Athlon/K7-Opimisation problems

Athlon/K7-Opimisation problems

[Carsten Leonhardt]

Hi,

here's my data regarding the K7 optimisation problem.

Until last week I had a 1GHz Athlon with 133MHz FSB. I then bought a
1.4GHz Athlon, again 133MHz FSB.

I never had any problems with the 1GHz processor, but as soon as I
stuck the 1.4GHz processor in, the kernel oopsed itself to
oblivion. (That was with kernel 2.4.5-ac5, approximately)

I also noticed that the computer booted ok with the Debian
bootfloppies, which use a kernel compiled for i386. So after several
kernel compile/boot cycles, I found out that I had to disable K7
optimisation to make the system boot again.

The mainboard is a Tyan Trinity KT-A (S2390B) with a VIA KT133A
chipset.

After reading here that it may be the PSU, I upgraded my 300W PSU to a
431W Enermax, which made no difference.

The only difference I can make out between the working and the
non-working CPU is the internal clockspeed of the CPU and the
stepping (old: 2, new: 4).

Appended is the output of /proc/cpuinfo (I sold the old CPU, but the
guy who bought it sent me the output).

I really hope this helps someone. If anyone needs any more data, or
wants me to test something, contact me.


Yours,
 leo



Old CPU:

processor       : 0
vendor_id       : AuthenticAMD
cpu family      : 6
model           : 4
model name      : AMD Athlon(tm) Processor
stepping        : 2
cpu MHz         : 1012.530
cache size      : 256 KB
fdiv_bug        : no
hlt_bug         : no
sep_bug         : no
f00f_bug        : no
coma_bug        : no
fpu             : yes
fpu_exception   : yes
cpuid level     : 1
wp              : yes
flags           : fpu vme de pse tsc msr 6 mce cx8 sep mtrr pge 14 cmov
pat 17 psn mmxext mmx fxsr 3dnowext 3dnow
bogomips        : 2018.51

New CPU:

processor       : 0
vendor_id       : AuthenticAMD
cpu family      : 6
model           : 4
model name      : AMD Athlon(tm) Processor
stepping        : 4
cpu MHz         : 1400.098
cache size      : 256 KB
fdiv_bug        : no
hlt_bug         : no
f00f_bug        : no
coma_bug        : no
fpu             : yes
fpu_exception   : yes
cpuid level     : 1
wp              : yes
flags           : fpu vme de pse tsc msr pae mce cx8 sep mtrr pge mca cmov pat pse36 mmx fxsr syscall mmxext 3dnowext 3dnow
bogomips        : 2791.83

Re: Athlon/K7-Opimisation problems

[Nicholas Knight]

On Sunday 09 September 2001 10:07 am, Carsten Leonhardt wrote:
> Hi,
>
> here's my data regarding the K7 optimisation problem.
>
> Until last week I had a 1GHz Athlon with 133MHz FSB. I then bought a
> 1.4GHz Athlon, again 133MHz FSB.
>
> I never had any problems with the 1GHz processor, but as soon as I
> stuck the 1.4GHz processor in, the kernel oopsed itself to
> oblivion. (That was with kernel 2.4.5-ac5, approximately)

This is common enough it's becoming absurd.

>
> I also noticed that the computer booted ok with the Debian
> bootfloppies, which use a kernel compiled for i386. So after several
> kernel compile/boot cycles, I found out that I had to disable K7
> optimisation to make the system boot again.
>
> The mainboard is a Tyan Trinity KT-A (S2390B) with a VIA KT133A
> chipset.
>
> After reading here that it may be the PSU, I upgraded my 300W PSU to a
> 431W Enermax, which made no difference.
>
> The only difference I can make out between the working and the
> non-working CPU is the internal clockspeed of the CPU and the
> stepping (old: 2, new: 4).

Heat anyone? stepping 4 hasn't seemed to be the problem, at least not 
directly
What's the tempature difference between your 1Ghz and 1.4Ghz? both CPU 
and system? What kind of cooling do you have?

Re: Athlon/K7-Opimisation problems

[J. Dow]

From: "Nicholas Knight" <tegeran@home.com>

> > The only difference I can make out between the working and the
> > non-working CPU is the internal clockspeed of the CPU and the
> > stepping (old: 2, new: 4).
> 
> Heat anyone? stepping 4 hasn't seemed to be the problem, at least not 
> directly
> What's the tempature difference between your 1Ghz and 1.4Ghz? both CPU 
> and system? What kind of cooling do you have?

