I'm hearing rumours about my University wanting to set up a cluster with AMD
Athlon XP+DDR computers, so I wonder what chipset is most stable under
Linux?
I assume it's the AMD DDR chipset, but I want to be pretty sure.
Btw, do compilators currently optimize for the third floating-point unit in
Athlon XP processors?
_____________________________________________________
| Martin Eriksson <[email protected]>
| MSc CSE student, department of Computing Science
| Ume? University, Sweden
On Tue, 13 Nov 2001, Martin Eriksson wrote:
> I'm hearing rumours about my University wanting to set up a cluster with AMD
> Athlon XP+DDR computers, so I wonder what chipset is most stable under
> Linux?
> I assume it's the AMD DDR chipset, but I want to be pretty sure.
I have no problems with the ALi Magik 1 chipset. I'm running a 1 GHz Duron
with 512 MB DDR-RAM at the moment.
--
Ciao, Pascal
-<[ [email protected], netmail 2:241/215.72, home http://cobol.cjb.net/) ]>-
On Tue, 2001-11-13 at 05:19, Martin Eriksson wrote:
> I'm hearing rumours about my University wanting to set up a cluster with AMD
> Athlon XP+DDR computers, so I wonder what chipset is most stable under
> Linux?
>
> I assume it's the AMD DDR chipset, but I want to be pretty sure.
I have no problems with the 1400MHz Athlon on AMD 760, including DRI
using the AGP bridge.
-jwb
-----Original Message-----
From: Jeffrey W. Baker [mailto:[email protected]]
>On Tue, 2001-11-13 at 05:19, Martin Eriksson wrote:
>> I'm hearing rumours about my University wanting to set up a cluster with
AMD
>> Athlon XP+DDR computers, so I wonder what chipset is most stable under
>> Linux?
>>
>> I assume it's the AMD DDR chipset, but I want to be pretty sure.
>I have no problems with the 1400MHz Athlon on AMD 760, including DRI
>using the AGP bridge.
It's my understanding that among boards with the same chipset, some may work
while others do not. I have tested an Iwill KK266+Raid, with KT133A, and it
would not run with athlon optimizations. I also have tried an Epox 8KHA+,
with KT266A, and it has been perfectly stable so far, for three days.
Giles Tyson wrote:
>
> -----Original Message-----
> From: Jeffrey W. Baker [mailto:[email protected]]
> >On Tue, 2001-11-13 at 05:19, Martin Eriksson wrote:
> >> I'm hearing rumours about my University wanting to set up a cluster with
> AMD
> >> Athlon XP+DDR computers, so I wonder what chipset is most stable under
> >> Linux?
> >>
> >> I assume it's the AMD DDR chipset, but I want to be pretty sure.
>
> >I have no problems with the 1400MHz Athlon on AMD 760, including DRI
> >using the AGP bridge.
>
> It's my understanding that among boards with the same chipset, some may work
> while others do not. I have tested an Iwill KK266+Raid, with KT133A, and it
> would not run with athlon optimizations. I also have tried an Epox 8KHA+,
> with KT266A, and it has been perfectly stable so far, for three days.
I have yet to try the latest kernels, but I have an Asus A7V133 (PC133,
not DDR) and a Tyan K7 Thunder (dual). Both are running 1.4GHz Athlons
and both have had problems. In both cases the IDE has been a problem. On
the Tyan, I've had to turn off DMA or it will lock up. I have not messed
with the Asus system yet to see exactly where the problem is.
I'm looking forward to seeing how a newer kernel works (I am running
2.4.9ac10 on my Tyan and a stock Red Hat 7.1 kernel on the Asus until I
upgrade it)
PGA
--
Paul G. Allen
UNIX Admin II/Programmer
Akamai Technologies, Inc.
http://www.akamai.com
On Tue, 13 Nov 2001, Martin Eriksson wrote:
> I'm hearing rumours about my University wanting to set up a cluster with AMD
> Athlon XP+DDR computers, so I wonder what chipset is most stable under
> Linux?
>
> I assume it's the AMD DDR chipset, but I want to be pretty sure.
>
> Btw, do compilators currently optimize for the third floating-point unit in
> Athlon XP processors?
Well, here's my little anecdote:
We bought 33 1.4 GHz AMD Athlons (non-XP) with the slightly deprecated VIA
KT266 Chipset (Spacewalker AK31 motherboards.. not exactly the Lexus of
the M/B world but oh well)..
Anyway, after trying various (2.4) kernel versions, both with and without
the new VM, both with and without Alan's magik, I can say that the only
way we got 99% uptime on these systems (as opposed to like the 70% uptime
I was getting from random kernel oopses) was to turn any Athlon and/or
Pentium optimizations off when compiling the kernel. With any form of
compilation for a CPU >386, the kernel would crash on at least 2 of the
boxes per day. The oops stack trace seemed to always indicate a crash
when in the paging code.. so it was a virtual memory problem? (I can only
speculate and I haven't bothered to investigate much further). I am not
sure if I encountered some unknown bug in my motherboard that needs some
yet undiscovered workaround or what. All I can say is stay away from the
KT266 chipset (however the newer KT266A seems to work fine based on what I
have seen and am told) if you can. But then again I may be crazy and/or
there may be workarounds or bios fixes for my problem, if it was really
kernel-related. (I suspect it was, but haven't bothered to figure out
exactly how to reproduce it so as to submit a patch.. the systems would
crash randomly and invesitigating it futher seemed onerous).
