Skip to main content
Apple M3

Apple M3

Central processing unitWikipedia

Search complete. 36 mentions across 10 episodes found for "Apple M3".

Sep 24, 2026

Federico ViticciHOST
39:45
Okay.
Federico ViticciHOST
39:48
Coming from the M3 Ultra Mac Studio, which I still have with 512 gigs of RAM, the main thing that this computer has allowed is... being able to use the best local models that fit on this machine with the kind of performance that makes them nice to use in apps.
Federico ViticciHOST
40:16
Like these local models could be used before in terms of fitting on the computer, right? And even on the M3 Ultra, I could even fit bigger models because I have the 512 gig version.
Federico ViticciHOST
40:31
But even if they technically fit in the machine, in memory and in storage, they were not necessarily nice to use because they were slow.
Federico ViticciHOST
40:43
And they were slow in two different metrics.

12 MINS LATER

Federico ViticciHOST
52:59
thing right now and I think if they're serious about this like when I when I when I mentioned like you gotta package it up in a way that makes sense they have a lot of work to do instead of just saying well just wait for you know the developer the open source community to figure it out But that's not a way to build a sustainable product over time.
Federico ViticciHOST
53:24
So I think they should maybe take a look at what NVIDIA is doing with the Sparks and be a little more invested into reaching out to the community.
Federico ViticciHOST
53:38
For example, when a new computer comes out, having a more prepared package to give out to people like me or other content creators that I saw on YouTube that have the same review unit and be like, here's this thing that you can test that we know is already optimized for the M5 Ultra and we know is going to give you the kind of performance that is way above the M3 Ultra.
Federico ViticciHOST
39:45
Okay.
Federico ViticciHOST
39:48
Coming from the M3 Ultra Mac Studio, which I still have with 512 gigs of RAM, the main thing that this computer has allowed is... being able to use the best local models that fit on this machine with the kind of performance that makes them nice to use in apps.
Federico ViticciHOST
40:16
Like these local models could be used before in terms of fitting on the computer, right? And even on the M3 Ultra, I could even fit bigger models because I have the 512 gig version.
Federico ViticciHOST
40:31
But even if they technically fit in the machine, in memory and in storage, they were not necessarily nice to use because they were slow.
Federico ViticciHOST
40:43
And they were slow in two different metrics.

