MP 7,1 GPU rendering on the new Mac Pro. News, updates, chat.

[MarkC426]

So reading more into this, the info suggests Octane X is ‘optimised’ for Navi and Vega gpu’s, but does not specify only those.
By the looks of it I could get a refurb 2019 macbook pro from Apple for under 1k and get €699 license of Octane for free..... ?

If it works with Vega, why not an eGPU with a mini?

 

[deconstruct60]

will be made available – for free – to all new users activating Octane X on 2019 or later models of Mac Pro, iMac and MacBook Pro running macOS Big Sur.

Do you think that is a typo.
iMac has basic gpu options, I think it should iMacPro.

There is no 2019 or newer model of the iMac Pro.

MacRumors Buyer's Guide: Know When to Buy iPhone, Mac, iPad

This page provides a product summary for each Apple model. The intent is to provide our best recommendations regarding current product cycles, and to provide a summary of currently available rumors for each model. This page is based on rumors and speculation and we provide no guarantee to its...

buyersguide.macrumors.com

The basic criteria here is not any Mac. It is newest Macs (as in have been refreshed since around the June 2019 time frame or later ). [ Not really for giving free software to folks buying off of the used system market. Or helping Apple flush out semi-comatose inventories. ]

Reportedly that will loop in Apple Silicon Macs by the end of the year. But it also includes some Intel iGPUs for now. The notion that this is only aimed and the highest, of high end GPUs isn't the objective.

When/if there is a refresh of iMac Pro then it probably should be added to the list.

 

[sirio76]

Vega 2 Duo performance is not what I'd hoped. 415 for that vs 400 for a Titan V? It is a more expensive card with higher specs. I sincerely hope this is just alpha stuff and that the benchmark will significantly improve.

That TitanV benchmark is highly suspicious or comes from an hugely OC system, as a matter of facts the newer Titan RTX score far less than that (323 on average) and it seems extremely unlikely that a newer card runs significantly slower compared to a 3 years old product. The RTX 8000 is the Nvidia flagship product and score about the same as the Titan RTX, it comes with less VRAM compared to the MP VII duo and costs more(8000$ on Newegg). You can argue that technically the VII duo is a dual GPU but price/performance is still very nice if you consider the huge amount of VRAM.
If you are on a budget and your scenes can fit in 16GB just buy a couple old VII and you will get the same performance for very cheap.
About performance optimization, according to developers they should be able to get the software up to 15% faster with the next OS, that may rise the score to about 475 points.

 

[MarkC426]

Ahh, just looked through the otoy forums, didn’t realise there was support for the lower spec gpu’s in laptops etc..... The possibilities look promising.

 

[skippermonkey]

Couldn't wait for Big Sur (stupid name) so signed up for a monthly sub, until the OS arrives. So far, so…

Good:
Running a single RX5700XT performance is fine. Actually better than I expected.
Stability seems appreciably better than when I last played with a GTX 9800Ti a few years back.
I can load and render huge scenes; I did mange this back then, but it was crash prone. This already feels better.
The Mac Pro with lots of cores and lots of RAM is great for out-of-core memory, according to OTOY.

Bad:
RX580 not supported yet, so they're useless for the time being. Will probably upgrade at least one to something beefier – or maybe just throw in another cheap 5700XT.
It's not totally crash-resistant. But development is ongoing and OTOY says this is the year of adding stability not features. it all bodes well.
OpenVDB import seems to be broken. Tried with a number of files and always says 'VDB not found!'
Octane does not run as well on AMD as Nvidia/CUDA, but Big Sur will supposedly add several % points of performance, and we have yet to see Big Navi. Can't wait!

 

[deconstruct60]

So reading more into this, the info suggests Octane X is ‘optimised’ for Navi and Vega gpu’s, but does not specify only those.
By the looks of it I could get a refurb 2019 macbook pro from Apple for under 1k and get €699 license of Octane for free..... ?

If it works with Vega, why not an eGPU with a mini?

It is optimized for Metal which means not just AMD. That would include Apple's GPUs when they start to roll out . For Macs the most widely deployed GPU running Metal at the moment is Intel's iGPU.

The press release says

"... Near perfect linear scaling of rendering speed with multiple GPU configurations, including eGPUs connected over Thunderbolt 3. ."