-Calin
>
> _____________________________________________________
> | Martin Eriksson <[email protected]>
> | MSc CSE student, department of Computing Science
> | Ume? University, Sweden
>
>
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/
>
On Tue, 13 Nov 2001, Martin Eriksson wrote:
> I'm hearing rumours about my University wanting to set up a cluster with AMD
> Athlon XP+DDR computers, so I wonder what chipset is most stable under
> Linux?
> I assume it's the AMD DDR chipset, but I want to be pretty sure.
AMD761 or AMD760MP(|X) is what you want. They seem to have by far the
least bugs of all the Athlon chipsets (and coming from AMD, you would
certainly hope so).
-Dan
--
[-] Omae no subete no kichi wa ore no mono da. [-]
We've tried a number of boards for our application servers and the only UP
AMD DDR board I trust right now is the Gigabyte GA-7DX. They are rock
solid.
Other AMD 761 boards may work, but I've made too many late night trips to
the colo to stray from what I know works. DDR support seems to be the
breaking point on most boards.
-- Brian
On Tuesday 13 November 2001 04:08 pm, Calin A. Culianu wrote:
> On Tue, 13 Nov 2001, Martin Eriksson wrote:
> > I'm hearing rumours about my University wanting to set up a cluster
> > with AMD Athlon XP+DDR computers, so I wonder what chipset is most
> > stable under Linux?
> >
> > I assume it's the AMD DDR chipset, but I want to be pretty sure.
> >
> > Btw, do compilators currently optimize for the third floating-point
> > unit in Athlon XP processors?
>
> Well, here's my little anecdote:
>
> We bought 33 1.4 GHz AMD Athlons (non-XP) with the slightly deprecated
> VIA KT266 Chipset (Spacewalker AK31 motherboards.. not exactly the Lexus
> of the M/B world but oh well)..
Brian wrote:
>
> We've tried a number of boards for our application servers and the only UP
> AMD DDR board I trust right now is the Gigabyte GA-7DX. They are rock
> solid.
>
> Other AMD 761 boards may work, but I've made too many late night trips to
> the colo to stray from what I know works. DDR support seems to be the
> breaking point on most boards.
>
Another thing to remember about Athlons is they need power and cooling. I've seen many a system with either a cheap power supply or a poorly ventilated case,
and often both. Athlons WILL push the hardware harder, not to mention the power they suck down themselves, and need a power supply that can handle the load as
well as the fast switching transitions. They also require more cooling, and not just on the CPU, but also on the chipset and throughout the case.
My dual 1.4GHz (K7 Thunder) has 12 fans in it. My single 1.4GHz has 8. They both have 400W+ power supplies.
PGA
--
Paul G. Allen
UNIX Admin II/Programmer
Akamai Technologies, Inc.
http://www.akamai.com
Work: (858)909-3630
> I'm looking forward to seeing how a newer kernel works (I am running
> 2.4.9ac10 on my Tyan and a stock Red Hat 7.1 kernel on the Asus until I
> upgrade it)
All the Athlon relevant patches for -ac are in Linus tree now
On Tue, 13 Nov 2001, Paul G. Allen wrote:
> Brian wrote:
> >
> > We've tried a number of boards for our application servers and the only UP
> > AMD DDR board I trust right now is the Gigabyte GA-7DX. They are rock
> > solid.
> >
> > Other AMD 761 boards may work, but I've made too many late night trips to
> > the colo to stray from what I know works. DDR support seems to be the
> > breaking point on most boards.
> >
>
> Another thing to remember about Athlons is they need power and cooling. I've seen many a system with either a cheap power supply or a poorly ventilated case,
> and often both. Athlons WILL push the hardware harder, not to mention the power they suck down themselves, and need a power supply that can handle the load as
> well as the fast switching transitions. They also require more cooling, and not just on the CPU, but also on the chipset and throughout the case.
Very true. I think good numbers to shoot for are like around 40-50
degrees C for the CPU temp, probably a bit lower for the M/B temp (like
around high 30's?).
>
> My dual 1.4GHz (K7 Thunder) has 12 fans in it. My single 1.4GHz has 8. They both have 400W+ power supplies.
>
> PGA
>
At 05:39 PM 13/11/01 -0500, Calin A. Culianu wrote:
> > > Other AMD 761 boards may work, but I've made too many late night trips to
> > > the colo to stray from what I know works. DDR support seems to be the
> > > breaking point on most boards.
> >
> > Another thing to remember about Athlons is they need power and cooling.
>
>Very true. I think good numbers to shoot for are like around 40-50
>degrees C for the CPU temp, probably a bit lower for the M/B temp (like
>around high 30's?).
I get from 26-36 (Celsius) for the chipset, and 32-45 for the CPU at the
moment. Average of 6-9 degrees difference between chipset and CPU. Main
thing that changes this is ambient air temp, and not process load. I don't
use APM or ACPI to put the cpu into power saving stuff in idle or shutting
down drives or what not, so my temp stays quite stable.
Cases are in general badly designed, and do not allow for any decent
airflow, wether convectional or forced. I've noticed this leading to all
sorts of failure, particularly in hard drives, which do not like heat at
all (and if someone could please shoot the nutcase at WD who started
wrapping their drives in rubber cases.. I'd be immensely grateful *grin*).
