Mac Studio & M1 Ultra

[zoomer]

Strange my aging 3950X is only 22637 multi core points.

Also my single core points are lower than in the second table.

24000 for the Ultra would be at least a bit more.

Which would be OK for my last CPU Render attempts.

 

OK, more would be always welcome.

 

[Mark Aceto]

8 hours ago, zoomer said:

24000 for the Ultra would be at least a bit more.

Which would be OK for my last CPU Render attempts.

 

 

That's what I'm hoping, and eager to test with some old and current projects (as soon as finish this manual migration).

 

8 hours ago, zoomer said:

OK, more would be always welcome.

 

Will be interesting to see what's announced with the Mac Pro... 

[Mark Aceto]

My 3Dconnexion devices work as expected (no issues):

• CadMouse (wired first gen)
• SpaceMouse Wireless (recent purchase) works in both wired and wireless modes
  • Except with Universal Control on the secondary Mac (not an issue for me but interesting nonetheless)

It sounds like there were initially some issues with M1 chips but 3Dc have finally ironed out the kinks. Btw they never notify customers when new drivers are released (that I know of). The most recent driver was released on April 29, and eliminates kernel extensions.

 

For now, these 2 devices are taking up both USB ports. I may be able to get the SpaceMouse Wireless to work by plugging the dongle into the monitor if I continue to use the TB 4 cable for signal... Or a wired keyboard (that would take its place). I'll test this later when the need arises. Note: Logitech Webcam C925e is also plugged into the monitor and signal is piped in through that same TB 4 cable. However, the first gen CadMouse doesn't like hubs (typically has to be unplugged and replugged after restarts). That's not an issue with the newer CadMouse devices though.

[zoomer]

I use both CADMouse + Enterprise with a (powered!) USB Hub.

(Well, I need one of my M1 only 2 USB Ports for my USB Hub, the

other for my Wacom Cintiq anyway)

That works but a non powered USB Hub did sometimes not work

with the Enterprise power consume.

 

But for the Ports, I always thought that 3DConnection delivered a

a dual USB adapter for one Port ?

I do not use it and am not sure, maybe I mixed it up with any other

devices ....

[Mark Aceto]

5 hours ago, zoomer said:

But for the Ports, I always thought that 3DConnection delivered a

a dual USB adapter for one Port ?

I do not use it and am not sure, maybe I mixed it up with any other

devices ....

 

Welcome to Dongletown!

[Mark Aceto]

OK this is my first, "Holy **** this thing is fast!!!!" moment.

 

The last time I installed and launched Unreal Editor, it took somewhere around 2 hours (it's a first launch thing).

 

Last week, I watched this YouTube video about the same thing on a base model Mac Studio, and the guy said it took 90 minutes.

 

Today, I set the stopwatch on my iPhone when I first launched UE on this Mac Studio... 

 

 

8 MINUTES

 

 

I checked to confirm that this is still an Intel-compiled install (it is) running on Rosetta.

 

 

I'll come back later when I bring some Spotlight Lighting Devices into UE to report how it does.

 

Next up: Twinmotion... 

[Mark Aceto]

Finally got to scratch the render itch.

 

Rule of thumb for these specs: Realistic is as fast as Preview used to be on the Intel MBP in my sig.

In this case, Preview Spotlight takes about 30 seconds for this scene, and Realistic Spotlight took about a minute and half (previously 5-10 min). Time stamp is in the screenshot below. This is on the design layer for a projection clipping study, so it's not a client-facing published viewport with RW camera, 300 DPI, and all the fixin's.

 

I was hoping that Preview Spotlight would feel like "real time" but it doesn't really function that way. It's more like a stop-start... rotate or pan view... stop-start.... vs what we're used to with Shaded or TM/UE. In hindsight, that makes sense because each scene is "baked" with RW. If anything moves are changes (even a class visibility), the scene has to bake again.

 

Looking forward to testing this with more challenging renders this weekend... 

 

Ugh... upload failed (again).

