dual amd64 cpus ???

  • Thread starter Thread starter Lynn
  • Start date Start date
Lynn said:
Does any of the ASUS motherboards support dual AMD64 cpus ?

Thanks,
Lynn
Click to expand...


Ehmm, they announced the CPUs just the last days.... Wait a little more time
and you will see. For now you can tell that all 939 boards will be able to
use the AMD Dual CORE CPUs with a bios update if the manufacturer will
programm one.

Frank
 
Do you mean two physical CPUs or a dual core processor?

I would prefer two physical cpus.

Lynn
 
Lynn said:
I would prefer two physical cpus.
Click to expand...

There should be little difference between two physical CPUs and a Dual
Core (Except availability) in many ways. Of course, a board with 2
sockets has the potential for 4 cores are 8 DIMMs...

I can't find any Dual Socket boards for AMD64, by Asus.

One of the best configurations you could have would be a couple of
nForce4 Pros with a couple of socket 939/940s. I guess that all depends
on what you want to do.

Ben
 
I would prefer two physical cpus.

Lynn
Click to expand...

So let's say you have a dual-core per CPU, dual physical CPU system.
That's four very fast CPUs!

How much memory would be needed to keep a system list that busy?

(hate to ask this but ) Does anyone really need a four CPU core
system, unless you are very heavy into video or Photoshop?

And, can Windows XP utilize all four CPU cores? Would you need to
wait until Longhorn becomes available?
 
Lady said:
So let's say you have a dual-core per CPU, dual physical CPU system.
That's four very fast CPUs!
Yeah.

How much memory would be needed to keep a system list that busy?
Click to expand...

Depends what you're doing.

But ideally, 2 DIMMS per socket. 4 DIMMS.
(hate to ask this but ) Does anyone really need a four CPU core
system, unless you are very heavy into video or Photoshop?
Click to expand...

Good for multitasking... If the operations are highly threaded, then
great, provided they don't rely on each other too much. (Although AMDs
architecture minimises the latency, even for inter-socket communication)
And, can Windows XP utilize all four CPU cores? Would you need to
wait until Longhorn becomes available?
Click to expand...

Windows NT 4 could utilise 4 CPUs ok. By the time you went to 8, given
the architectures back then, and the OS, you gained practically nothing
over 4 CPUS.

You might require a Server/Professional licence for XP though. I'm not
sure what Windows licencing is with respect to Dual Core CPUs.

Ben
 
Ben Pope said:
Depends what you're doing.

But ideally, 2 DIMMS per socket. 4 DIMMS.
Click to expand...

feast on this rack full of dual AMD-64's http://tinyurl.com/5uhd9
closeup of a (different) motherboard http://tinyurl.com/amv4t pulled from a
similar rack.
these systems all ran linux but don't tell SCO


--
=======================================================================
Beemer Biker (e-mail address removed)
http://TipsForTheComputingImpaired.com
http://ResearchRiders.org Ask about my 99'R1100RT
=======================================================================
 
Lynn said:
Found an ASUS 940 dual opteron MB http://shop.store.yahoo.com/directron/k8ndl.html
for $289

The 2.0 GHZ Opteron CPUs http://shop.store.yahoo.com/directron/opteron246.html
are $360 each

The 2.2 GHz Opteron CPUs http://shop.store.yahoo.com/directron/opteron248.html
are $529 each

Arent the Athlon AMD64 cpus suppose to be faster than the Opterons ? I think
that the Athlon already goes to 2.4 GHz ???
Click to expand...

I think that the original Opterons were 133*6 = 800MHz Hypertransport.
Athlon64s are 200FSB * 5 = 100MHz Hypertransport.

In terms of speed, yeah, the Athlon64s outperform the Opterons.

I think they are bringing the Opterons up to date though.

You can see a comparison of the Opterons here, including dual core:
http://www.amd.com/us-en/Processors/ProductInformation/0,,30_118_8796_9240,00.html?redir=CPOS14

The Opterons have more coherant Hypertransport links for multiprocessor
inter-communication.

Ben
 
Lynn said:
I would prefer two physical cpus.
Click to expand...

Why ? So far all of the benchmarks show that when you compare a
single 2.2 GHz dual-core Opteron 275 to a pair of 2.2 GHz
single-core Opterons 248, the dual-core wins *every* time.
Sometimes the margin is quite small, but sometimes it gets up to
10 or 15 percent.

Not only that, a 2.2 GHz dual-core Opty 275 uses less power than
a single-core Opty 248.
 
Why ? So far all of the benchmarks show that when you compare a single 2.2 GHz dual-core Opteron 275 to a pair of 2.2 GHz
single-core Opterons 248, the dual-core wins *every* time. Sometimes the margin is quite small, but sometimes it gets up to 10 or
15 percent.