It's not too hard to augment some normal cases to have some decent airflow.
Just make sure to turn off all the power saving features (APM, ACPI, etc)
otherwise you won't get reliable results that you can compare against.
> > My dual 1.4GHz (K7 Thunder) has 12 fans in it. My single 1.4GHz has 8.
> They both have 400W+ power supplies.
I'm running a 1.4Ghz fine with 4 x 80mm fans (really nice case with very
good airflow - about 120cfm in my case - it's a rackmount 3RU case known as
a Spinserver) and a 7000rpm fan on the CPU heatsink that has a copper core
(in which almost all the air that goes through the case goes past the CPU
heatsink). The m/board in this is an Asus A7M (AMD761 & Via 686b). Also in
the damn thing is a GeForce3, SBLive! and a 3Com 3c905B, so they add to the
extra juice draw (and the heat), and I run quite happily on a 300W PSU.
This system of mine was previously stable with 2.4.7, then 2.4.9, then it
wouldn't work properly with anything till 2.4.14. Now I get great disk
throughput, fast mem IO, and even better frame rates in OpenGL apps (using
the binary Nvidia driver no less, telling it to use AGPgart) than any
previous kernel. And it's the most stable system I've run as a user system
(and I do hit it pretty hard sometimes).
And yeah, a user (desktop) machine in rackmount case may not make sense
some people, but it's fine for me. *grin*
AMC Enterprises P/L - Stuart Young
First Floor - Network and Systems Admin
3 Chesterville Rd - [email protected]
Cheltenham Vic 3192 - Ph: (03) 9584-2700
http://www.amc.com.au/ - Fax: (03) 9584-2755
From: Brian <[email protected]>
Date: Tue, 13 Nov 2001 16:37:28 -0500
We've tried a number of boards for our application servers and the only UP
AMD DDR board I trust right now is the Gigabyte GA-7DX. They are rock
solid.
Try to use the AGP slot with a Radeon of GeForce card, do something
as simple as playing some quake with com_maxfps > 85 and the machine
will hang solidly.
From: Dan Hollis <[email protected]>
Date: Tue, 13 Nov 2001 13:19:07 -0800 (PST)
AMD761 ... is what you want.
Unless you actually plan on actually using the AGP slot without
crashes/hangs.
Franks a lot,
David S. Miller
[email protected]
The original question was for a cluster (of, presumably, servers).
If you're playing a quake client on an application server, you deserve
what you get.
-- Brian
On Tuesday 13 November 2001 09:32 pm, David S. Miller wrote:
> From: Brian <[email protected]>
> Date: Tue, 13 Nov 2001 16:37:28 -0500
>
> We've tried a number of boards for our application servers and the
> only UP AMD DDR board I trust right now is the Gigabyte GA-7DX. They
> are rock solid.
>
> Try to use the AGP slot with a Radeon of GeForce card, do something
> as simple as playing some quake with com_maxfps > 85 and the machine
> will hang solidly.
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel"
> in the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/
On Tue, 13 Nov 2001, David S. Miller wrote:
> Try to use the AGP slot with a Radeon of GeForce card, do something
> as simple as playing some quake with com_maxfps > 85 and the machine
> will hang solidly.
I thought there was a northbridge configuration workaroud for this errata.
(And no smarty pants replies like "yeah, dont use AGP")
-Dan
--
[-] Omae no subete no kichi wa ore no mono da. [-]
On Tue, 13 Nov 2001, David S. Miller wrote:
> Try to use the AGP slot with a Radeon of GeForce card, do something
> as simple as playing some quake with com_maxfps > 85 and the machine
> will hang solidly.
BTW this bug apparently doesnt affect AMD760MP as I am able to use
geforce2 with quake and unreal tournament for hours straight without any
problems.
Havent tried radeon on 760MP. But geforce2 works fine.
-Dan
--
[-] Omae no subete no kichi wa ore no mono da. [-]
From: Dan Hollis <[email protected]>
Date: Tue, 13 Nov 2001 19:05:11 -0800 (PST)
On Tue, 13 Nov 2001, David S. Miller wrote:
> Try to use the AGP slot with a Radeon of GeForce card, do something
> as simple as playing some quake with com_maxfps > 85 and the machine
> will hang solidly.
I thought there was a northbridge configuration workaroud for this errata.
(And no smarty pants replies like "yeah, dont use AGP")
No, the AMD761 AGP problems have no known fixes (outside of AMD).
AMD's windows drivers have the workarounds.
We've acutally had next to no hangs on the Thunder K7 using a FireGL2 card
--- if you don't mind spending $1000+ for high end 3D graphics :-) Stay away
from nVidia for now; though I understand they're working on fixes for this
chipset for their next driver release.
On Tuesday 13 November 2001 08:32 pm, David S. Miller wrote:
> From: Brian <[email protected]>
> Date: Tue, 13 Nov 2001 16:37:28 -0500
>
> We've tried a number of boards for our application servers and the only
> UP AMD DDR board I trust right now is the Gigabyte GA-7DX. They are rock
> solid.
>
> Try to use the AGP slot with a Radeon of GeForce card, do something
> as simple as playing some quake with com_maxfps > 85 and the machine
> will hang solidly.