 

 

Edited June 17, 2022 by Mark Aceto

[RonMan]

On 6/11/2022 at 10:16 AM, J. Miller said:

I agree RonMan.
I drank the kook-aid and have a similar model. The  render times scream. All 20 cores are used at once. A 30 minute custom render is like 5 minutes now. 

Are you using one or 2 monitors?  I run 2 and the only drawback so far has been a random screen freeze on the main monitor. The 2nd monitor remains active. I can’t duplicate the problem and it usually happens with a move type  event. ( redraw) maybe. The repair is to unplug the monitor and plug it back in. No data loss just a big PIA.  I’m working with Apple about this but not resolved yet.

Jeff 

I go away for tea or a bio break and four times so far I come back and all my windows are on one screen. I am unable to figure out why. I use Moom, with one click everything is back where it belongs. If I find the source I will post.

[Mark Aceto]

9 hours ago, RonMan said:

Moom

 

I would be lost without it:

[line-weight]

On 3/10/2022 at 10:41 AM, line-weight said:

I'd like to give a couple of examples of things where I have found the performance of Vectorworks 2022 on M1 disappointing.

 

These are both "everyday" tasks which, having been told that VW2022 has native support for Apple Silicon, I might have hoped would see improved performance compared to VW2021 on the same computer. But in fact what I see is either no improvement or actually *worse* performance.

 

They are both tasks that I might be commonly carrying out in a repetitive manner and therefore have a significant impact on speed of working. And although this is a large and complex file, I don't see that the tasks themselves are ones that ought to be computationally difficult.

 

This is VW2022 SP2.1 on an M1 mac mini with 16GB RAM. The VWX file size is around 2GB.

 

First, moving a single viewport on a sheet layer, then undoing the action:

 

 

 

Second, double clicking on a group to edit within that group, then double clicking a symbol to edit the symbol, then exiting the symbol, and exiting the group. This doesn't happen as painfully slowly as the previous example, but in real world use it's something you might do repeatedly and it all adds up. To me it doesn't feel unreasonable to expect that something like this (entering and existing a container object) should happen instantaneously.

 

 

Someone may tell me that this is all down to lack of memory.

 

You can see that VW is occupying a lot of memory - about 30GB, which in my limited understanding means it's swapping a lot of stuff in and out of some kind of cache somewhere. But why is it using so much memory when the exact same file in VW2021 seems to use up about half that amount?

 

I meet the "mid level profile" hardware requirements listed here

https://www.vectorworks.net/sysreq

although I don't meet the "high level profile".

 

Therefore I might expect that I am pushing things a bit with this file which is quite complex, so I would forgive things being a bit sluggish if I want to navigate the whole model in 3d with shadows on, or something like that. But these seem quite simple tasks to me. Or is it that the file size is simply one that an M1 machine with 16GB RAM can't be expected to handle efficiently?

 

I've just repeated these tests on VW2022 SP4 (again on my 16GB M1 mini) and am sad to say that I see no improvement.

[line-weight]

On 3/13/2022 at 9:54 PM, line-weight said:

Final test for this evening - same file, again with "save viewport cache" and "save VGM graphics cache" unticked.

 

This time, instead of cycling through (3d shaded perspective) saved views to see what happens with the memory, I looked at a sheet layer with several viewports each containing a fair bit of geometry (not section viewports, 3d perspective renderworks viewports, but displaying in wireframe because I haven't updated them since opening the file).

 

In VW2021, this prompts the memory to go up to about 33GB. Zooming and panning is sluggish but not completely unusable.

 

But in VW2022, the memory goes off up to 40GB and beyond, everything reacts incredibly slowly with a lot of beacballs and it's essentially unusable.

 

So this seems to be where 2022 is different, and worse. If the root cause is a "memory leak" problem in Monterey, then VW2022 seems to be more badly affected than VW2021, at least working in sheet layers.

 

I've also repeated this test in VW2022 SP4.

 

Things are perhaps a little better - such that VW2022 behaves not significantly worse than VW2021.

