A8V Deluxe memory advice needed

[formerprof]

I have an A8V Deluxe & FX-55 processor. It runs very well with 2 512 mb
Kingston RAM chips. I also have two Corsair 512 mb sticks with timings
identical to the Kingstons. They also run perfectly. I've made no BIOS
changes other than to run at 1T (i.e., "disable" the 2T setting). The board
revision is 2.0.

What I can't seem to do is use all four RAM sticks at the rated speed. When
I install all four the board throttles down from DDR 400 to DDR333. If I try
to set the memory speed manually (as ASUS web site suggests) for DDR 400 I
start to get memory errors.

It's driving me crazy. If anyone can help me with these questions I'd be
most grateful:

1. Has anyone solved this problem & with what settings?

2. How much of a performance hit do I take for DDR333 as opposed to DDR 400?
I do a lot of continuous number crunching and some gaming.

3. If I can get all the memory working how much of a performance hit do I
take if I have to drop back to 2T?

All suggestions gratefully received.

formerprof

(e-mail address removed) (change the "x" to an "h" for e-mail)

 

[Paul]

I have an A8V Deluxe & FX-55 processor. It runs very well with 2 512 mb
Kingston RAM chips. I also have two Corsair 512 mb sticks with timings
identical to the Kingstons. They also run perfectly. I've made no BIOS
changes other than to run at 1T (i.e., "disable" the 2T setting). The board
revision is 2.0.

What I can't seem to do is use all four RAM sticks at the rated speed. When
I install all four the board throttles down from DDR 400 to DDR333. If I try
to set the memory speed manually (as ASUS web site suggests) for DDR 400 I
start to get memory errors.

It's driving me crazy. If anyone can help me with these questions I'd be
most grateful:

1. Has anyone solved this problem & with what settings?

2. How much of a performance hit do I take for DDR333 as opposed to DDR 400?
I do a lot of continuous number crunching and some gaming.

3. If I can get all the memory working how much of a performance hit do I
take if I have to drop back to 2T?

All suggestions gratefully received.

formerprof

(e-mail address removed) (change the "x" to an "h" for e-mail)

A wild ass guess, is that 4x512 will run DDR333 1T or DDR400 2T,
which is roughly the same penalty. Try both of those first, to
see if either option is stable. (Bump up the Vdimm to 2.65 or
2.75, so the memory gets enough juice for DDR400. It makes a
difference stability wise.)

I somehow doubt you'll manage DDR400 1T, until a better
processor design from AMD starts shipping. DDR400 1T is
probably 20% more memory bandwidth than the other options,
which would be a 6% application performance difference,
in very rough terms.

Benchmark both options, using your favorite application. If
you are a gamer, run 3DMark and have it gauge the impact of
the two options. It shouldn't take more than an hour to get
an answer.

And once you think it is really stable, run Prime95 for
some number of hours, as that will identify any remaining
weakness in the setup.

Paul

 

[formerprof]

A wild ass guess, is that 4x512 will run DDR333 1T or DDR400 2T,
which is roughly the same penalty. Try both of those first, to
see if either option is stable. (Bump up the Vdimm to 2.65 or
2.75, so the memory gets enough juice for DDR400. It makes a
difference stability wise.)

I somehow doubt you'll manage DDR400 1T, until a better
processor design from AMD starts shipping. DDR400 1T is
probably 20% more memory bandwidth than the other options,
which would be a 6% application performance difference,
in very rough terms.

Benchmark both options, using your favorite application. If
you are a gamer, run 3DMark and have it gauge the impact of
the two options. It shouldn't take more than an hour to get
an answer.

And once you think it is really stable, run Prime95 for
some number of hours, as that will identify any remaining
weakness in the setup.

Paul

Paul,

Thanks so much. With the new A8V beta BIOS (1011.005) I was able to run
4x512, as you had suggested, at 333 1T and 400 2T. Neither would accept any
overclocking. The latter was slightly better on 3dMark, but both were quite
a bit below 2x512, 1T, 3% overclock. As a result I pulled the extra sticks
out -- I don't think that much of what I do requires 2 gb of memory.

