Mac: M3 - *Hardware accelerated RT (Part 1)

There is a simple solution to all of this – re-enable eGPU!

2 Likes

Please stop. Have you ever used an eGPU? Don’t believe pseudorendering benchmarks. This prosthesis kills graphics card performance in most cases. Especially in realtime graphics. The people at Apple would have to be complete idiots if they brought back eGPU support. This is a dead end. A complete negation of the Apple Silicon concept.

i had what you might call a “bad” experience with it myself, although that could as much have been about the overall bad laptop hardware that apple was shipping 2016-2019.

the conceptual benefit was there. it did give me a meaningful boost in certain instances where the (frankly awful) internal gpu wouldn’t cut it, and i almost certainly would have abandoned my 2018 machine (and apple, probably) much sooner without it. the main issues were stability and support. i was using a razer core x rather than apple’s official offering, which might have also contributed to issues.

overall i think the implementation was the bigger issue than the idea. personally, i have been much happier with decent but far from chart-topping performance in a simple, elegant package like the m1 max, but i wouldn’t discourage well implemented expansion options for people who need it.

The overall quality of Apple’s 2016-2019 computers doesn’t matter much. What disqualifies the eGPU is the poor bandwidth of the Thunderbolt interface and even if Apple uses TB5, it won’t change much. There are other issues, such as bad user experience (having to close programs before disconnecting the eGPU, or having some conflicts happen in the system that don’t allow you to close apps or even the system, etc. etc.).
Nor does the model of the eGPU device matter much. In fact, the Razer is a solid device.

Regarding what users need - some people probably need RS232 or Centronics, that doesn’t mean Apple should develop those interfaces.

Please let the eGPU rest in peace, and use Thunderbolt for what it’s great at - storage, display, ethernet, audio etc.

If you are using a MacBook and need more GPU power then you could try to buy an additional desktop computer with several Radeon graphics cards (RX 6800, RX 6800XT, RX 6900XT) and a Thunderbolt Ethernet connection (MacBook <–> Mac Pro) or 10Gb network card (MacBook <–> Hackintosh). Use the two computers almost as a single device using Universal Control or Synergy. This way you have a mobile workstation for working on the road and a powerful setup for desktop work. That’s how I work.

If money is not an issue, you can wait a while for the Mac Pro 2023 :slight_smile:

If they call it Mac Pro 2023… it will not arrive before 2024 …

1 Like

Beside reliability and such things …
barefeats.com tested eGPU up and down over years and checked
it against default GPUs for any kind of Mac hardware, Apps or
benchmark they could get into their hands …

For some it may be an advantage to have 60% of RT 7950 power
versus having no GPU power at all.

1 Like

yeah that was my point, its still something, and if implemented well, to me is less hassle than managing two systems (although thats a totally valid approach too - i’ll probably have that setup with my m1 max and a pc shortly for unreal work).

for kicks i could probably still fire up my old egpu setup and see how the vega64 over thunderbolt compares in a blender benchmark to the ‘raw’ numbers for that card. i can guarantee it’ll be tons better than the garbage 560x in the laptop itself at least!

the real issue was shortly after i got it, opencl rendering was almost immediately deprecated in cycles and it was only usable for eevee/viewport (as well as unity/unreal work) until recent improvements on AMD/Apple’s part.

the absolute worst part of that setup was how buggy the basic operation became. the ‘eject’ option stopped working completely after a few macos point updates (and looking online it was a common issue.) so my only option was to fully shut down the computer whenever i wanted to eject it. it also progressively stopped working on various thunderbolt ports. i think i’m down to only one on the system that will even properly boot with it connected.

but like… i’m pretty sure if someone actually cared to make sure all of that wasn’t so broken, it’d have been a fine tool for my needs (and it was for about a month before literally everything started falling apart)

Hi, the transfer speed does not influence render speed on a eGPU, many Octane user use them. If the scene is in the GPU ram all work is done on the GPU. Thunderbold should fill up any GPU ram in < 1 second.
Viewport render is a different story.

Cheers, mib

3 Likes

… and animations rendering is different story, and eGPU limitations in some apps (e.g. Apple Motion can’t use more than one GPU when eGPU is connected while easily supports multiple dGPUs) is different story, etc.

Thunderbolt max throughput in practice is around 2,5 - 3 GB/s, and this means that it can take a few seconds for a scene to load, while a fraction of a second for a dGPU. Yes, it matters, and in the case of animation this is a disaster for the eGPU.

I would suggest that you rely on your own experience rather than rumors from “many Octane user” because the eGPU may only work for certain niche situations.

I still feel that many saw substantial benefits from adding an eGPU to their systems vs. none.

This idea that a Mac Pro with a starting price of $6K+ for the base configuration will somehow solve all of our problems is ludicrous. The idea of running an external Hackintosh/Mac Pro in tandem with your main machine is also ludicrous.

Apple needs a proper GPU solution and not this insanity that they’ve latched on. I mean what is the Apple GPU-core count equivalent for an RTX 3080ti? 500 cores? 1000 cores?