 

Navigating a large sheet layer with multiple viewports remains painfully slow.

[Mark Aceto]

Experiencing my first sinking feeling "gut punch" with the Studio... 

 

Exporting to 3DM is hardly taking advantage of the hardware, so it's taking forever (not to mention collapsing everything to 1 layer for my design team to sort out).

 

Here's Activity Monitor CPU underutilization:

 

[Mark Aceto]

FBX export was slightly faster.

 

I suspect if I tried exporting the same file on my 2019 MBP, I either would have had to wait even longer or force quit.

 

For some reason / reasons file import/export has always been an issue with VW, especially on a Mac.

 

 

Edited June 29, 2022 by Mark Aceto

[zoomer]

38 minutes ago, Mark Aceto said:

For some reason / reasons file import/export has always been an issue with VW, especially on a Mac.

 

I agree.

I can import RVT or IFC much faster in my other CAD.

FBX export from VW was usually OK for me, beside that if the file is too large,

VW will save an empty FBX file.

[Mark Aceto]

On 6/29/2022 at 10:54 AM, zoomer said:

I agree.

I can import RVT or IFC much faster in my other CAD.

FBX export from VW was usually OK for me, beside that if the file is too large,

VW will save an empty FBX file.

 

Yeah, so I think the takeaway here this that the additional cores in the Ultra aren't helping but the additional memory might be.

Keep in mind, single core performance (of any of the cores) is pretty much the same across all M1 SoC's (from base model to Ultra).

 

Edited June 30, 2022 by Mark Aceto

[line-weight]

So frustrating that it's not possible to add memory to the M1 minis.

[zoomer]

16 minutes ago, Mark Aceto said:

Keep in mind, single core performance (of any of the cores) is pretty much the same across all M1 SoC's (from base model to Ultra).

 

 

Yes.

At least the M2 is 18% (?) faster.

But Intel, good in single core in the past, nearly overtaken by AMD lately,

good much better again too. M1 and better M2 are still on the OK side

but far from leading, which M1 let us hope for, nearly 2 years ago.

 

But if rumors turn out to be true and larger M2s (Pro/Max/Ultra/Extreme/Ludicrous/...)

will already use 3 nm production, it could look a bit better.

 

Nevertheless, my M1 Mini is already 6 moths over its expected life time, to bridge

waiting for something (really) better. So I am curious how long I further have to wait ....

 

[Mark Aceto]

What I'm unclear on is that the M1 Ultra has 20 cores but my Intel 10-core chip had 20 threads. In Activity Monitor, the graphical display of those 20 cores/threads looks identical. For context, I was expecting to see 40 threads when I jumped from 10 cores (20 threads) to 20 cores.

 

The nice things about this are:

• Every M1 core is the fastest single core (not just one of them that has been boosted)
• Oftentimes VW could only use 1 thread of each core, so it would be utilizing 10 threads (but the other 10 threads were idling)
  • Well, thinking this through a bit more... no change here

I'm not an expert but I think this sort of explains why the Ultra is getting beaten by the Intel 12900K (both single and multi core) and destroyed by 3-year old AMD Threadrippers in Cinebench (which will use every thread). Put another way, the M1 Ultra is more like a 10-core CPU with 20 threads when comparing Apples to oranges.

[zoomer]

In later x86 most cores were multithreaded (1 core = 2 Threads)

Multithreading for consumers started around 2005 (?)

 

The acceleration was about 40% vs single threaded.

So far from 200%, but a nice boost that, at Intel iMac 27" times,

was worth the Apple Upgrade Tax.

 

But not all CPUs had multithreading.

Like cheaper i3 and i5 usually had multithreading disabled.

AFAIK also the newer efficiency cores are often only single threaded

while only power cores are multithreaded.

 

Multithreaded mainly meant dual-threaded.

But could be much more on larger CISC dies. AFAIK it was long time speculated

that AMD may switch to 4 threads for their Threadrippers and Server CPUs.

But that did not happen so far.

 

ARM CPUs are usually single threaded.

