Sidebar Sidebar
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

M4 Max (16-Core CPU, 128GB RAM) vs. M3 Ultra (28-Core CPU, 96GB RAM)

seastar said:
For example, someone who buys an M3 Ultra today might end up with an outdated system much sooner than expected especially with the release of the M5 and possibly even the M6 not long after.
Click to expand...

I think this is already the case for M1 Ultra and M2 Ultra. It won't be much different for M3 Ultra.

The only difference is that when M1/M2 Ultra existed, there were demands for >128GB of RAM and M1/M2 Ultra just didn't have it. So even if some people may consider M1/M2 Ultra still quite capable today, it just doesn't cut it for those who need >128GB of RAM. Now M3 Ultra has up to 512 GB of RAM. Even if soon it won't be the fastest Apple CPU/GPU, it will remain a not-so-slow option for a very long while, especially for those who wait for it for its large amount of RAM. To me, that's what separates M3 Ultra from M1/M2 Ultra.

I plan to purchase an M3 Ultra Studio with 256 GB of RAM. But since Studio won't be available for another month or two in my country, I will wait until there are enough benchmark testing or even wait until WWDC to make the final decision.
 
Last edited:
  • Like
Reactions: picpicmac and Zubba
asiga said:
Mine has shipped!! (128GB 2TB). Pick up date has changed from March 25 to March 21!!
Click to expand...
It arrived to the store today, even before March 21!! I’ll be picking it up in a couple of hours. It seems they are speeding up Mac Studio orders, my initial pickup estimate was March 25, so this is one week earlier than initially expected.
 
  • Like
Reactions: xraydoc and JSRinUK
seastar said:
I purchased the base model of the Mac Studio and canceled my previously ordered M4 Max Mac Studio (128GB RAM and 1TB SSD).

As I mentioned in my previous reply, Apple releases new models frequently, so I decided to buy only what I truly need without any extras.

36GB of RAM will be sufficient for me because even if I allocate 16GB of RAM per VM and run two VMs simultaneously, I will still have 4GB of RAM left for the main system. Additionally, Docker doesn’t consume much RAM, so this setup should work well for my needs.

By the way, this is the most CPU and RAM intensive scenario for my use cases, which is why I shared it.
View attachment 2492952
Click to expand...
I also considered this option. But my MBP (from 2018, I make my Macs last ;) ) is a maxed out i9 with 32GB, and over the years I found some tasks where I wished I had more than 32GB, specially when building LLVM (which I usually build from source, and flang takes a lot of RAM to build), and also when creating ramdisks (a technique I sometimes use for speeding up some disk-based computing). But apart from that, 32GB have been fine for me during these 7 years since I bought that MBP, and I don’t plan to update my MBP until at least they release the rumored thinner version in 2026, or even later than that, because it’s still a very heavy duty MBP. So, for my new desktop, I really wanted more than 32GB. I guess this M4 128 GB will also last me at least 10 years. If new hardware is released for AI before 10 years, I’ll likely be getting a Linux box with a NVIDIA monster inside, and use it as a server from the Mac.
 
whwang said:
I think this is already the case for M1 Ultra and M2 Ultra. It won't be much different for M3 Ultra.

The only difference is that when M1/M2 Ultra existed, there were demands for >128GB of RAM and M1/M2 Ultra just didn't have it. So even if some people may consider M1/M2 Ultra still quite capable today, it just doesn't cut it for those who need >128GB of RAM. Now M3 Ultra has up to 512 GB of RAM. Even if soon it won't be the fastest Apple CPU/GPU, it will remain a not-so-slow option for a very long while, especially for those who wait for it for its large amount of RAM. To me, that's what separates M3 Ultra from M1/M2 Ultra.

I plan to purchase an M3 Ultra Studio with 256 GB of RAM. But since Studio won't be available for another month or two in my country, I will wait until there are enough benchmark testing or even wait until WWDC to make the final decision.
Click to expand...
To counter that, if you watch Artisright on YouTube he continues to suggest older Mac Studios because they hold up incredibly well performance-wise, The M1U and M2U going toe to toe pretty well with each other and even the M4M & M3U. If you focus on the performance over the resale market and psychology of having an "older chip", it will likely serve your needs for just as long.
 
