Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
16GB RAM Xserve Graphic?
- Thread starter MacRumors
- Start date
- Sort by reaction score
Um....
I'm a CS major and from what I have learned in my Computer Engineering classes, a 64 bit processor can only handle memory addresses for up to 8 gigabytes of ram, and a 32 bit can only go up to 4 gigs of ram...
I don't know where all this terrabyte stuff came from but I think its a dream...
As for 16 gigs, it is remotely possible in a machine with two 64 bit processors, although it would be a pain to implement and slow(since the processors DO have to communicate*), even if it were extremely expensive...
*=I mean think about it, if there are 2 processors, it has to output any computing it does to cache, and then from cache to main memory. If it did not output it to main memory (RAM) then, since each processor generally has its own cache, after one processor did something with some data, if the other processor was doing something with the data the other processor just used, then it would be using old data or something else entirely and crash...as for 16 gigs, it is possible that each processor would address 8 gigs of RAM and they communicate, but then if one processor needs data thats in the other main memory, it would have to ask the other processor if its in its memory and halt all processes going on at the moment, causing chaos throughout, the other way is to not have them communicate with eachother but thats just like having 2 machines in one case sharing data ports, powersupply, etc...
I'm a CS major and from what I have learned in my Computer Engineering classes, a 64 bit processor can only handle memory addresses for up to 8 gigabytes of ram, and a 32 bit can only go up to 4 gigs of ram...
I don't know where all this terrabyte stuff came from but I think its a dream...
As for 16 gigs, it is remotely possible in a machine with two 64 bit processors, although it would be a pain to implement and slow(since the processors DO have to communicate*), even if it were extremely expensive...
*=I mean think about it, if there are 2 processors, it has to output any computing it does to cache, and then from cache to main memory. If it did not output it to main memory (RAM) then, since each processor generally has its own cache, after one processor did something with some data, if the other processor was doing something with the data the other processor just used, then it would be using old data or something else entirely and crash...as for 16 gigs, it is possible that each processor would address 8 gigs of RAM and they communicate, but then if one processor needs data thats in the other main memory, it would have to ask the other processor if its in its memory and halt all processes going on at the moment, causing chaos throughout, the other way is to not have them communicate with eachother but thats just like having 2 machines in one case sharing data ports, powersupply, etc...
no
I would ask your comp eng teacher to reexplain.
64 can address 2^64 bytes of ram
32 can address 2^32 etc
this translates into roughly 4GB for 32 bit systems, If the hardware supports it.
64 bit systems can handle *much* more than 8GB. 16 Exabytes, as mentioned previously. 2 processors does not change that, as Apple has had dual processor 32 bit G4s and they are still limited to 4GB.
-Peter Thorson
Zaphoyd Studios
http://www.zaphoyd.com
I would ask your comp eng teacher to reexplain.
64 can address 2^64 bytes of ram
32 can address 2^32 etc
this translates into roughly 4GB for 32 bit systems, If the hardware supports it.
64 bit systems can handle *much* more than 8GB. 16 Exabytes, as mentioned previously. 2 processors does not change that, as Apple has had dual processor 32 bit G4s and they are still limited to 4GB.
-Peter Thorson
Zaphoyd Studios
http://www.zaphoyd.com
Wouldn't that be googleplexbyte RAM chips?
There will come a point in time when all this RAM won't make a darn bit of difference. When you start talking about PBs and EBs of RAM, I don't think any data on this Earth as we know it could fill up that much! Any math whiz out there care to prove/disprove that? I certainly don't have the knowledge myself... I just like saying yottabytes and zetabytes.
16 GB is plausible
Yoman said:
Yoman,
There are server applications that can utitlize quite a lot of ram. A midrange server from Sun Microsystems can hold up to 192 GB of ram. Keep in mind that Xserves are probably gearned more towards enterprises.
link here
moo083, zaphoyd is correct. In a 32 bit Operating system, you have 32 bits of address space. 2^32 = 4GB
2^33 = 8GB, so you can imagine the address space with 2^64. It grows exponentially.