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/
--
Marvin Justice
Software Developer
BOXX Technologies, Inc.
http://www.boxxtech.com
512-235-6318 (V)
512-835-0434 (F)
From: Brian <[email protected]>
Date: Tue, 13 Nov 2001 22:11:29 -0500
The original question was for a cluster (of, presumably, servers).
If you're playing a quake client on an application server, you deserve
what you get.
Yes, but the "is rock solid" statement sounded rather
generic and I wanted to correct that :-)
Franks a lot,
David S. Miller
[email protected]
From: Dan Hollis <[email protected]>
Date: Tue, 13 Nov 2001 19:11:56 -0800 (PST)
BTW this bug apparently doesnt affect AMD760MP as I am able to use
geforce2 with quake and unreal tournament for hours straight without any
problems.
What is your quake3 com_maxfps set to? By default it is 85, and
that can hide the bug. Set it to 130 or something like that.
Just bring down the quake3 console (with ') and type
/com_maxfps 130
Try that for a while.
I'm rather sure the AMD761 problems are motherboard vendor
independant, because I have 2 systems so far, using totally different
AMD761 based motherboards, which both hang pretty reliably with AGP.
Franks a lot,
David S. Miller
[email protected]
I have three asus a7v266 with 40-60 day uptimes under load at this point.
these have the older version of the via kt266 chipset.
joelja
On Tue, 13 Nov 2001, Brian wrote:
> The original question was for a cluster (of, presumably, servers).
> If you're playing a quake client on an application server, you deserve
> what you get.
>
> -- Brian
>
> On Tuesday 13 November 2001 09:32 pm, David S. Miller wrote:
> > From: Brian <[email protected]>
> > Date: Tue, 13 Nov 2001 16:37:28 -0500
> >
> > We've tried a number of boards for our application servers and the
> > only UP AMD DDR board I trust right now is the Gigabyte GA-7DX. They
> > are rock solid.
> >
> > Try to use the AGP slot with a Radeon of GeForce card, do something
> > as simple as playing some quake with com_maxfps > 85 and the machine
> > will hang solidly.
> > -
> > To unsubscribe from this list: send the line "unsubscribe linux-kernel"
> > in the body of a message to [email protected]
> > More majordomo info at http://vger.kernel.org/majordomo-info.html
> > Please read the FAQ at http://www.tux.org/lkml/
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/
>
--
--------------------------------------------------------------------------
Joel Jaeggli [email protected]
Academic User Services [email protected]
PGP Key Fingerprint: 1DE9 8FCA 51FB 4195 B42A 9C32 A30D 121E
--------------------------------------------------------------------------
It is clear that the arm of criticism cannot replace the criticism of
arms. Karl Marx -- Introduction to the critique of Hegel's Philosophy of
the right, 1843.
I've been pushing my AMD Abit KG7 board to the limits without crashing,
using a Thunderbird 1.4.
Don't forget that cooling is also an important issue with these
processors/motherboards!!
(Mobo is using an AMD 761 AGP Bridge and the rest is from Via ... (don't
recall, not behind pc)
DK
At 16:37 13/11/2001 -0500, Brian wrote:
>We've tried a number of boards for our application servers and the only UP
>AMD DDR board I trust right now is the Gigabyte GA-7DX. They are rock
>solid.
>
>Other AMD 761 boards may work, but I've made too many late night trips to
>the colo to stray from what I know works. DDR support seems to be the
>breaking point on most boards.
>
> -- Brian
>
>On Tuesday 13 November 2001 04:08 pm, Calin A. Culianu wrote:
> > On Tue, 13 Nov 2001, Martin Eriksson wrote:
> > > I'm hearing rumours about my University wanting to set up a cluster
> > > with AMD Athlon XP+DDR computers, so I wonder what chipset is most
> > > stable under Linux?
> > >
> > > I assume it's the AMD DDR chipset, but I want to be pretty sure.
> > >
> > > Btw, do compilators currently optimize for the third floating-point
> > > unit in Athlon XP processors?
> >
> > Well, here's my little anecdote:
> >
> > We bought 33 1.4 GHz AMD Athlons (non-XP) with the slightly deprecated
> > VIA KT266 Chipset (Spacewalker AK31 motherboards.. not exactly the Lexus
> > of the M/B world but oh well)..
> From: Dan Hollis <[email protected]>
> Date: Tue, 13 Nov 2001 13:19:07 -0800 (PST)
>
> AMD761 ... is what you want.
>
> Unless you actually plan on actually using the AGP slot without
> crashes/hangs.
Only if your card hits the AMD errata, and that specifically claims the
card is the problem. It mostly appears to afflict nvidia users so its not
a problem ;)
From: Alan Cox <[email protected]>
Date: Wed, 14 Nov 2001 10:51:49 +0000 (GMT)
Only if your card hits the AMD errata, and that specifically claims
the card is the problem. It mostly appears to afflict nvidia users
so its not a problem ;)
And radeon users...
Franks a lot,
David S. Miller
[email protected]
On Tue, 13 Nov 2001, Marvin Justice wrote:
> We've acutally had next to no hangs on the Thunder K7 using a FireGL2 card
> --- if you don't mind spending $1000+ for high end 3D graphics :-) Stay away
> from nVidia for now; though I understand they're working on fixes for this
> chipset for their next driver release.
