At least those have some, if not all, actual "Pro" features like user upgradable RAM, user upgradable SSD, Kensington lock, ECC memory and/or built-in Ethernet.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
M3 Chip Generation - Discussion Megathread
- Thread starter scottrichardson
- Start date
- Sort by reaction score
One other interesting bit that I haven't seen discussed much yet - Did Apple retreat on cache in the M3 generation?
One of the M1's great strengths was the massive cache it had, relative to contemporary chips from Intel and AMD. The M2 continued this.
But now the M3 appears to have 6 cores per cluster instead of 4. Did the L2 cache size also go up by 50%?
If not, that suggests Apple rebalancing resources to take into account the drastically uneven scaling between logic and SRAM going from N5 to N3. And if they did... perhaps that might explain some of the gains in multithread scores we're seeing in GB6 (and why there's much less gain with the base M3).
Anyone know?
One of the M1's great strengths was the massive cache it had, relative to contemporary chips from Intel and AMD. The M2 continued this.
But now the M3 appears to have 6 cores per cluster instead of 4. Did the L2 cache size also go up by 50%?
If not, that suggests Apple rebalancing resources to take into account the drastically uneven scaling between logic and SRAM going from N5 to N3. And if they did... perhaps that might explain some of the gains in multithread scores we're seeing in GB6 (and why there's much less gain with the base M3).
Anyone know?
They don’t have ECC memory. As to the rest, performance and ergonomy is much more important to me than the ability to upgrade RAM. Why would I ever buy a slower and objectively crappier computer just because the RAM can be upgraded?
The “upgradable RAM” issue always bugged me.
I’ve had failed hard drives, failed gpus, failed motherboard components, failed screens, but never failed RAM.
It’s not an issue. It’s just a bunch of loud people forcing their misconstrued worldview onto others. Last time I’ve cared about upgradeable components was twenty years ago when I was a poor student with no money and was building a gaming PC from scraps. Believe it or not, I also used to hate Macs. It’s a social group phenomenon.
I feel like upgradable RAM on consumer market is dead but server/workstation is another story and margins on that are probably the biggest in the whole PC sector.
Kinda weird that Apple isnt trying to milk it, as power usage is main drive force for a lot of server wrokloads and M series CPUs are pretty good at that, maybe we'll see some sort of extrem socket in the future.
I am sure they looked at it and decided it’s not their business. It would also mean abandoning vertical integration which is kind of their A and O.
I miss the ability to upgrade the SSD personally, or at least add more storage. I can understand that upgradable RAM involves some compromises, but I'm sure Apple could make internal storage expandable if they really wanted. I don't want to carry an external SSD around.
The LPDDR5 memory modules include bit error-correction, so there is some ability to correct memory errors, but not in the usual sense of “ECC” memory where multi-bit errors can be corrected across the memory bus.
If the primary improvements in the GPU are in
- functionality (eg better atomics) and
- in the ability to run MORE different kernels simultaneously,
then microbenchmarks that do not test these will not show impressive results.
What will show these results are presumably games, but also "widespread" app benchmarks
Compare eg PugetBench Photoshop as a rough example of an app that tries to pack a lot of different functionality onto the GPU, not just one thing:
M3:
Overall Score
1101
GPU Score
100.9
General Score
127.6
Filter Score
92.6
M2:
Overall Score
720
GPU Score
72.9
General Score
83.8
Filter Score
60.1
Unfortunately gfxBench is so pathetic compared to modern GPUs (everything runs at like 400fps) that it's no longer a useful test of the sort of issue I'm talking about.
Very interesting. There is a score for an M2 Pro and the GPU score for the M3 is higher than it!
PugetBench Results
Build your own computer, workstation, mini pc, or server at Puget Custom Computers.
I agree. Of course, one would need to buy the SSDs from Apple directly, and they would likely be very pricey, but I’d still prefer it to the soldered-on SSDs we have now. There is just no technical reason for that…
I actually had failed RAM on a Windows workstation once... all the way back in 2005 or so. It took a couple days to find the bad stick as a failure would happen anywhere from minutes to hours.
Not a concern I have on a modern machine, and it's by and large something that should be evident from the factory.
They used to, yes, but it doesn’t look like any newer Precisions have Xeons in them. In fact, it doesn’t seem like Intel is making any mobile Xeons anymore.
P.S. And I just had another look and the new 16” Precisions use soldered-on LPDDR5.
I didn't think photoshop was using the GPU in more "complex" ways than any 3D game engine, even GFXBench.
The GPU score doesn't appear to scale very well for instance, the M2 Max being only marginally faster than the M2 Pro (and the M2 Ultra bringing very little improvements).
EDIT: and scores appear all over the place. For instance this result places the M2 much closer to the M3: https://benchmarks.pugetsystems.com/benchmarks/view.php?id=149916
But the Photoshop version is different (which is also the case for the comparison you posted). The results you showed was for an M2 with 8GB of RAM, while the M3 had 24GB. The result linked above is for an M2 with 16GB.
Not counting the fact that we don't know the testing conditions. The M2 scores are from fanless MacBook Airs that could be throttling.
Can we really interpret these scores?
Last edited:
True, but do we know if it’s because the benchmark doesn’t scale, or because the M2 didn’t scale? Certainly the M3 has seen a very big improvement. Depending on the M3 Max scores, we might have a clearer picture.
Maybe the philosophy guys over at Stanford need to add a No-True-Pro fallacy to their encyclopedia.
I keep repeating: The Day of the DIMM is over.
Hmmm... maybe "Day of the DIMM" would make a good movie title?
This generalization is bad, multi core tests, more than single core, are workload dependent, in the Clang subtlest for example the M2 Ultra is about 12% faster than the M3 Max with 33% more cores. Basically in line with per core performance improvements. This does not bode well for code compilation performance of the M3 Pro when compared with it’s predecessor.M3 Max supposedly beats M2 Ultra and Intel 14th gen in multicore.
M3 Max Beats Apple’s Desktop-Class M2 Ultra In Multi-Threaded Results By A Small Margin While Having Fewer CPU Cores
The M3 Max is now the fastest Apple Silicon in existence as it is able to beat the M2 Ultra in multi-core tests, albeit by a small marginwccftech.com
View attachment 2305686
Hi! I’m gonna give you my humble opinion.
Both the M3 and the M2 Pro are pretty close in performance, with the M3 outperforming the M2 Pro in single core and GPU (Metal) according to the first benchmarks.
With the M3 you also gain Ray Tracing, AV1 decoding and a few more details… but honestly, it boils down to wether you want a 24” all in 1, or a bigger 27” display (with lower pixel density) with a Mac mini on the desk.
If you can live with 24”, I’d personally go with the iMac because Apple LCD displays are gorgeous (I love glossy screens). But if you like the Dell mate 27” screen better, then the M2 Pro Mac mini is a really good device.
I will probably go with an Apple Studio Display (which is 27”) and a Mac mini in the future, but that probably falls out of your budget.
This is really interesting and valuable to me since I spend the majority of my time in Photoshop. That seems like a genuine upgrade over the M2. I'm on M2 Max 38 core so keen to see what the M3 Max gets!
Same here. Black looks like all other cheap plastic windows laptops incl. Lenovo etc.
I guess we really need to just wait until someone has these in-hand to test things like Photoshop, Games, and Blender Rendering - which is three areas I believe should give us a good overall portrayal of performance improvements.