The Inquirer-Home

Qimonda DDR3 works fast at low voltages

First INQpressions DDR3-1900 at stock 1.5 volts!
Mon Sep 29 2008, 10:52

THE FOCUS ON MEMORY overclocking was, up until now, how to get the maximum bandwidth at minimum latency at the highest workable voltage. Since, on Intel platforms, you could independently set the North Bridge and DDR memory controller voltage - often way higher than the CPU one - there was no real limit on pumping up the DRAM volts as long as the modules could take it.

So, on DDR3, it wasn't uncommon to see 2 volt memory despite the official voltage only being 1.5 volts. Not exactly a healthy setting, but yes many quality modules can run this way for months at least.

alt='qimondaxtune'

However, the Nehalem CPU generation, despite its separate memory controller voltage planes, can't seemingly drive memory at the same high voltages. Also, as the latencies are significantly reduced by the internal memory controller, it is not so critical to squeeze every bit of DRAM latency settings this time.

So, while the frequency and associated bandwidth are still critical for both module branding and those synthetic benchmark results, the ability to reach good performance at stock voltage becomes more important. You gain less heat, power consumption and stress on the CPU memcon voltage, while aiding overall system reliability.

Qimonda was the 'dark horse' to steal the limelight here, with their Aeneon XTUNE modules. Seen first during the IDF San Francisco, these little sticks provide default DDR3-1866 CL10 at 1.5 volts only! Intel's chief overclocker, Francois Piednoel, used these modules in his Nehalem Bloomfield extreme desktop demos. Here we have a look at the standard flavour, dual-channel kit.

alt='qimondarampagemobo'

Before we run them on our Nehalem test bed, I gave the two 2 GB DIMMs a quick boot shot at two of the best mobos around from Asus: Rampage Extreme using the X48 chipset, and Nforce 790i Ultra based Striker II Extreme. Both can do above FSB1800 on quad-core Penryns without much tweaking, with command rate 1, the fastest possible for these two sticks. I used the newest BIOS updates from the Asus download page as well.

alt='qimondastrikermobo'

Both mobos used the Intel QX9650 CPU running at 8X multiplier and FSB1866, i.e. 3.73 GHz default speed. The purpose of the test was to get the best initial stable setting that boots Vista64, yet keeping the voltage memory setting at the 1.5 volts all the time (1.51 volts on the Rampage, as that was its lowest supported voltage). No benchmarks this round, it will be in the next update where we fiddle with the voltages too.

alt='qimondarampagebiosvolt'

As expected, the default DDR3-1866 CL10-10-10-24 setting worked fine on both boards, the only (non critical) issue being that this SLI-ready memory EPP parameters were disabled by the Striker II Extreme BIOS.

alt='qimondarampagebiossetting'
On the Rampage board, I managed to get a noticeably better CL9-9-9-20 latency setting work as well at DDR3-1866 and boot Windows too. For 1.5 volts, this is a great deal. I'd say this by itself would be more than enough for any Nehalem configuration, as it'd provide over 40 GB/s theoretical memory bandwidth on a Bloomfield, with respectable latency at stock voltage.

alt='qimondastrikerbiossetting'

The Striker II mobo fared a little worse at the latency front, enabling " only" CL10-9-9-20 - however, it managed to do that at DDR-1900 and FSB1900 though, all again at stock 1.5 volts. All this without any major tweaking yet.

alt='qimondastrikerbios'

The DIMMs, with their plain vanilla heat spreaders, felt barely warm to the touch. Keep in mind it was a very hot pre Formula 1 afternoon in Singapore, with my home thermometer measuring 36 C - yes, it shows the wrong date, but its temperature estimates are trusty up to now.

The test environs had airconditioner turned off and no assisted ventilation at the mobo besides the CPU fan. The final benchmark session will of course have some of that stuff in, but even then these initial results are pretty awesome.

alt='qimondatemp'
Now, other makers like Elpida are jumping on the same bandwagon - the Japanese maker was talking about DDR3-2500 speed grade at the 1.5 v stock, and I believe Qimonda will be able to have DDR3-1600++ speeds soon at the new power saving 1.35 volts, far closer to the CPU voltage levels isn't it? Not to mention the power savings when six of these DIMMs work together as 12 GB for a Bloomfield on, say, Asus P6T Deluxe X58 mobo.

In summary, a good set of modules that give very decent performance with lower voltage and overall power usage. Look forward to our O/C benches on these - Qimonda will have more news at this front soon. And yes, they do have the 3-channel triple module Bloomfield packs coming in a few weeks, too. µ

Share this:

Comments
Oh,NO There Goes Tokyo.

Memory can be corrupted by anything more than minimum voltage. If it plays at 1.5v, that is overclocking in way, its just better. If actually turning voltage up is faster with 100% of memory still working, good play.

40 gb/s is greater, that may be real final bottom line, here.
drashek

posted by : Oops, 29 September 2008 Complain about this comment
aboutus
Advertisement
Subscribe to INQ newsletters
Advertisement
INQ Poll

Authorities in several countries raided Megaupload recently, shut down all of its services, seized hundreds of servers and arrested several of its executives on criminal charges.

Do you think the move was justified?