See how insane and cost inefficient that sounds?

2 Likes

“Ludicrous” for the moment is what you write, because you don’t give any arguments. What is ludicrous is the idea of bringing back eGPU support. The bandwidth of an eGPU is 3GB/s and the bandwidth of Apple Silicon is >100x more than that crap. The eGPU is a totally idiotic idea these days.

Only one reply, if you have to render several minutes for a frame/still it does not matter if it take 1 or 10 seconds to load the scene.
I don´t need own experience to know how fast a Porsche is, I simply have to read the manual.

Cheers, mib

… but if you are using denoising, a few seconds are often enough to render one frame.

eGPU is not a Porsche. eGPU is Citroen 2CV or VW Käfer.

Right…and what we have right now is a bicycle (not even a eBike!). I’ll take the Citroen thank you!

You are just speculating and not pointing to any actual data to support your point other than your limited personal experience that eGPU is crap.

Perhaps you’re right, but it’s also true that eGPU can be improved on and Thunderbolt can get faster to the point where it approaches or even surpasses PCI bus speeds.

I don’t see you offering any practical solutions other than basically buy a Mac Pro and fill it with…GPU’s! :rofl:

2 Likes

Drive whatever you want, just don’t talk nonsense.

I have used eGPUs long enough for my experience to be much more than ‘limited’. In complete contrast to you - you have no experience at all.

What a BS. Thunderbolt is just limited external version of PCIe so by definition it will always be slower. In practice, many times slower because it is limited by the physical external connection.

I have proposed a very reliable solution that works for me. You have not proposed anything that makes sense. Unfortunately, you don’t have enough knowledge on this subject and in addition you are resistant to substantive arguments. EOT.

i just ran the blender benchmark. my egpu gets a score 726, a dedicated system with the same card gets 803 according to the database. thats 90% of the benefit of the card. the internal gpu of the macbook pro was 111.

if you want to get the maximum performance out of your card, a dedicated system is the way to go. but an egpu still gives a massive performance boost to weaker systems. and for many, that slight performance penalty is worth the convenience of only maintaining a single system.

7 Likes

2007-2012 Apple offered a Mac Pro at a reasonable price that
was on par or better than PCs. OK Apple was a bit slow with
providing GPUs.
2014 came the expensive 2013 Trash Can with not bad double GPU,
that weren’t supported over its live time, but weak and expensive
Xeons. And it was never updated.

Forget about iMac Pro, 2019 finally came the new Mac Pro. At ridiculous
prices, with weak single Xeons. Did not look like Apple really learned a lot.
At that time you could get PC Performance with good parts for around $3k
where a comparable Mac Pro would start at $12k.

And until now, we get at max a $6k Studio with performance comparable
to a Ryzen 5950X and a RTX 3080 (?) - at least at only double the price.

I am all in for the ARM architecture.
But some would like to invest 3 times the power consumption to get their
jobs done at reasonable time frames.

And now a disappointing M2 successor after 2 years, while x86 prospers
now under a working competition …
Sorry, I do currently not see an Apple performance wonder exhausting
soon out of Mr. Cooks pipeline. And if there would, there is no sign of
availability for anything else than ludicrous off prices.

If you don’t like to work with Windows, can’t switch to Linux in near future,
you have to live with Apples hardware offers. And compete with your x86
pragmatic competitors …

So even if an eGPU would have only a 50% efficiency and 600W power
consumptions, some could profit immediately.
Versus hoping for all Apps to optimize for Apple ARM and slight gains
by Apple from adding more cores and voltage, and hopefully a new
shrinking process soon (?).

2 Likes

Well, you only need performance when rendering or work with realtime GPU. As long as this is not the usual case, the Mac can be used. But, it won’t stay this way. I already feel like I can’t render or preview my shaders on the Mac, because all is so slow. From Texture loading till response, it takes like 10 times longer. And no, no M2 studio here. To expensive in these times.
The new Apple CEO has one serious problem, after a number of failures, you should retire. But, he doesn’t and that can be a problem for the future.

1 Like

If I would be ready for Linux (ElementaryOS over all) and had
ALL my needed Apps,
I think I would still (?) be more than happy with my (2 year upgrade old)
Ryzen 3950X for CPU Rendering, my RX 6800 16 GB VRAM and maybe
would not even upgrade my slow (?) 2400 MHz 64 GB RAM.
Although I could find a budget for a PC upgrade.

And I think my NZXT midi case looks more architectonical and pleasing
than my Trash Can and is not louder.
Not a real problem for me as a desktop user, that the PC has 50 time the
volume of my M1 Mini.

From old Modo times,
I always preferred to work on a capable Workstation. Because while work you want
already fast previews, all other related Apps open beside, …
If it can also finally render your stuff over night in only 2 hours, although you would
have 8 night hours, even better.

But it doesn’t look like Apple will ever offer better or at least competitive CPU
performance, and for GPU, I don’t see much than adding more cores, still falling
behind against real IPC gains on the x86 side.