Have Q9550 & DDR2-800, which motherboard?

Which Gigabyte motherboard would you buy if you had a Q9550, four 1GB
modules of DDR2-800, and a GeForce 7950GT/500, and an SSD main drive?

Would you ditch the DDR2 memory and go for a GA-EP43T-UD3L and buy
DDR3-1600 memory modules?

I play Forged Alliance and the upcoming Supreme Commander 2, and do
some flight simulation FS 9/10.

Thanks.

John Doe wrote:
Which Gigabyte motherboard would you buy if you had a Q9550, four 1GB
modules of DDR2-800, and a GeForce 7950GT/500, and an SSD main drive?

Would you ditch the DDR2 memory and go for a GA-EP43T-UD3L and buy
DDR3-1600 memory modules?

I play Forged Alliance and the upcoming Supreme Commander 2, and do
some flight simulation FS 9/10.

Thanks.

You can buy DDR3-1600 memory modules if you want, but you should look at their
tested memory table first. The DDR3-1600 entries are all single sided
modules. I don't know if they did that, so they could tick all the
columns in the table ? Or whether they really tried 2x2GB DS modules
(one stick per channel) and it didn't work ? In any case, the implication is,
that operation at DDR3-1600 is near the limits.

http://www.gigabyte.com.tw/FileList/...ep43t-ud3l.pdf

As for memory speed and its effect on performance, the bottom table here
gives a small sampling. From DDR3-800 CAS5 to DDR3-1600 CAS6, the
benchmark time goes from 40.95 seconds to 39.16 seconds. That is a
4.6% difference caused by the memory alone (as they're tweaking FSB
and multiplier for constant core speed).

http://www.anandtech.com/memory/showdoc.aspx?i=3053&p=5

So while you asked about DDR2 versus DDR3, the results in that
table should give you some idea what changing the memory would do.
DDR3-800 CAS5 is industry standard, and isn't even an enthusiast
memory. While DDR3-1600 CAS6 is bleeding edge, and it was the lowest
CAS I cound find on Newegg for that memory.

Memory has both latency and frequency as a consideration. Low latency
memory could add a bit to the purchase price, and it is a factor too.
So when you ask about DDR3-1600, is it CAS6 or CAS9 ? CAS6 would be
better.

In addition to the basics, as shown in that Anandtech article, there
is another article here. This discusses tuning and the difference a
good BIOS can make. The GA-EP43T-UD3L board does have a BIOS
page with tRD in it. And the "(G)MCH Frequency Latch" appears to
be support for setting the "strap".

http://www.anandtech.com/mb/showdoc.aspx?i=3208&p=1

I can't even tell you, looking at that article, what you should
buy now. Using the Auto settings, the BIOS will set tRD according
to whatever memory you install, so you don't actually have to do
any math to use the motherboard. The main purpose of that Anandtech
article, is if you wanted a math basis for making a selection.

The lowest latency DDR3-1600 memory I could find, was CAS6.
It meets those specs, at 1.65V (nominal is 1.5V), so the RAM
doesn't need to be punished to do well. 2x2GB $170. A faster
memory should be able to run at a slower speed, if the DDR3-1333
setting is needed.

http://www.newegg.com/Product/Produc...82E16820226120

The fastest memory you can get now, is this DDR3-2400 CAS9. The latency,
when scaled for frequency, is the same as the previous product (at
DDR3-1600 it would run CAS6). The interesting question, is what computer
a memory like that would make sense on ? What FSB and dividers would you
need to hit that speed ? Another cute aspect of this product, is
it comes with its own cooling fan. I wonder how long that'll last.

http://www.newegg.com/Product/Produc...82E16820231338

Paul

Paul wrote:
As for memory speed and its effect on performance, the bottom table
here gives a small sampling. From DDR3-800 CAS5 to DDR3-1600 CAS6,
the benchmark time goes from 40.95 seconds to 39.16 seconds. That is a
4.6% difference caused by the memory alone (as they're tweaking FSB
and multiplier for constant core speed).

http://www.anandtech.com/memory/showdoc.aspx?i=3053&p=5

That particular benchmark is very dependent upon memory bandwidth.
In your average real world application it probably matters not as much.

Recently, I swapped two 1GB sticks of DDR2-667 for the one 2GB stick of
DDR2-800 I had put in my mom's computer. The SuperPi Mod 1M time
increased by nearly ten seconds! Apparently the old memory had been
running asynchronously at 400 MHz and I set the new stuff for just 200 MHz,
which was synchronous for this CPU (E2200). When I set it to run at 333 and
lowered the CAS latency, most of the performance came back. I never
imagined it would matter that much. This is on a single channel ECS board
(671T-M) with a SIS chipset that came essentially free with my Q6600 at Fry's.