Saw in your sig, I know who Boxx Tech is, they build machines like I build
for myself, but for less money. :P
Anyway, to keep this kinda kernel related. How well supported are the
FireGL cards? That is DRM wise, and of course X to go with that.
-Chris
--
Two penguins were walking on an iceberg. The first penguin said to the
second, "you look like you are wearing a tuxedo." The second penguin
said, "I might be..." --David Lynch, Twin Peaks
On Tue, 13 Nov 2001, David S. Miller wrote:
> From: Brian <[email protected]>
> Date: Tue, 13 Nov 2001 22:11:29 -0500
>
> The original question was for a cluster (of, presumably, servers).
> If you're playing a quake client on an application server, you deserve
> what you get.
>
> Yes, but the "is rock solid" statement sounded rather
> generic and I wanted to correct that :-)
Talcum, or diamond?
Mike
On Wednesday 14 November 2001 07:27 am, Chris Meadors wrote:
> Anyway, to keep this kinda kernel related. How well supported are the
> FireGL cards? That is DRM wise, and of course X to go with that.
>
> -Chris
Drivers are distributed pretty much the same way as nVidia, ie., as a binary
core along with the necessary wrappers to compile for new kernels, private
libGL.so, etc. Aside from that, we've generally had pretty positive
experiences.
These aren't quake cards; in fact the demo1 fps are surprisingly low. In
serious OpenGL benchmarks, however,
(http://www.specbench.org/gpc/opc.static/opcview.htm ) they're currently the
fastest cards available for Linux by a long shot.
I think FGL is what people at ILM, Dreamworks, Digital Domain etc. are using
at the moment in the gradual shift toward Linux on the artist's desktop. Not
cheap --- unless you've already forked over $16K for Maya ;-)
--
Marvin Justice
Software Developer
BOXX Technologies, Inc.
http://www.boxxtech.com
512-235-6318 (V)
512-835-0434 (F)
On Tue, 13 Nov 2001 19:16:07 -0800 (PST), David S. Miller <[email protected]> wrote:
> From: Dan Hollis <[email protected]>
> Date: Tue, 13 Nov 2001 19:11:56 -0800 (PST)
>
> BTW this bug apparently doesnt affect AMD760MP as I am able to use
> geforce2 with quake and unreal tournament for hours straight without any
> problems.
>
> I'm rather sure the AMD761 problems are motherboard vendor
> independant, because I have 2 systems so far, using totally different
> AMD761 based motherboards, which both hang pretty reliably with AGP.
As far as I know, the 760MP chipset uses a 762 North Bridge, not a 761.
That might explain why the 760MP is stable and the 760 is not.
Ion
--
It is better to keep your mouth shut and be thought a fool,
than to open it and remove all doubt.
The 762 North Bridge definitely has AGP issues and will lock up with GeForce
3 and nVidia's latest official drivers. I just got my hands on beta drivers
and the lockups appear to have gone away --- so far ;-) Their binary only
kernel module has functions with names ike "AMD_FixupGART",
"AMD_ApplyChipsetUpdates" etc.
-Marvin
On Wednesday 14 November 2001 10:54 am, Ion Badulescu wrote:
> On Tue, 13 Nov 2001 19:16:07 -0800 (PST), David S. Miller
<[email protected]> wrote:
> >?? From: Dan Hollis <[email protected]>
> >?? Date: Tue, 13 Nov 2001 19:11:56 -0800 (PST)
> >
> >?? BTW this bug apparently doesnt affect AMD760MP as I am able to use
> >?? geforce2 with quake and unreal tournament for hours straight without
> > any problems.
> >
> > I'm rather sure the AMD761 problems are motherboard vendor
> > independant, because I have 2 systems so far, using totally different
> > AMD761 based motherboards, which both hang pretty reliably with AGP.
>
> As far as I know, the 760MP chipset uses a 762 North Bridge, not a 761.
> That might explain why the 760MP is stable and the 760 is not.
>
> Ion
On Tue, 13 Nov 2001, Joel Jaeggli wrote:
> I have three asus a7v266 with 40-60 day uptimes under load at this point.
> these have the older version of the via kt266 chipset.
Ok, actually i was a little drastic there. I think maybe I should have
said "shuttle AK31 motherboards with the kt266 seem problematic at leasts
in my experience". Yeah many people seem to be using the kt266 just fine.
Are you running with Athlon or pentium optimizations turned on?
-Calin
>
> joelja
>
>
> On Tue, 13 Nov 2001, Brian wrote:
>
> > The original question was for a cluster (of, presumably, servers).
> > If you're playing a quake client on an application server, you deserve
> > what you get.
> >
> > -- Brian
> >
> > On Tuesday 13 November 2001 09:32 pm, David S. Miller wrote:
> > > From: Brian <[email protected]>
> > > Date: Tue, 13 Nov 2001 16:37:28 -0500
> > >
> > > We've tried a number of boards for our application servers and the
> > > only UP AMD DDR board I trust right now is the Gigabyte GA-7DX. They
> > > are rock solid.
> > >
> > > Try to use the AGP slot with a Radeon of GeForce card, do something
> > > as simple as playing some quake with com_maxfps > 85 and the machine
> > > will hang solidly.
