[Pc_Support] Re: Nforce kernel info

[Bryan J. Smith]

On Tue, 2005-03-15 at 23:29 -0800, coreyfro at coreyfro.com wrote:
> I have been reading "nVidia nForce4 Linux Compatibility Pre-Sale
> Evaluation" thread and I wanted to thank you for the wealth of
> information.  You thread is the single best source of info regarding the
> chipset I have seen and it is easily worth publication.

I currently write for CMP Media's "Sys Admin," the #1 UNIX or Linux
magazine in print circulation.

Here is a sample of my work:  
  "Dissecting PC Server Performance"  
  http://www.samag.com/documents/sam0411b/  

It has no less than 7 diagrams.

> I think you should consider that, google bombing said publication,
> and making sure this fantastic info is up there for all to see.
> While I am complimenting you, I was wondering, in the past 2 months, have
> you had any more info regarding the nForce4.  I am strongly considering
> getting the first Opteron nForce 4 board that comes to market to work my
> way to a dualcore dualsocket monster.  I was wating for VIA (the chipset I
> usually prefer) but the K8T890 has been delayed forever.

There are a few K8T890 chipsets out now.  But the problem with both the
nForce4 and K8T890 is that they PCI I/O sucks and _all_ the I/O is on 1
CPU.  You want to distribute the I/O across 2 CPUs -- especially for
SLI.

That's where this mainboard comes in (nForce Professional 2200+2050):  
http://lists.leap-cf.org/pipermail/pc_support/2005-February/000214.html 

It has 3x the I/O of the K8T890/nForce4, 8x the PCI I/O of them.
It's similar to Figure 7 in my Sys Admin article:  
http://www.samag.com/documents/sam0411b/0411b_f7.htm  

Except the AMD8131 is on CPU1, and there is another CK8"Pro" (2050) chip
on CPU2 (in addition to the CK8"Pro", 2200, on CPU1).
You get _true_ SLI at _full_ PCI x16 _each_, plus the PCI-X channels for
high-end storage (I prefer 3Ware Escalade 8000/9000 and NetCell SR5000
series cards and 32-bit PCI is a massive bottleneck for them).

If you're going to spend the money for dual-CPUs, possibly dual-cores in
the future, spend the extra $250-300 ($450) for an nForce Pro 2200+2050
and AMD8131 mainboard like the Tyan S2895.

Also note the power supply requirements ...
http://lists.leap-cf.org/pipermail/pc_support/2005-March/000236.html  


> While on the subject, SATA II with NCQ is something I am strongly looking
> forward to.  Any idea how that is coming along under linux?

Advanced Host Controller Interface (ACHI) is the software-driven, upto
32-device, NCQ-capable SerialATA interface standard.
Here is the current status of SerialATA support in Linux:  
  http://linux.yyz.us/sata/sata-status.html  

Unfortunately it's still not "intelligent" so it may not be worth it.

I'd recommend you look at a 3Ware Escalade 8506-4 (4-channel, 64-bit
ASIC + 2MB SRAM), 9500S-4 (4-channel, 64-bit ASIC + 2MB SRAM + 128MB
SDRAM) or NetCell SR5000 (5-channel, 64-bit ASIC + 128MB SDRAM) for
$200-350 if you _really_ want performance.  I'm working on updating
NetCell's driver for 2.6.11 (just haven't had time).

> Sorry for bursting out of now where and asking these random questions. 
> You might be the only person capable and willing to answer these
> questions.

Well, there are plenty of people who _think_ they know the answers.  I
pride myself on getting down into the engineering specification sheets
and reading stuff.  I'm wrong on occasion, and I'm not too prideful when
someone points out I'm wrong -- but I do get occasionally sick of people
completely ignorant on system design trying to cross me when they don't
realize they have absolutely no clue what I'm talking about.

Because system design is a mesh of dozens of variables, and a single
"specification" _never_ tells all.  E.g., the concept of Front Side Bus
(FSB) is _dead_ (because A64/Opterons have 2-5 of them! ;-), and PCxxxx
memory throughput is typically _not_ the most important specification
anymore (latency is).


-- 
Bryan J. Smith                                  b.j.smith at ieee.org 
---------------------------------------------------------------- 
Community software is all about choice, choice of technology.
Unfortunately, too many Linux advocates port over the so-called
"choice" from the commercial software world, brand name marketing.
The result is false assumptions, failure to focus on the real
technical similarities, but loyalty to blind vendor alignments.