Many people are dumping their M1 Max and M1 Ultra Mac Studio models, just look on FB Marketplace or eBay to see so many for sale currently. The price has sure dropped also. You may want to keep yours, or sell it and upgrade to a M4 Max or M3 Ultra Mac Studio model, while you can still get a decent price for yours. I have a feeling the M1 Max and M1 Ultra Mac Studio models will soon become dinosaurs and not command that much of a price.
Yes, prices are dropping and there are many good deals to be found on M1 Mac Studio models. I just picked up on Facebook Marketplace locally (Saint Louis) a M1 Max 32 Core GPU 32GB RAM Mac Studio with 1 TB SSD for $850, and I also bought a M1 Ultra 64 Core GPU 64GB RAM Mac Studio with 2TB SSD for $1,900.
They are completely correct in saying "up tp.." the prime example is watching a 12 minute YouTube cat video. The hyper-fast M7-Ultra will still take 12 minutes to do the cat video task. But it will be 15 times faster for 3D rendering. So it is "up to 15 times faster".
I purchased a Linksys 7 last year. biggest benefit was replacing 3 nodes with just 1. And it is located at the end of my house.
Perhaps you have a negative view on the framing but it's reasonable and accurate - I don't think your suspicious are warranted given the context is annotated. It might be nice to have sustained performance numbers but marketing generally focuses on peak for fairly obviously reasons - humans are obsessed with potential and that's reflected in how things are marketed.
🤦♂️
[…] It also now features up to 512GB of memory of 16GB of storage, as well as Thunderbolt 5 connectivity.
These things are not holding their prices as well as I might have expected. I would have expected the M1 Max 32/512 Mac Studio to be around US$1000 at this time, so yes $850 is a nice deal. I guess it's a testament to just how quickly the M series chips have improved in the past several years.
www.macworld.com
.. okay, I will try again being more direct. The M3 Ultra is ~36 tops, not 76 tops if the M4 series is 38 tops. I did not say that I used Geekbench 6. I said I used Geekbench AI specifically targeting the neural engines running Core ML which is what the person was asking about. I just ran it on my base model M4 mini. Nearly the same scores as the M4 Max and faster than the M3 Ultra. That is what would be expected if the M3 Ultra is using two sets of M3 neural engines. The article you linked doesn't even talk about the neural engines. I'm not saying the M3 Ultra is not faster in many tasks because it is but it is also slower at some things.
My Mac Mini M4 routinely runs over 100C with the fan manually set to 100%. Is the Mac Studio just as poorly designed?
I think the coil whine was more prominent with the max models from what I remember but I've never had one. It's a somewhat common thing when using heat pipes though. I've owned M1, M2, M3 Ultra model studios and have never heard it. I frequently hammer them for days at a time doing AI upscaling work too so there would be ample time for it to crop up if it were going to.
Right there is that. M3 Ultra GPU is clearly faster so even with the neural engines being slightly slower it is generally the faster option for AI without even considering the memory differences. There are cases though where things do run primarily on the NPU. In those cases an iPhone 16, barring any throttling, would have pretty similar performance to the M4 series. Mac Whisper, for example, uses only NPU by default. You can force Topaz Video AI to mostly use the NPU by turning on low power mode and in that case that should level out the performance between devices that have similar TOPS.
I read on another forum that while Geekbench cannot max out M3 Ultra's CPU, it's actually more reflective of most real world CPU-dependent workloads than benchmarks like Cinebench. Most real world third party software cannot max out the M3 Ultra's CPU either. Thus, there is no desire to design Geekbench in such a way as to consistently max out all CPU cores, because according to some, that would actually make it much less useful as a CPU benchmark. Also, you have the option of looking at the specific subtests in the bench. Some are more parallelized than others.
I get your skepticism. I'm not an M3 Ultra hater. Like I said, I have one. I just have a base model right now, but depending on WWDC, I may get a beefed one with more RAM. Anyway, in this case, I think the math from the benchmark indicates it's using them or at least using all of the cores pretty equally between the different machines. I'm saying this because as far as we know, the M3 Ultra is using M3 NPUs, and the M3 series with 16 cores was said to be 18 tops, so 32 of those would be 36 tops with 100% scaling. We know Apple has said the M4 NPU is 38 tops, which is more than 36 tops. If the M3 Ultra was using the M4 NPUs, at the very least, it should have matched the benchmark performance of base M4/M4 Max, which it did not. It also was not less than half of the performance, which if it was just using one bank of the NPUs (16 cores), it should have been, so that indicates it is using more than 16 of the NPU cores. Just knowing that it is using the M3 NPU cores is really all you need to know to say M3 Ultra is not 76 tops because that would be equivalent to 32 M4 NPU cores. Doubling the M4 NPU number instead of the M3 NPUs is where the mistake in the math was likely made by the original overview posted.