Not only that, a 2.2 GHz dual-core Opty 275 uses less power than a single-core Opty 248.
Click to expand...

Hmmm. Do you have the URL for that performance test. I was
impressed that it was quite the other way around.

THanks,
Lynn
 
Lynn said:
Hmmm. Do you have the URL for that performance test. I was
impressed that it was quite the other way around.
Click to expand...

I guess it depends slightly on what you are doing. With 2 seperate
Opterons, you have 2 memory busses, with a dual core, you have one.

I guess if the 2 CPUs need to work on the same data, the dual core has
some advantages in terms of latency, otherwise they have some
disadvantages in terms of bandwidth.

Ben
 
Anandtech are testing 2x 252s (2 cores @2.6GHz) againt 1x 875 processor (2 cores @ 2.2GHz) and 2x 875s (4 cores @ 2.2GHz). Shame
about the clock speed differences:
Click to expand...

I sure would like to see a dual core cpu test vs. a dual cpu test of the same
cpus. It is my impression that the 875 series of the opteron have much
higher memory bandwidth than the 252 series.

Lynn
 
Ben said:
I guess it depends slightly on what you are doing.
Click to expand...

You seem to have found the TechReport benchmarks.

From those it seems clear that it does *not* depend on what you
are doing. An Opty 275 beats a single or pair of Opty 248's in
*everything* even though both the 275 and the 248 are 2.2 GHz chips.
With 2 seperate
Opterons, you have 2 memory busses, with a dual core, you have one.
Click to expand...

True enough. However the memory controller in the dual-core
Opterons is supposed to be better and faster.

As well, whether you have two single-core chips or one dual-core
chip, in order to maintain cache coherency each core has to snoop
the caches on the other core before accessing main memory - which
adds substantial latencies to the RAM accesses. In a dual-core
Opteron those additional latencies should be substantially
reduced because there is no need for one core to go off the die
to take a peek at what is in the other core's caches.

Hence, a pair of single-core Opterons should have a substantial
memory bandwidth advantage over one dual-core Opteron. However,
the daul-core Opteron will have a big latency advantage. And, as
we have seen in AMD64 vs (P4 and Xeon), reductions in latencies
almost always do more for your performance than bandwidth increases.

A search at AnandTech will get you latency numbers for the
single-core Opterons but it will probably be a while longer
before they have similar numbers for the dual-core chips.

I expect a better explanation than my stumblings to also show up
soon at places like AnandTech and Ace's Hardware. I'm still
struggling to digest all I have read over the past few days.
 
Rob said:
You seem to have found the TechReport benchmarks.
Click to expand...

After my guessing, yes, I did.
From those it seems clear that it does *not* depend on what you are
doing. An Opty 275 beats a single or pair of Opty 248's in *everything*
even though both the 275 and the 248 are 2.2 GHz chips.
Click to expand...

Except raw memory bandwidth tests (not that it actually 'counts' in real
life).
True enough. However the memory controller in the dual-core Opterons is
supposed to be better and faster.
Click to expand...

Yes, something I hadn't accounted for.
As well, whether you have two single-core chips or one dual-core chip,
in order to maintain cache coherency each core has to snoop the caches
on the other core before accessing main memory - which adds substantial
latencies to the RAM accesses. In a dual-core Opteron those additional
latencies should be substantially reduced because there is no need for
one core to go off the die to take a peek at what is in the other core's
caches.
Click to expand...

Or indeed, even over the Hypertransport link. SRI sorts that one out at
full whack. Something else I didn't know about. AMD have really
thought about their design.
Hence, a pair of single-core Opterons should have a substantial memory
bandwidth advantage over one dual-core Opteron. However, the daul-core
Opteron will have a big latency advantage. And, as we have seen in AMD64
vs (P4 and Xeon), reductions in latencies almost always do more for your
performance than bandwidth increases.
Indeed.

A search at AnandTech will get you latency numbers for the single-core
Opterons but it will probably be a while longer before they have similar
numbers for the dual-core chips.

I expect a better explanation than my stumblings to also show up soon at
places like AnandTech and Ace's Hardware. I'm still struggling to
digest all I have read over the past few days.
Click to expand...

The Techreport article is actually very well written and informative, I
felt. Will have to start going there more often.

Looks like the Athlon FX2 will probably arrive at around the £500 mark,
which will probably put me off for a while. Enough speculation, anyway.

I grabbed my Winchester a month ago and now there are Venice cores
available... the duals have the same advantages as the Venice, namely
SSE3. I like.

Ben
 
Back
Top