What about the power supply. If it is at all marginal the power consumption
boost going to 1.4G is likely a killer.

{^_^}    Joanne Dow, jdow@earthlink.net

Re: Athlon/K7-Opimisation problems

[Nicholas Knight]

On Sunday 09 September 2001 03:23 pm, J. Dow wrote:
> From: "Nicholas Knight" <tegeran@home.com>
>
> > > The only difference I can make out between the working and the
> > > non-working CPU is the internal clockspeed of the CPU and the
> > > stepping (old: 2, new: 4).
> >
> > Heat anyone? stepping 4 hasn't seemed to be the problem, at least not
> > directly
> > What's the tempature difference between your 1Ghz and 1.4Ghz? both
> > CPU and system? What kind of cooling do you have?
>
> What about the power supply. If it is at all marginal the power
> consumption boost going to 1.4G is likely a killer.

Well, he didn't mention the amperage outputs, but he said 431W Enermax, 
from what I hear Enermax PSU's are good.
I still have trouble dealing with the idea that the optimizations cause 
power consumption like this, but then, I have trouble with my own idea 
that it causes sufficient heat increase in the chipset that soon after 
boot.

Do most people that experience this problem also experience after a 
cold-boot where the system had been off for at least 10-15 minutes? And 
has ANYONE sucsesfully cured this problem by changing power supplies?

Re: Athlon/K7-Opimisation problems

[J. Dow]

From: "Nicholas Knight" <tegeran@home.com>

> > What about the power supply. If it is at all marginal the power
> > consumption boost going to 1.4G is likely a killer.
> 
> Well, he didn't mention the amperage outputs, but he said 431W Enermax, 
> from what I hear Enermax PSU's are good.
> I still have trouble dealing with the idea that the optimizations cause 
> power consumption like this, but then, I have trouble with my own idea 
> that it causes sufficient heat increase in the chipset that soon after 
> boot.
> 
> Do most people that experience this problem also experience after a 
> cold-boot where the system had been off for at least 10-15 minutes? And 
> has ANYONE sucsesfully cured this problem by changing power supplies?

Don't forget that there are two regulators involved. First there is the
primary power supply's regulator down to either 3.3 or 5 volts. Then there
is the motherboard regulator down to the 1.7 volt range. If THAT one is
not up to handling the required oompf during certain CPU loads that is a
sure way to glitch the machine.

{^_^}

Re: Athlon/K7-Opimisation problems

[Nicholas Knight]

On Sunday 09 September 2001 03:49 pm, J. Dow wrote:
> From: "Nicholas Knight" <tegeran@home.com>
>
> > > What about the power supply. If it is at all marginal the power
> > > consumption boost going to 1.4G is likely a killer.
> >
> > Well, he didn't mention the amperage outputs, but he said 431W
> > Enermax, from what I hear Enermax PSU's are good.
> > I still have trouble dealing with the idea that the optimizations
> > cause power consumption like this, but then, I have trouble with my
> > own idea that it causes sufficient heat increase in the chipset that
> > soon after boot.
> >
> > Do most people that experience this problem also experience after a
> > cold-boot where the system had been off for at least 10-15 minutes?
> > And has ANYONE sucsesfully cured this problem by changing power
> > supplies?
>
> Don't forget that there are two regulators involved. First there is the
> primary power supply's regulator down to either 3.3 or 5 volts. Then
> there is the motherboard regulator down to the 1.7 volt range. If THAT
> one is not up to handling the required oompf during certain CPU loads
> that is a sure way to glitch the machine.

Now THAT I'll buy... It would certainly explain why some KT133A 
motherboards work and some don't, but the relation would be a little 
complex, the chipset might be exposing general problems with regulators 
used in motherboards.
Verifying this would require both extensive analysis of the motherboards 
and chipsets, and the regulators.
You'd also need to verify the manufacturing place, batch and time/date 
for at least 50 to 100 motherboards to begin verifying this.

Re: Athlon/K7-Opimisation problems

[Carsten Leonhardt]

Nicholas Knight <tegeran@home.com> writes:

> On Sunday 09 September 2001 03:23 pm, J. Dow wrote:
> > From: "Nicholas Knight" <tegeran@home.com>

> > > Heat anyone? stepping 4 hasn't seemed to be the problem, at least not
> > > directly
> > > What's the tempature difference between your 1Ghz and 1.4Ghz? both
> > > CPU and system? What kind of cooling do you have?

