Guys,

I'm currently running an Opteron system, and am thinking of experimenting
with a second, Athlon 64 based system, for cost reasons. It is much cheaper
to put together an Athlon 64 system, as the CPU is cheaper per MHz, and the
non-ECC memory is cheaper than the ECC memory needed by the Opteron chip.

However, the system will be doing critical data calculations. I can afford
to have it crash with a memory parity error and then be rebooted. I cannot
afford to have bad data from an undetected parity error get into the output
data.

So, is anyone currently running an Athlon 64 system with parity memory and
with the motherboard checking for parity errors? If so, what brand and
model of motherboard?

Thanks,

Paul Missman

Paul Missman wrote:

> Guys,
>
> I'm currently running an Opteron system, and am thinking of experimenting
> with a second, Athlon 64 based system, for cost reasons. It is much cheaper
> to put together an Athlon 64 system,

Not really, as for example an Opteron 150 is only $615, while the Athlon 64
FX-53 socket 939 is $839. The Athlon 64 FX-51(this is also socket 940 though!)
is $745, while the Opteron 148 is $350. If your application doesn't benefit much
from a meg L2 vs
512K L2, then you could go for Athlon 64 3500+ at around $370. If I was in your
place,
I would probably choose an Opteron 148.


> as the CPU is cheaper per MHz, and the
> non-ECC memory is cheaper than the ECC memory needed by the Opteron chip.
>
> However, the system will be doing critical data calculations.

That would be even more of a reason to go for an Opteron.

> I can afford
> to have it crash with a memory parity error and then be rebooted. I cannot
> afford to have bad data from an undetected parity error get into the output
> data.
>
> So, is anyone currently running an Athlon 64 system with parity memory and
> with the motherboard checking for parity errors? If so, what brand and
> model of motherboard?
>
> Thanks,
>
> Paul Missman

On Fri, 10 Sep 2004 00:05:57 -0400, "Paul Missman" <(E-Mail Removed)>
wrote:
>I'm currently running an Opteron system, and am thinking of experimenting
>with a second, Athlon 64 based system, for cost reasons. It is much cheaper
>to put together an Athlon 64 system, as the CPU is cheaper per MHz, and the
>non-ECC memory is cheaper than the ECC memory needed by the Opteron chip.
>
>However, the system will be doing critical data calculations. I can afford
>to have it crash with a memory parity error and then be rebooted. I cannot
>afford to have bad data from an undetected parity error get into the output
>data.
>
>So, is anyone currently running an Athlon 64 system with parity memory and
>with the motherboard checking for parity errors? If so, what brand and
>model of motherboard?

Motherboard doesn't play any part of this, it's all in the chip, and
all Athlon64 chips support ECC fully. That's part of the beauty of
bringing the memory controller on-die, AMD doesn't have to depend on
low-budget motherboard manufacturing doing sketchy implementations
here, it all just works.

Note: I'm assuming by "parity" you mean ECC, as parity memory hasn't
actually been used in about 15 years. It was replaced long ago by
ECC. Also note that the Opteron does *NOT* require ECC memory (though
it's HIGHLY recommended), but they do require *registered* memory.
Registered and ECC are two totally different things.

-------------
Tony Hill
hilla <underscore> 20 <at> yahoo <dot> ca

Paul Missman wrote:

> I'm currently running an Opteron system, and am thinking of
> experimenting with a second, Athlon 64 based system, for cost
> reasons. It is much cheaper to put together an Athlon 64 system,
> as the CPU is cheaper per MHz, and the non-ECC memory is cheaper
> than the ECC memory needed by the Opteron chip.

I may be wrong, but I think registered/buffered and ECC are two separate
issues, e.g. Kingston sells ECC Unbuffered DDR SDRAM DIMMs.

http://www.ec.kingston.com/ecom/conf...R400X72C3A/512

You'll find a list of compatible motherboards.

For example, the ASUS K8N-E Deluxe (socket 754, thus single channel)
supports both ECC and non-ECC unbuffered DDR SDRAM.

http://www.asus.com/prog/spec.asp?m=K8N-E%20Deluxe

> However, the system will be doing critical data calculations. I can
> afford to have it crash with a memory parity error and then be
> rebooted. I cannot afford to have bad data from an undetected parity
> error get into the output data.

I suppose ECC would help.

--
Regards, Grumble

Tony Hill <(E-Mail Removed)> writes:

