Redshift (CUDA GPU render) devs - we are not leaving the Mac

[deconstruct60]

....
Does anyone know how the processing power of a Radeon VII compares to a Radeon Pro Vega II or Radeon Pro Vega II Duo?

The VII is a slightly kneecapped Instinct MI50. the Pro Vega II is a modified MI60. There are 4 more CUs ( 256 more stream processors ) and twice as much HBM2 ( 32GB )


https://www.amd.com/en/graphics/workstations-radeon-pro-vega-ii
https://www.amd.com/en/products/professional-graphics/instinct-mi50
https://www.amd.com/en/products/professional-graphics/instinct-mi60

If look at the theoretical peak performance numbers for MI50 and MI60 that is probably illustrative of the gap. ( 13.3 vs 14.7 TFLOP single precision so around ~10% )

If need to load a larger data set into HBM RAM then it is pragmatically a larger gap (e.g, if blow past the 16GB limit of the VII ) . If two GPUs need to swap data then the Duo can have ever larger edge ( since can do that over Infinity Fabric).

He says
2 x Radeon Pro Vega II Duo = 57 teraflops
2 x Nvidia RTX 2080 ti = 27 teraflops

TFLOPs of what? FP16? But FP32, no. Not even AMD is claiming that on their marketing bragging pages. The MI60 is 29.5 TFLOP FP16. ( 2 x is 59 so 57 is ).

So maybe a Radeon Pro Vega II is equivalent to a Nvidia RTX 2080 ti, so I'm wishfully thinking they will render in Redshift & Octane in similar speeds. If thats true hoping they will be similarly priced also.

Not sure how they would be similarly priced at all. Pro Vega II has 32GB of HBMv2 memory and 2080 ti has 11GB of GDDR6. Vega II an large interposer to hook that all together versus VRAM soldered to logic board. (i.e., is more space efficient but that costs money) That's a huge price gap right there. Vega II is 7nm (bleeding edge fab for that die size ) and 2080 Ti is mature 12nm. AMD stripped Infinity Fabric from the Radeon VII so it is up in the air to see what price AMD and Apple are going to charge to leave it on,

So maybe a Radeon Pro Vega II MPX would be $1200, Radeon Pro Vega II Duo MPX would be $2400, and two Radeon Pro Vega II Duo MPX would be $4800.

I think you are letting the Founders edition price for the RTX 2080 Ti price anchor the outlook here rather than starting with what AMD's bill of materials costs are and going forward from there with margins that AMD and Apple are likely to impose.

This is probably not going to be huge winner on best $/performance. it will more so be on the raw perf number coupled to what runs on macOS. The Pro Vega II is probably closer to $2K than $1k. The Duo is probably closer to $4K than $2K. And two Duo is probably closer to $8K than $4K (i.e., more than the base entry entire system price ).

 

[Jack Burton]

I'll be interested to see how Redshift and Octane do on Metal compared to CUDA. RTX hardware really speeds up some scenes, supposedly.

 

[bsbeamer]

Maxon is now merging with Red Giant, just announced. So they've acquired Redshift and now this. Will be interesting to see what happens at NAB this year. Getting a sense it's going to be a year of changing workflow patterns.

 

[Jack Burton]

I don't get that merger at all. Lots of big moves happening.

I saw some accomplished 3d artists marvel at the odd acquisitions - Substance and Mixamo by Adobe, Red Giant by Maxon - they were thinking that the reverse made more sense.

Other people that theorized that Maxon is priming itself be acquired by Adobe.

Weird times. Maxon is behind in a lot of areas, so I wonder how long they can say "we hear you!" and then not deliver.

 

[bsbeamer]

The Maxon/Adobe acquisition has been rumored and theorized for several years. In some ways it makes a lot of sense, but in others it's almost baffling that Adobe can't make their own 3D video software application to compete in the space. Their 3D tools are getting better, but it's not C4D/3DS/Maya level and mostly just for AR/VR "junk" in my opinion.