I sold the 1GHz, so I can't tell anymore.  I have a huge heatsink with
a Papst-fan. I have to admit that it was sold to me as being good
enough for 1.33GHz (by the manufacturer himself), but that shouldn't
be the problem, see below.

(If someone wants to know exactly:
http://www.tiger-electronics.de/english/cooler2.htm and look at
PDA38000BC, for a picture look at cooler1.htm SDA38000BC and pretend
you see the Papst-Fan on the top -- the pages seem to be outdated, the
invoice says the fan is good for a 1.33GHz Athlon).

> > What about the power supply. If it is at all marginal the power
> > consumption boost going to 1.4G is likely a killer.
> 
> Well, he didn't mention the amperage outputs, but he said 431W Enermax, 
> from what I hear Enermax PSU's are good.

Here's the data from the PSU's label:

Model: EG465P-VE
DC Output: +3.3V: 38A
           +5V  : 44A
           +12V : 20A
           -5V  : 2A
           -12V : 1A
           +5Vsb: 2.2A
     +3.3V & 5V : 220W max
    total power : 431W

> Do most people that experience this problem also experience after a 
> cold-boot where the system had been off for at least 10-15 minutes?

I just switched the machine off for about 10min, and for good measure,
jumpered the board to 100MHz FSB, so the CPU runs at 1050MHz (thus
being well below the spec for the heatsink). It may have gotten a tad
further in the boot process, but if so, not much. Luckily, it never
reaches the point where filesystems get mounted rw...

Without the optimisation the machine runs smoothly with steady
cpu-load generated by a distributed-net client.
Currently I added a burnK7 and a burnMMX to the load, I'll see if the
machine is still alive in the morning...

leo

Re: Athlon/K7-Opimisation problems

[David Hollister]

Nicholas Knight wrote:
> 
> Do most people that experience this problem also experience after a 
> cold-boot where the system had been off for at least 10-15 minutes? And 
> has ANYONE sucsesfully cured this problem by changing power supplies?

I can get it to oops before the rc scripts are done, even from cold boot.  I 
have not swapped power supplies.  I seriously doubt that's the problem.



-- 
David Hollister
Driversoft Engineering:  http://devicedrivers.com
Digital Audio Resources: http://digitalaudioresources.org

Re: Athlon/K7-Opimisation problems

[J. Dow]

From: "Carsten Leonhardt" <leo@arioch.oche.de>

> Here's the data from the PSU's label:
> 
> Model: EG465P-VE
> DC Output: +3.3V: 38A
>            +5V  : 44A
>            +12V : 20A
>            -5V  : 2A
>            -12V : 1A
>            +5Vsb: 2.2A
>      +3.3V & 5V : 220W max

Hm, with CPUs running out close to 100 watts of late that number sounds
just a wee bit low of the rest of the system is a full set of power hungry
boards.

> I just switched the machine off for about 10min, and for good measure,
> jumpered the board to 100MHz FSB, so the CPU runs at 1050MHz (thus
> being well below the spec for the heatsink). It may have gotten a tad
> further in the boot process, but if so, not much. Luckily, it never
> reaches the point where filesystems get mounted rw...

10 minutes is not enough time to run temperatures WAY down. Leave it off
over night and power it up while cool in the AM. If it is temperature that
should so it up pretty clearly.

> Without the optimisation the machine runs smoothly with steady
> cpu-load generated by a distributed-net client.
> Currently I added a burnK7 and a burnMMX to the load, I'll see if the
> machine is still alive in the morning...

Far better is run burnK7 and burnMMX alternately in one second on one second
off cycles alternating the loads with each on second. This gives you the
dubious benefit of the voltage drops and raises that will happen when the
CPU load alters appreciably. With a constant load the voltage regulators can
compensate. It may be that the problem is related to voltage regulators that
cannot respond to transient loads properly. (I've been there. I have the tee-
shirt. I try to forget the experience. {^_-})

{^_^}

Re: Athlon/K7-Opimisation problems

[Jan Niehusmann]

On Sun, Sep 09, 2001 at 11:35:52PM -0700, J. Dow wrote:
> From: "Carsten Leonhardt" <leo@arioch.oche.de>
> > Currently I added a burnK7 and a burnMMX to the load, I'll see if the
> > machine is still alive in the morning...
> 
> Far better is run burnK7 and burnMMX alternately in one second on one second
> off cycles alternating the loads with each on second. This gives you the

The one Duron I have access to which doesn't like the athlon optimsed 
kernels did crash some seconds after starting a little test program which
does use the kernel copy code in user space.
Note that neither burnK7 nor burnMMX use the kernel copy code, but 
Robert Redelmeier (author of these two programs) has a burnMMX3 program
which uses kernel-equivalent code.

