> 
> The current 40-bit physical memory address space limit of the amd64 arch
> is somewhere well into the multi-terabyte range (IDR the exact numbers and
> am too lazy to do the math or look it up, but someone else might have them
> handy), and the virtual memory isn't limited even to that, as it can use
> the full 64-bit address space, so as I said, our arch isn't practically
> affected by this limit at the moment.  By the time the 40-bit physical
> memory constraint is reached, they will have upped that likely to 48 bit,
> and it can eventually be upped to 64-bit, ...


For the Opteron, it's 1 TB.  More a limit of memory controller than physical
address.  Beyond that, 40-bits is too limiting.  There are two main reasons
Intel's Itanium hasn't died on the shelf - FPU performance, and physical
memory addressing - 48-bits.  There are several systems that have modest
processor counts - 64 cpus, but sport 4 TB of physical memory in the hands of
real users doing serious work.  While AMD lost out on this rather small,
unique niche, it is an important one.

The third limitation is Hypertransport V1.xx and V2.xx only support a maximum
of 32 sockets, and that is only when a rather expensive board design that
includes additional logic and an L3 cache to deal with coherency in a NUMA
environment.

The other issue with Hypertransport is that it's an in-box bus.  It was designed
for chip-to-chip connect, not a box-to-box.  Trying to scale the number
of cpus, PCI-X slots, and AGP/PCI-E slots up tends to require multiple
boxes - some dedicated to a specific function, which require connecting
cables.  For certain types of work, a lose network of computers won't
work,  it neeeds to be a single system - not a cluster.

And you don't want to know how much work it takes to get 16 graphics pipes
to work together in a reliable, seamless, fashion that can deal
with a 4 TB database of textures.

For reference - a 2 cpu, 2 graphics pipe, 16 GB workstation is
a pretty normal entry level workstation in certain industries.

Bob
-- 
-  
-- 
[email protected] mailing list

Reply via email to