Red Giant certainly adds value. Their new "everything" plugin subscription would fit in well with Adobe's model, if it can be added or bundled for a decent price... but those with full licenses would (again) pay more.

Avid seems to continue to trim down their product offerings and offer more "value" in their lower end tiers, following very closely to what Autodesk did with their higher end tools.

Will at least be interesting watching the transition.

 

[fuchsdh]

I really would love to know what Apple feels like they’re gaining by locking out Nvidia besides a small but potential security hole.

They don’t have to even ship Nvidia GPUs.

This both-sides-pointing-at-the-other stuff is childish and doesn’t serve end users.

The Maxon/Adobe acquisition has been rumored and theorized for several years. In some ways it makes a lot of sense, but in others it's almost baffling that Adobe can't make their own 3D video software application to compete in the space. Their 3D tools are getting better, but it's not C4D/3DS/Maya level and mostly just for AR/VR "junk" in my opinion.

Red Giant certainly adds value. Their new "everything" plugin subscription would fit in well with Adobe's model, if it can be added or bundled for a decent price... but those with full licenses would (again) pay more.

Avid seems to continue to trim down their product offerings and offer more "value" in their lower end tiers, following very closely to what Autodesk did with their higher end tools.

Will at least be interesting watching the transition.

I still remember Adobe's 3D modeling software. Had a copy too, I think.

 

[mattspace]

The Maxon/Adobe acquisition has been rumored and theorized for several years. In some ways it makes a lot of sense, but in others it's almost baffling that Adobe can't make their own 3D video software application to compete in the space. Their 3D tools are getting better, but it's not C4D/3DS/Maya level and mostly just for AR/VR "junk" in my opinion.

They just bought Medium from Oculus - though that'll only really effect people who love Facebook enough to strap it on their heads, unless they're planning on bringing it to other VR headsets.

 

[AWW_13]

Are buyers of the new Mac Pro getting a free license for the 'Octane X Enterprise Edition' as promised in OTOY news in June 2019?

And would that mean they have successfully implemented Metal2 support? (And maybe RedShift will follow in 2020?)

 

[H. Flower]

Wow......didn't hear about this!

Anymore details from anywhere?

 

[AWW_13]

It's stated down near the bottom of this post but I haven't heard it repeated anywhere..

OTOY Announces Octane X for macOS and iOS at WWDC 2019

OTOY, Inc. unveiled Octane X - the 10th anniversary edition of OTOY’s industry leading GPU renderer OctaneRender® at Apple’s World Wide Developer Conference. Octane X has been rebuilt from the ground up in Metal for the all-new Mac Pro and the latest generation iPad Pro. It is the culmination of...

home.otoy.com

 

[tommy chen]

free license or one year!

 

[koyoot]

I really would love to know what Apple feels like they’re gaining by locking out Nvidia besides a small but potential security hole.

They don’t have to even ship Nvidia GPUs.

This both-sides-pointing-at-the-other stuff is childish and doesn’t serve end users.



I still remember Adobe's 3D modeling software. Had a copy too, I think.

Its funny that people complain that Apple locked out their users from Nvidia options, but they do not see how the users are locked in to Nvidia options, because of proprietary nature of CUDA.

Open Source is the key, guys. Follow Open Source initiatives, if you want to be able to freely switch software, and not be locked down to any solution.

 

[blackadde]

Its funny that people complain that Apple locked out their users from Nvidia options, but they do not see how the users are locked in to Nvidia options, because of proprietary nature of CUDA.

Open Source is the key, guys. Follow Open Source initiatives, if you want to be able to freely switch software, and not be locked down to any solution.

I’m not a 3D guy (at all) but isn’t Metal locked down to iOS and MacOS? I looked and didn’t find any resources or documentation for a cross platform implementation on Linux or Windows.

