Hacker News Reader

Comment by internobody | Hacker News Reader

3D modelers want NVIDIA cards, not Radeon, as NVIDIA supports CUDA. This has been a longstanding issue with 3D folks; the old trashcan Mac Pro had NVIDIA GPUs even with otherwise-outdated internals. When rumblings came of new Mac Pros a few years ago there was excitement that they could get some fast machines again; alas this appears not to be the case.

View on HN

MacOS does not support vendor-supplied drivers which has been the bane to every graphics engineer's existence. Supporting mac for games as a result, is a completely inane task we avoid like the plague. To me, because of the GPU, this machine is a giant paperweight.

I’d argue this is good for users, because vendor-supplied drivers always seem to have crappy updater software running at all times and send pop-ups at all hours on Windows. Vendors really need to better support open-source drivers for user experience purposes, but they don’t because they want more control over their market.

Don't conflate the updater with the driver itself. Apple could have enforced a policy around integration with the system update functionality or something, but the driver supplied by the vendor will always be better/faster because it can exploit internal knowledge about the hardware.

Yes, I get that. All I’m trying to say is that the current user experience with third-party drivers for anything (especially nVidia graphics cards) is not the kind of thing Apple would want to inflict on their users.

It's not good for users when vendor drivers work and Apple's drivers don't. I know OpenGL is deprecated, but a lot of apps still use it because it's cross platform, and Apple has the worst GL drivers in the industry. Even Metal is missing many features from Vulkan [1].

[1]: https://github.com/KhronosGroup/MoltenVK/blob/master/Docs/Mo...

Right, I'm not an expert on the Metal API but I noticed there didn't seem to be a way to submit split barriers to the command queues.

I don't know much more about Metal beyond that though.

That's not true. If you let Windows manage your drivers, you won't have any popups. Drivers will be updated as part of standard Windows update. Now if you want latest drivers and you're intalling GeForce Experience application, then yeah, it'll notify you of new drivers once in a month. It's just a standard Windows notification and if you installed that application, probably you want to stay on latest drivers, so why would you complain about that?

Often when I plug one of my Razer mice into a different USB port I get a big pop up with a prompt to install their Synapse software that takes two clicks to dismiss.

Yep, nothing better for users than a platform no one develops for with great update infrastructure.

I don't think any $5000 machine is a good choice for gaming.

Nvidia's web driver works for pre-Mojave macs

Most major 3D modeling programs support OpenCL, as well, not just CUDA. In particularly any of them with MacOS ports are not CUDA-only (since, you know, MacOS hasn't supported CUDA-capable GPUs for many many years now), making the argument for CUDA specifically rather moot.

In terms of just GPGPU performance Vega is pretty competitive against Nvidia. It's not nearly as one-sided as the gaming space is. That is, in fact, the one saving grace of the Radeon VII in the first place vs. the RTX 2080 - its stronger professional workload capabilities.

Gaming space is not one-sided either, the original Vega delivers excellent 4K60 performance if you don't do stupid things like MSAA.

The programming model for OpenCL is not near what CUDA is.

> 3D modelers want NVIDIA cards, not Radeon, as NVIDIA supports CUDA

What do 3D modelers use CUDA for? I know it gets used for machine learning, but I'm not familiar with what else.

I've heard that CUDA can help decrease render times on Adobe Premiere Pro and several other programs. I'm guessing that includes 3D modeling software like Fusion 360 or something to help with various molding and simulation features.

FWIW, I've done some pretty complex part simulations in F360, and it's not that slow, on my Vega MacBook pro 2018. But, if it is, Fusion has online-sim calculator or renderers that are pretty affordable.

Probably has to do with other software like Maya or other simulation like programs. In fact F360 worked flawlessly on my 2016 dual-core 2.0 ghz CPU with 8gm of RAM.

Not F360 that I'm aware of, but for 3D, 3DS Max, C4D, and Maya both support GPU-based rendering using CUDA, which can be a big performance boost. On the video side, Premiere Pro, Vegas, and AVID tend to have better support for CUDA as well. However, AMD support has been catching up in the last couple years.

tin foil hat thoughts what if CUDA on MacOS allows Adobe products to outperform Final Cut rendering times and Apple don’t want to jeopardize it’s offering?

Could be possible, but given Apple optimizes like crazy using its own hardware and software, Apple products get crazy performance. On top of that Nvidia and Apple aren't on good terms, so they obviously didn't want to build a very powerful workstation.

I used to develop professional 3D tools and CUDA is used to accellerate features in 3D software including: scene manipulation, simulation, real time rendering, ray tracing, rendering shader effects and video encoding etc.

Thank you!

Maybe Apple is trying to replace that with that weird accelerator card they mentioned? If anything is going to be orphaned fast it is that card.

I wonder if that card support metal2, which AFAIK is still the recommended compute (openCL/CUDA replacement) technology on macos?

The video says the accelerator card is for 4k & 8k RAW video. So I wouldn't expect it to do anything at all other than decode 4K & 8K RAW footage, kinda like the RED ROCKET-X.

What does CUDA support that Metal 2 cannot do? Recently I've been implementing some 3D mesh algorithms in Metal 2, and it was quite easy to do.

Support for all the CUDA code that's already out there. Not that I've used Metal, but that's what seems to be the reason behind OpenCL's lack of adoption

Sure. But there is also a lot of Metal code out there in order to make iPads fly. And they really do fly, I was amazed what kind of performance I could get out of them. For example, I had an algorithm running on my iMac Pro, programmed in Swift, utilising all of its 18 cores fully. It took about 5 minutes on a typical example. Then I recoded the algorithm for Metal, and it runs on the same example on my iPad Pro in under 10 seconds.

So, my bold prediction: Apple is going to shred Nvidia to pieces within the next 10 years.

Raytracing, rendering, certain kinds of vector operations, physical simulations.

CUDA/OpenCL is all about writing super parallelism code to be run on the GPU instead of CPU, so many trivially parallel problems are usually outsourced to it.

Did the trashcan really support NVIDIA GPUs? Wikipedia only lists NVIDIA GPUs for 2009 and earlier, which makes it the tower models, not the trashcan.

https://en.wikipedia.org/wiki/Mac_Pro#Specifications

Nope, only AMD.

Source: https://everymac.com/systems/apple/mac_pro/mac-pro-cylinder-...

The old Mac Pro has AMD FirePro GPUs. Not Nvidia

Depends on your definition of "old". From 2006 to 2009 they were nvidia, and then in 2010 they switched.

Trash can is trash can and it had Radeons, before that was cheese grater with Nvidia GPUs

In theory you can stick any nvidia cards you want into this machine. Don't have to stick with the stock AMD GPUs.

Unfortunately for reasons that remain murky Nvidia still hasn’t put drivers out for Mojave and their absence from the keynote leads me to believe they won’t have drivers for Catalina. Nvidia blames Apple saying they’re not approving drivers. So yes you can slap a 2080ti in your Mac Pro but without drivers it’s useless.

Anybody know what Apple's problem with nVidia is? Are they still feeling burned from the BGE failures several generations back?

This is a professional machine. No one wants to buy it and use hacky work-arounds, they want good drivers. Which aren't an option with nvidia on macos.

Reply on news.ycombinator.com