Yoman said:
Yoman,
There are server applications that can utitlize quite a lot of ram. A midrange server from Sun Microsystems can hold up to 192 GB of ram. Keep in mind that Xserves are probably gearned more towards enterprises.
link here
moo083, zaphoyd is correct. In a 32 bit Operating system, you have 32 bits of address space. 2^32 = 4GB
2^33 = 8GB, so you can imagine the address space with 2^64. It grows exponentially.
Re: Um....
Would you mind pointing out which engineering school you are going to. Those of us with freinds looking for schools or those of us with kids would really like to know.
The reason being is what you describe below is wrong. As others have stated, 64 bit machines imply 2^64 addressable memeory locations. The 970 does not implement all of that address range, but that address range is far greater than 8 gigabytes. A particular machine may have addressing limits far below the hardware addressing range of the CPU driving the computer. Such address range limitations are often hardware issues outside the CPU.
Thanks
Dave
Would you mind pointing out which engineering school you are going to. Those of us with freinds looking for schools or those of us with kids would really like to know.
The reason being is what you describe below is wrong. As others have stated, 64 bit machines imply 2^64 addressable memeory locations. The 970 does not implement all of that address range, but that address range is far greater than 8 gigabytes. A particular machine may have addressing limits far below the hardware addressing range of the CPU driving the computer. Such address range limitations are often hardware issues outside the CPU.
Thanks
Dave
Iam currently testing the 2GB PC3200 ECC Registered Crucial chips in my TYAN board so far so good at 16GB. The PC2100 4GB is a completly different story as the chips are high density (1Gbit/64Mbyte parts) and stacked and only run at 266MHz vs 400MHz and also the $7000 cost. Most likely you would be using these with Itaniums and 800 series Opterons.
Clarifications about addressing
a 64 bit processor can address 2^64 bits.. not bytes.. effectively 2^61 bytes. There are ways around this as 32 can only address 512 megs... so the roofs have some leway.. yet in a pure sense, the correct amount is 2^61 bytes, which translates into 2.3058e+18 bytes. by my pen that looks like 18,096,912 MB.
a 64 bit processor can address 2^64 bits.. not bytes.. effectively 2^61 bytes. There are ways around this as 32 can only address 512 megs... so the roofs have some leway.. yet in a pure sense, the correct amount is 2^61 bytes, which translates into 2.3058e+18 bytes. by my pen that looks like 18,096,912 MB.
I'm sorry that's wrong, too. The way processors work is by addressing chunks of bits, it so happens that both the Power architecture and the x86 architecture smallest chunks are 8 bits = 1 byte. That's why a 64 bit processor in theory could address 2^64 bytes, not bits.pjo898 said:
This is also the reason why a boolean value (yes/no, true/false, 1/0) declared in any programming language on these architectures take up 1 byte of memory, not just one bit.
gekko513 said:
I don't think you are right about this. Whilst it is true that a CPU will tend you read groups of bits (actually the width of it's data bus) and will deal with those groups as bytes it stores then as a set of bits. It can address each bit individually so the amount of memory space is equal to 2^(width of address bus). Note that memory space != total maximum RAM as some devices are likely to be mapped into the memory space which reduces the amount of RAM you can have (without nasty tricks).
gekko513 said:
OK, maybe my honors degree in Artificial Intelligence and Computer Science doesn't count for much. Read this: Intro to 64 bit computing Now lets talk.
So, I read it, and it confirmed my statement: "it's fairly common knowledge that a 32-bit processor can address at most 4GB of memory. (Remember our 232 = 4.3 billion number? That 4.3 billion bytes is about 4GB.) A 64-bit architecture could theoretically, by contrast, address up to 18 million terabytes."robbieduncan said:
If the cpus were addressing individual bits, then a 32-bit processor would only be able to address 2^32 bits = 4Gb = 512MB. We all know this is wrong.
(Oh, and you're not the only one with a Computer Science degree. And having a degree doesn't prevent any of us from being misinformed.)
Hemingray said:Wouldn't that be googleplexbyte RAM chips?
