Re: Athlon64 is a true 64 bits processor?

From: Anton Ertl (anton_at_mips.complang.tuwien.ac.at)
Date: 08/31/04


Date: Tue, 31 Aug 2004 13:04:15 GMT

General Schvantzkoph <schvantzkoph@yahoo.com> writes:
>On Mon, 30 Aug 2004 23:40:20 +0100, VP wrote:
>
>> I read here
>> http://www.geek.com/news/geeknews/2002Dec/bch20021218017827.htm
>>
>> ......
>> The K8's use 48bit memory addressing (4TB or 1TB, I forget).

256TB. The K8 has 64-bit registers, 48 bits virtual address space,
and 40 bits physical address space.

>> I believe AMD
>> will add full 64bit memory
>> addressing to the K9 series, and hopefully at that point they can increase
>> the memory bitpath from 128 to
>> 256bit, as the K8 currently only deals with 128.

My K8 (Socket 754 Athlon 64) has a 64-bit memory interface, and I am
happy with that.

>> Don't think that's a big
>> factor? Look at the GFFX5900 and
>> it's predecessor.

The K8 is not a GPU, but a CPU. Caches work for CPUs (for most
applications). Compare the results for the socket 754 Newcastle with
the Socket 939 Newcastle; not a big difference.

>> This is true ? Athlon64(K8) isn't a real 64bit processor ?

It's just as real as any other 64-bit CPU (various Alpha, SPARCv9,
IA-64, MIPS-IV, HPPA-2.0 and PPC64 chips), none of which do 64-bit
virtual and 64-bit physical addressing (supposedly Power5 has 64-bit
virtual addresses, though). They all check the upper bits of the
addresses, so making the virtual addresses bigger (up to 64 bits)
won't be a problem. That's why they are called 64-bit architectures.

>Don't confuse physical addressing with virtual addressing. The AMD64 is a
>true 64 bit architecture which means that it computes 64 bit virtual
>addresses.

Sure it does. But the MMU can only translate addresses where the
upper 16 bit are all the same as bit 47 (everything else gives an
error). And Linux with its current three-level page table limit can
only deal with 41 bits or so.

- anton

-- 
M. Anton Ertl                    Some things have to be seen to be believed
anton@mips.complang.tuwien.ac.at Most things have to be believed to be seen
http://www.complang.tuwien.ac.at/anton/home.html


Relevant Pages

  • BIND 9.4-ESVb1 is now available.
    ... BIND 9.4-ESVb1 is a extended release version beta for BIND 9.4. ... API and glibc hides parts of the IPv6 Advanced Socket ... WARNING: API CHANGE: over memory callback ... If allow-recursion is not set in named.conf then ...
    (comp.protocols.dns.bind)
  • Re: AMD : Phenom vs Opteron ; or then Intel ?
    ... I am going to buy a new database server which will have to work ... support registered memory and Opterons do. ... dual socket motherboards on the desktop, ... server motherboard, try this Tyan datasheet. ...
    (alt.comp.hardware.pc-homebuilt)
  • Re: Optiplex GX300 Memory
    ... sockets, and the second socket seems to have a dummy module. ... with unused memory sockets will have them. ... P3 RDRAM systems with the 820 chipset can handle two different ... mismatched RAMBUS modules. ...
    (alt.sys.pc-clone.dell)
  • Re: Garbage Collection - Strange Results.
    ... I have created a private member for each socket ... that has a Buffer for reading and a ObjectBuffer for completing. ... it goes anywhere between 0 to 2000 pinned objects on each update. ... >> When I look in Windows Performance Monitor and under .NET Memory ...
    (microsoft.public.dotnet.framework.clr)
  • Re: Any comparisons of Opteron F and Woodcrest yet?
    ... memory capacity. ... What do L, SMP and ECC mean? ... Socket F and AM2 are very similar, as are Conroe and Woodcrest. ... Each Woodcrest/Conroe and Socket F/AM2 processor contains 2 cores. ...
    (comp.arch)