Sidebar Sidebar
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

How long until... you can plug in external RAM/CPU/GPU's?

M

MBX

macrumors 68020
Original poster
Sep 14, 2006
2,030
817
I'm curious:

As we have seen with the MBP Air and being able to use external optical drives, how long do you think until the same is possible with RAM/CPU/GPU?

Let's say you have a MBP or MBA and you want more than 4 or 8 GB Ram, you could plugin 32GB of Ram via external module somehow via USB-like plug or something.

Same with CPU/GPU.

Do you think it will be possible?
 
simsaladimbamba said:
When there are interfaces, that can transmit 6 Gigabyte per second and much, much more.
Click to expand...

Thing is, by then, RAM, CPU and GPU speeds will have increased to the point where 6Gb/s is worth about as much (relatively) as USB is now.

MBX- basically, anything you can plug in via USB, or ESATA, or firewire, or anything else currently available as an external interface, has a transfer speed that is a teeny tiny fraction of the speed RAM, CPUs and GPUs use to shift data around.
Only slight exception being those USB decoder card things that will let you mess around with video without hogging your CPU.
 
You can already do external GPUs Apple just doesn't support it. Many other laptops have a dock or expressport that you can plug an external GPU.
 
DoghouseMike said:
Thing is, by then, RAM, CPU and GPU speeds will have increased to the point where 6Gb/s is worth about as much (relatively) as USB is now.

MBX- basically, anything you can plug in via USB, or ESATA, or firewire, or anything else currently available as an external interface, has a transfer speed that is a teeny tiny fraction of the speed RAM, CPUs and GPUs use to shift data around.
Only slight exception being those USB decoder card things that will let you mess around with video without hogging your CPU.
Click to expand...

hm im wondering, can you use a part (like 8gb) of your SSD as ram? ssd is fast, not like the HDD, so why not? i know its not as fast as ram but its nice for occasions you need more ram than what you have.
 
orfeas0 said:
hm im wondering, can you use a part (like 8gb) of your SSD as ram? ssd is fast, not like the HDD, so why not? i know its not as fast as ram but its nice for occasions you need more ram than what you have.
Click to expand...

RAM is at least 30 times faster nowadays than an SSD. And an SSD gets used for RAM, when the RAM has to swap some of its contents.
 
I can't imagine running RAM through USB or even firewire...do you have any idea how ridiculously fast RAM is? Your computer would be seriously pissed off if it tried to access RAM and it wasn't there in 0.0000000000000000000000000000000000001 seconds (a little inflated but you get the point). as for CPU oh God you would need some interesting set up.
 
I do not think in the near future any how that an external CPU or RAM will be able to do the tasks that are done internally. The speeds at which these two communicate is astounding when compared to the current technologies for attaching peripherals to the machine.
 
orfeas0 said:
hm im wondering, can you use a part (like 8gb) of your SSD as ram? ssd is fast, not like the HDD, so why not? i know its not as fast as ram but its nice for occasions you need more ram than what you have.
Click to expand...

SSD is fast, but RAM is written to and read from constantly. Meaning that would wear down those sectors really quickly meaning it would have to change sectors until the drive runs out of physical space.

Mac OS X already does this though with a page file. It isn't the same as using RAM, essentially when a RAM page is not used for a while it is recorded to the hard disk (SSD or HDD). When you need information from that page it copies it back into the physical RAM. It effectively gives you "unlimited" RAM but the cost of copying pages makes it slower even if you have a fast SSD.

It gets even more complicated with the CPU's L1 and L2 cache which are both much faster to access than RAM. So essentially the CPU tries to use memory in this order:
L1 cache
L2 cache
RAM
Swap space on the hard disk

Things that are getting used often are copied up a level in order to improve performance.
 
Never.

Before that would happen cloud computing would become mainstream and all of the processing would reside in a data center somewhere. Need more CPU/RAM? Pay an access fee and your allocated processor/RAM bandwidth is bumped. The computer you interface with would be little more than a dumb terminal.

Long way from becoming the norm but there is more money in the industry behind cloud based computing than there is finding an external i/o solution for processors and memory.
 
As mentioned there have been some experiments to try attaching external GPUs. The idea being that you could use a much faster GPU that would need better cooling than what you want to carry around. The only thing that exists are more or less low end cards or those USB adapters that just give you additional screen real estate. Over USB they compress the picture to put it out.
The external GPUs usually are connected over PCIe x1 lanes which als limits there potential quite a bit.

If anything CPU, GPU, RAM move even closer together. Now in SandyBridge the GPU moved so close to the CPU that the whole PCIe x8 or QPI is only necessary for dedicated GPUs. RAM also moved closer with integrated memory controllers. A long time ago CPU Cache (like the 6MB L3 Cache) were on a seperate chip. They are planning to put stacked DRAM on to the CPU/GPU package. I think they want to add fast parallel VRAM this way.