Jan

Re: Athlon/K7-Opimisation problems

[Liakakis Kostas]

On Sun, 9 Sep 2001, J. Dow wrote:

> From: "Carsten Leonhardt" <leo@arioch.oche.de>
> 
> > Here's the data from the PSU's label:
> > 
> > Model: EG465P-VE
> > DC Output: +3.3V: 38A
> >            +5V  : 44A
> >            +12V : 20A
> >            -5V  : 2A
> >            -12V : 1A
> >            +5Vsb: 2.2A
> >      +3.3V & 5V : 220W max
> 
> Hm, with CPUs running out close to 100 watts of late that number sounds
> just a wee bit low of the rest of the system is a full set of power hungry
> boards.

Maybe. But your model is not recommened by AMD. Only the EG365P-VE is from
Enermax, which is 350W/185W combined.

Check their site.

-K.

Re: Athlon/K7-Opimisation problems

[bill davidsen]

In article <01090915115400.00173@c779218-a> you write:
| On Sunday 09 September 2001 10:07 am, Carsten Leonhardt wrote:
| > Hi,
| >
| > here's my data regarding the K7 optimisation problem.
| >
| > Until last week I had a 1GHz Athlon with 133MHz FSB. I then bought a
| > 1.4GHz Athlon, again 133MHz FSB.
| >
| > I never had any problems with the 1GHz processor, but as soon as I
| > stuck the 1.4GHz processor in, the kernel oopsed itself to
| > oblivion. (That was with kernel 2.4.5-ac5, approximately)
| 
| This is common enough it's becoming absurd.

Never had a slower Athlon than 1.4, but I suppose the system could be
underclocked as well;-) However, I have to think there's other stuff
happen, see below.

| >
| > I also noticed that the computer booted ok with the Debian
| > bootfloppies, which use a kernel compiled for i386. So after several
| > kernel compile/boot cycles, I found out that I had to disable K7
| > optimisation to make the system boot again.
| >
| > The mainboard is a Tyan Trinity KT-A (S2390B) with a VIA KT133A
| > chipset.
| >
| > After reading here that it may be the PSU, I upgraded my 300W PSU to a
| > 431W Enermax, which made no difference.
| >
| > The only difference I can make out between the working and the
| > non-working CPU is the internal clockspeed of the CPU and the
| > stepping (old: 2, new: 4).
| 
| Heat anyone? stepping 4 hasn't seemed to be the problem, at least not 
| directly
| What's the tempature difference between your 1Ghz and 1.4Ghz? both CPU 
| and system? What kind of cooling do you have?

In my case, if I compile a 686 kernel it is nice and stable, while an
Athlon kernel is not. With all due respect I bet the temperature is dead
the same regardless of kernel, so I can't blame temperature.

The problem I have is seen only when I use Athlon code enabled with
menuconfig, so the question is if the code is bad (as in timing
sensitive), or if the compiler might be generating bad code for Athlon.

Just a data point, I believe that the problems (at least those which are
bugging me) are releated to something other than temperature.

-- 
bill davidsen <davidsen@tmr.com>
 "If I were a diplomat, in the best case I'd go hungry.  In the worst
  case, people would die."
		-- Robert Lipe

Re: Athlon/K7-Opimisation problems

[Bill Davidsen]

On Fri, 21 Sep 2001, Mark Hahn wrote:

> > The problem I have is seen only when I use Athlon code enabled with
> > menuconfig, so the question is if the code is bad (as in timing
> > sensitive), or if the compiler might be generating bad code for Athlon.
> 
> or the chipset/motherboard/dram can't handle the 2-3x bandwidth
> demanded by the athlon code in the kernel.  the latter is conventional
> wisdom for the past 6-8 months.

The so-called Athlon patch fixed the problem for me on my Abit KK266R
board, I stress tested it at 146MHz FSB (10% o/c) just to see if it was
stable, and it seems all the problems have gone away in both kernel and
user land.

It could be that the bit cleared slows memory, I haven't checked to see if
playing with memory parameters will do the same thing.

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