> On Fri, 10 Sep 2004 00:05:57 -0400, "Paul Missman" <(E-Mail Removed)>
> wrote:
> >I'm currently running an Opteron system, and am thinking of experimenting
> >with a second, Athlon 64 based system, for cost reasons. It is much cheaper
> >to put together an Athlon 64 system, as the CPU is cheaper per MHz, and the
> >non-ECC memory is cheaper than the ECC memory needed by the Opteron chip.
> >
> >However, the system will be doing critical data calculations. I can afford
> >to have it crash with a memory parity error and then be rebooted. I cannot
> >afford to have bad data from an undetected parity error get into the output
> >data.
> >
> >So, is anyone currently running an Athlon 64 system with parity memory and
> >with the motherboard checking for parity errors? If so, what brand and
> >model of motherboard?
>
> Motherboard doesn't play any part of this, it's all in the chip, and
> all Athlon64 chips support ECC fully. That's part of the beauty of
> bringing the memory controller on-die, AMD doesn't have to depend on
> low-budget motherboard manufacturing doing sketchy implementations
> here, it all just works.
>
> Note: I'm assuming by "parity" you mean ECC, as parity memory hasn't
> actually been used in about 15 years. It was replaced long ago by
> ECC. Also note that the Opteron does *NOT* require ECC memory (though
> it's HIGHLY recommended), but they do require *registered* memory.
> Registered and ECC are two totally different things.
>
> -------------
> Tony Hill
> hilla <underscore> 20 <at> yahoo <dot> ca

Dear Tony,

So, you're saying if I put (unregistered?) ECC memory in an Athlon 64
system it will work? What about 754 vs 940 sockets? Will they both work?
Could you suggest a high quality board (Tyan, Asus, etc) for Athlon 64
that has known support for ECC?

Also, what happens if I put registered (ECC) memory in a socket 754 or 940
board? Will it work or fail. In the past (Intel chips) I could usually
use registered memory in any machine, but that may have changed.

I absolutely need ECC. So I was considering getting a Tyan dual 939 chipset
board (single or dual) and using an single Opteron 14[0-8] chip in it. But it
would be a lot cheaper to use the more popular 754/940 socket designs.

Thanks,
Richard

(E-Mail Removed) wrote:

> Tony Hill wrote:
>
>> Paul Missman wrote:
>>
>>> I'm currently running an Opteron system, and am thinking of
>>> experimenting with a second, Athlon 64 based system, for cost
>>> reasons. It is much cheaper to put together an Athlon 64 system,
>>> as the CPU is cheaper per MHz, and the non-ECC memory is cheaper
>>> than the ECC memory needed by the Opteron chip.
>>>
>>> However, the system will be doing critical data calculations. I
>>> can afford to have it crash with a memory parity error and then
>>> be rebooted. I cannot afford to have bad data from an undetected
>>> parity error get into the output data.
>>>
>>> So, is anyone currently running an Athlon 64 system with parity
>>> memory and with the motherboard checking for parity errors? If
>>> so, what brand and model of motherboard?
>>
>> Motherboard doesn't play any part of this, it's all in the chip,
>> and all Athlon64 chips support ECC fully. That's part of the
>> beauty of bringing the memory controller on-die, AMD doesn't have
>> to depend on low-budget motherboard manufacturing doing sketchy
>> implementations here, it all just works.
>>
>> Note: I'm assuming by "parity" you mean ECC, as parity memory
>> hasn't actually been used in about 15 years. It was replaced long
>> ago by ECC. Also note that the Opteron does *NOT* require ECC
>> memory (though it's HIGHLY recommended), but they do require
>> *registered* memory. Registered and ECC are two totally different
>> things.
>
> So, you're saying if I put (unregistered?) ECC memory in an Athlon 64
> system it will work?

Several motherboards do support both ECC and non-ECC unbuffered DDR
SDRAM DIMMs (are we swimming in a sea of acronyms, or what!). I cited
the ASUS K8N-E Deluxe (socket 754) earlier. The user manual states:

You may install 64MB, 128MB, 256MB, 512MB, and 1GB unbuffered ECC
and non-ECC DDR DIMMs into the DIMM sockets using the memory
configurations in this section.

Master ECC Enable
Enables or disables support on all nodes for ECC [...]

> What about 754 vs 940 sockets? Will they both work?

Socket 940 is for Opteron which requires registered memory. I don't
think anybody (??) manufactures non-ECC registered memory.

> Could you suggest a high quality board (Tyan, Asus, etc) for Athlon 64
> that has known support for ECC?

Socket 754 or 939?

You could start here:
http://www.ec.kingston.com/ecom/conf...R400X72C3A/512

> I absolutely need ECC. So I was considering getting a Tyan dual 939 chipset
> board (single or dual) and using an single Opteron 14[0-8] chip in it. But it
> would be a lot cheaper to use the more popular 754/940 socket designs.

It seems you have 939 and 940 mixed up.

Socket 754 = Athlon 64, single channel memory controller
Socket 939 = Athlon 64, dual channel memory controller
Socket 940 = Opteron

--
Regards, Grumble

Grumble wrote:

> Socket 940 is for Opteron which requires registered memory. I don't
> think anybody (??) manufactures non-ECC registered memory.

http://www.lostcircuits.com/memory/reg_ddr/

Briefly, ECC has nothing to do with Registered DIMMs, it is mere
coincidence that the two features are usually implemented together for
the simple reason that the boards that require Registered DIMMs are
usually working non-stop and have a need for ECC to avoid accumulation
and propagation of soft errors that occur statistically over time. For
the memory vendors, the trade-off of carrying Registered non-ECC modules
is a double inventory and, therefore, the savings incurred by leaving
off one chip per bank are not worth carrying two separate lines of
memory, especially, if the non-ECC modules are only geared towards a
splinter cell of user.

