There is a HUGE difference between DDR3 and DDR5 both in performance and energy savings.
DDR5 is about 5 times faster, runs cooler and is more power efficient.
The leap from DDR3 to DDR4 is very minor and is pretty much unnoticeable.
DDR5 is a BIG performance boost. That is what increases the HD4870's bandwidth over the HD4850 by such a large margin.
I read an article on the HD4870 today. I think for $100.00 more (roughly) you gain about 23% more performance with the HD4870 over the HD4850, However they still cited the HD4850 as the best bang for your buck.
I want to get one really bad but I am waiting for better Aftermarket coolers to be available for them. I want an ultra quiet quality cooler that exits the hot air outside of the case. (DUAL SLOT.)
GDDR4 had a potential for performance boosts, but GDDR5 came around before prices of higher-end chips could come down; and coupled with the advent of memory interfaces wider than 256 bits, it ceased to be the only way to get extremely high memory bandwidth, especially as GDDR3 managed to push into speeds that previously were what shipped GDDR4 were used for. (i.e, 2.0+ GHz) Were cards shipped that had the higher speeds tested used (such as around 2.8 GHz) then we would've seen it prove to be effective; yet cost reasons apparently kept AMD from pursuing that route.
GDDR5 is effectively a quad-data-rate memory rather than a dual-data-rate memory. As a result, it can lead to potentially absurd gains; 1.1ns is the SLOWEST GDDR5 comes in; that's an effective 3.6 GHz clock rate, which on a 256-bit memory bus provides memory bandwidth nearly on a par with more conventional GDDR3 setups on a 512-bit memory interface as with the GTX 280.
There apparently were plans for a GDDR5 version of the 4850, but those were apparently scrapped. People with 4870s have apparently found the RAM very overclockable; the three memory speeds being produced are 900, 1000, and 1250 MHz, and apparently, it seems most owners manage to readily push well past 4.0GHz.
1100 is the max allowed by the CCC right now, but there might be a way around that with a third party application. I've had it cranked up that high with no stability problems, normally I run it at 1000 though. Memory bandwidth is already high enough for me thanks