12 MINS LATER

Federico ViticciHOST
52:59
thing right now and I think if they're serious about this like when I when I when I mentioned like you gotta package it up in a way that makes sense they have a lot of work to do instead of just saying well just wait for you know the developer the open source community to figure it out But that's not a way to build a sustainable product over time.
Federico ViticciHOST
53:24
So I think they should maybe take a look at what NVIDIA is doing with the Sparks and be a little more invested into reaching out to the community.
Federico ViticciHOST
53:38
For example, when a new computer comes out, having a more prepared package to give out to people like me or other content creators that I saw on YouTube that have the same review unit and be like, here's this thing that you can test that we know is already optimized for the M5 Ultra and we know is going to give you the kind of performance that is way above the M3 Ultra.
Vin StoneHOST
8:32
Cool.
Jill BryantHOST
8:34
And also the Mac Studio powered by the M3 Ultra is not yet supported.
Jill BryantHOST
8:40
And so that'll probably be their next big task after they get the other M3 and M3 Pro and M3 Max support.
Jill BryantHOST
8:52
But the Asahi Linux M3 update can easily be installed in the macOS terminal.
David ImelHOST
2:05
Next, our $12,300 M5 Ultra Max Studio review is out, and as expected, it absolutely tore through our benchmark tests.
David ImelHOST
2:14
Our nearly maxed out M5 Ultra came out about 30% faster than our similarly specced M3 Ultra from last year in both single-core and multi-core synthetic CPU tests.
David ImelHOST
2:25
On a heavier GPU rendering test Cinebench, the M5 Ultra scored 63% higher.
David ImelHOST
2:30
And finally, the Wall Street Journal reported that Google's Gemini model broke containment in May and hacked three separate companies.
Federico ViticciHOST
9:52
And you're looking at this new GPU with 80 cores, and inside each core, there's a neural accelerator with a new architecture.
Federico ViticciHOST
9:59
So this is... something that apple started doing as we saw before in the earlier in the m5 generation putting a neural accelerator inside each gpu core so you got 80 of those uh in the in the m5 ultra in addition to the new gpu uh which you know on average apple says should perform like up to four times you know better in terms of compute than the m3 ultra and like nine times better or something than the m1 ultra So I think sort of the angle of my story was I've been using the Mac Studio as a local server for all kinds of things for the past year.
Federico ViticciHOST
10:37
But especially this summer, I went all in with local AI on the Mac Studio because I had this big project, the iOS 27 review.
Federico ViticciHOST
10:48
that I structured in such a way that required an agent to be running at all times, meaning every five minutes, an agent would actually fire up and do some processing.
Federico ViticciHOST
11:00
Multiple of those agents with different schedules.
Federico ViticciHOST
11:02
So effectively, it was always running something.
Federico ViticciHOST
11:05
In that case, the M3 Ultra Max Studio was running DeepSeek V4 Flash.
Federico ViticciHOST
11:10
And I structured it in a way that I had to be using a local model.
Luke MianiHOST
41:20
Oh, this was pretty crazy.
Luke MianiHOST
41:21
The M5 Ultra absolutely smoking the M3 Ultra.
Luke MianiHOST
41:28
Not surprisingly, given that the M5 Max was at 225,000 in the medal score, the M3 Ultra was at 230,000.
Luke MianiHOST
41:36
So you're already 99% of the way there with the max chip.
Luke MianiHOST
41:42
You put two of them together, yeah, 360K.
Luke MianiHOST
41:44
That's an insane jump over the M3 Ultra.
Luke MianiHOST
41:50
But where's the M6? Oh, yeah, the M6...
Luke MianiHOST
41:53
Wow, so the M6 is scoring in between an M2 and an M4 Pro.
Casey LissHOST
20:21
And then finally, the M5 Ultra, the CPU is roughly the same as the M5 Max for single core, which is to be expected, I think.
Casey LissHOST
20:31
But multi-core, it's almost 40% better than the M3 Ultra.
Casey LissHOST
20:36
So again, M3 to M5 is an almost 40% increase.
Casey LissHOST
20:40
M5 Max to M5 Ultra, a 60% increase.
Casey LissHOST
20:45
That is astonishing to me.
Casey LissHOST
20:47
So again, M3 Ultra to M5 Ultra, uh, 40% increase, but M5 Max to M5 Ultra, 60, almost 60% increase.
John SiracusaHOST
20:56
Yeah, I mean, you would, uh, it's t- it's tricky because like, hey, isn't it two M5 Maxes put together? Shouldn't it be 100% increase 'cause you double everything? And that's where you get into some of the-- obviously, there's overhead of the, you know, the, the way they do the interposer and blah, blah, blah.
John SiracusaHOST
21:08
But also, are these benchmarks like, uh, what kind of data dependencies are in these benchmarks? Like, there's a whole bunch of things that limit the exploitable parallelism in CPU stuff.
Will BettsHOST
59:53
You're living inside this house with the studio in an open space, lots of natural light, big windows, but it's still cozy and homely.
Will BettsHOST
60:03
Your computer is a souped-up Mac Studio M3 Ultra with 96 gigabytes of RAM.
Will BettsHOST
60:09
Your interface is an Apogee Symphony I/O Mark 2.
Will BettsHOST
60:13
Your DAW is Ableton 12 Suite.
Jake EllenbogenHOST
6:09
So I have like, um, like stands for like my laptop, but I don't use a laptop anymore.
Jake EllenbogenHOST
6:18
As you know, I have this giant computer, this M3 Ultra.
Jake EllenbogenHOST
6:22
And, um, I put other things on the stands.
Jake EllenbogenHOST
6:26
So like, I'll literally put like, like things to remember.

16 MINS LATER

Gary Sheffield Jr.HOST
22:29
Well, so do you, Jake.
Gary Sheffield Jr.HOST
22:30
You just have bad Wi-Fi in your area.
Gary Sheffield Jr.HOST
22:32
And then two, I'm in a basement on my computer with a MacBook Pro microphone.
Gary Sheffield Jr.HOST
22:41
So, we're grinding.
Max SmartHOST
47:44
So not just the AI engine that used to be separate.
Max SmartHOST
47:48
It's up to 4.5 times faster than with AI than, say, the M3 Ultra.
Max SmartHOST
47:55
It's 40% faster graphics than the M3 Ultra.
Max SmartHOST
48:00
It does include 32 core neural engine, and that's separate.
Max SmartHOST
48:05
So you've got the neural accelerators in the GPU, and you still have those neural engines.
Max SmartHOST
48:31
If you have $12,000, you can get the M5 Ultra Max Studio with 512 gig of unified memory combined.
Max SmartHOST
48:42
By the way, that has 1.2 terabytes per second of memory bandwidth.
Max SmartHOST
48:47
That's 50% more than the M3 Ultra.

We value your privacy

We use cookies to understand how you use our platform and to improve your experience. Click “Accept All” to consent, or “Decline non-essential” to opt out of non-essential cookies. Read our Privacy Policy.