I may have stated it in a way that gave a misconception.
this is more of sales against technical.
it would be assumed that when a computer says it has 2 gigs of memory
for a 64 bit process that would be for the 64 bit addressing space.
however marketing wants to make it look bigger so there reference to 2
gigs is 32 bit wide memory modules.
so 2 gigs of 32 bit wide memory modules, equals 1 gigbyte of 64 bit
addressing.

Porter Woodward sent the following on 3/29/2007 12:00 PM:
> My understanding as far as 64 bit Java is concerned is that the JVM is
> available in 64 bit versions for Solaris, Linux, and Windows - typically
> for x64 (64 bit x86 processor) architecture.  My home PC runs Windows XP
> 64 bit edition, and the 64 bit JVM with no problems.  OFBiz even fires
> up and runs fine on it.  I would believe that the 64 bit JVM would have
> a few advantages over the 64 bit version - not in terms of memory
> consumption (and the assertion that a 64 bit system will use double the
> memory is incorrect*) - but in terms of actual instructions needed to
> accomplish work.
> 
> Internally Java has used pretty wide (bitwise) representations of
> primitive data types.  For example a Java "int" is 32 bits on all JVMs. 
> At the time of it's introduction that was pretty radical as typically a
> 'C' style "int" was typically 16 bits wide (depending on the vendor of
> your compiler).  So you can imagine the "book-keeping" the JVM might
> have had to do on earlier operating systems and architectures - needing
> to use two 16 bit memory locations to store the value of a 32 bit
> integer, and also needing to keep track of which two 16 bit values went
> together, etc.  The Java "long" type is 64 bits wide.  Up until recently
> this has meant that the JVM has had to provide "emulation" in a manner
> similar to that described for 32 bit values on older systems for 64 bit
> values.  No longer.  A 64 bit JVM should be using native operations
> capable of 64 bit wide operands to handle 64 bit values.
> 
> * as to why 64 bit processors don't suddenly see 1/2 the memory... (a) a
> byte is defined as 8 bits.
> (b) a kilobyte is 1000 bytes.  (actually 1024)
> (b) a megabyte is 1000 KB. (actually 1024)
> 
> So the sizing of memory values is external to the bit-width of
> operations and operands on a CPU.
> 
> Additionally because x86 processors - even the x64 variants are still
> "CISC" style chips, they have variable bit length instructions.  An ADD
> instruction might be 4 bits wide (and still work on 64 bit operands),
> whereas an SSE instruction might be 24 or 32 bits wide.  Hence x86 CISC
> chips have to do more work to "decode" instructions as they come to the
> CPU...  RISC was viewed as an improvement as all operations, and
> operands were fixed width.  So a 32 bit RISC chip all instructions were
> 32 bits wide, and worked on 32 bit wide operands.  Hence binary sizes
> for RISC chips were typically larger than the equivalent CISC binary -
> which might be the source of misapprehension concerning memory usage
> when increasing bit width of a CPU.
> 
> 
> BJ Freeman wrote:
>> the is slightly off topic:
>> when you increase the bits of a processor, the available ram is
>> decreased.
>> so 2 gig on a 32 bit is only usable 1 gig on 64 bit.
>> so unless the code in javaVM at the machine level is taking advantage of
>> the 64 bit instruction set, you are not gaining anything.
>> What usually happens it the complied runtime is still using the 32 bit
>> instruction set, on a 64 bit cpu. until they go in and optimize the
>> machine code generated.
>>
>>
>> Walter Vaughan sent the following on 3/29/2007 8:00 AM:
>>  
>>> Is anyone working with ofBiz in 64 bit mode? I'm looking at
>>> configuring/leasing a server (dual quad core box), and to get the javaVM
>>> to use more than 2Gig of memory from what I've read, I need to run a 64
>>> bit JavaVM.
>>>
>>> I know this probably is way overkill, but the cost is reasonable right
>>> now for these  servers, and we've got time to 'speriment, and if we have
>>> to,we can still run this stuff in 32bit mode.
>>>
>>> -- 
>>> Walter
>>>
>>>
>>>
>>>     
>>
>>   
> 
> 
> 

Reply via email to