General Schvantzkopf writes:
>What I
>care about is compute speed, memory size and stability. Any
>recommendations will be appreciated.
>
>Is there any support for overclocking on a server motherboard. There are
>a couple of SuperMicro motherboards that look good, the MBD-X7DWA-N with
>8 FB-DIMM sockets and the MBD-X7DWA-N+ that has a whopping 16 FB_DIMM
>sockets. Can these boards be overclocked?

We have a dual-Xeon 5160 box with 24GB of RAM based on the SuperMicro
X7DBE+ and are happy with it. The only funny thing is that, when we
turn it on, it does not show any sign of working for about one minute.

Concerning RAM, I recommend that you avoid FB-DIMMs if you can:
FB-DIMMs consume much more power (our machine consumes 280W when
idle), and require special cooling (we have a 5300rpm 80mm fan for our
FB-DIMMs, otherwise the machine would not work). Instead, go for
registered DIMMs (e.g. PC5300R instead of PC5300F DIMMs); if you want
to use a dual-Xeon board, that would mean the 5100 chipset.
Supermicro seems to have only a 6-DIMM board (X7DCL-i), whereas Tyan
has an 8-DIMM board in the Tempest i5100X (S5375); you might also want
to look at what Intel has to offer.

Concerning overclocking, there's little demand for that in servers, so
server boards usually do not support it. Looking at the manual of the
X7DBE+, I see no way to overclock the FSB. You can set the
multiplier, but that's probably locked on the CPU. One thing that may
be possible is to get a motherboard with a FSB1600, buy CPUs with
FSB1333, and try to run it at FSB1600. That would overclock a 3GHz
CPU to 3.6GHz. Unfortunately the 5100 chipset I recommended above
seems to be limited to FSB1333, and FSB1600 seems to be limited to the
5400 chipset, which requires FB-DIMMs.

- anton
--
M. Anton Ertl Some things have to be seen to be believed
anton@mips.complang.tuwien.ac.at Most things have to be believed to be seen
http://www.complang.tuwien.ac.at/anton/home.html

On Fri, 25 Jan 2008 12:54:46 +0000, Anton Ertl wrote:

> General Schvantzkopf writes:
>>What I
>>care about is compute speed, memory size and stability. Any
>>recommendations will be appreciated.
>>
>>Is there any support for overclocking on a server motherboard. There are
>>a couple of SuperMicro motherboards that look good, the MBD-X7DWA-N with
>>8 FB-DIMM sockets and the MBD-X7DWA-N+ that has a whopping 16 FB_DIMM
>>sockets. Can these boards be overclocked?
>
> We have a dual-Xeon 5160 box with 24GB of RAM based on the SuperMicro
> X7DBE+ and are happy with it. The only funny thing is that, when we
> turn it on, it does not show any sign of working for about one minute.
>
> Concerning RAM, I recommend that you avoid FB-DIMMs if you can: FB-DIMMs
> consume much more power (our machine consumes 280W when idle), and
> require special cooling (we have a 5300rpm 80mm fan for our FB-DIMMs,
> otherwise the machine would not work). Instead, go for registered DIMMs
> (e.g. PC5300R instead of PC5300F DIMMs); if you want to use a dual-Xeon
> board, that would mean the 5100 chipset. Supermicro seems to have only a
> 6-DIMM board (X7DCL-i), whereas Tyan has an 8-DIMM board in the Tempest
> i5100X (S5375); you might also want to look at what Intel has to offer.
>
> Concerning overclocking, there's little demand for that in servers, so
> server boards usually do not support it. Looking at the manual of the
> X7DBE+, I see no way to overclock the FSB. You can set the multiplier,
> but that's probably locked on the CPU. One thing that may be possible
> is to get a motherboard with a FSB1600, buy CPUs with FSB1333, and try
> to run it at FSB1600. That would overclock a 3GHz CPU to 3.6GHz.
> Unfortunately the 5100 chipset I recommended above seems to be limited
> to FSB1333, and FSB1600 seems to be limited to the 5400 chipset, which
> requires FB-DIMMs.
>
> - anton

I read somewhere that SuperMicro does have some overclocking support on
their boards but they don't advertise the fact.

I probably won't build this system until March or April. I'm hoping that
4G DDR2 DIMMs will be available by then which would allow me to use a
desktop motherboard. I was looking at the Samsung and Micron sites and it
appears that they are both sampling 2G DDR2 RAMs which makes a 4G DIMM
possible, in fact Samsung was also sampling a 4G unbuffered DIMM.

