Re: AMD MP, SMP, Tyan 2466

Timothy Miller (miller@techsource.com)
Wed, 25 Jun 2003 19:26:15 -0400


Edward Tandi wrote:
> On Wed, 2003-06-25 at 23:59, Timothy Miller wrote:
>
>>
>>DDR memory works very much like single data rate, except that data is
>>transferred (in whichever direction it's going) on both edges of the
>>clock, thus doubling the transfer rate. The memory does not switch
>>between reading and writing as you describe it.
>>
>>I believe registering is for reliability. Data is transferred one clock
>>cycle later but reduces signal loading.
>
>
> Thanks for the clarification. I do not profess to be an expert in the
> technology. Two writes or a read+write per clock cycle is close enough
> for the purpose of the discussion.
>
> The point I was trying to make is that the registers are there to deal
> with an SMP race condition of some sort. Athlon MP motherboards fitted
> with two processors will not work properly without 'registered' RAM. I
> have hard experience of this and it this experience I am sharing with
> someone who is seeing the same symptoms.

It is my understanding that the registered memory requirement has
nothing to do with SMP but instead with the amount of memory you have.
The more memory chips you have, the greater the signal loading on the
memory bus. More input drivers means more capacitance which means you
need your output drivers to put out data sooner (relative to the clock
edge, so registered delays by one clock) and stronger (greater drive
strength).

In an SMP system (besides NUMA), multiple processors will talk to the
same memory through a shared memory controller (like in a Northbridge),
so although there are multiple processors, there is still only one
memory bus. Pulling off one CPU isn't going to change that situation.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/