AFAIK ARM could do MT too, if it would be worth in overall efficiency and energy

consumption, for the typical devices that use ARM hardware.

Maybe it is still better to just use 40% more but simpler ST cores.

 

So for now the Studio Ultra has 20 cores / 20 Threads.

But beside other cores, the 20 CPU cores of the Ultra scale well, similar to what

we were used from our Intel Macs.

 

 

11 hours ago, Mark Aceto said:

Every M1 core is the fastest single core (not just one of them that has been boosted)

 

That is true for AMD or Intel CPU too.

Usually for ST tasks, they switch through all cores, to better distribute the heat over

the whole die, or prefer one of the better cores, to reach highest ST Boost frequency.

 

If you have MT and more and more cores in use at a time, the max Boost Frequency decreases

again at certain steps for power and thermal reasons. The marketed standard CPU frequency

is what the CPU should reach if all cores are in use, at least

under propper cooling.

 

 

 

11 hours ago, Mark Aceto said:

Oftentimes VW could only use 1 thread of each core, so it would be utilizing 10 threads (but the other 10 threads were idling)

 

No, VW offentimes can use only a single core, as there are threads in CAD that can't be

separated into multiple threads as they need to be calculated linearly and everything

else has to wait for that result.

But you can use those other free CPU Cores/Threads for iTunes, Mail, Safari, .... during

that time.

 

 

 

11 hours ago, Mark Aceto said:

explains why the Ultra is getting beaten by the Intel 12900K (both single and multi core) and destroyed by 3-year old AMD Threadrippers in Cinebench (which will use every thread).

 

 

All M1 cores/threads will be used when doing CPU Rendering like RW or Redshift CPU.

An Ultra has 20 cores = 20 Threads, these will be used.

 

Ultra is getting beaten because it is by far not yet the strongest CPU availabe.

Also because AMD or Intel lets 12900K use much more power.

(My Ryzen 3950X peaks at about 170 W. The Ultra is about same as fast but needs max 60(?) W)

[zeno]

I apologize in advance. For some things I'm interested in deepening, for others I'm not.

 

If I may, ask a question. I have a M1max with 64GB of RAM. Sometimes I see that I struggle to do navigation and, above all, sheet layers and section viewports operations.

 

Does it make sense to spend almost 8000 euros to take an M1ultra and wait for it 3 months, in the meantime comes ultra m2 or m3, extreme, planetarium, extramporalis, super Sayan level V processors?

 

I know it's a simple question, maybe from newbies, but sometimes simple things are the best.

 

(Obviously if you want to tell me which trades is best for and for which it wouldn't change much)

 

Thank you

 

@zoomer @Mark Aceto

 

[zoomer]

If possible I would go on using your M1 Max 64 GB.

Basically that should work well.

 

I would wait 6-12 (?) months until larger M2s settled.

 

 

I also had large problems with SLVP Viewport rendering or editing.

(But on Windows too. I think it is ok to say that is a VW 2022 regression)

So far I tested only VP Rendering and Sections. Not Editing.

 

Since SP4 that works much better for me on Mac. And as I see it

needs only half the RAM on PC vs SP3.1.

Overall SP4 seems to better swap memory to SSD or give it back

after usage.

Looks like even my M1 Mini is now capable of a bit larger files

than before, but of course still too limited for what I really need.

 

 

I also see lags in View Navigation, when I exceed my memory.

(Yellow memory pressure in Activity Monitor)

until everything lags and it looks like every work of VW is swapped

to SSD and shifted back and forth.

 

But meanwhile I think that would not happen that often if I had 64

instead of only 16 GB. Like I also use easily 32 GB for VW alone

on my PC for larger projects.

 

 

So i would first test again with SP4.

(If you have the ITA VW version you may need to wait another 1-2

weeks for SP4 ?)

If it runs better - great. If not, I would also wait 10 weeks and hope

for VW 2023 improvements (?)

[dtheory]

How are folks feeling about the Max vs Ultra question at this point? 

