|
Home > Archive > alt.os.linux > October 2002 > Maximum Number of CPUs
You are viewing an archived Text-only version of the thread.
To view this thread in it's original format and/or if you want to reply to
this thread please [click here]
| Author |
Maximum Number of CPUs
|
|
| cadilacowboy 2002-10-03, 1:25 pm |
| Can anyone find documentation regarding the maximum number of CPUs the 2.4.x
kernels will support? I need to know the max numbers of both physical CPUs,
and logical CPUs (hyperthreading).
| |
| D. Stussy 2002-10-03, 4:25 pm |
| On Thu, 3 Oct 2002, cadilacowboy wrote:
>Can anyone find documentation regarding the maximum number of CPUs the 2.4.x
>kernels will support? I need to know the max numbers of both physical CPUs,
>and logical CPUs (hyperthreading).
I think I've heard that it is 16. Why not scan the kernel source and see if it
has a programmed maximum.
The most that I have heard of so far on a single motherboard is 4. If there are
motherboards with more, I haven't seen or heard of them.
| |
|
| cadilacowboy wrote:
> Can anyone find documentation regarding the maximum number of CPUs the
> 2.4.x
> kernels will support? I need to know the max numbers of both physical
> CPUs, and logical CPUs (hyperthreading).
Hey Cadilac Cowboy:
Few days ago I saw a linux news article about a supercomputer being set
up at some national Lab with 1024 CPU's run by Linux.
Don't quote me on it though.
| |
| Joachim Feise 2002-10-03, 10:25 pm |
| Seyed wrote:
> cadilacowboy wrote:
>
>
>>Can anyone find documentation regarding the maximum number of CPUs the
>>2.4.x
>>kernels will support? I need to know the max numbers of both physical
>>CPUs, and logical CPUs (hyperthreading).
>
>
> Hey Cadilac Cowboy:
>
> Few days ago I saw a linux news article about a supercomputer being set
> up at some national Lab with 1024 CPU's run by Linux.
That's probably a Beowulf cluster and has nothing to do with the number
of CPUs in *one* machine.
As usual, Google to the rescue:
According to <http://yara.ecn.purdue.edu/~pplinux...to-2.html#ss2.1>,
"SMP Linux supports most Intel MPS version 1.1 or 1.4 compliant machines with up to sixteen 486DX,
Pentium, Pentium MMX, Pentium Pro, or Pentium II processors."
-Joe
| |
| Juergen Pfann 2002-10-04, 2:25 am |
| Joachim Feise wrote:
>
> As usual, Google to the rescue:
> According to <http://yara.ecn.purdue.edu/~pplinux...to-2.html#ss2.1>,
> "SMP Linux supports most Intel MPS version 1.1 or 1.4 compliant machines with up to sixteen 486DX,
> Pentium, Pentium MMX, Pentium Pro, or Pentium II processors."
>
Where did the author get this number from?
From the observation, there simply has been no IA32 machine
with more than 16 (or even = 16, AFAIK) CPUs yet?
This would be a cvery weak argument...
By a quick glance / grep in the kernel source (of 2.2.20,
I admit, because it was lying around anyway...), I get
the following impressions instead:
The symbol MAX_CPUS doesn't seem to be defined in IA32 (but PPC).
There is, however, 'max_cpus', apparently serving as boot
parameter to "switch off" CPU(s) if needed.
Peeping into arch/i386/kernel/smp.c, I don't find any explicit
limit number, except the (arch-independent) definition of "NR_CPUS"
in include/linux/tasks.h:
-----8<----
#ifdef __SMP__
#define NR_CPUS 32 /* Max processors that can be running in SMP */
#else
#define NR_CPUS 1
#endif
-----8<-----
OK, I'm no kernel-hacker, so maybe I just didn't happen to
find any hard-coded limit for IA32 (or other architectures).
But I tend to believe, there simply isn't one, except the
definition above - which should be easily altered, isn't it?
Juergen
| |
| Joachim Feise 2002-10-04, 10:25 am |
| Joachim Feise wrote:
> Seyed wrote:
>
>> cadilacowboy wrote:
>>
>>
>>> Can anyone find documentation regarding the maximum number of CPUs the
>>> 2.4.x
>>> kernels will support? I need to know the max numbers of both physical
>>> CPUs, and logical CPUs (hyperthreading).
>>
>>
>>
>> Hey Cadilac Cowboy:
>>
>> Few days ago I saw a linux news article about a supercomputer being set
>> up at some national Lab with 1024 CPU's run by Linux.
>
>
> That's probably a Beowulf cluster and has nothing to do with the number
> of CPUs in *one* machine.
> As usual, Google to the rescue:
> According to
> <http://yara.ecn.purdue.edu/~pplinux...to-2.html#ss2.1>,
> "SMP Linux supports most Intel MPS version 1.1 or 1.4 compliant machines
> with up to sixteen 486DX, Pentium, Pentium MMX, Pentium Pro, or Pentium
> II processors."
I have to correct myself.
The book "Understanding the Linux Kernel" says that the maximum number of
CPUs is defined by NR_CPUS, which in 2.4.19 is defined as 32.
| |
| D. Stussy 2002-10-04, 4:25 pm |
| On Fri, 4 Oct 2002, Juergen Pfann wrote:
>Joachim Feise wrote:
>> As usual, Google to the rescue:
>> According to <http://yara.ecn.purdue.edu/~pplinux...to-2.html#ss2.1>,
>> "SMP Linux supports most Intel MPS version 1.1 or 1.4 compliant machines with up to sixteen 486DX,
>> Pentium, Pentium MMX, Pentium Pro, or Pentium II processors."
>
>Where did the author get this number from?
>From the observation, there simply has been no IA32 machine
>with more than 16 (or even = 16, AFAIK) CPUs yet?
>This would be a cvery weak argument...
>
>By a quick glance / grep in the kernel source (of 2.2.20,
>I admit, because it was lying around anyway...), I get
>the following impressions instead:
>The symbol MAX_CPUS doesn't seem to be defined in IA32 (but PPC).
>There is, however, 'max_cpus', apparently serving as boot
>parameter to "switch off" CPU(s) if needed.
>Peeping into arch/i386/kernel/smp.c, I don't find any explicit
>limit number, except the (arch-independent) definition of "NR_CPUS"
>in include/linux/tasks.h:
>
>-----8<----
>#ifdef __SMP__
>#define NR_CPUS 32 /* Max processors that can be running in SMP */
>#else
>#define NR_CPUS 1
>#endif
>-----8<-----
>
>OK, I'm no kernel-hacker, so maybe I just didn't happen to
>find any hard-coded limit for IA32 (or other architectures).
>But I tend to believe, there simply isn't one, except the
>definition above - which should be easily altered, isn't it?
I believe what it comes down to is how many "processor" slots (as in the
kernel's control memory structure) is the maximum it will allocate.
Two and four processor motherboards do occur now. Beyond that is stretching it.
A 1024 processor board (32x32 square) would be quite large, and if laid out
planar, would be 8'x8' (assumes each CPU takes 3"x3") not counting support
circuitry, memory, etc.... 1024 doesn't have an integral cubic root, but 512
does ( = 8 ). 16 CPU's x 16 CPU's x 4 boards makes more sense for heat
dissipation. However, whoever dreamed up that system must be out of their mind.
|
|
|
|
|