News:

Willkommen im Notebookcheck.com Forum! Hier können sie über alle unsere Artikel und allgemein über Notebook relevante Dinge disuktieren. Viel Spass!

Main Menu

Post reply

Other options
Verification:
Please leave this box empty:

Shortcuts: ALT+S post or ALT+P preview

Topic summary

Posted by Dave Haynie
 - March 19, 2025, 00:04:40
The simple fact that an M3 Ultra system with twice as many cores, some hellacious main memory speed (as long as the GPU's not too busy), etc. only sees 25-30% performance a is a testament to the wisdom of AMD maintaining multi threading. How about a core to core test? Or a price to price test? Why claim these AMD and Intel chips are peak, when there are Threadrippers, Threadrippers Pro, Xeons, and Epyx processors... Some well in the same price range as an M3 Ultra system.

Sure, the M3 Ultra will use less power. But both systems are plugged in. The power efficiency in a laptop matters, but every laptop is an inherent cost, performance, and usability disadvantage.
Posted by No one
 - March 16, 2025, 07:28:35
CR2024 isn't meant to be compared cross system, it's meant to compare similar systems.
Posted by Memalcolm
 - March 15, 2025, 16:29:54
This article is trash. 14k for the max spec m3 ultra? You're comparing workstation hardware to home pcs. Compare this to a threadripper workstation and see how it stacks up.
Posted by davidm
 - March 14, 2025, 15:04:20
Quote from: Phil.k on March 13, 2025, 23:06:37Says tested against highest end x86 chips. Ignores middle range threadripper with the same core count and better cinebench scores

That's like the "25% improvement" comment. Test it comprehensively on a range of benchmarks, including price:performance, wattage:performance, db:performance. And test it with up to 500GB of RAM. The M-chip is not going to win all of them, but some of them will be quite revealing.
Posted by Phil.k
 - March 13, 2025, 23:06:37
Says tested against highest end x86 chips. Ignores middle range threadripper with the same core count and better cinebench scores
Posted by Oum
 - March 13, 2025, 17:22:18
Twice the cores but 25% improvement only ? Sorry I think Apple lost here
Posted by davidm
 - March 13, 2025, 17:03:24
Quote from: Sasquach on March 13, 2025, 16:28:17
Quote from: umbrella on March 13, 2025, 08:50:30everyone keeps mentioning core count but that's somewhat irrelevant here, AMD still uses hyper threading meaning 32 threads. Intel ditched hyper threading and has 24 cores. Either way though there's no cheap way to obtain an m3 ultra, it's apple, so maybe it is fair to compare it to server parts.
At 14k price point I can have dell workstation sporting 2TB ram and thread ripper 7995wx with 96 cores.
That thing will run circles around M3 ultra. Especially with an option for 4 GPUs....

I don't know what kind of deals you get, but a Dell Precision 7875 with 7995wx & 2TB would be more than seven times the price of the top end Mac Ultra in my country, not including GPUs. And its RAM speed is still around half the M3 Ultra's.

Whether it runs circles depends on the task, for many tasks a $600 entry M4 would be faster. To run an LLM larger than 70b, you're going to need multiple professional GPU$$$ (about $60k worth), and you don't want that sitting on your desk. Though, M3 Ultra's startup time is pretty slow for larger LLMs, and is ultimately limited, but we're talking about something that sits on the corner of a desk in a quiet office.

As for software, a lot of Windows & Linux software looks like crap. When I have to use a Mac, I just turn all the colours and widgets off, and I'm left with a terminal (though I'd still 100% prefer to be using Linux). It's whether it runs the right software quickly.

As I said in the AMD thread, I'd like to see nbc run head to head tests of the m3 ultra against the top end x86 chips, not because I prefer Apple, but because moments of humiliation is the only way x86 is going to pull ahead.

Obviously, each system will have its own advantages, unless you're coming from a bizarre ungrounded "twice the cores but 25% improvement" perspective, Apple's lines of chip are an incredible advancement.
 

Posted by Sasquach
 - March 13, 2025, 16:28:17
Quote from: umbrella on March 13, 2025, 08:50:30everyone keeps mentioning core count but that's somewhat irrelevant here, AMD still uses hyper threading meaning 32 threads. Intel ditched hyper threading and has 24 cores. Either way though there's no cheap way to obtain an m3 ultra, it's apple, so maybe it is fair to compare it to server parts.
At 14k price point I can have dell workstation sporting 2TB ram and thread ripper 7995wx with 96 cores.
That thing will run circles around M3 ultra. Especially with an option for 4 GPUs....
Posted by Shaun
 - March 13, 2025, 12:07:30
The comments section is why most PC gamers are degens.

Happy to compare a £€2500 PC to a £€600 or less console...then they cry that people compare a super high end Mac that's probably £€10k+

(Gamed on both for most of my life)
Posted by Tecnical
 - March 13, 2025, 11:53:01
What is this !? Comparing a $15000 mac with a $2500 pc ! LoL!
This isn't a fair comparison. You should compare it with a Threadripper system.
Posted by Ayush
 - March 13, 2025, 10:27:27
Imagine comparing an ARM CPU with absolute garbage software support to full fledged x86 cpus and that too after paying 10x higher price
Posted by umbrella
 - March 13, 2025, 08:50:30
everyone keeps mentioning core count but that's somewhat irrelevant here, AMD still uses hyper threading meaning 32 threads. Intel ditched hyper threading and has 24 cores. Either way though there's no cheap way to obtain an m3 ultra, it's apple, so maybe it is fair to compare it to server parts.
Posted by Jason Rogers
 - March 13, 2025, 08:17:35
Quote from: MonkeySpanx on March 13, 2025, 08:08:44At that price you should be comparing it to threadripper systems, with way more cores.

That would be suicidal 🙂
Posted by MonkeySpanx
 - March 13, 2025, 08:08:44
At that price you should be comparing it to threadripper systems, with way more cores.
Posted by RobertJasiek
 - March 13, 2025, 08:01:41
Quote from: cyberwizrd on March 13, 2025, 01:09:53for local AI they are monsters

Not in general for local AI. The main criteria are: How much VRAM / assigned unified memory does the software need? Is the software written for Apple M or Nvidia dGPU? Depending on the software, it might or might not work on both systems, and speeds might differ by a factor of several dozens.