On Wednesday 28 January 2004 02:42 pm, John Richard Smith wrote:
> Tom Brinkman wrote:
> >If your MSI board is Revision 2.0C, it's AMD approved. What
> >PSU and DDR Ram are you usin?  If you have lm_sensors running,
> >are the voltages at or a little over spec, and are they
> > steady, specially under load (ie, with mprime, 17 running) ?
>
> Yep

    Again, what ram?  What power supply?   If either is decent, 
they'll have a label on them with manufacture and specs.

> >BTW, you probly won't be assured you're ready for Trancode and
> >Mencoder till you can run 'burnK7'
>
> what's burn K7 ?

     The correct Cpuburn module for your CPU.  The rpm is on your 
CD's.  I'm usin    cpuburn-1.4-5mdk

> >for at least 10 or 15 minutes and stay properly cooled.... at
> > 133x11.5.
>
> OK so I've spent best part of the day bobbing back and forth
> into bios and back to mprime .
> I've varied the memory settings from the least to the most
> demanding, according to the mobo manual.I manually reset the
> cpu and memory voltages as suggested.
>
> I don't actually note any significant variance .

     You won't. On average you'll get about 7% better performance 
with the tightest ram timings. BUT only in ram intensive 
proccesses like encoding. Better to do with 93% efficiency 
successfully, no?

> These are the most demanding settings used.
>
> CHIPSET FEATURES SETUP
> Configure SDRAM by user          : user

                                     SPD

> SDRAM frequency                       : HCLK
> Cas latency                                    : 2

                                                   3

> Row precharge time                     : 3T

      no disable?

> Ras pulse width                            : 5T
> RAS to CAS delay                        : 2T

                                             3 

> Bank interleave                            : 4 - way

      You should definetly have an option to disable this

> Burst length                                  : 8QW

    This is best disabled in any event. It provides practically no 
benefit at any setting, 8QW could definitely cause problems

> SDRAM 1T Command                 : Enable  = 1T
> Fast command                              : enable

   disable both

> AGP mode                                    :auto
> AGP  Comp driving                      : auto
> Manual AGP Comp driving         : [not applicable]
> AGP Fast write                             : enable
> AGP aperture size                        : 128MB ( var between
> 8/256MB)(S/B 64?)

    A safe setting is 4mb. It effectively disables sidebanding. 
About AGP, for the gain (not much) over PCI, it also imposes 
extra load on cpu/cache/ram and motherboard/PSU.

> AGP Master 1 w/s write               : enable
> AGP Master1 w/s read                 : enable
> Write to read delay                       : enable
> read to write delay
>   (manual says enable/dis)           :  0   (choice 0/1/2/3)
> AGP read synchronisation           : enabled

                 disable all these AGP options

> Looks to me like I have some pretty useless memory if stressing
> it between most and least demanding setting appears to have so
> little effect. Unless there can be something else having an
> overall effect on performance.

     Just tell me what it is.  PCxxx (should be at least 2100, I'd 
use the next label up tho). CAS and ns rating, brand would be 
more helpful.  Hint: Samsung chips on a generic PCB = generic ram
The PCB (board the chips are on) is as important as the chips 
used.  When you find name brand chips on a generic PCB, it's most 
likely the chips were rejects from the first line.

    As to ram, performance and reliability has 4 major factors. 
The ram itself, the motherboard, IO/voltage to the ram, and the 
PSU.  One weak link and you've got instability, errors.

> I have geforce3 video card

    If you're usin the nVidia driver, remove it. Then you'll also 
need to reinstall your kernel....
    'rpm -ivh --force your_current_kernel_version.rpm'
...in order to remove the nVidia's kernel taints the kernel warned 
you about during the nVidia driver install.

> The "cans look normal"
> I honestly don't think there is a heat problem with either cpu
> or memory.

   I suspect either is the problem.  BTW, your problem is not 
transcode, it's needin to underclock to boot and run Linux.

> I think I had better start investigating what memory
> manufacture best suits my
> mobo.

   Usually the mobo manufacturer makes recommends and non- 
recommends. Otherwise use high quality reputable DDR PCxxxx one 
or more label rating above the required (PC2100 for your board).  
Lower CAS and ns spec's are much more important than the PCxxxx 
label.  Your board requires a minimum of 1000/133.3 = 7.5ns ram.
Cas3 is OK, lower is better. Probly can't use the lower cas 
timing, but it goes along way towards insuring ram quality.

> Here is sensors output,
> sensors
> lm84-i2c-0-18
> Adapter: SMBus Via Pro adapter at 0400
> Algorithm: Non-I2C SMBus adapter
> Board:       +15�C  (min =   -1�C, max =  -49�C)ALARM (LOW)
> CPU:          +0�C  (min =  +28�C, max =   -1�C)

    Both of those are ridiculous enough as to make all the rest 
very suspect.  Your Board (usually gathered at or near the 
chipset) should be in the 30C to 50C range, CPU 40 to 60C range.
That's nominal ranges, of course lower is better.  But I can 
assure you your CPU is not cold enough to freeze water (0C).

    Boot a warm machine and look in bios for the outputs. Expect 
CPU temp to rise about 5C to 10C from what you saw in bios after 
the OS is up and running, low load.  With an Athlon then add 
another 10C to 20C to approximate actual core temp.

     Don't attempt cpuburn till you get accurate realtime CPU temp 
reporting, at 11.5x133
-- 
      Tom Brinkman                 Corpus Christi, Texas

Want to buy your Pack or Services from MandrakeSoft? 
Go to http://www.mandrakestore.com

Reply via email to