More RAM more important than the better processor? ornot?

[Mark Aceto]

5 hours ago, zoomer said:

17 hours ago, Mark Aceto said:

Oftentimes VW could only use 1 thread of each core, so it would be utilizing 10 threads (but the other 10 threads were idling)

 

No, VW offentimes can use only a single core, as there are threads in CAD that can't be

separated into multiple threads as they need to be calculated linearly and everything

else has to wait for that result.

But you can use those other free CPU Cores/Threads for iTunes, Mail, Safari, .... during

that time.

 

I'd have to dig through screenshots but I can definitely show you VW using every other thread.

[Mark Aceto]

3 hours ago, zeno said:

I apologize in advance. For some things I'm interested in deepening, for others I'm not.

 

If I may, ask a question. I have a M1max with 64GB of RAM. Sometimes I see that I struggle to do navigation and, above all, sheet layers and section viewports operations.

 

Does it make sense to spend almost 8000 euros to take an M1ultra and wait for it 3 months, in the meantime comes ultra m2 or m3, extreme, planetarium, extramporalis, super Sayan level V processors?

 

I know it's a simple question, maybe from newbies, but sometimes simple things are the best.

 

(Obviously if you want to tell me which trades is best for and for which it wouldn't change much)

 

Thank you

 

@zoomer @Mark Aceto

 

 

I'm still getting to know this Ultra but I'd say you're in a good spot to wait and see what's announced in the second half of this year. What you're missing right now is faster renders with RW. I would say the rest of your workflows are probably the same speed / performance as this studio because of how the ARM SoC's scale.

 

Personally, I decided to commit to the Studio because the Pro seems to be delayed, and I needed the fast Mac I could get my hands on right now. I don't have any buyer's remorse, and honestly, it's more about software developers optimizing for M series, so our M1 computers should get relatively faster as the software we use everyday gets more optimized (kind of like when NVIDIA release a new GPU on day 1 vs a year later).

 

The other factor that went into my decision is that I only need a laptop 10% of the time, so I'll keep the old MBP around until there's a 15" M2 Air or something. Everyone's needs are different but I'm glad I doubled my multicore performance 90% of the time.

 

My next purchase will hopefully be a M2 Extreme with 40 cores (in an enclosure that fits my budget). However, what's more critical to me is hardware accelerated ray tracing for UE and TM. So, until that's available, I'll probably stick with this Studio.

 

Btw there's an eponymous law (like Murphy's or Moore's) about perpetually waiting for the next tech (particularly Apple devices). The name escapes me but if anyone is still on an Intel machine from 2015, just get the best option that's available right now. Especially with the chip shortage and general supply chain issues. I sold my unboxed MBP M1 Max for the same as I paid for it 4 months later, and the buyer said that people were scalping them for $2000 USD above list price. Mac's hold their value, so you can always sell it when something better comes along.

[Mark Aceto]

6 hours ago, dtheory said:

How are folks feeling about the Max vs Ultra question at this point? 

More RAM more important than the better processor? ornot?

 

TL;DR

• 32gb minimum
• 48gb good
• 64gb better
• 128 best

And, remember, that number is combined RAM and GPU, so you can't compare 16gb M1 to 16gb Intel (for a lot of reasons). And I'm still waiting for someone at VW to officially confirm how many gigs the GPU can pull from RAM. My wild guess based on absolutely nothing is that it's split down the middle: 50/50.

 

I would say the additional RAM in the Ultra is a nice-to-have. A maxed out Max is a great Mac for most VW users.

 

If you're a heavy RW user that can take advantage of (double) the cores, you know who you are.

 

Btw I have iStat Menus running, and I would say no app developers have figured out how to optimize memory usage in these things (including Apple). Sometimes I'm using 50% when I'm barely doing anything (VW is closed). Meanwhile, right now I'm using 22% and VW is open in the background (has been for a few days). It's kind of a new frontier right now. I imagine a year or two from now, things will level off.

 

Edited July 1, 2022 by Mark Aceto