It's annoying that I can't get the board and processor to run at their
advertised speeds with all four memory slots filled. I have no doubt that if
I complained to ASUS and/or AMD I would learn from each that the other was
at fault. The prevalence of this problem (there's been much discussion on
some of the forums as I have discovered) suggests that even if AMD and ASUS
were to replace my CPU & board there would be no substantial change. The
new BIOS _is_ an improvement over 1009, but still not up to spec.

I'm most grateful for your assistance. All good wishes.


formerprof

 

[CapeGuy]

Thanks so much. With the new A8V beta BIOS (1011.005) I was able to run
4x512, as you had suggested, at 333 1T and 400 2T. Neither would accept
any overclocking. The latter was slightly better on 3dMark, but both were
quite a bit below 2x512, 1T, 3% overclock.

Prof, how much was "slightly better" and "quite a bit below"?
Are we talking 5%, 10%, 20% ???

Plus, I've always wondered - what's the big deal about a 3%
(or even a 5-10%) overclock? Seems like inviting complications
and trouble for a performance gain that won't even be perceptible
except in benchmark tests. Is it just a matter of bragging rights?

 

[Ed]

Plus, I've always wondered - what's the big deal about a 3%
(or even a 5-10%) overclock? Seems like inviting complications
and trouble for a performance gain that won't even be perceptible
except in benchmark tests. Is it just a matter of bragging rights?

Why pay more for the faster CPU when the cheaper slower one gets you
basically the same performance? As a FAB matures the later production
runs essentially put out better chips, the slower CPUs are just
re-marked faster capable cores to fill market demand.

I rem back when AMD was using Slot-A, the demand for the lower end chips
was so great AMD was putting faster cores in their place, buy a 500 and
get a 650 that would run at 850 (with little effort), The 850 was still
months away from release and was going to cost a lot more cash.

Ed

 

[aberger]

Paul,

Thanks so much. With the new A8V beta BIOS (1011.005) I was able to run
4x512, as you had suggested, at 333 1T and 400 2T. Neither would accept any
overclocking. The latter was slightly better on 3dMark, but both were quite
a bit below 2x512, 1T, 3% overclock. As a result I pulled the extra sticks
out -- I don't think that much of what I do requires 2 gb of memory.

It's annoying that I can't get the board and processor to run at their
advertised speeds with all four memory slots filled. I have no doubt that if
I complained to ASUS and/or AMD I would learn from each that the other was
at fault. The prevalence of this problem (there's been much discussion on
some of the forums as I have discovered) suggests that even if AMD and ASUS
were to replace my CPU & board there would be no substantial change. The
new BIOS _is_ an improvement over 1009, but still not up to spec.

I'm most grateful for your assistance. All good wishes.


formerprof

The problem lies with the chpset vendors, Via and nVidia. The chipsets
do not have the drive capability to drive the capacitance loads of 4
memory SIMMS at the 400 MHz double data rate. For my KN8E-DLX the limit
is 2 512M simms at 400 MHz, even though I have 3 slots available. I
could back off the speed and then get 3 simms in there.

Arnie (currentprof) Berger

 

[signmeuptoo]

So your saying that it has to do with capacitors? Or do you mean that it
is simply the fault of the chipsets, which it is, right? But also, the AMD
memory controller is part of the issue as well, so that is why people are
looking forward to the new Venice chips because they are hoping that they
will have a better memory controller...

 

[Robert Hancock]

So your saying that it has to do with capacitors? Or do you mean that it
is simply the fault of the chipsets, which it is, right? But also, the AMD
memory controller is part of the issue as well, so that is why people are
looking forward to the new Venice chips because they are hoping that they
will have a better memory controller...

