Before you all think you can replace RAM with SSDs: SSD has a limited number of read/write cycles before it wears out, RAM is written to much more frequently than mass storage, so unless there are orders of magnitudes of improvements in the wear of SSD, that doesn't sound like a good option, especially since it's non-trivial to replace an aging SSD from one of these super-slim devices. If you can afford to toss a $5k computer every 18-24 months, and the OS and hardware controllers are doing an excellent job at wear leveling, then you might be OK, but otherwise 16GB just isn't enough.
We might be OK once reading to SSD is fast enough and hardware is advanced enough, that code can be memory mapped and executed off the mass storage, never even hitting any RAM other than maybe CPU cache, and RAM being used strictly for data only; but I don't think we're there yet.
So when judging Apple's current offerings, it's not relevant what can be done in the future, but what can be done now, and now, 16GB isn't enough. If battery life is an issue, they could shrink available RAM during battery operations by switching off one bank, and halving it from 32GB to 16GB, with some engineering efforts; heck, if they can switch GPUs on the fly....
Before you all think you can replace RAM with SSDs: SSD has a limited number of read/write cycles before it wears out, RAM is written to much more frequently than mass storage, so unless there are orders of magnitudes of improvements in the wear of SSD, that doesn't sound like a good option, especially since it's non-trivial to replace an aging SSD from one of these super-slim devices. If you can afford to toss a $5k computer every 18-24 months, and the OS and hardware controllers are doing an excellent job at wear leveling, then you might be OK, but otherwise 16GB just isn't enough.
We might be OK once reading to SSD is fast enough and hardware is advanced enough, that code can be memory mapped and executed off the mass storage, never even hitting any RAM other than maybe CPU cache, and RAM being used strictly for data only; but I don't think we're there yet.
So when judging Apple's current offerings, it's not relevant what can be done in the future, but what can be done now, and now, 16GB isn't enough. If battery life is an issue, they could shrink available RAM during battery operations by switching off one bank, and halving it from 32GB to 16GB, with some engineering efforts; heck, if they can switch GPUs on the fly....
The battery life is not the only issue Apple has mentioned in their message. Performance is the other reason: "...wouldn't be efficient enough for a laptop..." I read this as using 32 GB of desktop RAM in a MBP notebook would introduce much more compromises than using a single module of soldered 16 GB of true notebook RAM. We cannot go into more details because available documentation from Intel and RAM producers (Samsung, Hynix..) is limited and even if such documentation exists, these must be available to OEMs only. So it appears futile to dig further. As a power user I can only state my personal preferences:
I would like 32 GB but without compromising the lightness and thinness of a MBP. I don't want to carry a bulky machine, and since a bulky machine would retain much more heat, I'd never accept CPU throttling because of the increased temperature. I tried even laptop cooler stands, they have no effect except 1-2°. My 2011 15" MBP was running at 203° F / 95° C under heavy load, my current rMBP 15" 2015 runs at max 147 ° F / 64° C without any throttling and I'd never want such a delicate thermal balance be compromised because of bigger RAM.
I wouldn't torture myself by opening several VMs at once, I would buy a 27" iMac instead with 64 GB memory, or use cloud VMs.
I would never pay extra half of a laptop for a 32 GB RAM option. I would buy a whole second laptop instead.
I thought the armchair engineers had determined that Apple's SSDs cost too much and the RAM was way too limited.
Shocking to imagine that Apple might actually HAVE A PLAN to move the state of the art forward.
This is why I don't engage in the suggestions Apple is 'doing it wrong' with specific hardware choices.
Apple always has a plan. Armchair engineers always think they know better.
I can just imagine the old muscle car guys who may have thought a 2016 performance vehicle would have an 8 liter engine.
Rather, the state of the art (outside the exotics) is turbocharging, electric drivetrains, ultralight structural materials and other outside the box design concepts that certainly defy the old-style conventions of how to build a powerful vehicle.
Sorry to trot out the tired car metaphor again.
"Never apologize for a good metaphor." --Socrates.
to leave the same comment i left 4 years ago about the then-released MBP:
now if we could just get video performance that isn't on par with windows laptops from 4 years ago......
to leave the same comment i left 4 years ago about the then-released MBP:
now if we could just get video performance that isn't on par with windows laptops from 4 years ago......
