All of lore.kernel.org
 help / color / mirror / Atom feed
* Dual Core vs. Dual CPU
@ 2006-04-23 18:22 Urs Thuermann
  2006-04-24  0:37 ` Robert M. Hyatt
  2006-04-24  4:07 ` Bill Davidsen
  0 siblings, 2 replies; 5+ messages in thread
From: Urs Thuermann @ 2006-04-23 18:22 UTC (permalink / raw)
  To: linux-smp

I'd like to know what is the difference between a system with a dual
core CPU compared to two (single core) CPUs.  I know dual core means
two CPUs on one chip, but is it really two complete CPUs or do they
share some component which is used by both CPU cores?

For example HT duplicates only those parts of CPU which hold the state
of the program in execution (i.e. program counter, registers, etc.)
but they share the execution logic which often leads to performance
loss. 

Is there some performance difference between dual core and two real
CPUs, too?  I assume the dual cores on one chip must share the bus
interface, but since in a multi CPU system only one CPU can access the
bus at a time, I think this doesn't make a performance difference,
right?

Can anybody give some clarifiactions on these issue?

urs

^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: Dual Core vs. Dual CPU
  2006-04-23 18:22 Dual Core vs. Dual CPU Urs Thuermann
@ 2006-04-24  0:37 ` Robert M. Hyatt
  2006-04-24  4:07 ` Bill Davidsen
  1 sibling, 0 replies; 5+ messages in thread
From: Robert M. Hyatt @ 2006-04-24  0:37 UTC (permalink / raw)
  To: Urs Thuermann; +Cc: linux-smp


On Sun, 23 Apr 2006, Urs Thuermann wrote:

> I'd like to know what is the difference between a system with a dual
> core CPU compared to two (single core) CPUs.  I know dual core means
> two CPUs on one chip, but is it really two complete CPUs or do they
> share some component which is used by both CPU cores?

In the case of AMD, the processors, L1 and L2 cache are replicated for a 
dual-core chip.  But the hypertransport to get off-chip is not 
replicated, nor is the memory controller.  So there could be a 
bottleneck on the dual-core that two single-cores don't have.

But then there is price.  The dual-core system will be cheaper than the 
dual cpu system...




>
> For example HT duplicates only those parts of CPU which hold the state
> of the program in execution (i.e. program counter, registers, etc.)
> but they share the execution logic which often leads to performance
> loss.
>
> Is there some performance difference between dual core and two real
> CPUs, too?  I assume the dual cores on one chip must share the bus
> interface, but since in a multi CPU system only one CPU can access the
> bus at a time, I think this doesn't make a performance difference,
> right?
>
> Can anybody give some clarifiactions on these issue?
>
> urs
> -
> To unsubscribe from this list: send the line "unsubscribe linux-smp" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>

^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: Dual Core vs. Dual CPU
  2006-04-23 18:22 Dual Core vs. Dual CPU Urs Thuermann
  2006-04-24  0:37 ` Robert M. Hyatt
@ 2006-04-24  4:07 ` Bill Davidsen
  2006-04-26 19:58   ` Urs Thuermann
  1 sibling, 1 reply; 5+ messages in thread
From: Bill Davidsen @ 2006-04-24  4:07 UTC (permalink / raw)
  To: Urs Thuermann; +Cc: linux-smp

Urs Thuermann wrote:

>I'd like to know what is the difference between a system with a dual
>core CPU compared to two (single core) CPUs.  I know dual core means
>two CPUs on one chip, but is it really two complete CPUs or do they
>share some component which is used by both CPU cores?
>
>For example HT duplicates only those parts of CPU which hold the state
>of the program in execution (i.e. program counter, registers, etc.)
>but they share the execution logic which often leads to performance
>loss. 
>
>Is there some performance difference between dual core and two real
>CPUs, too?  I assume the dual cores on one chip must share the bus
>interface, but since in a multi CPU system only one CPU can access the
>bus at a time, I think this doesn't make a performance difference,
>right?
>
>Can anybody give some clarifiactions on these issue?
>
The bad part is that the cores share a bus connection, the good part is 
that it appears that some IPC can be done on chip.

RELATED: has anyone done a measurements on the 955 chip, which is dual 
core and HT? On heavily threaded or parallel tasks I find 20-30% drop in 
clock time for HT processors, but I wonder if the CPU runs out of memory 
bandwidth. My standard test is a kernel compile.

-- 
bill davidsen <davidsen@tmr.com>
  CTO TMR Associates, Inc
  Doing interesting things with small computers since 1979


^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: Dual Core vs. Dual CPU
  2006-04-24  4:07 ` Bill Davidsen
@ 2006-04-26 19:58   ` Urs Thuermann
  2006-04-26 20:47     ` Robert M. Hyatt
  0 siblings, 1 reply; 5+ messages in thread
From: Urs Thuermann @ 2006-04-26 19:58 UTC (permalink / raw)
  To: linux-smp

Bill Davidsen <davidsen@tmr.com> writes:

> The bad part is that the cores share a bus connection, the good part
> is that it appears that some IPC can be done on chip.

Why is that bad?  Is that different from a dual CPU system?  I thought
in a dual CPU system, too, both CPUs share the bus and only one CPU
can use the bus at a time (as long we're not talking about NUMA, of
course).

urs

^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: Dual Core vs. Dual CPU
  2006-04-26 19:58   ` Urs Thuermann
@ 2006-04-26 20:47     ` Robert M. Hyatt
  0 siblings, 0 replies; 5+ messages in thread
From: Robert M. Hyatt @ 2006-04-26 20:47 UTC (permalink / raw)
  To: Urs Thuermann; +Cc: linux-smp



Depends on the architecture.  In an AMD opteron-based dual, each cpu has 
a local memory controller and local memory, connected through the 
hypertransport to the other processor through its hypertransport 
interface.  Both processors can access their own local memory with no 
interference whatsoever.  On a dual core, there is a single memory 
controller, shared by two processors, leading to a potential bottleneck. 
There is also a single hypertransport to get to other processors on 
other chips, and that can also be a bottleneck.

But all in all, dual-cores are dynomite for performance, but one needs 
to be careful to avoid the memory bottleneck...



Robert M. Hyatt, Ph.D.          Computer and Information Sciences
hyatt@uab.edu                   University of Alabama at Birmingham
(205) 934-2213                  136A Campbell Hall
(205) 934-5473 FAX              Birmingham, AL 35294-1170

On Wed, 26 Apr 2006, Urs Thuermann wrote:

> Bill Davidsen <davidsen@tmr.com> writes:
>
>> The bad part is that the cores share a bus connection, the good part
>> is that it appears that some IPC can be done on chip.
>
> Why is that bad?  Is that different from a dual CPU system?  I thought
> in a dual CPU system, too, both CPUs share the bus and only one CPU
> can use the bus at a time (as long we're not talking about NUMA, of
> course).
>
> urs
> -
> To unsubscribe from this list: send the line "unsubscribe linux-smp" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>

^ permalink raw reply	[flat|nested] 5+ messages in thread

end of thread, other threads:[~2006-04-26 20:47 UTC | newest]

Thread overview: 5+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2006-04-23 18:22 Dual Core vs. Dual CPU Urs Thuermann
2006-04-24  0:37 ` Robert M. Hyatt
2006-04-24  4:07 ` Bill Davidsen
2006-04-26 19:58   ` Urs Thuermann
2006-04-26 20:47     ` Robert M. Hyatt

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.