get a system for After Effects + Cinema4D CPU rendering. Then later when Octane and Redshift are out get VEGA II or Radeon VII cards depending on GPU performance with those GPU renderers.
I kinda ran out of money to get a VEGA II, so all good to start saving up for one and upgrade later.
Is this Mac Pro just going to be a money pit for us though!?
After much deliberation I went with 16C / 96GB / 580X / 1TB / Nano XDR VESA + AppleCare.
My thinking for the time being is to get a system for After Effects + Cinema4D CPU rendering. Then later when Octane and Redshift are out get VEGA II or Radeon VII cards depending on GPU performance with those GPU renderers.
I kinda ran out of money to get a VEGA II, so all good to start saving up for one and upgrade later.
Is this Mac Pro just going to be a money pit for us though!?
ok that cost you short under 10k, right ?
the question i'm asking myself is... this performance is actually only needed for 1-2 applications (in your case after effects, cpu rendering) and these apps exist and work exactly the same on pc. why not buy a pc? for 6-8k you get a 32core threadripper with 2 x nvidia 2080 11gb. that is a way faster system. ok it is not macos, but you'd still have the money to buy an imac for mail, office, photoshop and stuff and put it aside the pc. all in all this seems much wiser, because already in 2020 you have the option to upgrade to the coming 64core threadripper and you can upgrade the gpus at any time to the state of the art, which wont be possible with apple
I really really sincerely hope that Apple is gonna flank this super impressive hardware with proper software support both on the drivers side but also by working closely together with software vendors as they promised, to make this thing a bigger success than the trashcan. I remember how disappointing it was after it turned out that all those great promises that they made about the dual GPUs back then really didnt come to much fruition and I saw all my 3d artist friends switch from old cheesegrater to PC...
I'm a big fan of the Mac platform (I have a PC and an iMac Pro), but the Mac Pro is a very questionable choice for certain computing applications.
If you focus on After Effects work, a Xeon CPU is not the best option at all, and adding more cores to it typically gives you worse performance, not more. See these benches:
After Effects CPU Benchmarks
On the 3D side, it's possible it may show some gains over the competition with a Pro Vega II if Metal shows the performance that's promised, but until the big render engines are released with Metal support, it's still kind of dead in the water as a 3D machine. I guess we will see next year...
Dont live with illusions... the whole 3d (render) industry relies on CUDA and this wont change so soon.
I dont believe a word of Apples marketing that they all support Metal now. Some maybe will, because Apple supports them, but the main development will stay with CUDA. why invest for a hand full of Mac Pro users?
That’s not entirely true, a lot of low to mid-end rendering relies on CUDA, yes, but any high end rendering is still CPU, and probably done in a render farm.
@vinegarshots : this thread is about GPU rendering...
for CPU rendering i would go with a threadripper
for CPU rendering i would go with a threadripper
1. Pooled memory isn't based solely on the NVLINK or Infinity Fabric. It is more so the memory controller. Can do pooled over PCI-e but that is far more Non Uniform Memory Access ( NUMA) than most folks want to deal with . Vega has some support for this too. Not exact match to Nvidia.
The memory controller and library foundation can "flatten things out" in terms of access.
2. Infinity Fabric isn't fixed width. You can vary the width of the bus between two chips/cards. From AMD's Pro card specs.
Vega II 84GB/s ( 672 Gb/s )
MI50 "up to" 184GB/s ( 1,472 Gb/s ) [ I think AMD is tap dancing a bit here. ]
MI60 marketing page seems to be missing.
On the compare for the pro card though between MI50 and MI60
https://www.amd.com/en/products/specifications/compare/professional-graphics/8291,8286
there are two Infinity Fabric lines on MI50 and MI60 and each link is 92GB/s ( 2*92 ... 184 )
However, Nvidia does some similar tap dancing ( in + out to get to biggest number. ) .