And yet, it seems to work fine for this person .....
Of course, Apple laptops have never been popular with gamers since they try to hit a balance (and battery life of about 10+ hours) instead of top of the line graphics.
Of course people like me do not need top of the line graphics -- just needs to be able to drive multiple monitors.....
But, but, but, I need more than 16GB of RAM. (whining, crying baby emoticon)
On a $3000 machine that people may hope to keep for 4-5 yrs, it's not very future proof. Especially if you use a virtual machine, like many professionals do, which effectively can cut your available RAM in half.
Which is why it's nice to know that, for all practical purposes, there will be no difference between RAM and hard storage.
That's what this is about. You can whine about needing more RAM all you want, but if you have storage that's as fast AS RAM (or close to it), forget about your piddling 32 GB; you can buy your laptop with what amounts to TWO TERABYTES of memory.
Sit on that.
How do you figure? SSDs are not close to the speed of ram, and if a page resides on the disk, the OS needs to move it to ram first anyway, assuming a lack of resulting pageouts. I realize the discussion centers around 32 vs 16 GB, but where did you get the idea that one is almost as fast as the other?
But, but, but, I need more than 16GB of RAM. (whining, crying baby emoticon)
On a $3000 machine that people may hope to keep for 4-5 yrs, it's not very future proof. Especially if you use a virtual machine, like many professionals do, which effectively can cut your available RAM in half.
Which is why it's nice to know that, for all practical purposes, there will be no difference between RAM and hard storage.
That's what this is about. You can whine about needing more RAM all you want, but if you have storage that's as fast AS RAM (or close to it), forget about your piddling 32 GB; you can buy your laptop with what amounts to TWO TERABYTES of memory.
Sit on that.
How do you figure? SSDs are not close to the speed of ram, and if a page resides on the disk, the OS needs to move it to ram first anyway, assuming a lack of resulting pageouts. I realize the discussion centers around 32 vs 16 GB, but where did you get the idea that one is almost as fast as the other?
They are still quite far apart but they are close enough that swapping shouldn't cause the kind of problems seen on old computers. Theoretical memory bandwidth calculation is here:
Clock speed x transfers per clock x bus width x number of interfaces. For the Mac Pro, which has the highest Mac memory bandwidth, this is: 1866MHz x 2 (DDR memory) x 64-bit x 2 interfaces / 8 (to get Bytes) = 59.7GB/s
SSD performance in the latest machines is around 2-3GB/s. This is 1/5th-1/10th the speed of modern RAM, it's a little faster than DDR1 memory but it's not having to swap in and out all the time. GPUs do memory swapping too. Iris Pro uses a small fast memory cache and swaps data with the system memory:
Dedicated GPUs have memory that can be 10x the speed of system memory. As long as the swap can happen fast enough, it doesn't matter. In 1/60th of a frame, you could swap about 256MB of GPU data without noticing.
Say a video app like Da Vinci or After Effects has 4k 10-bit frames in memory that have been modified, they'd be 3840x2160 x 40-bit = 41MB per frame. At 30FPS, this is 1.2GB/s sustained bandwidth. Storage has never been this fast before so it would get held in memory. With SSD, it's easily fast enough to use frames from the disk in real-time.
Even if you had a large memory allocation like 8GB, it would write to an SSD in 4 seconds. On old machines with HDDs, this would take 2.5 minutes. This is what used to cause the beachball hangs because if you have to wait that long for the machine to continue, it's effectively locked up. Not so with a 4 second swap time and Apple can even compress the 8GB before writing it.
Really high-end VFX jobs that are reading 3D data use lots of memory:
"Because of the intensely interconnected nature of the light, everything had to be processed at once, which resulted in incredible RAM demands. Describing one scene, Hery says, "We had the full campus. We had the trees, the grass, the people, the crowds, all held in memory at once. At some point, the rendering engine can try to be clever" about what doesn’t need to be rendered, "but you can only hide so much."