> > > -
> > > To unsubscribe from this list: send the line "unsubscribe linux-kernel"
> > > in the body of a message to [email protected]
> > > More majordomo info at http://vger.kernel.org/majordomo-info.html
> > > Please read the FAQ at http://www.tux.org/lkml/
> > -
> > To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> > the body of a message to [email protected]
> > More majordomo info at http://vger.kernel.org/majordomo-info.html
> > Please read the FAQ at http://www.tux.org/lkml/
> >
>
>
On Wed, 14 Nov 2001, Stuart Young wrote:
> At 05:39 PM 13/11/01 -0500, Calin A. Culianu wrote:
> > > > Other AMD 761 boards may work, but I've made too many late night trips to
> > > > the colo to stray from what I know works. DDR support seems to be the
> > > > breaking point on most boards.
> > >
> > > Another thing to remember about Athlons is they need power and cooling.
> >
> >Very true. I think good numbers to shoot for are like around 40-50
> >degrees C for the CPU temp, probably a bit lower for the M/B temp (like
> >around high 30's?).
>
> I get from 26-36 (Celsius) for the chipset, and 32-45 for the CPU at the
> moment. Average of 6-9 degrees difference between chipset and CPU. Main
> thing that changes this is ambient air temp, and not process load. I don't
> use APM or ACPI to put the cpu into power saving stuff in idle or shutting
> down drives or what not, so my temp stays quite stable.
>
> Cases are in general badly designed, and do not allow for any decent
> airflow, wether convectional or forced. I've noticed this leading to all
> sorts of failure, particularly in hard drives, which do not like heat at
> all (and if someone could please shoot the nutcase at WD who started
> wrapping their drives in rubber cases.. I'd be immensely grateful *grin*).
> It's not too hard to augment some normal cases to have some decent airflow.
> Just make sure to turn off all the power saving features (APM, ACPI, etc)
> otherwise you won't get reliable results that you can compare against.
Yes, true. At least while testing out your cooling setup, it helps to
have the cpu heating up as much as it can. That pesky 'hlt'
instruction... :)
>
> > > My dual 1.4GHz (K7 Thunder) has 12 fans in it. My single 1.4GHz has 8.
> > They both have 400W+ power supplies.
>
> I'm running a 1.4Ghz fine with 4 x 80mm fans (really nice case with very
> good airflow - about 120cfm in my case - it's a rackmount 3RU case known as
> a Spinserver) and a 7000rpm fan on the CPU heatsink that has a copper core
> (in which almost all the air that goes through the case goes past the CPU
> heatsink). The m/board in this is an Asus A7M (AMD761 & Via 686b). Also in
> the damn thing is a GeForce3, SBLive! and a 3Com 3c905B, so they add to the
> extra juice draw (and the heat), and I run quite happily on a 300W PSU.
>
> This system of mine was previously stable with 2.4.7, then 2.4.9, then it
> wouldn't work properly with anything till 2.4.14. Now I get great disk
> throughput, fast mem IO, and even better frame rates in OpenGL apps (using
> the binary Nvidia driver no less, telling it to use AGPgart) than any
> previous kernel. And it's the most stable system I've run as a user system
> (and I do hit it pretty hard sometimes).
>
> And yeah, a user (desktop) machine in rackmount case may not make sense
> some people, but it's fine for me. *grin*
>
>
> AMC Enterprises P/L - Stuart Young
> First Floor - Network and Systems Admin
> 3 Chesterville Rd - [email protected]
> Cheltenham Vic 3192 - Ph: (03) 9584-2700
> http://www.amc.com.au/ - Fax: (03) 9584-2755
>
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/
>
At 07:16 PM 13/11/01 -0800, David S. Miller wrote:
>What is your quake3 com_maxfps set to? By default it is 85, and
>that can hide the bug. Set it to 130 or something like that.
>
><snip!>
>
>I'm rather sure the AMD761 problems are motherboard vendor
>independant, because I have 2 systems so far, using totally different
>AMD761 based motherboards, which both hang pretty reliably with AGP.
Last night I decided to test this out on my Asus A7M (AMD761 and VIA 686b),
which has a 1.4 Ghz Athlon, Creative SBLive!, 3Com 3c905B, and an Asus
V8200 (GeForce3 64Mb DDR). The case has very good airflow, at about 120
Cubic Feet per Minute, and most of this passes past the CPU. PSU is a 300W
Enhance (brand name) power supply - use a lot of this brand, and they are
good quality. Linux kernel is 2.4.14 vanilla, with the Nvidia 1.0-1541
binary driver (using kernel AGPgart, and Nvidia module options for speed),
XFree86 4.1.0 that is currently in Debian sid. Was running at 1024x768 at
24 bit, defaults for Q3A texture settings, sound, etc, with /com_maxfps set
to 255. I played the single player setup against the bots (in the medium
hardness setting - I didn't lose a round, which kept the 3D going as
continuous as possible) from the start to the finish (all 6 tiers and the
final Z tier) without a lockup. Note: I do not run the Riva FrameBuffer
module, nor do I enable ANY of the APM options (eg: Make CPU Idle calls,
Enable console blanking, etc) in the kernel, just the generic APM support
itself - mainly so the machine shuts off automatically after shutdown).
AGP and bad Power Supplies are always a problem, especially when you try
and exploit anything greater than 1x AGP, and the power drain really rises.
Heat is always an issue, but with good cooling, most problems vanish.