Thus it is the exact oposite. The only reason the ODD left the MBA is because few people ever use or only rarely and it is not worth to be carried around. An ODD needs about a 6 or 10 MB/s transfer speed to work fine. For a sick comparison the theoretical aggregate ring bus bandwidth between Sandy Bridge Core1-4-iGPU-LLC-IMC is 446464 MB/s.
Even Lightpeak will not reach nearly high enough bandwidth in time and even once it does it will still suffer from bad Latency which is actually much more important and the true reason why everything is getting more and more tightly packed. Unlike a GPU a CPU needs more than pure bandwidth it needs quick access or it starves and just consumes power without getting anything done. They praise the great latency of SSD's compared to HDD's but compared to the 55ns of main memory it is still slow for a CPU.
 
It's not just bandwidth, it's latency. Light travels about 1 foot in a nanosecond. So for every foot the signal has to travel, along the wires in the computer, out the computer, into the device, out the device, back into the computer, thru the wires to the CPU, it takes 1 nanosecond. And that's in a perfect world where buffer and driver chips have zero propagation delay, wires have zero capacitance and inductance, RAM has zero access time, and all signals travel at the speed of light in a vacuum.

So in the perfect case, with a signal path of 2 feet from the CPU to the RAM in the external device, that's 4 nanoseconds of latency (2 feet out, 2 feet back). 4 nanoseconds is 250 MHz.
 
orfeas0 said:
hm im wondering, can you use a part (like 8gb) of your SSD as ram? ssd is fast, not like the HDD, so why not? i know its not as fast as ram but its nice for occasions you need more ram than what you have.
Click to expand...


Yes it's called Virtual Memory, and your operating system does it automatically ;)
 
Just as everyone else said, plugin GPU's have been available for other computers since the 90's... I had an older Thinkpad that had a graphics card it could be docked to.
In Windows, you can use a flash drive as RAM, via "Windows Readyboost"
I don't think Mac OS has something similar, because of USB 2.0's limited access speeds, as well as the limited write speeds of USB flash drive, renders the feature all but useful in terms of actual usability.
CPU might be a long time until it comes around, because of such the high transfer speed required. 6GB/s, Maybe. 200gb/s Probably, and 1TB/s, Definitely. (IMO)
 
Possible? I would imagine it is now.

Profitable? I highly doubt it. Most people don't even know what a GPU is...selling them individually would only appeal to a very specialized market. They would also have to be cheap/effective enough to justify spending that kind of money versus a new laptop. Given you can get a quad SB laptop for under a grand, that will be hard.
 
chown33 said:
4 nanoseconds is 250 MHz.
Click to expand...
I think you have your numbers wrong. 1 seconds has 1 mil ns that means it would be 62,5 Thz if 4ns meant 250Mhz. It is more like 12khz assuming a 3Ghz CPU core.
It is still a lot considering what a cpu can do in 1 clock today and 12 thousand of them is a lot of waiting time. L2 Cache is only 11 clocks according to Intel.
 
dusk007 said:
I think you have your numbers wrong. 1 seconds has 1 mil ns
Click to expand...

No. 1 second has 1000 million ns.
One millionth of a second is 1 microsecond.
There are 1000 ns in 1 microsecond.
http://en.wikipedia.org/wiki/Nanosecond
"1.017 nanoseconds (approximately) – time taken for light to travel 1 foot."
​

http://en.wikipedia.org/wiki/Microsecond

that means it would be 62,5 Thz if 4ns meant 250Mhz. It is more like 12khz assuming a 3Ghz CPU core.
It is still a lot considering what a cpu can do in 1 clock today and 12 thousand of them is a lot of waiting time. L2 Cache is only 11 clocks according to Intel.
Click to expand...
12 khz? Really? Audio tape and vinyl records have more bandwidth than that.
 
chown33 said:
It's not just bandwidth, it's latency. Light travels about 1 foot in a nanosecond. So for every foot the signal has to travel, along the wires in the computer, out the computer, into the device, out the device, back into the computer, thru the wires to the CPU, it takes 1 nanosecond. And that's in a perfect world where buffer and driver chips have zero propagation delay, wires have zero capacitance and inductance, RAM has zero access time, and all signals travel at the speed of light in a vacuum.

So in the perfect case, with a signal path of 2 feet from the CPU to the RAM in the external device, that's 4 nanoseconds of latency (2 feet out, 2 feet back). 4 nanoseconds is 250 MHz.
Click to expand...

Nerd. :cool:
 
Asus and someone else experimented with an external GPU but at the time it just a low end built in GPU + an external mid range one so not much of an improvement then. I guess it never took off.
 
Expresscard can't do CPU/GPU but it can do PCI and actually any PCI-E card that'll fit in an enclosure.
 
mulo said:
so your telling me there are expresscard to pci adapters?
Click to expand...

Yah. Well, I meant "it can do PCI-E" but apparently one can do PCI as well.

cb1h.gif


It's expensive as heck though, but one could probably homebrew their own solution. CB1H + EC34 for 1059 bucks.
 
You must log in or register to reply here.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Top Bottom
Back