I'm not sure sure that technically the "lame" Mini refresh counts as a real 2019-20 product. It is basically the same 2018 hardware

MacRumors Buyer's Guide: Know When to Buy iPhone, Mac, iPad

This page provides a product summary for each Apple model. The intent is to provide our best recommendations regarding current product cycles, and to provide a summary of currently available rumors for each model. This page is based on rumors and speculation and we provide no guarantee to its...

buyersguide.macrumors.com

with some very small configuration "tweaks" Apple slapped on top. The mini being a 2020 product is more Apple reality distortion field than substance. [ Mini would work just not 'free' . ]



But an MBP 13" four port 2020 with an eGPU should technically fit the criteria. (although they probably have the MBP 16" more so in mind. )
[automerge]1595282274[/automerge]

....

Bad:
RX5800 not supported yet, so they're useless for the time being. Will probably upgrade at least one to something beefier – or maybe just throw in another cheap 5700XT.

RX 5800 RDNA2 cards that aren't shipping yet probably wouldn't be supported.

 

[skippermonkey]

Sorry, RX580 and 580X. Threw in an extra zero there.

 

[robinp]

Any info on a rhino plugin? Been waiting some time for a decent render plugin for rhino and hoping this might be it.

 

[Boil]

Current AMD Vega dGPUs in select 2019 iMacs & iMac Pros...

iMac:

Pro Vega 20 - 4GB HBM2
Pro Vega 48 - 8GB HBM2

iMac Pro:

Pro Vega 56 - 8GB HBM2
Pro Vega 64 - 16GB HBM2
Pro Vega 64X - 16GB HBM2

 

[Hps1]

That TitanV benchmark is highly suspicious or comes from an hugely OC system, as a matter of facts the newer Titan RTX score far less than that (323 on average) and it seems extremely unlikely that a newer card runs significantly slower compared to a 3 years old product. The RTX 8000 is the Nvidia flagship product and score about the same as the Titan RTX, it comes with less VRAM compared to the MP VII duo and costs more(8000$ on Newegg). You can argue that technically the VII duo is a dual GPU but price/performance is still very nice if you consider the huge amount of VRAM.
If you are on a budget and your scenes can fit in 16GB just buy a couple old VII and you will get the same performance for very cheap.
About performance optimization, according to developers they should be able to get the software up to 15% faster with the next OS, that may rise the score to about 475 points.

I agree with what you’re saying here. But I also need to point out that the ob bench doesn’t use RTX. With RTX on a single 2080 hits 600 something.
I sincerely hope big navi and improvements to octane/macos result in a Mac system that can offer a 4x 2080ti equivalent system. Even if it’s more expensive.

 

[th0masp]

Wow, a single 2080 Ti with RTX manages 1.5x Vega-duo results, really? That sounds quite impressive.

How does the Apple Vega card stack up compared to a regular Radeon VII anyway?

 

[Hps1]

I misread the results. But I dont understand what I'm looking at, as it says 2080ti = 869, 2080 = 1175?

Here: https://render.otoy.com/forum/viewtopic.php?f=7&t=70705


...Vega 2 duo = 415 ish
Radeon VII = 200 ish

 

[shuto]

From https://render.otoy.com/forum/viewtopic.php?f=7&t=75411

The OB score of all the AMD/Intel GPUs on 10.15 now...

Radeon Pro Vega II Duo - 412 OB
Radeon Pro Vega II - 206 OB
Radeon VII - 200 OB
5700 XT - 170 OB
Vega FE - ~161 OB
Vega 64 -~148 OB
Vega 56 - ~117 OB
Vega 48 - 108 OB
5600M - 108 OB
5500M - 77 OB
Vega 20 - ~50 OB
Intel iris 640 - ~17 OB
Intel iris 630 - ~15 OB
Apple A13 - ~15 OB

I'm pretty excited to try Octane X on my MacBook Pro when its out.

On the PC side for comparison RTX 2080 Ti scores a Octane Bench score of 302 OB (£1050 a card). From Hps1's link above, with RTX on its scores 869 OB, so that's a crazy improvement!

I feel like OTOY are kind of on purpose over hyping the Mac version, saying things like a Vega II Duo is the fastest card ever, which sure it is, but its two cards really. In a PC you could cram 4 x 2080 ti, and in a Mac Pro you can cram 2 x Vega II Duos (maybe plus some thunderbolt eGPU boxes?), so kinda not a big point. Plus not including RTX octane bench results makes the Mac cards appear to render faster comparatively.

To be honest I'm biased on the Mac side and hate working on my PC, and choose to work on my noisy 16" MacBook Pro instead. So this is pretty exciting that its eventually happening.