That's a lot to keep in RAM. The biggest struggle was to keep each frame down to 20GB of memory. Pixar’s render farm uses machines with 96GB of RAM each, but since the machines need to process four frames in parallel, anything much higher than 20GB was dangerous. That might seem like a lot for a single frame, but it has to hold every object in the scene, on-and-off camera, right down to the hairs of each monster’s fur. As long as it’s reflecting light, it needs to be in memory. And if any of those hairs move, the light has to be completely recalibrated — usually in another overnight turn through the render farm."
Here, there are some sys admins for VFX studios talking about running out of 128GB:
"even with 128GB RAM + 128GB swap we still get OOM'ed (out of memory)"
They say this happens due to a mistake by the artists. You can max out any resources with poor software or workflows.
This memory discussion also concerns laptops, those are servers. DDR4 memory will mostly solve the problem for people doing memory intensive operations on a laptop and Intel's XPoint/Optane DIMMs would be a further solution for any computer running out of memory. The DIMMs will run at around 6GB/s per channel.
Any idea of the difference in speed between the memory the MacBook Pro has now and the speed of the "slower" memory they would have to use if they had a different chipset with 32GB?
I cannot find it now but several articles mentioned that the 32GB option would not work under the current low power memory and it would require an additional 3 watts of energy as a minimum -- in addition to thermal. It also mentioned that it would run slower. I don't know if it is the chipset support or whether faster memory options would require even more power to support ....
As far as I can tell the memory in the last release of the MacBook Pro laptops runs at 2,333MHz.
What I think the argument about RAM needs to be, is not what Apple should have done to THESE machines in regards to RAM, or whatever. The real argument to be made here is why Apple isn't interested in customers with the highest performance needs.
we see this with the delayed, or doscontinued Mac Pro, monitors, and now, routers.
if they wanted to, they could have come out with another 4.5 pound Macbook Pro that has all of what some want. Upgradable RAM, upgradable SSDs, etc. but they're not interested. It's really that simple. It's lighter, and longer battery life. That's the mantra. I don't know what would convince them differently.
Before you all think you can replace RAM with SSDs: SSD has a limited number of read/write cycles before it wears out, RAM is written to much more frequently than mass storage, so unless there are orders of magnitudes of improvements in the wear of SSD, that doesn't sound like a good option,
Comments
SSD has a limited number of read/write cycles before it wears out, RAM is written to much more frequently than mass storage, so unless there are orders of magnitudes of improvements in the wear of SSD, that doesn't sound like a good option, especially since it's non-trivial to replace an aging SSD from one of these super-slim devices.
If you can afford to toss a $5k computer every 18-24 months, and the OS and hardware controllers are doing an excellent job at wear leveling, then you might be OK, but otherwise 16GB just isn't enough.
We might be OK once reading to SSD is fast enough and hardware is advanced enough, that code can be memory mapped and executed off the mass storage, never even hitting any RAM other than maybe CPU cache, and RAM being used strictly for data only; but I don't think we're there yet.
So when judging Apple's current offerings, it's not relevant what can be done in the future, but what can be done now, and now, 16GB isn't enough.
If battery life is an issue, they could shrink available RAM during battery operations by switching off one bank, and halving it from 32GB to 16GB, with some engineering efforts; heck, if they can switch GPUs on the fly....
http://www.zdnet.com/article/why-macbook-pros-dont-need-32gb-of-ram
Of course, Apple laptops have never been popular with gamers since they try to hit a balance (and battery life of about 10+ hours) instead of top of the line graphics.
Of course people like me do not need top of the line graphics -- just needs to be able to drive multiple monitors.....
https://en.wikipedia.org/wiki/Memory_bandwidth
Clock speed x transfers per clock x bus width x number of interfaces.
For the Mac Pro, which has the highest Mac memory bandwidth, this is:
1866MHz x 2 (DDR memory) x 64-bit x 2 interfaces / 8 (to get Bytes) = 59.7GB/s
The Mac Pro memory might be capable of more (3 interfaces?) but the CPU is limited to 59.7GB/s:
http://ark.intel.com/products/75283/Intel-Xeon-Processor-E5-2697-v2-30M-Cache-2_70-GHz
The new one is 76.8GB/s:
http://ark.intel.com/products/91755/Intel-Xeon-Processor-E5-2697-v4-45M-Cache-2_30-GHz
The laptop CPUs are 25.6GB/s (newer ones are 34.1GB/s).
