[Info-vax] OpenVMS I64 V8.1 "Evaluation Release"?

[Johnny Billquist]

On 2012-03-21 01.25, Michael Kraemer wrote:
> John Wallace schrieb:
>
>> 64bit application design does not require >4GB of RAM, though more RAM
>> may often be helpful for performance.
>
> For a high-end chip like the Alpha (and the other RISCs too),
> it's the *only* way to get performance for large problem sizes,
> isn't it?

More physical memory always helps. Don't matter if you have a 16-bit, 
32-bit or 64-bit architecture.

>> But even with less than 4GB of
>> memory, it's often easier for some apps to be designed to assume a
>> 64bit virtual address
>
> such as?

Lots of times, when you have programs that have threads, dynamically 
loaded libraries, and so on. Each thread needs its own stack, which you 
want to be able to grow when needed. So, how and where do you allocate 
the stack for each thread? Possibly (probably) at large distance from 
each other so that there is room to grow when needed. And with such 
sparsely allocated memory, you quickly run out of your 32-bit address 
space, even if you are not actually using much memory.

>> and let a decent OS worry about mapping virtual
>> memory to physical memory.
>
> That's what 32bit systems do too.
> But what happens when you do a calloc( 5GB ) on a 64bit system which
> has only 256MB real RAM?

Same thing as on a 32-bit system. Why do you assume it would be any 
different?

>> "If one really desperately needed >4GB, one could just as well have
>> chosen one of the mature 64bit RISC platforms."
>>
>> Such as? With what OS? If you really wanted flat 64bit addressing, for
>> a long time Tru64 was the answer (even if it wasn't called that at the
>> time).
>
> By the time Itanics became at least marginally usable (around 2001),
> all RISC platforms had 64bit for quite some time, both in hardware
> and OS. HP and IBM turned 64bit around 1997/98, i.e. long before
> Merced even existed in silicon. SGI's IRIX went 64bit around 1994/95,
> and their hardware had 64bit CPUs even before Alpha.
> But early 64bit features are more are less moot,
> since there was no appropriate hardware to support them,
> i.e. enough (and affordable) RAM. For example, in 1993 DEC
> offered us bread-and-butter Alpha's like the AXP 3000
> with 256MB and 2GB disk(s). That's far from 64bit
> and not any different from what the 32bit competition had back then
> (and not faster nor cheaper btw).

You totally miss the point with regards to "appropriate". It don't 
matter how much/little physical ram you have. The point is that with a 
larger virtual address space, you can do some things in a better/easier 
way than with a smaller. Physical ram have *nothing* to do with this.

But yes, Itanic is a loosing proposition from many points of views.

> So if one wanted/needed 64bit, it wasn't necessary to wait
> for IA64, nor was Tru64 the only horse in town.

Right.

	Johnny