The chipset has little or nothing to do with it on an Athlon 64 board.
Signal routing on the board may have some impact, but likely the
limitations of the CPU's on-die memory controller are most significant.
Unless you are running some very special RAM I don't think it's usually
possible to run 4 DIMMs at DDR400 with 1T command rate on a current
Athlon 64..

 

[Not]

Paul,

Thanks so much. With the new A8V beta BIOS (1011.005) I was able to run
4x512, as you had suggested, at 333 1T and 400 2T. Neither would accept any
overclocking. The latter was slightly better on 3dMark, but both were quite
a bit below 2x512, 1T, 3% overclock. As a result I pulled the extra sticks
out -- I don't think that much of what I do requires 2 gb of memory.

It's annoying that I can't get the board and processor to run at their
advertised speeds with all four memory slots filled. I have no doubt that if
I complained to ASUS and/or AMD I would learn from each that the other was
at fault. The prevalence of this problem (there's been much discussion on
some of the forums as I have discovered) suggests that even if AMD and ASUS
were to replace my CPU & board there would be no substantial change. The
new BIOS _is_ an improvement over 1009, but still not up to spec.

I'm most grateful for your assistance. All good wishes.


formerprof

Were is this beta Bios for the A8V Deluxe. Can only find Ver1009 on
website.

 

[Paul]

Were is this beta Bios for the A8V Deluxe. Can only find Ver1009 on
website.

You can sometimes find BIOS on the Asus Germany site, before their
official release on the download webpages. For example, I would look
here for your board:

ftp://ftp.asuscom.de/pub/asuscom/BIOS/Socket_939/VIA_Chipset/K8T800Pro/A8V_Deluxe/

Paul

 

[CapeGuy]

$6l.341830@pd7tw2no...

The chipset has little or nothing to do with it on an Athlon 64 board.
Signal routing on the board may have some impact, but likely the
limitations of the CPU's on-die memory controller are most significant.
Unless you are running some very special RAM I don't think it's usually
possible to run 4 DIMMs at DDR400 with 1T command rate on a current Athlon
64..

Robert, have you heard/read anything as to whether a fix for
this might be high on AMD's list of improvements for near-
term new Athlons?

How large is the actual computation performance penalty
that one incurs when running in 2T mode?

 

[Robert Hancock]

$6l.341830@pd7tw2no...



Robert, have you heard/read anything as to whether a fix for
this might be high on AMD's list of improvements for near-
term new Athlons?

How large is the actual computation performance penalty
that one incurs when running in 2T mode?

I've heard rumors that improving the drive capability of the memory
controller is something that AMD is working on, but I'm not sure if
anything definitive has been said on the subject.

I've seen the figure of 20% difference in memory bandwidth between 2T
and 1T command rate. The difference in overall application performance
will be quite a bit less than that however.

 

[Paul]

The problem lies with the chpset vendors, Via and nVidia. The chipsets
do not have the drive capability to drive the capacitance loads of 4
memory SIMMS at the 400 MHz double data rate. For my KN8E-DLX the limit
is 2 512M simms at 400 MHz, even though I have 3 slots available. I
could back off the speed and then get 3 simms in there.

Arnie (currentprof) Berger

S754 processor datasheet. Block diagram Pg.11
The single channel memory controller is on the processor.
Two address busses (complementary), plus one data bus.
http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/31410.pdf

One address bus drives one slot, the other address bus
drives two slots.

S939 processor datasheet. Block diagram Pg.9
The dual channel memory controller is on the processor.
Two address busses and the data bus runs in either 64bit or
128bit mode.
http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/31411.pdf

It is _not_ the fault of the chipset. Memory drive problems are
AMD's responsibility. Maybe the next revision will fix the
drive problem. When you put a couple hundred I/O on a processor
like that, you cannot use too much current per I/O, before the
switching noise becomes too great. Simultaneous switching noise
is one of the down sides of the AMD integrated memory controller
approach.

Paul

 

[signmeuptoo]

Paul, you sure are knowledable, but you read too. That takes patience.
You must have a high IQ! Are you a professor by any chance?