Ask HN: Intel Mac Pro Users with > 192GB of RAM, Views on New Machine? | Hacker News Reader

Ask HN: Intel Mac Pro Users with > 192GB of RAM, Views on New Machine?

34 pointsjmkni39 comments

The old Intel Mac Pro maxxed out at 1.5TB of RAM, the new one maxxes out at 192GB.

I've been wondering recently about those with > 192GB of RAM in the Intel model, what are your views on the new Apple Silicon version?

Is it still useful for you?

Does the less RAM matter less because it's faster (as is the SSD when it comes to swapping etc)?

Are you screwed because you need all of that RAM, are dependent on macOS specific software and now there is no upgrade path?

Or are you going to move to a PC running Windows or Linux with all of the RAM you need?

OR are you just sticking with the machine you have because the build quality is great and it does everything you need + will last for a long time to come?

Also, what do you need all of that RAM for anyway?

Cheers

View on HN

Why not just use Ubuntu and a custom desktop? It’s basically intended for use as something server like at that point anyway. Then ssh into it with your ultraportable laptop. You can mail me the money you saved, I’ll spend it on something good. Don’t worry

I did exactly this. Initially i wanted a mac pro but went for a custom amd nvidia combo and couldn't be happier. Plan slapping a second 3090 with nvlink, might even upgrade my ram to 6.5 ghz.

Ever run into issues with not having a static ip when out of the house? Do you use a dynamic ip service? Any security precautions, besides using ssh, you take? I’ve only used mine from home so far over lan

One option is to use ngrok, but in my case i leased a dedicated ip address. Another option is the cheapest available vps and reverse ssh tunnelling.

In regards to security i only ever enable ssh. While that could be an issue i dont worry that much. With ssh enabled you can forward ports to your local machine.

One thing that sucks is that x11 forwarding is super slow and as such i am stuck with vnc for when i need desktop apps.

You can always use a tunneling service such as https://pinggy.io or zrok for self hosted or ngrok.

“$2.50/month” is a very good price. I think your product should be advertised a bit more as i didnt stumble upon it when googling ngrok alternatives. Also i love that you dont need to install tools, really clever. Thanks for sharing.

FWIW it’s extremely easy to get a domain despite having a dynamic IP with Google domains. I just have a cron job on my linux server that sends its dynamic IP address to a Google API endpoint.

https://support.google.com/domains/answer/6147083?hl=en

I have a similar setup and decided to use Tailscale. It's been working seamlessly so far.

> The old Intel Mac Pro maxxed out at 1.5TB of RAM, the new one maxxes out at 192GB.

I mean if this is not a regression, then what. I could live with 192GB just as I lived with 12MB back in the day. Now there might be some workloads which benefit from more RAM (like some concurent youtube sessions) but besides that, if i stay away from Teams, i think i can make it. All in all, i can always use the excess RAM as a RAM drive. /s

My question would be why not hackintosh? easy to build these days, more customizable and able to get even better performance than Apple silicon...

You get no support whatsoever with a hackintosh and generally the people that need these computers don’t also want to spend time doing their own IT support work as well.

To be honest, my hackintosh from the Haswell era was the most stable machine I have had; certainly better than whatever the shitshow my current machine has been with OpenSuse or Manjaro on a 12th gen intel, and slightly ahead of my 5800x machine.

If you need your machine for $DayJob, you're not going to trust that to a hobbiest solution. And now that the transition to Apple Silicon is complete, don't expect future version of macOS to have an Intel version for much longer.

For anyone curious, this seems to be one of the more popular easy Hackintosh-in-a-VM setup projects these days:

* https://github.com/notAperson535/OneClick-macOS-Simple-KVM

Works "out of the box" on Ubuntu, and apparently on Windows too thought I've not tried that.

I've used it on Ubuntu occasionally when trying to diagnose intel specific macOS weirdness for a side project, when the weirdness doesn't show up on the M1 mac mini.

Does hackintosh still require custom kernel level drivers with no available source code?

Nope, OpenCore and most drivers are open source on GitHub

Are there any good GPU options for Hackintosh? Seems like the top of the line is still based on what Nvidia and AMD were selling 5-10 years ago.

My rx6900xt works fine through opencore and whatever red.

Oh cool, I didn't realize modern GPUs were usable

My 2019 mac pro has 384gb of ram I believe - I honestly don't need this much RAM but I'm keeping it because for my work (devops) and for my creative endeavours (music and video production) most of my apps and plugins don't support arm yet in fact even the ones that support arm perform better under x86 still for some reason

Out of curiosity, how much of that do you typically use during a heavy workload, and what apps use most of it?