There will come a point in time when all this RAM won't make a darn bit of difference. When you start talking about PBs and EBs of RAM, I don't think any data on this Earth as we know it could fill up that much! Any math whiz out there care to prove/disprove that? I certainly don't have the knowledge myself... I just like saying yottabytes and zetabytes.
thats what they said about 4GB of ram back when 32bit was new but a googleplexabyte is a bit overkill.
let me put it like this you work out how long it would take you to write it out taking 1 second per every didgit?
(googleplex = (10^10)^100 I think)
In all due honesty, 16 GBytes of RAM is quite tame. The company I work for has an IBM SP/2 with well over a terabyte of RAM. Granted, it's a massively multiprocessor machine where RAM is split between processors (and not shared as in an SMP box like the G5 XServes), but you get the idea: some tasks require astonishing amounts of memory. Databases and 3D rendering being two of these tasks, and typical bioinformatics applications (which AltiVec is surprisingly adept at) being a third.
Contrary to what somebody misguidedly stated, 16 GBytes is well within the addressable memory space of the 64-bit G5. The PowerPC is a word-aligned architecture (as opposed to byte-aligned architecure such as x86); words presumably being either 32- or 64-bit (4 and 8 bytes, respectively). Whether or not the G5 is limited to memory on the order of TBytes or not, it could theoretically address 2^64 bytes as correctly stated. This is far in excess of 16 GBytes.
As a mark of amusement, I'd like to note that here in China and Hong Kong 2 GByte memory modules are starting to appear on the market, albeit at a high premium. Nonetheless one particular performance freak I know has already loaded four such modules into his dual Opteron box, thus achieving the key 16 GBytes of RAM figure. I seem to remember him having to recompile his Linux kernel to accomodate this figure, but as far as I know all is now good and well. Myself, I have no need for such copious amounts of memory.
So there are certainly no barriers to putting 16 GBytes of RAM into an XServe G5, nor, for that matter, a PowerMac G5. Since these are SMP (Symmetric MultiProcessing, a form of "shared memory architecture") calculations of the form "each processor can address X amount of RAM, and there are two processors, so total amount of RAM can be 2*X" are flawed. I'd also like to note that whereas the G5 PowerMac takes "regular RAM", the XServe does indeed take registered (error-correcting) memory. This is, if I remember correctly, the main reason for which VirginaTech has replaced its PowerMacs with XServes. Thus availability of 2 GByte RAM modules to the "registered" variety would not preclude their installation in a G5 XServe.
Contrary to what somebody misguidedly stated, 16 GBytes is well within the addressable memory space of the 64-bit G5. The PowerPC is a word-aligned architecture (as opposed to byte-aligned architecure such as x86); words presumably being either 32- or 64-bit (4 and 8 bytes, respectively). Whether or not the G5 is limited to memory on the order of TBytes or not, it could theoretically address 2^64 bytes as correctly stated. This is far in excess of 16 GBytes.
As a mark of amusement, I'd like to note that here in China and Hong Kong 2 GByte memory modules are starting to appear on the market, albeit at a high premium. Nonetheless one particular performance freak I know has already loaded four such modules into his dual Opteron box, thus achieving the key 16 GBytes of RAM figure. I seem to remember him having to recompile his Linux kernel to accomodate this figure, but as far as I know all is now good and well. Myself, I have no need for such copious amounts of memory.
So there are certainly no barriers to putting 16 GBytes of RAM into an XServe G5, nor, for that matter, a PowerMac G5. Since these are SMP (Symmetric MultiProcessing, a form of "shared memory architecture") calculations of the form "each processor can address X amount of RAM, and there are two processors, so total amount of RAM can be 2*X" are flawed. I'd also like to note that whereas the G5 PowerMac takes "regular RAM", the XServe does indeed take registered (error-correcting) memory. This is, if I remember correctly, the main reason for which VirginaTech has replaced its PowerMacs with XServes. Thus availability of 2 GByte RAM modules to the "registered" variety would not preclude their installation in a G5 XServe.