DDR3 comes shy into the scene

[Micutzu]

No, it's not like with the CPU's. Not even like with the VGA's. When a new type of RAM appears it's usually too expensive and can't even match the top performers of the previous generation. No boom, no hype, the next gen just enters very shy into the market and as the volume increases and the technology develops it just moves upfront and replaces the old stuff as a performance leader. So don't expect to get your hats blown off in the next few posts because it won't happen.

So, what are we exactly talking about ? Some kind of comparation, as fair as it gets, between DDR2 memory and the DDR3 newcomers. The testing platforms ... will definetely not be limiting the results , for DDR3 the Asus P5K3 Deluxe proved to be a good match and DDR2 was paired with one of the few i965 motherboards that can keep up with the P35'ers, the abit AB9 QuadGT.

Setup

DDR3 platform: Asus P5K3 Deluxe + Kingston KVR1066D3N7K2/2G (Elpida chips)
DDR2 platform: abit AB9 QuadGT + TEAM Xtreem TXDD2048M1200HC5DC (Micron D9GKX)

Intel E6700 + Asetek Vapochill LS
EVGA 8800Ultra SuperClocked
Seventeam ST-600EAD


Right, the DDR3's are lowend and DDR2 and top notch performers ... even if the prices would fool you. Relax, i didn't walk the Xtreem's into the park with no muzzle and leash, the settings i've used are reasonable for any average DDR2.

The tests won't make things clear for everybody, i can tell you this right now, especially as i've mostly ran sinthetic benchmarks; to be honest, this is the best way you could see a consistent difference, as in real life the memory performance is not THAT important. The Lobby High test from 3DM2001 is still a very good indicator of memory performance (mostly latency), and SiSoft Sandra Lite XI SP2 and Everest 2006 are already too well known for me to say anything else about them. The CPU ran at two different frequencies (3.6Ghz and 4.0Ghz) just to point out any scaling; VGA drivers were 158.19 set on performance and with LOD = 4.9. VGA clocks were the default 8800Ultra SuperClocked ones, 655/1125, and the PCI-e bus was at 120Mhz.

The timings for DDR3 can be seen in the images below, and for DDR2 i've used 4-4-4-10 2T with tRFC = 20.

[Micutzu]

To get things started, i've chosen the bandwidth benchmarks from SiSoft Sandra. Both Integer and Floating graphs show a very slight advantage for DDR2.

[Micutzu]

Sandra - Cache and Memory Index - memory + cache subsistem performance, higher is better. DDR3 catches up.

[Micutzu]

Both Sandra and Everest have latency tests, and even if the values are different the graphs look very similar. 2 juicy points for DDR2, wich can also run much tighter .

[Micutzu]

Everest's memory read and write tests are very similar to the bandwidth ones from Sandra, but with less of a synthetic bench stink. Even if the performance seems to vary alot depending on the settings, the two types of memory are very close overall, forcing me to give them equal points. As we're getting close to the end of this match, the score DDR3 vs DDR2 is 3 to 5.

[Micutzu]

The Everest Memory Copy test looks a bit weird, even if i guarantee the results are as consistent as they get. I don't really understand the idea that performance gets worse as the CPU clock increases, even if there might be a technical reason somewhere. Anyway, on average the two setups are again on par, keeping the 2 point lead for DDR2 before the final test.

[Micutzu]

3DMark2001 - the best benchmark ever. I even use it for HDD testing ... by copying it from one partition to the other .

Seriously, 2k1 remains the best bench all-round, the whole secret being in the result interpretation. Here we mush watch very closely the Lobby High Detail test, but i will show you all the scores so you can get a better idea.

[Micutzu]

First, 3.6Ghz CPU testing.

[Micutzu]

And 4.0Ghz testing, for your enjoyment.

[Micutzu]

So far it seems very simple, DDR2 is better. Really ?

Not quite. DDR3 just entered the market and offers a few advantages over DDR2. Lower consumption, higher density and higher performance potential, will all will converge in better performance per price in the end. It is a step ahead and as soon as DDR3 will gain a little momentum it will show what it can really do. The transition is pretty easy, the P35 chipset didn't seem imature at all in the few days i've played with it so far and the Asus implementation is very good.

It's not the moment to jump all over DDR3 as the the price/performance ratio is simply sad for now, but the platform has alot of potential and I can bet that we will have some reasonable DDR3 available from September or so. Cheers !

[kiwi]

Thanks

Any chance to see DDR3 32M @ 3.6Ghz to compare?

[Micutzu]

I've sent the samples to a magazine for review so i won't have them for a week or so. Maybe i'll get something better in the meantime, KHX stuff, who knows.
From the feeling i got so far, it should be pretty close to DDR2 in 32M ... 2k1 loves latency and it's still reserved to DDR2, but i think good DDR3 can put up fairly decent scores in 32M.

[Anemone]

Nice tests

A little observed tidbit on the 3 vs 2 issue is that when you bump the memory up to 4gb, either via 4x1 or 2x2, the DDR2 controller begins to get beyond it's comfort zone and many folks seem to end up in the 800-900 range at tighter timings, unable to get to the higher frequencies at such densities. That's generalizing, and not meant to be "everyone's" experience.

DDR3 can hit 1333 with more memory enabled, and probably it can do better than that. So the comparison here I was looking at was 1333 CL6 vs 800 CL4 which is a fairly normal 4gb to 4gb comparison.

Only intended this as food for thought, for those who are actually populating their systems with 4gb

[over@locker886]

Thank you tests Micutzu!

Good comparison, because I think Abit now one of the fastest mobo with ddr2 platform. I hope so can test some ddr3 soon

[saaya]

thx for sharing Micutzu