On Thu, 24 Jan 2008 15:57:40 -0600, General Schvantzkopf wrote:
>I'm starting to plan my next system. I would like to have support for >
>8G of RAM which puts me on a server motherboard. I'll probably use a pair
>of 3GHz dual core 45nm Xeons which I would like to overclock if possible.
>My current fastest compute server is an 8G Core2 E6700 overclocked to

Why would you select a server motherboard for increased reliability, and
then overclock it to destroy the system's reliability?

Pick a faster processor if you need it to run faster.

> Why would you select a server motherboard for increased reliability, and
> then overclock it to destroy the system's reliability?
>
> Pick a faster processor if you need it to run faster.

The only reason that I'm looking at server motherboards is for the large
memory support. Desktop motherboards are limited to 8G which is starting
to be a problem for me, I would like to have 16G in my next system. If 4G
unbuffered DIMMs get announced in the next couple of months then I'll
stick with a desktop motherboard.

My current Core2 is overclocked and it's been completely reliable. I've
been running it 24/7 for about 15 months without an issue. My server room
is air conditioned in the summer and in the winter I keep the heat down
to 60 so I don't have to worry about running in a hot room. My Core2 is
using stock Intel cooling, I plan on using a heavy duty cooler like the
Zalman 9700 in the next box which should allow me to get another .5GHz.
With my Core2 system I used sys_basher to find it's breaking point, then
I backed off by about 15%. I also underclocked the memory system, the
RAMS are DDR2800 but I'm running them at 600 which gives the memory
system plenty of margin.

On 2008-01-25, General Schvantzkoph hit the keyboard and wrote:
>


> My current Core2 is overclocked and it's been completely reliable. I've
> been running it 24/7 for about 15 months without an issue. My server room
> is air conditioned in the summer and in the winter I keep the heat down
> to 60 so I don't have to worry about running in a hot room. My Core2 is
> using stock Intel cooling, I plan on using a heavy duty cooler like the
> Zalman 9700 in the next box which should allow me to get another .5GHz.

Liquid cooling is then no option for you I take? :-)
Overclockers like to use it...

> With my Core2 system I used sys_basher to find it's breaking point, then
> I backed off by about 15%. I also underclocked the memory system, the
> RAMS are DDR2800 but I'm running them at 600 which gives the memory
> system plenty of margin.




Dragomir Kollaric
--
Q: Do you know what the death-rate around here is?
A: One per person!

On Sat, 26 Jan 2008 17:24:12 +0000, Dragomir Kollaric wrote:

> On 2008-01-25, General Schvantzkoph hit the keyboard and wrote:
>>
>
>
>> My current Core2 is overclocked and it's been completely reliable. I've
>> been running it 24/7 for about 15 months without an issue. My server
>> room is air conditioned in the summer and in the winter I keep the heat
>> down to 60 so I don't have to worry about running in a hot room. My
>> Core2 is using stock Intel cooling, I plan on using a heavy duty cooler
>> like the Zalman 9700 in the next box which should allow me to get
>> another .5GHz.
>
> Liquid cooling is then no option for you I take? :-) Overclockers like
> to use it...
>
>> With my Core2 system I used sys_basher to find it's breaking point,
>> then I backed off by about 15%. I also underclocked the memory system,
>> the RAMS are DDR2800 but I'm running them at 600 which gives the memory
>> system plenty of margin.
>
>
>
>
> Dragomir Kollaric

Cooling is not the problem, there are very good air coolers that will do
the job. The issue is BIOS support for overclocking which allows you to
control the clock rate, DDR clock rate, and voltages. Desktop
motherboards generally have these features, server motherboards mostly
don't. Overclocking is a selling feature for desktop motherboards, it
isn't for servers. However some server motherboards have included some
overclocking features, probably because the BIOS codes are basically the
same as for the desktops and the engineers developing the server
motherboards need to be able to tweak the same parameters anyway. The
decision to enable those features in a shipping BIOS is a subjective one
so some manufacturers may choose to do it on the grounds that it doesn't
cost them anything and that it might give them some selling advantage,
and others may choose not to because it could effect their reputation for
reliability. What I was asking was if anyone knew of a server motherboard
that supports the 45nm Xeons and was overclockable. As I've said earlier
what I'm really interested in is 16G of memory. A desktop board that
could use registered DIMMs would satisfy my needs. The availability of
4G unbuffered DIMMs would also satisfy my needs.