 

[gabrielefx]

Octane X is out and it works! I tested it on my Mac Book Pro late 2019. It's a public preview but runs smooth.
I think this is a revolution for Apple because OTOY is going to release the plugins for all 3d apps.
Gpu rendering now is possible also on Macs without Cuda.
That's incredible...
It's very strange that Pixar, Solid Angle, Red Shift, Chaos Group, Blender and Luxion haven't released yet their gpu rendering engine for Metal.
Now Apple has to design its own powerful gpus.

 

[Hps1]

On the PC side for comparison RTX 2080 Ti scores a Octane Bench score of 302 OB (£1050 a card). From Hps1's link above, with RTX on its scores 869 OB, so that's a crazy improvement!

The jury is out on this for me as i don't get what they're saying. Is 869 the base OB score with RTX on, or is it just an infopass ob score?

From personal experience i do not see a 2x improvement with RTX on. More like 20%. This is with Redshift though.

And yes you can only fit 2x vega 2 duos - but you can fit 4x double width cards. So if big navi brings a 2080ti/3080ti level card we might be laughing. Whether in a happy sense or a maniacal "why is this so slow" sense i dunno. Worst case gpu rendering on the mac is here for the time being and the performance hit might not be so bad.

...it'll be better than running 1080tis in a 5,1 that's for sure!

 

[th0masp]

The jury is out on this for me as i don't get what they're saying. Is 869 the base OB score with RTX on, or is it just an infopass ob score?

From what I gather it's the base score for RTX. They do say that it gives something like 2 to 3 times of a speedup in their testing, depending on the scene. On the last page of the thread you linked they are saying that at some point this will be merged into the benchmark.

So how many 2080 Ti's can you fit in the Mac Pro?

 

[sirio76]

RTX performance may vary wildly depending on the scene setup, when you see 2x(or more) boost it's a best case scenario.

 

[Hps1]

You can fit four in - whether you can power them I dont know.

I have never seen 2x boost but it might just be my scenes

 

[MarkC426]

But there is no Nvidia support......

 

[Hps1]

The point is you can fit four double wide cards.

Hopefully with big navi we see some 2080ti level cards, and can use four of those.

I wasn’t saying I had 2080tis in a Mac Pro.

 

[Boil]

But there is no Nvidia support......

Why would there be Nvidia support for Octane X, a render engine designed specifically to work with macOS & AMD GPUs...?!?

 

[Hps1]

There was a communications error during the last few posts, ignore it

 

[deconstruct60]

You can fit four in - whether you can power them I dont know.

Probably not at the highest class card. Each MPX Bay only has two 8-pin connectors. Cards which require 8+6 or 8+8 to power up aren't going to provision out of the MPX Bay's two double wide slots. There is one more 6-pin just above MPX Bay 2 . (near slot 5).

The "max" GPU power delivery for a count of 4 GPU packages is via MPX modules. Perhaps get away with 3 cards, but four is a stretch. Just two higher end, double wide "mainstream" cards is what Apple's baseline design is for.

"...
Alternatively, each MPX bay can support:

• One full-length, double-wide x16 gen 3 slot and one full-length, double-wide x8 gen 3 slot (MPX bay 1)
• Or two full-length, double-wide x16 gen 3 slots (MPX bay 2)
• Up to 300W auxiliary power via two 8-pin connectors

..."

Mac Pro (Rack, 2019) - Technical Specifications - Apple Support

support.apple.com

Multiple 250W cards per bay isn't matching up with those specs. Can "hack" around it for one bay but it will be tough to repeat that hack for the second bay. You can get to higher counts if 'slim down" the GPU cards power profile wise. (e.g., e.g each take one 8-pin + bus power (+75 ) )

The high add-in card count is far more geared toward Audio/video capture andn/or storage I/O . Not GPGPU bulk nodes.


The other big leap is whether "Big Navi" is really going to be a big leap with Metal in this particular niche area ( ray tracing. ). I won't bet on that short term after those cards general release. It wouldn't be surprising that it wasn't until next WWDC (2021) to see pragmatic traction there.

 

[Hps1]

Hmm. What if we do the old 5,1 extra psu trick? That’ll fix the power issue at least. Only question is whether macos will address four cards I suppose.

 

[sirio76]

MacOS already worked with 4 cards even using eGPU setup (that's what Otoy developers were using on their iMacPro) so there's no reason why it should not work the same on the MP.