Slashdot Mirror
Samsung Develops Power-Sipping DDR4 Memory
Alex writes with this excerpt from TechSpot: "Samsung Electronics has announced that it completed development of the industry's first DDR4 DRAM module last month, using 30nm class process technology, and provided 1.2V 2GB DDR4 unbuffered dual in-line memory modules (UDIMM) to a controller maker for testing. The new DDR4 DRAM module can achieve data transfer rates of 2.133Gbps at 1.2V, compared to 1.35V and 1.5V DDR3 DRAM at an equivalent 30nm-class process technology, with speeds of up to 1.6Gbps. In a notebook, the DDR4 module reduces power consumption by 40 percent compared to a 1.5V DDR3 module. The module makes use of Pseudo Open Drain (POD) technology, which allows DDR4 DRAM to consume just half the electric current of DDR3 when reading and writing data."
In a typical notebook, how much power does memory actually consume compared to other components (CPU, HD, screen, wireless transmitter etc..)?
I just bought my Sandy Bridge rig, now they announce this?! Ffffffuuuuuuuuu-
Now, lets pair this with a ARM core and hope we get a reasonable hack that allows a wireless that does not eat power like the current ones.....
Then lets enjoy our ARM-puter: Portable, powerful, and battery for more than a day of use.
Does anyone else besides me hate that term?
PatPending (talking to friend on phone during a bash help session): It's called Pseudo Open Drain (POD) technology
Friend: Okay, I'll try that...
Friend(typing): sudo open drain
Friend: Argh! I hate this command line bullshit!
What one fool can do, another can. (Ancient Simian Proverb)
What's up with the pseudo-open drain? Is that new and exciting or just marketing speak? I know what open drain is, but how do you have a "pseudo" open drain?
Visit the
You forgot some things: in particular, the GPU and DVD
The full story: Revisiting "How Much Power Does My Laptop Really Use"?
Graph
What one fool can do, another can. (Ancient Simian Proverb)
One thing not mentioned in the article or summary is whether or not this technology reduces standby power consumption in DRAM. Under normal use, especially if you have a lot of memory in your system, the standby power consumption is going to matter as much as read/write, if not more.
Monstar L
If they decide the decreased power draw isn't as important, they could could increase performance significantly by running DDR4 at current DDR3 voltages.
I'd rather have them finally mass-produce 8 and 16 GB modules for the desktop market.
What about latency?
POD addresses this by actively pulling up at the beginning of a rising edge, then releasing the pullup to avoid a bus contention later. This reduces the termination current (at some cost in impedance mismatch, but it's already a sloppy line) and improved switching symmetry.
Lacking <sarcasm> tags,
POD by itself doesn't reduce power consumption in standby, since both POD and SSTL turn off the bus drivers then. The older POD technologies from the GDDR families use Thevenin termination, though, so the terminators draw a lot of unnecessary current when they're enabled (as distinct from the result with a dedicated termination supply.)
If you really want to know how this all works, JEDEC has the DDR4 standard available for free download. Follow the "free standards" link.
Lacking <sarcasm> tags,
Yeah, if you're looking at laptops or desktops. But when you start talking server class machines and you start jamming DIMMs in them for large database, virtualization clusters, or other high memory applications, a few watts can make a big difference. The current generation of DDR3 low power DIMMs usable with the Westmere chipsets have brought server power consumption down 30-50 watts (72 GB RAM configuration). With advancements like this, you can further increase memory density within servers and/or server density within data centers. A 5-10% improvement in server power utilization has significant financial impacts. Many data centers are power constrained not space constrained. Better utilized data centers means longer time between needing new physical space. Data center space is kind of expensive, especially in large server environments.
You measure power consumption in W, and power consumed in kWh, so your pedantic correction is also incorrect. However, assuming the resistance is equal (incorrect, but a good enough approximation), then the more Volts, the more Watts, and the fewer Volts, the fewer Watts. So you don't measure power or power consumption in Volts, but you can compare power consumption in Volts if resistances remain similar. And that's more convenient for this discussion.
Learn to love Alaska
So you don't measure power or power consumption in Volts, but you can compare power consumption in Volts if resistances remain similar.
Cool, there aren't so many of us left who still use ECL. A recent project had a single flip-flop consuming a constant 1/8 watt IIRC. A nice little one bit memory, with jitter so low you can hardly measure it. I'm pretty sure I read a data sheet on a SiGe LVPECL part with cycle to cycle switching jitter specified in femtoseconds. The beauty of constant power draw.
Aside from that, I'm not sure why we're talking about resistive-load DRAM.