Asking because I have 64GB on Ubuntu and rarely exceed 30 or so of actual usage. 384GB of RAM is just wild to me. Maybe it's because I don't do video/audio production, but I can't imagine what one could possibly run on a non-server that would use that much.

Not the parent but the biggest use of more ram I've seen in extreme situations like this is having VMs open. Still, I can't imagine what I could use 384GB for (not that it wouldn't be a blast to have).

IMO there is some lack of information on this. Especially in how the on-socket RAM performs in combination with the GPU.

If anybody using llama or Stable diffusion with such a setup could share their experience vs a more traditional Nvidia GPU setup it would be very interesting.

I am sorry, I am out of memory :p

This question seems to ignore how RAM works in the Apple Silicon machines. It's part of a system--as available, as quickly, to the CPU as to the GPU.

It does not. OP asks if the increase in speed makes up for the limited quantity. They are asking for people's real world experience with the tradeoff.

yeah, that's a bunch of crap. If you have something that needs lots of RAM (say audio uses where samples and such must be in RAM for latency reasons) swapping to SSD is NOT a solution.

And swapping to SSD is pretty stupid - SSDs have finite write lifetimes. Yeah, you can get data center capable SSDs that are over provisioned and cached to the hilt - but at that point RAM would just be cheaper; and still be faster.

I get why Apple is sticking with the tightly coupled RAM on SOC - it's part of the equation that gets them their blistering performance (love my maxed out M1 Max - but kind of nuts 64GB is the most it can have). Some use cases just need more memory - period.

I'm still hoping they will have a true desktop chip at some time where RAM could be externalized even if it causes a slight performance hit. The problem is the number of users who truly need that much RAM is so small I doubt they can justify diverging from their existing architecture that severely.

  > I'm still hoping they will have a true desktop chip at some time where RAM could be externalized even if it causes a slight performance hit.

I believe it was John Ternus on the WWDC23 episode of The Talk Show that said this isn’t going to happen.

They were uncharacteristically firm about this. Same with supporting external GPUs. The indication was this is not going to happen.

How is that relevant?

Spends a lot fewer cpu cycles waiting for data. Same is true with how much faster the main storage moves. Won’t be enough for all workflows (hence it’s worth the OPs question) but will go further than expected.

Edit-latest i9s seem to get around 90GB/s to ram while they promise the M2 ultra 800GB/s. So quarter second to access all of the ram?

Edit2-the ssd is higher end nvme speeds.

Bandwidth doesn't tell us about how many clock cycles it's waiting for the data though (latency), does it?

For example with paging, I don't think those bandwidths make much difference and it's instead the latency which matters. Or am I wrong?

What’s going to have the faster latency? Memory chiplets on the same substrate as the cpu cores or memory dimms sitting in a housing a few inches away?

I’m sure there’s more than a few factors making latency faster for those chiplets.

https://www.anandtech.com/show/17431/apple-announces-m2-soc-...

Edit-but I don’t think this question is about latency exactly. If you’re editing a large file that fits in 196gb ram all at once then those edits should benefit by apparently an order of magnitude higher bandwidth. It’s going to be able to address all of its ram several times per second while the i9 will take much longer (roughly). That and similar speedups from the whole systems engineering should make up some for not having more ram. Think how 8 gig of ram in the m1 was widely reported to be more usable than expected when announced.

Memory chiplets on the same substrate as the cpu cores or memory dimms sitting in a housing a few inches away?

Where are you getting this information on the memory latency? What is the actual number?

I don’t think this question is about latency exactly.

You were the one the brought up latency

those edits should benefit by apparently an order of magnitude higher bandwidth

Most software is not written well enough to be memory bandwidth limited, even graphics software. It is mostly games that end up actually needing the memory bandwidth.

You mean like how 8GB RAM in my Mini results in prolonged swap pauses with 3 apps open?

To be fair, M1 MacBook Pro with 32GB is smooth as silk with ~20 apps open.

Which 3 apps on the Mini? I’ve actually been really impressed with my base model M1 Mini. I’ve been able to use XCode, Safari, iMessage, and Music playing without many noticeable hiccups at all.

Chrome, PHPstorm, OpenSCAD, Cura. I lied, 4 apps.

My point though is, doesn’t matter how fast your RAM is if you run out and have to swap out.

Oh yeah JetBrains IDEs definitely need more RAM than Xcode. Likewise with Chrome versus Safari. I’m not familiar with OpenSCAD and Cura.

OpenSCAD is a problem even with 96GB ram.

Something is very wrong with their build on Apple Silicon. It works fine for me on other platforms.

I've been having acceptable experience with it on my 32GB M1 MBP with the Apple Silicon build. I admit though, I don't use it as much on my work laptop as my home mini.

Reply on news.ycombinator.com