Sure that beanie of yours isn't cutting off the circulation, or that you've
been drinking too much V again Dave? *joke*
BTW: As a member of a group called the LGL (or Linux Gamers League) here in
Australia, I help out a lot of people who want to get their systems running
games, particularly 3D stuff like Q3A, UT, and Tribes2. I've been able to
track down most of the crash issues with AMD 76x chipsets to either 1) bad
quality components, 2) heat or 3) power supplies. Admittedly AMD 760
(vanilla, not sure about MP) chipsets don't like doing more than 1x AGP,
but since they are a bit old in the tooth now, this isn't so much of a
problem for us gamers. If there was a huge issue with this, I'm sure you'd
hear a hell of a lot of the guys and girls (yes, girls play games under
Linux too!) in the LGL complaining about it. *grin*
Take care.
AMC Enterprises P/L - Stuart Young
First Floor - Network and Systems Admin
3 Chesterville Rd - [email protected]
Cheltenham Vic 3192 - Ph: (03) 9584-2700
http://www.amc.com.au/ - Fax: (03) 9584-2755
"David S. Miller" wrote:
>
> From: Dan Hollis <[email protected]>
> Date: Tue, 13 Nov 2001 19:11:56 -0800 (PST)
>
> BTW this bug apparently doesnt affect AMD760MP as I am able to use
> geforce2 with quake and unreal tournament for hours straight without any
> problems.
>
> What is your quake3 com_maxfps set to? By default it is 85, and
> that can hide the bug. Set it to 130 or something like that.
>
> Just bring down the quake3 console (with ') and type
>
> /com_maxfps 130
>
> Try that for a while.
>
> I'm rather sure the AMD761 problems are motherboard vendor
> independant, because I have 2 systems so far, using totally different
> AMD761 based motherboards, which both hang pretty reliably with AGP.
>
My dual Tyan runs Q3A, UT, and Tribes 2 at over 100fps at times with no problems UNLESS I enable DMA for the IDE drive. Q3a will go above 130fps (map and game
dependent), UT will go even higher, and Tribes 2 will hit right around 140fps if I'm in a room (any outdoor areas slow WAY down).
I am running 2.4.9ac10 with a few minor tweaks, agpgart slightly tweaked compiled in, and a tweaked Detonator 3 nVidia driver. I plan to upgrade all these soon
and see what happens.
My A7V133 however is crap when it comes to playing games. It's been demoted to straight server duty. :)
PGA
--
Paul G. Allen
UNIX Admin II/Programmer
Akamai Technologies, Inc.
http://www.akamai.com
Work: (858)909-3630
On Wed, 14 Nov 2001, Paul G. Allen wrote:
> I am running 2.4.9ac10 with a few minor tweaks, agpgart slightly tweaked
> compiled in, and a tweaked Detonator 3 nVidia driver
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
???
-Dan
--
[-] Omae no subete no kichi wa ore no mono da. [-]
At 05:15 PM 14/11/01 -0800, Dan Hollis wrote:
>On Wed, 14 Nov 2001, Paul G. Allen wrote:
> > I am running 2.4.9ac10 with a few minor tweaks, agpgart slightly tweaked
> > compiled in, and a tweaked Detonator 3 nVidia driver
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>???
Damn binary thing.
It's the same core code as the Windows Detonator 3 driver. Hence the mention.
The next one will be based on the Detonator 4 core code, so things like
OpenGL 1.3 (which is supposedly available with the Detonator 4 drivers
under Windows) will be available.
Paul, you may want to see my post earlier re 2.4.14 vanilla and my
experiences so far, which have all been good.
AMC Enterprises P/L - Stuart Young
First Floor - Network and Systems Admin
3 Chesterville Rd - [email protected]
Cheltenham Vic 3192 - Ph: (03) 9584-2700
http://www.amc.com.au/ - Fax: (03) 9584-2755
Stuart Young wrote:
>
> At 05:15 PM 14/11/01 -0800, Dan Hollis wrote:
> >On Wed, 14 Nov 2001, Paul G. Allen wrote:
> > > I am running 2.4.9ac10 with a few minor tweaks, agpgart slightly tweaked
> > > compiled in, and a tweaked Detonator 3 nVidia driver
> > ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> >???
>
I made a couple small changes so that it would recognize the MP chipset and not drop to a generic operational mode. The generic mode did not work properly and
would cause occasional system hangs. Fast writes still won't work, but I think a BIOS upgrade might fix that.
> Damn binary thing.
>
> It's the same core code as the Windows Detonator 3 driver. Hence the mention.
>
> The next one will be based on the Detonator 4 core code, so things like
> OpenGL 1.3 (which is supposedly available with the Detonator 4 drivers
> under Windows) will be available.
>
> Paul, you may want to see my post earlier re 2.4.14 vanilla and my
> experiences so far, which have all been good.
>
I just saw it. I am a registered nVidia developer, but have been too busy with other things (mainly working on the V12 game engine) to look into updating my
actual system software and ping on nVidia about their Linux driver. I asked them once a while back, and they said "A new driver will be released soon, so wait
for that."
I'm getting ready to do it though, since I need to wait for another developer to fix some code (in the game engine) before I can continue.
PGA
--
Paul G. Allen
UNIX Admin II/Programmer
Akamai Technologies, Inc.
http://www.akamai.com
Work: (858)909-3630
You won't see the problem with the GeForce3 because it has stricter
AGP timings.