 

[goMac]

I’m not a 3D guy (at all) but isn’t Metal locked down to iOS and MacOS? I looked and didn’t find any resources or documentation for a cross platform implementation on Linux or Windows.

It's complicated.

Metal is meant to be something low level that other things are used on top of. You can still use it directly if you want. But typically you're using something else on top of it, like Unreal, Unity, etc etc. It doesn't offer the conveniences of OpenGL that was meant to be used directly.

You can also put other 3D environments on top of Metal. On iOS, OpenGL is now done on top of Metal. Vulkan can be done on top of Metal through MoltenVK.

So yeah, while Metal is macOS only, it's generally meant to host something else on top of it. Which means it's not really a competitor to traditional DirectX or OpenGL. It's the layer that another environment can be built on top of.

It's kind of like a driver layer for 3D graphics systems. You can use it directly, but it was written with being an adapter to other things in mind. Also a handy way for Apple to shove OpenGL support out to the wider community instead of doing it themselves.

Before Metal, people would put things on top of OpenGL. For example, there are layers that will translate OpenGL to Metal. But OpenGL is such a bulky, bloated API that the slowdowns in doing this could be significant. Metal gets rid of all that bloat letting you have a lot more raw access and optimization to the GPU, at the cost of not being as easy to work with as OpenGL.

My guess is this is why Apple is working with a lot of companies directly on adopting Metal. For a lot of apps that have been using OpenGL, using Metal is like jumping to a more raw GPU layer. But that means you've lost a lot of the handholding GL did.

Nvidia could totally get CUDA running on top of Metal, using Metal as a driver-ish layer. But that would allow CUDA to run on AMD GPUs. So they won't.

 

[fuchsdh]

Its funny that people complain that Apple locked out their users from Nvidia options, but they do not see how the users are locked in to Nvidia options, because of proprietary nature of CUDA.

Open Source is the key, guys. Follow Open Source initiatives, if you want to be able to freely switch software, and not be locked down to any solution.

I generally agree, but that's really not relevant pragmatically. If you have software that requires CUDA, you need CUDA. Apple has its own "lock-in" effects as well, but the problem from a pro standpoint is if you have parts of your workflow that need or otherwise benefit from CUDA, you're better off just not considering Macs, period.

I wouldn't expect or even want Apple to put Nvidia cards back in their computers, but at least allowing end users to add it after the fact if they need it makes the Mac more versatile.

I don't think Apple really has any leverage to try and stop CUDA, especially since Metal isn't cross-platform and since they've abandoned standards like OpenGL and OpenCL in favor of it.

 

[OkiRun]

It's complicated.

Metal is meant to be something low level that other things are used on top of. You can still use it directly if you want. But typically you're using something else on top of it, like Unreal, Unity, etc etc. It doesn't offer the conveniences of OpenGL that was meant to be used directly.

You can also put other 3D environments on top of Metal. On iOS, OpenGL is now done on top of Metal. Vulkan can be done on top of Metal through MoltenVK.

So yeah, while Metal is macOS only, it's generally meant to host something else on top of it. Which means it's not really a competitor to traditional DirectX or OpenGL. It's the layer that another environment can be built on top of.

It's kind of like a driver layer for 3D graphics systems. You can use it directly, but it was written with being an adapter to other things in mind. Also a handy way for Apple to shove OpenGL support out to the wider community instead of doing it themselves.

Before Metal, people would put things on top of OpenGL. For example, there are layers that will translate OpenGL to Metal. But OpenGL is such a bulky, bloated API that the slowdowns in doing this could be significant. Metal gets rid of all that bloat letting you have a lot more raw access and optimization to the GPU, at the cost of not being as easy to work with as OpenGL.

My guess is this is why Apple is working with a lot of companies directly on adopting Metal. For a lot of apps that have been using OpenGL, using Metal is like jumping to a more raw GPU layer. But that means you've lost a lot of the handholding GL did.