These are theoretical maximum numbers, not actual performance. Geekbench measures real-world memory write speeds. The last MBP is here:
https://browser.geekbench.com/v4/cpu/924914
Scroll to the bottom and memory copy is 14.3GB/s. Mac Pro is here with 19.4GB/s:
https://browser.geekbench.com/v4/cpu/987544
Here's a few older models:
https://macperformanceguide.com/blog/2012/20120621_5-MacBookPro-Retina-memory-bandwidth.html
SSD performance in the latest machines is around 2-3GB/s. This is 1/5th-1/10th the speed of modern RAM, it's a little faster than DDR1 memory but it's not having to swap in and out all the time. GPUs do memory swapping too. Iris Pro uses a small fast memory cache and swaps data with the system memory:
http://www.anandtech.com/show/6993/intel-iris-pro-5200-graphics-review-core-i74950hq-tested/3
Dedicated GPUs have memory that can be 10x the speed of system memory. As long as the swap can happen fast enough, it doesn't matter. In 1/60th of a frame, you could swap about 256MB of GPU data without noticing.
Say a video app like Da Vinci or After Effects has 4k 10-bit frames in memory that have been modified, they'd be 3840x2160 x 40-bit = 41MB per frame. At 30FPS, this is 1.2GB/s sustained bandwidth. Storage has never been this fast before so it would get held in memory. With SSD, it's easily fast enough to use frames from the disk in real-time.
Even if you had a large memory allocation like 8GB, it would write to an SSD in 4 seconds. On old machines with HDDs, this would take 2.5 minutes. This is what used to cause the beachball hangs because if you have to wait that long for the machine to continue, it's effectively locked up. Not so with a 4 second swap time and Apple can even compress the 8GB before writing it.
Really high-end VFX jobs that are reading 3D data use lots of memory:
http://www.theverge.com/2013/6/21/4446606/how-pixar-changed-the-way-light-works-for-monsters-university
"Because of the intensely interconnected nature of the light, everything had to be processed at once, which resulted in incredible RAM demands. Describing one scene, Hery says, "We had the full campus. We had the trees, the grass, the people, the crowds, all held in memory at once. At some point, the rendering engine can try to be clever" about what doesn’t need to be rendered, "but you can only hide so much."
That's a lot to keep in RAM. The biggest struggle was to keep each frame down to 20GB of memory. Pixar’s render farm uses machines with 96GB of RAM each, but since the machines need to process four frames in parallel, anything much higher than 20GB was dangerous. That might seem like a lot for a single frame, but it has to hold every object in the scene, on-and-off camera, right down to the hairs of each monster’s fur. As long as it’s reflecting light, it needs to be in memory. And if any of those hairs move, the light has to be completely recalibrated — usually in another overnight turn through the render farm."
Here, there are some sys admins for VFX studios talking about running out of 128GB:
http://www.studiosysadmins.com/board/threadview/5761/
http://www.imdb.com/name/nm2085517/
"even with 128GB RAM + 128GB swap we still get OOM'ed (out of memory)"
They say this happens due to a mistake by the artists. You can max out any resources with poor software or workflows.
This memory discussion also concerns laptops, those are servers. DDR4 memory will mostly solve the problem for people doing memory intensive operations on a laptop and Intel's XPoint/Optane DIMMs would be a further solution for any computer running out of memory. The DIMMs will run at around 6GB/s per channel.
As far as I can tell the memory in the last release of the MacBook Pro laptops runs at 2,333MHz.
2133 MHz PC3-17000 LPDDR3 SDRAM
we see this with the delayed, or doscontinued Mac Pro, monitors, and now, routers.
if they wanted to, they could have come out with another 4.5 pound Macbook Pro that has all of what some want. Upgradable RAM, upgradable SSDs, etc. but they're not interested. It's really that simple. It's lighter, and longer battery life. That's the mantra. I don't know what would convince them differently.
i claim 1000X endurance on top of the speed access (latency)... I'll wait and see the test data including temperature range and reliability (and error rate) before take a shot it down for no reason...