"JK" <(E-Mail Removed)> wrote in message
news:(E-Mail Removed)...
>
>
> Not really, as for example an Opteron 150 is only $615, while the Athlon
> 64
> FX-53 socket 939 is $839. The Athlon 64 FX-51(this is also socket 940
> though!)
> is $745, while the Opteron 148 is $350. If your application doesn't
> benefit much
> from a meg L2 vs
> 512K L2, then you could go for Athlon 64 3500+ at around $370. If I was in
> your
> place,
> I would probably choose an Opteron 148.
>

Since it looks like simple parity isn't an option, I'll probably stick with
the Opteron CPU line.

The price difference I was looking at was an Opteron 146 with motherboard at
$460 where ECC memory is about $150 for 512 Meg, versus an Athlon 64 3000
with motherboard for $270 where 512 Meg of non-ECC memory is $100.

At least these prices are cheaper than they were about a year ago when I
purchased my first Opteron system.

Paul Missman

Paul Missman wrote:

> "JK" <(E-Mail Removed)> wrote in message
> news:(E-Mail Removed)...
> >
> >
> > Not really, as for example an Opteron 150 is only $615, while the Athlon
> > 64
> > FX-53 socket 939 is $839. The Athlon 64 FX-51(this is also socket 940
> > though!)
> > is $745, while the Opteron 148 is $350. If your application doesn't
> > benefit much
> > from a meg L2 vs
> > 512K L2, then you could go for Athlon 64 3500+ at around $370. If I was in
> > your
> > place,
> > I would probably choose an Opteron 148.
> >
>
> Since it looks like simple parity isn't an option, I'll probably stick with
> the Opteron CPU line.
>
> The price difference I was looking at was an Opteron 146 with motherboard at
> $460

That sounds a bit high. You could probably get that for around $400-420

> where ECC memory is about $150 for 512 Meg, versus an Athlon 64 3000
> with motherboard for $270

Okay, $240-270

> where 512 Meg of non-ECC memory is $100.

More like $75. The Athlon 64 3000+ will be much slower than an Opteron 146.
The dual channel memory controllers and 1 meg L2 cache vs the single
memory controller and 512K L2 of the Athlon 64 3000+ might make a big
difference in performance. Figure on average a speed grade for each,
so an Athlon 64 3400+ might on average come close to the Opteron 146.
When everything is factored in, the Opteron comes very close in price
to the Athlon 64, except for people who require large amounts of ram.
I mention the Athlon 64 FX-53 and Opteron 150 not because I thought
you should buy them, but that the chips are almost the same.

>
>
> At least these prices are cheaper than they were about a year ago when I
> purchased my first Opteron system.
>
> Paul Missman

On 10 Sep 2004 07:37:26 -0400, (E-Mail Removed) wrote:
>
>Tony Hill <(E-Mail Removed)> writes:
>
>> Motherboard doesn't play any part of this, it's all in the chip, and
>> all Athlon64 chips support ECC fully. That's part of the beauty of
>> bringing the memory controller on-die, AMD doesn't have to depend on
>> low-budget motherboard manufacturing doing sketchy implementations
>> here, it all just works.
>>
>> Note: I'm assuming by "parity" you mean ECC, as parity memory hasn't
>> actually been used in about 15 years. It was replaced long ago by
>> ECC. Also note that the Opteron does *NOT* require ECC memory (though
>> it's HIGHLY recommended), but they do require *registered* memory.
>> Registered and ECC are two totally different things.
>>
>
>Dear Tony,
>
>So, you're saying if I put (unregistered?) ECC memory in an Athlon 64
>system it will work?

Yup, unless someone motherboard manufacturer has really gone out of
their way to try and break ECC support.

> What about 754 vs 940 sockets?

Socket 940 is used for the Opteron and it requires registered memory.
Socket 754 and socket 939 are both used for the Athlon64 line and they
use unregistered memory. The difference between the latter two
sockets is that 754 using single channel memory while socket 939 uses
dual channel memory, ie it requires that you add memory in matched
pairs.

> Will they both work?
>Could you suggest a high quality board (Tyan, Asus, etc) for Athlon 64
>that has known support for ECC?

I know that Asus specifies that their Socket 754 K8N-E board supports
ECC memory, as does their Socket 939 A8V board.

>Also, what happens if I put registered (ECC) memory in a socket 754 or 940
>board? Will it work or fail. In the past (Intel chips) I could usually
>use registered memory in any machine, but that may have changed.

If you put registered memory in socket 754 or socket 939 it will
probably fail. If you put it in socket 940 it will work just fine.

>I absolutely need ECC. So I was considering getting a Tyan dual 939 chipset
>board (single or dual) and using an single Opteron 14[0-8] chip in it. But it
>would be a lot cheaper to use the more popular 754/940 socket designs.

Note that I think you've got you Socket 940 and 939 swapped here, but
otherwise you seem to be on the right track.

-------------
Tony Hill
hilla <underscore> 20 <at> yahoo <dot> ca