The memory dilemma
Remember that I said that I was happy to have purchased this M5 MBA with 1TB of storage instead of the base 512GB, but I wasn’t so sure about the 32GB of RAM?
And that if I had to choose now, I wouldn’t have doubts regarding this storage? Every new LLM model or media file that I download shows me how quickly the gigabytes shrink. A 14GB model, a 4GB movie file, a TV show that takes 23GB, maybe a couple of games that are 30 to 40GB each… yeah, 1TB was the right call.
However, I saw how the memory pressure stayed low, or how my machine was usually taking only 16GB out of the 32GB. That’s what led me to think that I should have bought the 24GB model and save 250€…
But that was with “normal” usage. Local LLM experimentation is a whole different thing. Don’t get me wrong, if it was now I’d still have doubts between 24 and 32GB of RAM, but I’m seeing how beneficial this amount of memory is for toying (and learning) with local inference.
Sure, most useful models in this range (for up to 32GB of unified RAM) run on a machine with 24GB of memory. But the context might become a bit tight sooner. And not only that, having 32GB of RAM allows me to choose a 32B Q4 model over a 24B Q4 model. Or a less quantized 24B model. And bigger context window. And more room for the system to breath. Or even room for a second model that does voice transcription. Or run Siri AI, who knows!
So, the conclusion is that, thanks to having bought before the price rise, I now have a 32GB of RAM machine that costed me less than what a MacBook Air with 24GB of RAM would cost me now. And that I can make use of that extra RAM. Even more if I manually allocate more RAM for the GPU.
The SoC dilemma
Honestly, I’m going to be tempted by the M6 MacBook Air next year. But it’s almost 8 months!!! 8 months of more wait, without knowing if the price is going up again (the RAM/NAND shortages are serious) if the memory upgrade price is going to be more expensive for the M6 machine… maybe they start charging 300-350€ for each extra step in RAM upgrade! And that, along with the already higher price of the MBA (with the base 512/16GB going from 1200€ to 1400€) would put that machine way too far from what I’m willing to pay.
Sure, the M6 MacBook Air is going to be tempting: a fresher SoC with two more GPU cores (from 10 to 12), and a much bigger memory bandwidth (from 153GB/s to ~200GB/s), but I think it is wiser to wait to see how the things be develop and only when Apple releases it, only when I see how it is priced, only then I may try to sell this M5 MBA minimizing the losses, and get the M6 model.
And what about the M7? Yes, it is said that the M7 is going to be quite a bigger upgrade compared to the M6. And that might be true. But if the M7 comes with a redesigned MBA with an OLED display, then I’m out, honestly. I don’t want OLED screens in my life, that’s why I bought the last LCD IPS iPhone, the last LCD IPS iPad Pro, and potentially, the last LCD IPS MacBook Air.
If by the time the M6 MacBook Air the rumors regarding the M7 MBA are more solid, I’ll make the decision then.
But until then, as I said, there’s almost a year ahead and I needed a new Mac already. I’ve been waiting for too long.
Remember that I said that I was happy to have purchased this M5 MBA with 1TB of storage instead of the base 512GB, but I wasn’t so sure about the 32GB of RAM?
And that if I had to choose now, I wouldn’t have doubts regarding this storage? Every new LLM model or media file that I download shows me how quickly the gigabytes shrink. A 14GB model, a 4GB movie file, a TV show that takes 23GB, maybe a couple of games that are 30 to 40GB each… yeah, 1TB was the right call.
However, I saw how the memory pressure stayed low, or how my machine was usually taking only 16GB out of the 32GB. That’s what led me to think that I should have bought the 24GB model and save 250€…
But that was with “normal” usage. Local LLM experimentation is a whole different thing. Don’t get me wrong, if it was now I’d still have doubts between 24 and 32GB of RAM, but I’m seeing how beneficial this amount of memory is for toying (and learning) with local inference.
Sure, most useful models in this range (for up to 32GB of unified RAM) run on a machine with 24GB of memory. But the context might become a bit tight sooner. And not only that, having 32GB of RAM allows me to choose a 32B Q4 model over a 24B Q4 model. Or a less quantized 24B model. And bigger context window. And more room for the system to breath. Or even room for a second model that does voice transcription. Or run Siri AI, who knows!
So, the conclusion is that, thanks to having bought before the price rise, I now have a 32GB of RAM machine that costed me less than what a MacBook Air with 24GB of RAM would cost me now. And that I can make use of that extra RAM. Even more if I manually allocate more RAM for the GPU.
The SoC dilemma
Honestly, I’m going to be tempted by the M6 MacBook Air next year. But it’s almost 8 months!!! 8 months of more wait, without knowing if the price is going up again (the RAM/NAND shortages are serious) if the memory upgrade price is going to be more expensive for the M6 machine… maybe they start charging 300-350€ for each extra step in RAM upgrade! And that, along with the already higher price of the MBA (with the base 512/16GB going from 1200€ to 1400€) would put that machine way too far from what I’m willing to pay.
Sure, the M6 MacBook Air is going to be tempting: a fresher SoC with two more GPU cores (from 10 to 12), and a much bigger memory bandwidth (from 153GB/s to ~200GB/s), but I think it is wiser to wait to see how the things be develop and only when Apple releases it, only when I see how it is priced, only then I may try to sell this M5 MBA minimizing the losses, and get the M6 model.
And what about the M7? Yes, it is said that the M7 is going to be quite a bigger upgrade compared to the M6. And that might be true. But if the M7 comes with a redesigned MBA with an OLED display, then I’m out, honestly. I don’t want OLED screens in my life, that’s why I bought the last LCD IPS iPhone, the last LCD IPS iPad Pro, and potentially, the last LCD IPS MacBook Air.
If by the time the M6 MacBook Air the rumors regarding the M7 MBA are more solid, I’ll make the decision then.
But until then, as I said, there’s almost a year ahead and I needed a new Mac already. I’ve been waiting for too long.