Hi John Walliker; Re "A long time ago when we were discussing relative power consumption, you insisted that notebook PCs would not need parallel termination to achieve acceptable signal integrity."
Go back and look through those links again. There was an example of a DDR bus with nothing but series resistors, and it looked great. The clincher is that it was done point to point, and used reduced drive power.
Re: "Data from Micron shows that at high data rates the power consumption of PC2100 DDR and PC800 RDRAM are almost identical. I suspect that if the RDRAM was operated at 60 ohms bus impedance (which is allowed for soldered chips not mounted on RIMMs) RDRAM would be considerably better."
Actually, the interface is a fairly small portion of the total power consumption. And you can also tweak the DDR interfaces. For instance, they are using x8 chips in the desktop calculations and x16 chips in the mobile. Power consumption figures for x16 and x32 chips would have been about 2/3 of the ones published.
Transmeta has signed up a lot of big guys to make DDR based mobile PCs. RDRAM was there before DDR, but still has no mobile wins. Any explanation?
I think it's simple. On the charts of power versus bandwidth, ignore the high bandwidth ends of the chart, where DDR and RDRAM have similar power consumption. The vast majority of mobile computers spend the vast majority of their operating time either idle or with relatively low memory bandwidth use. The same can be said of desktops. Modern cache memories prevent memory from having to provide full bandwidth all the time. That regime is where DDR kills RDRAM, and that is the important region. The regime of maximum bandwidth is great for a few synthetic benchmarks, but it has no use in the real world.
RDRAM would like to live in a world where memory is either ignored (and is in nap mode) or is in full bandwidth situations. But that is not how modern processors use memory. Processors tend to use memory enough to prevent its placement in full nap mode, but then only sporadically. RDRAM ends up spending a lot of time in Active, rather than Nap, and that is where its power consumption is worst, compared to DDR.
In the real world, actual memory bandwidths are around the 100MB/sec, not 1GB/sec, and that is a region completely dominated, as far as power efficiency, by DDR. Hence, no RDRAM wins in mobile computing. Now, let's hear your explanation for this paradox, plenty of DDR design wins in mobile, but even Intel cancelled its RDRAM mobile project.
-- Carl |