> We have a dual-Xeon 5160 box with 24GB of RAM based on the SuperMicro
> X7DBE+ and are happy with it. The only funny thing is that, when we
> turn it on, it does not show any sign of working for about one minute.

With ECC, it has to go through and zero out all of the memory so that ECC
operation can be reliable. If you took out most of the RAM, you'd find it
would take considerably less time.

"Steve Wolfe" writes:
>> We have a dual-Xeon 5160 box with 24GB of RAM based on the SuperMicro
>> X7DBE+ and are happy with it. The only funny thing is that, when we
>> turn it on, it does not show any sign of working for about one minute.
>
> With ECC, it has to go through and zero out all of the memory so that ECC
>operation can be reliable.

Good theory. I would expect it to initialize the graphics card and
the keyboard first, though, and only initialize most of the memory
afterwards, just like the classical RAM test does. We had no
indication that the machine was working at all, and turned it off
after a while the first few times, and checked that all the cables
were plugged in etc. We even thought that we would have to send it
back as defective, but luckily before we did that, we once left it on
long enough that we did get signs of life.

Also, a single process under Linux can zero the memory on that machine
at 1.4GB/s, so the zeroing could be done in 17s, but the BIOS takes
around a minute.

- anton
--
M. Anton Ertl Some things have to be seen to be believed
anton@mips.complang.tuwien.ac.at Most things have to be believed to be seen
http://www.complang.tuwien.ac.at/anton/home.html

> Also, a single process under Linux can zero the memory on that machine
> at 1.4GB/s, so the zeroing could be done in 17s, but the BIOS takes
> around a minute.

Yes, I don't know why it takes so long - but that's how it's been on every
ECC motherboard I've ever used. "Back in the day", when you'd get a machine
with lots of sloooooow memory, I saw it take as long as three minutes.

On Sun, 10 Feb 2008 16:59:44 -0700, Steve Wolfe wrote:

>> Also, a single process under Linux can zero the memory on that machine
>> at 1.4GB/s, so the zeroing could be done in 17s, but the BIOS takes
>> around a minute.
>
> Yes, I don't know why it takes so long - but that's how it's been on
> every
> ECC motherboard I've ever used. "Back in the day", when you'd get a
> machine with lots of sloooooow memory, I saw it take as long as three
> minutes.

Chances are they always run some sort of memory test. As you pointed out
you have to initialize ECC memory before you uses it, if you don't you
will get ECC errors when read uninitialized locations. Since you have to
write every location once a quick bring up isn't possible so you might as
well do at least one read back pass which means that the minimum bring up
time is 2X the time it takes to access the entire memory. However once
you've committed to doing a single pass of a memory test you might as
well do two, writing a checkerboard on one pass and then writing it's
complement on the second pass so that you checked every cell with both a
1 and a 0. Two passes is 4X the time it takes to access the entire memory
which corresponds to Anton's observation about the boot time on his board.

"Steve Wolfe" writes:
>> Also, a single process under Linux can zero the memory on that machine
>> at 1.4GB/s, so the zeroing could be done in 17s, but the BIOS takes
>> around a minute.
>
> Yes, I don't know why it takes so long - but that's how it's been on every
>ECC motherboard I've ever used.

Apart from this board and some RISC-based servers, we have not noticed
a machine doing nothing noticable for a significant time after turning
it on, and some of them have 8GB of ECC RAM, so when this board did
not do anything for a long time, it came as a surprise to us. Anyway,
I just wanted to pass the lesson on that sometimes it takes a while.

- anton
--
M. Anton Ertl Some things have to be seen to be believed
anton@mips.complang.tuwien.ac.at Most things have to be believed to be seen
http://www.complang.tuwien.ac.at/anton/home.html

On Mon, 11 Feb 2008 16:05:34 UTC in comp.os.linux.hardware,
anton@mips.complang.tuwien.ac.at (Anton Ertl) wrote:

> Apart from this board and some RISC-based servers, we have not noticed
> a machine doing nothing noticable for a significant time after turning
> it on,

My old Tyan S2466N-4M used to do this. Also ECC memory and took about 2 mins to
show you the BIOS splash after startup and by the time it did, it was too late
to press a key to get inside the CMOS setup.

--
Trevor Hemsley, Brighton, UK
Trevor dot Hemsley at ntlworld dot com