Originally Posted by Bageljoey
When did they start soldering in the RAM on MacBook Pros? How did I miss this this?
Is 16 GB the maximum anyway so user access is irrelevant?
They went soldered as soon as they made a Retina model. The old model used stacked RAM slots, which are too tall to fit into a Retina model. 32GB is the maximum that the CPUs support but the vast majority of tasks won't max out 16GB. It's possible to do but you'd have to be doing a pretty heavy workload.
Originally Posted by mpantone
There appears to be something wrong with your system.
My 2010 Mac mini (with 8GB RAM) running Mavericks uses 1.65GB after a boot.
Mavericks has a file cache now so it loads up more into RAM but it's not that you have less RAM available, it'll flush that extra content out when it needs to. The actual usage is the total minus the file cache and it's further complicated by compression as the total is higher than the physical amount used.
Originally Posted by mdrfitmeyer
That should be the baseline, but a BTO with a 32GB soldiered option.
They won't be able to fit 32GB in until DDR4 (likely Broadwell) as there's not enough room for the chips. Double density chips will let them use 16x 16 gigabit chips. There's even a single 128GB DDR4 module been made for high-end applications:
The next Mac Pro will move to DDR4 so it should be able to hold at least 128GB but maybe even up to 512GB. Anything above 128GB would be far too expensive though and probably won't be offered commercially.
Originally Posted by GrangerFX
A 200 MHZ upgrade at 2.8 GHZ is about 7%. What would the reaction be if the iPhone 6 is only 7% faster than the iPhone 5s?
If you consider the MBA, the entry 2008 model scored 1210 in Geekbench 32-bit at $1799. The current model starts at $899 and scores 4678. So 3.86x speedup in 7 years and the price is half. If Moore's Law were to be followed, it would be 2x every 18 months so it should be 16x speedup but it's not only the CPU that has sped up. The transistor count is being shared with the internal GPU. The X3100 in the original is slower than the GPU in the A6 chips for the iPhone.
The latest HD5000 is about 3x faster than the 9400M and the 9400M is easily double the X3100 so 6x faster GPU. The power consumption goes down too.
Moore's Law really has to take into account price because you can compare a 12-core Xeon otherwise. So the Air has 3.8x faster CPU, 6x faster GPU and 1/2 the price. This is pretty close to a 16x CPU speedup at the same price.
It would be nice to get larger speed jumps but this revision isn't a major one. The GPU manufacturers have been doing the same. The 750M was a rebadge of the 650M and even came out slower in some tests. Broadwell is supposed to drop 30% of the power for the same performance so they may boost performance a bit and lower power consumption.
Originally Posted by ascii
Hopefully the 850M will be available as BTO. It's only 3% slower than two 750Ms in SLI, a major upgrade.
It's a big upgrade but I very much doubt they'd do a BTO option. It would be disappointing if they didn't use it. Even if the consumption went up 5W, it's not the end of the world to get a 50-90% speed boost.
I think they are phasing out 3rd party GPUs because the power draw and temperatures will be much easier to lower with a single chip. A Broadwell MBP could run at 35-45W and stay much cooler than the old one.