Nvidia could totally get CUDA running on top of Metal, using Metal as a driver-ish layer. But that would allow CUDA to run on AMD GPUs. So they won't.

I learned a lot from this article:

Down to the Metal: what Apple’s new graphics API means for Mac apps and games

Get the official T3 lowdown on Apple's new Metal graphics API

www.t3.com

 

[goMac]

I learned a lot from this article:

Down to the Metal: what Apple’s new graphics API means for Mac apps and games

Get the official T3 lowdown on Apple's new Metal graphics API

www.t3.com

Another parallel is the relationship between Vulkan and OpenGL. Vulkan does not replace OpenGL. Vulkan is the more bare-to-the-Metal graphics library. And OpenGL is the easier-with-more-features-but-slower API.

When you need to squeeze out the very last bit of performance, you use Vulkan. When you don't care as much about performance but you want something easy to use, you use OpenGL. That's why OpenGL is fairly popular. A lot of hobbyists still use OpenGL because it provides more out of the box.

There is talk that OpenGL standard may be changed to get rid of OpenGL drivers completely, and have all versions of OpenGL be built on top of Vulkan. These means vendors would only have to ship Vulkan drivers, and OpenGL would just become an SDK on top of Vulkan.

Again, in that sort of world, OpenGL doesn't go away on the Mac either. OpenGL can just be maintained (likely by the community) on top of Metal.

I thought I heard of a similar thing happening on Windows. DirectX 12 is their bare metal API, and I thought I heard that instead of shipping drivers for Vulkan and OpenGL, that vendors were thinking about just putting those on top of DirectX 12 instead. With ARM, not all of Microsoft's environments have Vulkan drivers. And building everything on top of DirectX 12 means that only one set of drivers needs to be implemented everywhere.

Anyway, what I'm getting at is cross platform graphics is kind of an idea that's not a thing anywhere anymore. You have a platform specific layer, and then you build cross platform libraries on top of it. And Metal supports that kind of thing. Including running OpenGL apps on top of Metal.

 

[Jack Burton]

Well, it's January 1. Anyone know if Redshift or Octane announced anything regarding Mac support? They all said we would hear more at the end of the year back in 2019.

Otherwise I'm off to the 5th circle of hell that is Windows courtesy of the Ryzen 3950x as soon as it is in stock for retail price at a respectable retailer.

 

[AWW_13]

Well, it's January 1. Anyone know if Redshift or Octane announced anything regarding Mac support? They all said we would hear more at the end of the year back in 2019.

Otherwise I'm off to the 5th circle of hell that is Windows courtesy of the Ryzen 3950x as soon as it is in stock for retail price at a respectable retailer.

From ‘Company News’ Dec 31st.

[automerge]1577916677[/automerge]

[automerge]1577917156[/automerge]
OTOY Twitter (pinned tweet)

 

[AWW_13]

Wow......didn't hear about this!

Anymore details from anywhere?

Well they seemed to still want to honor the offer. Yesterday from their Twitter:

 

[H. Flower]

Wow, that’s awesome. Will definitely start using it octane x!

 

[vel0city]

OTOY want Octane to become a platform, they are already giving a free version away for Daz3D and the iPad Pro/iOS versions are very interesting, it's really exciting what they have planned for Octane - most of all I'm glad they've completely re-written it for the Mac, hopefully it will help with some of its stability issues. Octane is a beautiful renderer out of the box but can be unreliable. Looking forward to the release and reports about how it runs on the new Mac Pro. It's going to be a great year for 3D.

 

[Jack Burton]

Wow, thanks for the updates! I'm excited about it again. I still wonder if I should build that 16 core beast, though and use the Physical renderer.

I am curious about trading in my iMac for a 16 inch MBP and throwing a ton of eGPUs on it for Rendering, if needed. Someone on the CG talk forums hooked up 4 and didn't have to worry about cooling.