  • Like
Reactions: Toonartist and picpicmac
Joooeforlong said:
To counter that, if you watch Artisright on YouTube he continues to suggest older Mac Studios because they hold up incredibly well performance-wise, The M1U and M2U going toe to toe pretty well with each other and even the M4M & M3U. If you focus on the performance over the resale market and psychology of having an "older chip", it will likely serve your needs for just as long.
Click to expand...
What you said really just made my point stronger, M3U will remain strong in the era of M4 or even M5.
 
  • Like
Reactions: picpicmac and Joooeforlong
asiga said:
... I’ll likely be getting a Linux box with a NVIDIA monster inside, and use it as a server from the Mac.
Click to expand...
In my opinion, this is the best solution. Instead of spending nearly $4,000 on a machine with 128GB of RAM, you could go with this approach, and if you prefer a desktop over a laptop, simply get the base model of the M4 Max Mac Studio.
I find the base model of the Mac Studio to be an excellent choice. It feels more stable than the Mac Mini, probably due to its more powerful cooling system.
 
  • Like
Reactions: asiga
If the historical pattern of 30% gains in performance for each new M Series CPU holds, the M5 Max Studio will have geek bench scores of 5k+ for single core and about 35k for multicore speeds. That will be 2x the M2 Max performance and by next year TB5 external enclosures will be common (Maybe even with PCIe Gen 5 drives) and maybe a TB5 ASD will appear. Worth the wait.
 
datagov63 said:
If the historical pattern of 30% gains in performance for each new M Series CPU holds, the M5 Max Studio will have geek bench scores of 5k+ for single core and about 35k for multicore speeds. That will be 2x the M2 Max performance and by next year TB5 external enclosures will be common (Maybe even with PCIe Gen 5 drives) and maybe a TB5 ASD will appear. Worth the wait.
Click to expand...
M2 -> M3 was only about 12-13%
M3 -> M4 was about 25+%
 
  • Like
Reactions: picpicmac
whwang said:
What you said really just made my point stronger, M3U will remain strong in the era of M4 or even M5.
Click to expand...
What exactly makes the M3 Ultra so special that you believe it will remain stronger than even the M4 or M5?
 
My estimated delivery for my M3 Ultra has been showing "Arrives Mar 19 - Mar 21" for past week or so. Today is Mar 17, so I'm suspecting delivery to slide out some to be beyond Mar 21st. It must be when the progress bar moves to "Preparing to Ship" it will then take another 3 or 4 more days to actually deliver.
 
  • Like
Reactions: asiga
seastar said:
In my opinion, this is the best solution. Instead of spending nearly $4,000 on a machine with 128GB of RAM, you could go with this approach, and if you prefer a desktop over a laptop, simply get the base model of the M4 Max Mac Studio.
I find the base model of the Mac Studio to be an excellent choice. It feels more stable than the Mac Mini, probably due to its more powerful cooling system.
Click to expand...
Yes, I fully understand. Here we are at differences between our ways of work. I use both laptop and desktop. This purchase replaces my desktop (a hackintosh built in 2016 with a NVIDIA Pascal Titan 12GB VRAM, Skylake, and 64GB RAM). OTOH, the laptop (2018 maxed up 15’’ MBP) has still 2 or 4 years more of use until being replaced.

I didn’t discard running LLMs, but I’ll run only those whose complete model will run in 128GB. Also, I’ll be using more than 64GB in some development projects. And, most importantly, I won’t be buying another Mac desktop in 10 years or so. Therefore, the maxed up M4 Max is what fits the best (besides, that’s my usual way of buying Macs: buy the highest I can afford and make it last a decade).

My only concern is that the AI trend can change hardware a lot in the next 5 years. But that’s only for AI, and, if many things change (for example, if NVIDIA releases a Geforce with 1TB RAM at prosumer price), then I might consider building a Linux box with it. Otherwise, this M4 Max 128GB 2TB will provide me with everything I’ll need for 10 years.

BTW, got it, and unboxed it. Now set up fun :)
 
  • Like
Reactions: JaredJenkinsDesign and picpicmac
Have just rerun my speed tests for the 4TB OWC TB5 Envoy Ultra on my temporary M4 Pro mini while waiting for my M3 Ultra to deliver.

I compared the read/writing to the M4 Pro mini's internal 512 GB SSD and to the Envoy Ultra.

I used the Terminal command dd for doing these tests.

The Envoy Ultra was directly connected to the mini's backside TB5 port.

