DDR2 800 CL5 vs 1000 CL4 vs 1200 CL5, a comparison

[Nickel020]

I've told quite a few people in the Gigabyte forum here at TweakTown that RAM speed doesn't matter much, and I finally got around to run some benchmarks to prove my point.

Here's the system I used:
Core 2 Quad Q6600 @3.6GHz (400x9)
Gigabyte EP45 UD3P (BIOS F7)
EVGA GTX 260 Core 216 55nm
Transcend AxeRAM 1200 2x1GB (Micron D9GMH chips)
I only set the main timings manually, all the others except for tRead (set to 7 for 1000 and 1200) were on AUTO and Performance Enhance was on Turbo for all three speeds.

I did some memory bandwidth tests in Everest and the difference are quite significant:

DDR2 800 5-5-5-18 (strap B)

7368 MB/s Read

DDR2 1000 4-4-4-12 (strap A)

8762 MB/s Read

DDR2 1200 5-5-5-18

9256 MB/s Read

Looks good so far, almost 2000 MB/s more in read and write for DDR2 1200 than for 800.

Now, to check if this actually translates into any real performance benefit I used the CPU benchmark in the Crysis demo (same as in the retail game). I set all settings to Very High (DX10 mode) but only benched at 1280x1024 so the video card wouldn't completely limit the results.

DDR2 800

Average FPS 32.69 (best run)

DDR2 1000

Average FPS 32.65 (best run)

DDR2 1200

Average FPS 32.84 (best run)

Although the Everest results show a big difference in memory bandwidth we only see about 0.15 Average FPS more in Crysis with DDR2 1200 than with DDR2 800.

Edit:
Here are some Far Cry 2 benchmarks for 800 and 1200 as well:

DDR2 800


DDR2 1200


We have a difference of a little less than 5% here. That's not too bad, but the RAM speed difference is 400 MHz, if you use a 4GB kit you're not gonna reach 1200 MHz unless you do some extreme stuff. So with 4GB the difference between some relaxed DDR2 800 settings and faster settings with lower timings is maybe 3% at most, not anything you'll notice in a game.

[bius]

High FSB is the big deal...

[Nickel020]

Not necessarily, a too-high FSB will need a higher tRD which has a large impact on synthetic memory bandwidth results in Everest. Whether actual application is linked to Everest bandwifht results is another question though (i.e. is the setting with the highest Everest result also the fastest one for real applications?).
Take a look at this review, it's a great read:
AnandTech: ASUS ROG Rampage Formula: Why we were wrong about the Intel X48

[Lsdmeasap]

You should also compare 444-12 vs 555-15/18 here at the same exact FSB, Strap and Memory Multi/speed as I see you have linked to here for users to check out 444-12 vs 555-12 but that is not what this post is about really considering the Strap changes and Memory Speeds and other timings that are not the same

If you have the time that is. I just came to check out your results as I saw you mention a user to come and see the results of 444-12 vs 555-15, which are not posted here.

If you do run some tests to actually compare this, I think you should use 400 FSB and the 2.00B multi/strap, and then only 800Mhz comparison of 444-12 vs 555-15 as there is a large difference I think. But it does all depend on all other settings of course, with that said leave tRD as low as you can get it as well.

When you get time of course, and if you want to add that here

If you want, I can add some DDR3 comparisons as well, maybe 1333 vs 1600 and or 666-18 vs 777-20 vs 888-24. When I have some time too

[Nickel020]

You're right, I should also include some simple CL4 vs CL5 comparisons at DDR2 800. I'll add some benchmarks for Far Cry 2 since Crysis doesn't show a difference at all. A comparison how tRD affects memory bandwidth and whether it results in an actual performance benefit would also be interesting.

I just compared 1000 CL4 and 1200 CL5 to the JEDEC 800 CL5 to show that there was little difference. The difference may also partially have been caused by me manually setting tRD to 7 for 1000 and 1200, but 7 would not boot for DDR2 800 so I set it to auto which ended up as 10.

As for a DDR3 comparison, do you have a Q6600 or a GTX 260 Core216? We should both be running the same config to see if there's any difference. I also have other CPUs and video cards here, so I could just change my setup completely to match yours.

[Lsdmeasap]

That is odd, your comment about tRD of 7 works at 1000+1200 but not at 800. Sounds like you had some voltages or other settings off maybe

Na, all I have is a few E3110's with DDR2 boards, and a X58 setup. So we could not compare our setups and the only DDr3 results I could provide would be X58 i7 920 results. And I use ATI Based cards so we could not compare any of that either....

But ya, please do post some 444-12 vs 555-12 with all the same settings other then that, that is a valid CAS Comparison, when you get time of course and if you don't mind. I would but I have my DDR2 setups all apart right now doing some bench table remodeling

[fatguy1992]

Nice comparison, maybe u could test a game that replies on RAM more?