The Inquirer-Home

DDR2 vs DDR3 tested head to head

First INQpressions Asus mobos, Geil RAMs, and of course Intel CPUs
Tue Jun 12 2007, 19:57
WE HAVE run quite a few stories over the past months, covering the old DDR2 vs new DDR3 debate.

Beside the obvious-cum-obnoxious price difference affecting the newbie, we also learned that the latencies still have a lot of room to improve and that, despite the low voltage, the initial DDR3 DIMMs are just a hot to touch as well-overclocked DDR2 ones would be.

Put aside all that, how about a side-by-side comparison with as many things identical as possible, just the RAMs different? As the Computex trip ended with some good Geil DDR3 samples in the bag, the last missing piece of the puzzle was in - time to do it.

The configurations truly was an apples-to-apples one: Intel Xeon 3220 quad-core desktop Xeon running a 3.2 GHz with FSB1600, Asus P5K mobo for DDR2 vs P5K3 for DDR3, both Deluxe, both with newest BIOSes and both practically identical save for the memory, as well as Geil DDR2 high speed kit vs Geil DDR3 "whatever we got in hand" kit. Geil's RAMs are, together with Corsair's, the fastest generally on latency runs - soon we'll be checking the new, promising Gskill and OCZ parts too.

In the copper corner, DDR2
In the black corner, DDR3

Both configurations ran one same WinXP SP2 and WinXP 64 from a dual-boot PATA HD used for testing. The graphics card was an Asus ATI X1900XTX. I ran both Sandra XI and Everest Ultimate benchmarks, comparing both bandwidth and latency measured in either case.

But before that, BIOS tuning came up. For the DDR2 portion, Geil DIMMs ran perfect at CL 3-3-3-5 DDR2-800 speed at 2.15 volts, while, for the same vendor's DDR3, we managed to get CL 7-8-7-14 at DDR2-1600 at 1.7 volts. So, latency-wise, in pure nanoseconds, DDR2 would win that game outright for these parts. Does it translate to the benchmarks?

Here are the results:

CL 3-3-3-5
CL 7-8-7-14
Sandra Mem BW int MB/s
Sandra Mem BW FP MB/s
Sandra Mem Latency ns
Everest Mem Read MB/s
Everest Mem Latency ns

As you can see, DDR3 won over DDR2 between 3% and 19% in the results, more so in the bandwidth tests - somewhat expected, as after the first word, the rest of the words in the burst ticks at twice the rate, and shortest burst rate is now 8 transfers, compared to 4 in DDR2. This brings out the DDR3 bandwidth benefit further.

This also validates my point from the initial P35 memory benchmark - that the new memory controller is optimised to hide high latencies and make the best possible use of high spare bandwidth. Therefore, it won't work well with low-latency standard in-sync bandwidth DDR2 memories anymore.

Also, even after all the speedups, neither can still match or exceed Nforce 680i with DDR2-800 memory at same CL3-3-3-5 settings on the same FSB. Nvidia's memory controller still seems superb - what will Intel do in the X38 then to match it?

Finally, the DDR3 DIMM at 1.7 volts was noticeably hotter than the DDR2 DIMM at 2.15 volts, both Geils using the same standard heat spreader type. Geil usually has pretty good choice of dies (Micron in both cases here) and, even without going very high in voltage, I was able to push them to these pretty decent performance levels. But again, Corsair, OCZ and G.skill can pretty much do the same.

In summary, DDR3 is beginning to mature - this is still not such a bad latency, after all, and the results do benefit from it. Now, if just the price improves downward, as fast as the latency does. ?


Share this:

blog comments powered by Disqus
Subscribe to INQ newsletters

Sign up for INQbot – a weekly roundup of the best from the INQ

Existing User
Please fill in the field below to receive your profile link.
Sign-up for the INQBot weekly newsletter
Click here
INQ Poll

Microsoft Windows 10 poll

Which feature of Windows 10 are you most excited about?