Writing and reading a 20 GB file to/from the mini's internal 512GB SSD
============================================
Writing to internal SSD: 3.256482760 GB/sec
Reading from internal SSD: 6.624817598 GB/sec

Writing and reading a 20 GB file to/from the 4TB OWC TB5 Envoy Ultra
============================================
Writing to Envoy Ultra: 4.466570245 GB/sec
Read from Envoy Ultra: 7.054391160 GB/sec

Internal 512 GB SSD had 208 GB free space & Internal SSD has an SLC write cache of around ~40 GB
External 4TB Envoy Ultra SSD had 3.28 TB free space - Envoy Ultra has an SLC write cache of around ~60 GB

Recently the 4TB OWC TB5 Envoy Ultra has just been reduced by $50 to $549, and is on its 3rd pre-order phase.

Question: Setting up two Envoy Ultra's as RAID-0 will allow what kind of read and write data rates ? :) :eek::D
Question: Setting up two Envoy Ultra's as RAID-0 will allow what kind of combined SLC write cache size?
 
bxs said:
My estimated delivery for my M3 Ultra has been showing "Arrives Mar 19 - Mar 21" for past week or so. Today is Mar 17, so I'm suspecting delivery to slide out some to be beyond Mar 21st. It must be when the progress bar moves to "Preparing to Ship" it will then take another 3 or 4 more days to actually deliver.
Click to expand...
Hey, hey..... The progress bar just moved to the "Preparing to Ship" state..... so delivery may still make it between Mar 19 - Mar 21.:) ...plus my Apple Card with its 3% cash back has been charged for the M3 Ultra's purchase!

It's going to be 'beans on toast' for a few weeks now.... ;)
 
Last edited:
  • Like
Reactions: asiga
picpicmac said:
Not going to happen.
Click to expand...
Of course it will. The M4 Max has 4k single core and 27k multicore in Geekbench 6. The M3 Max had 3k and 20k. And the M2 Max is 2.6k and 15k. The clock ticks by the same increments as the chip names scale up. The M5 has architectural changes and by M6, Apple gets to 2nm.

Meanwhile, despite all the peer pressure to upgrade, my Studio M2 Max continues to do everything I need and there is yet much more software optimization for apple silicon to come.
 
  • Like
Reactions: seastar
datagov63 said:
and by M6, Apple gets to 2nm.
Click to expand...
If the M6 is 2nm then it is possible, maybe even likely, that an M6 Max will reach your goal.

But, if the M5 is just the next 3nm node along TSMC's roadmap then I expect it to only have a modest clock bump. And if there are increases in core counts I expect them to be minimal increases.
 
The jump from M2 Max to M3 Max isn't that huge, while the jump from M3 Max to M4 Max is very significant. Who knows what will happen from M4 to M5? Maybe again an incremental one, and the next big jump will be the next revision (M5 to M6).
 
picpicmac said:
If the M6 is 2nm then it is possible, maybe even likely, that an M6 Max will reach your goal.

But, if the M5 is just the next 3nm node along TSMC's roadmap then I expect it to only have a modest clock bump. And if there are increases in core counts I expect them to be minimal increases.
Click to expand...
Sure you could be right. But today, there are few apps that stress the M2 Max if you are not doing video or AI processing. And currently there are no TB5 monitors and very few external ssd enclosures. Waiting a year for an upgrade seems to me, for my needs, the prudent approach.
 
  • Like
Reactions: splifingate
As good as M1U or M2U is, they don't have latest TB5, which makes it not future proof enough. The TB5 does change a lot of things, you can see the TB5 products or DP2.1 products are more and more.
As for M3U, it is a hard choice for people who doesn't need that amount of RAM (256 or 512), the architecture fall behind M4 family. A bit sad we start seeing Mac Studio not getting latest and greatest...
Waiting for refurb M4 max with 128RAM, so happy you can buy this config with 512GB SSD (later can modify to upgrade to as much as 8TB)
 
Seem like M4 Max is the sweet spot base [if one is fine with 36GB, if not even knocking up chipset to the next level unbin is fine if need 48/64GB]. and leaving it at 512GB/1TB and hooking external storage. i plug a T9 4tb and a TB5 diy 4TB into it and its golden. 1000$ For 8TB worth of externals that is quite fast [for my use case that is, while TB5 speeds are impressive i probably can't even go to toilet while transferring files]. The most I'd go through is 15-30GB worth a session. and maybe 60GB worth of video clips.

I can't see an actual use case for M3 Ultra even though it seems powerful and gargantuan. Photo/Video editing with more emphasis towards Photos && Using Davinci Resolve for occasional video clips.
 
You must log in or register to reply here.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Top Bottom
Back