Re: Comparing Opteron to Xeon

From: Andrew Gideon (c172driver1_at_gideon.org)
Date: 11/11/05 

Date: Fri, 11 Nov 2005 12:32:15 -0500

Henrik Carlqvist wrote:

> AMD didn't do that adding of pipeline steps so their GHz are more
> efficeint, but they haven't reached as high frequencies.

It has been a *long* time since school (which is when I last really studied
hardware at this level {8^), so please forgive as I break down what you've
written. Is my understanding correct?

Where you write "adding of pipeline steps", you're not speaking of
processing more instructions concurrently but are speaking of the number of
steps involved in taking an instruction from start to finish? And each
step of this pipe consumes a single clock tick?

Yes, I can see that having stages in the pipe which achieve less work would
(1) permit a higher clock rate but (2) slow processing per clock tick.

[...]

> Two dual core CPUs are almost like four single-core CPUs, but there is one
> difference...
>
> Another reason that opteron based machines usually are fater than Xeon
> based machines is memory bandwidht. Usually Xeon based machines have a
> chipset where both CPUs share the same front side bus. This means that
> only one CPU is able to access RAM at a given time.
>
> Also some cheap dual opteron motherboards only have RAM connected to the
> hypertransport of one CPU, but most opteron motherboards have one memory
> bank for each CPU. This gives a lot more memory bandwidth. I have no
> experience from dual core opterons myself, but I suppose that they will
> lack some memory bandwidth performance compared to a 4-way opteron system
> where each CPU has its own bank of RAM connected to a hyptertransport.

So here's a particular motherboard that's under my consideration:

 http://www.supermicro.com/Aplus/motherboard/Opteron/8131/H8DAE.cfm

How can I tell whether this has RAM connected to the hypertransport of one
or both CPUs? This claims under "System Memory":

 Single or Dual (recommended) channel memory support

so it sounds to me like it could go either way. What makes the
determination? On the Xeon side theres:

 http://www.supermicro.com/products/motherboard/Xeon800/E7520/X6DHR-3G2.cfm

Under "memory capacity" this claims:

 Dual-channel memory bus

Would that indicate that each CPU has its own access to RAM?

Or there's the more extreme:

 http://www.supermicro.com/Aplus/motherboard/Opteron/nForce/H8QC8.cfm

Would this have four buses into memory? Under "CPU" this claims:

 8-way option has 2 HyperTransports using PCI-e, extra I/O
 or Pathscale communication device

I couldn't locate an available supermicro quad CPU system (the H8QC8 is
"coming soon"). But there's:

 http://www-03.ibm.com/servers/eserver/xseries/x366/specs.html

Apparently, this can run 4 dual core Xeon processors. But how can I
determine what it does for interaction between CPU and memory?

Thanks...

 Andrew

Relevant Pages

  • Next July 27: boot failure(hang) on x86_64 box.
    ... Freeing unused kernel memory: 1360k freed ... ACPI: PM-Timer IO Port: 0x488 ... CPU: L2 Cache: 1024K ... # AX.25 network device drivers ...
    (Linux-Kernel)
  • Re: V40z v.s. IBM hs40
    ... > superceeds IBM on this database run. ... To address your questions on the Opteron servers vs the Xeon servers. ... has a memory controller built into every CPU and ... every CPU you add also adds memory bandwidth. ...
    (comp.sys.sun.hardware)
  • Re: PowerPC or PARISC?
    ... While Intel NGMA supports 4 sockets and Opteron supports ... Let's start with Intel. ... Purely local memory accesses ... but an access by CPU A to CPU B's memory or a HT packet ...
    (comp.arch)
  • [PATCH] Document Linuxs memory barriers [try #3]
    ... The attached patch documents the Linux kernel's memory barriers. ... I've tried to get rid of the concept of memory accesses appearing on the bus; ... barring implicit enforcement by the CPU. ...
    (Linux-Kernel)
  • Oops in 2.6.28-rc9 and -rc8 -- mtrr issues / e1000e
    ... Bios 1.04beta did show correct memory sizing in dmidecode, ... I hope this is as simple as me doing something glaringly wrong in the kernel ... DMI present. ... CPU: L2 cache: 6144K ...
    (Linux-Kernel)