The problem is going to show up with cards that are a little bit out
of the AGP spec, this includes the Radeon and the GeForce2.
Franks a lot,
David S. Miller
[email protected]
On Wed, 14 Nov 2001 10:51:49 +0000 (GMT), Alan Cox wrote:
>> From: Dan Hollis <[email protected]>
>> Date: Tue, 13 Nov 2001 13:19:07 -0800 (PST)
>>
>> AMD761 ... is what you want.
>>
>> Unless you actually plan on actually using the AGP slot without
>> crashes/hangs.
>
>Only if your card hits the AMD errata, and that specifically claims the
>card is the problem. It mostly appears to afflict nvidia users so its not
>a problem ;)
Is Matrox G200 on this list ( or where can I find this document ? )
I am not addicted to Q3 :) , but testing my DRI configuration
Q3 locks my system in 2-3 min after game start .
sytem Abit KT7E
AMD DURON 800
256 MB RAM noname
Matrox G200
HD Maxtor 20 GB UDMA100
--------------------------------------------------------
Marek Mentel [email protected] 2:484/3.8
INSTITUTE FOR CHEMICAL PROCESSING OF COAL , Zabrze , POLAND
NOTE: my opinions are strictly my own and not those of my employer
From: "Marek Mentel" <[email protected]>
Date: Thu, 15 Nov 2001 14:57:31 -0500 (EST)
Is Matrox G200 on this list ( or where can I find this document ? )
I am not addicted to Q3 :) , but testing my DRI configuration
Q3 locks my system in 2-3 min after game start .
Ask AMD, they won't tell us...
My rule has been, if AGP usage makes it hang under Linux it must be on
their list. Does this answer your question? :-)
Chris Wedgwood wrote:
>
> On Wed, Nov 14, 2001 at 05:03:21PM -0800, Paul G. Allen wrote:
>
> I am running 2.4.9ac10 with a few minor tweaks, agpgart slightly
> tweaked compiled in, and a tweaked Detonator 3 nVidia driver. I
> plan to upgrade all these soon and see what happens.
>
> Is this different from 1541? If so, where might I find this
>
I just D/L, modified and compiled 1541. A cat of /proc/nv/card0 shows:
[root@keroon /root]# cat /proc/nv/card0
----- Driver Info -----
NVRM Version: 1.0-1541
------ Card Info ------
Model: GeForce3
IRQ: 17
Video BIOS: 03.20.00.10
------ AGP Info -------
AGP status: Enabled
AGP Driver: NVIDIA
Bridge: AMD Irongate MP
SBA: Supported [enabled]
FW: Supported [disabled]
Rates: 4x 2x 1x [4x]
Registers: 0x0f000217:0x00000304
[root@keroon /root]#
So, AGP 4x and Side Band Addressing is enabled, but for some reason Fast
Writes are not. I am still using the 2.06 Tyan BIOS (as shipped) and
need to upgrade to the latest (I've had trouble getting the BIOS file
from the Tyan web site). agpgart is not compiled into this kernel, so as
you can see it's using the NVIDIA driver instead. I e-mailed developer
support at nVidia to ask why FW is disabled even though it's supported
(it could very well be the BIOS).
I also installed thier GLX libraries.
PGA
--
Paul G. Allen
UNIX Admin II/Network Security
Akamai Technologies, Inc.
http://www.akamai.com
On Mon, 19 Nov 2001, Paul G. Allen wrote:
> Chris Wedgwood wrote:
> >
> > On Wed, Nov 14, 2001 at 05:03:21PM -0800, Paul G. Allen wrote:
> >
> > I am running 2.4.9ac10 with a few minor tweaks, agpgart slightly
> > tweaked compiled in, and a tweaked Detonator 3 nVidia driver. I
> > plan to upgrade all these soon and see what happens.
> >
> > Is this different from 1541? If so, where might I find this
> >
>
> I just D/L, modified and compiled 1541. A cat of /proc/nv/card0 shows:
>
> [root@keroon /root]# cat /proc/nv/card0
> ----- Driver Info -----
> NVRM Version: 1.0-1541
> ------ Card Info ------
> Model: GeForce3
> IRQ: 17
> Video BIOS: 03.20.00.10
> ------ AGP Info -------
> AGP status: Enabled
> AGP Driver: NVIDIA
> Bridge: AMD Irongate MP
> SBA: Supported [enabled]
> FW: Supported [disabled]
> Rates: 4x 2x 1x [4x]
> Registers: 0x0f000217:0x00000304
> [root@keroon /root]#
>
> So, AGP 4x and Side Band Addressing is enabled, but for some reason Fast
> Writes are not. I am still using the 2.06 Tyan BIOS (as shipped) and
> need to upgrade to the latest (I've had trouble getting the BIOS file
> from the Tyan web site). agpgart is not compiled into this kernel, so as
> you can see it's using the NVIDIA driver instead. I e-mailed developer
> support at nVidia to ask why FW is disabled even though it's supported
> (it could very well be the BIOS).
May-be, FW works quite well here, but they just want to enable it in some
future driver version and then claim 50% speed improvement. :-) :o)
To be serious, FW needs special handling in AGP, notably flow-control by
the target using WBF for example. As a result, it could well be broken for
your board forever due to some errata in signalling.
> I also installed thier GLX libraries.
You seem to like bloat. :)
G?rard.