The coherence latency overhead (and to some extent concurrency to hide latencies ) is going to be a factor at these speeds ( some of this "mine is bigger than yours" posturing is how fast go down hill and with a hurricane tail wind. Not real world app access patterns. )
They aren't that far apart for the data center cards. Apple/AMD tones it down a bit for the Vega II.
Both are better than PCI-e v4 ( x16 : 256Gb/s ) and PCI-e v5 ( x16 : 512Gb/s ). CXL ( what Intel is going to use and AMD may support in future is layered on top of PCI-e v5. [ A very high chunk of PCI-e v5 usage is probably mainly going to be CXL (and similar variants where used for package interconnect) ... not random add-in-card buy at the local computer store. ]
To put into context Intel QPI used between Xeon SP packages to do NUMA is in the 200-400 Gb/s range and major systems are run every day with that level of bandwidth between chips get real work down everyday.
Right, because of the lack of working Metal-based GPU renderers, and the current extortionate cost of the Pro Vega IIs, I've pulled the trigger on a 24-core with a Radeon 580X for the time being. Waiting to see how the W-series GPUs shape up in performance and cost – but that will be for next year. And I can still render nicely with lots of CPU core-power C4D/Keyshot.
Nvidias have linear scale of performance in Octane which means my benchmark would double if I add second identical GPU or triple when I add two and so on. My main rig has only two GPUs but extras are on server build that I use on LAN for final renders. I do the same for CPU only renders such as Keyshot as well as BMD Fusion for comping. If you are into video editing I strongly recommend conforming and outputting in Assimilate SCRATCH with dual GPUs instead of Final Cut, Premiere or Resolve. SCRATCH eats those RED Raw files for lunch.
Episode 213 of the Mograph Podcast features OTOY's CEO discussing Octane X and the journey of bringing Octane to Metal. https://www.brograph.com/podcast/2019/12/20/mograph-podcast-213
Feature complete, a few months of optimisation left to do, fast, waiting for Apple to release 10.15.5 to release the preview, works on MacBook Pro, iOS build is full version of Octane runs on iPhone and iPad Pro, touch sculpting features in Sculptron with Apple pencil is incoming, "way faster" than some NVidia desktop GPUs, "super exciting" possibilities in iPad Pro, Octane X scales 4x (linear) on new Mac Pro with 4x GPUs, same as Nvdia version. Stable version public March/April. "Proof of life" videos coming ASAP. Free one year Octane enterprise license with Mac Pro. CEO is Mac user himself. Headless rendering coming to Octane X via network rendering, network rendering also works on iPad.
Jules (The Octane CEO) is a great listen and talks in deep detail. Really worth a listen.
Feature complete, a few months of optimisation left to do, fast, waiting for Apple to release 10.15.5 to release the preview, works on MacBook Pro, iOS build is full version of Octane runs on iPhone and iPad Pro, touch sculpting features in Sculptron with Apple pencil is incoming, "way faster" than some NVidia desktop GPUs, "super exciting" possibilities in iPad Pro, Octane X scales 4x (linear) on new Mac Pro with 4x GPUs, same as Nvdia version. Stable version public March/April. "Proof of life" videos coming ASAP. Free one year Octane enterprise license with Mac Pro. CEO is Mac user himself. Headless rendering coming to Octane X via network rendering, network rendering also works on iPad.
Jules (The Octane CEO) is a great listen and talks in deep detail. Really worth a listen.
Last edited:
The crux of the biscuit is properly exploiting Metal. If Apple is willing to put serious resources into creating a vibrant ecosystem for Metal, from the most fundamental coding tools to leveraging Infinity Fabric for multi-GPU performance...
Apple/AMD - particularly with process shrunk silicon already in the pipeline - can compete with Nvidia/CUDA, but only if they make it a priority. That likely includes underwriting coding costs for 3rd party developers who would otherwise focus on Linux/Windows rather than the smaller Mac market.
Apple/AMD - particularly with process shrunk silicon already in the pipeline - can compete with Nvidia/CUDA, but only if they make it a priority. That likely includes underwriting coding costs for 3rd party developers who would otherwise focus on Linux/Windows rather than the smaller Mac market.