There's just no point getting this flash new RAM until the timings come down to something reasonable. It was the same with DDR2 - it was only when AMD started using it and C2D came out that it got any faster than DDR1.
My DDR2 at 900Mhz on a DS3 can do 7200MB/s in that test - and most of the difference is due to the better memory controller on that Bearlake chipset which gives 15% extra bandwidth.
2ghz ddr3 at cas 10 is the same latency as 1ghz ddr2 at cas 5, but with twice the bandwidth and better prefetch/buffering.
Quite whether an Intel FSB based system will benefit from additional bandwidth when the FSB is already fully saturated with ddr2 at lower than maximum speeds, is another question altogether.
Looks like a Nehelam timeframe purchase is looking good though.
Thats the FSB strangling the usable bandwidth. Once we have onboard DDR3 memory controllers/Higher FSB's/Integrated graphics DDR3 chipsets we'll see the real benefits of the added bandwidth.
I know the actual performance isn't that great....but its still nice to see the numbers And once dd3 becomes a bit more mainstream the timings will go down and frequencies will go up and it WILL be much faster
BTW this is kind of random, but I'm just curious...what kind of memory latencies does ddr3 and ddr4 have on most GPU's? I would imagine they are quite loose timings...why are loose timings not so much of an issue with graphics cards?
Thats the FSB strangling the usable bandwidth. Once we have onboard DDR3 memory controllers/Higher FSB's/Integrated graphics DDR3 chipsets we'll see the real benefits of the added bandwidth.
If thats all you get at 2000+Mhz FSB, PC3-16000 ( !), then there really is no point in DDR3 untill Nelhaem comes out with an intergrated mem controler, AMD release a DDR3 chipset, or latencies drop significantly.
My point was that with all that extra bandwidth that is currently untapped by the p35, the g33/g35 will be an application where ddr3 will see tangiable benefits over ddr2.
which begs the question why someone would spend a fortune on DDR3 ram to go in an intergrated graphics chipset rather than getting DDR2 and a discrete gfx.
Dedicated graphics will always perform better, but integrated graphics setups will benefit greatly from the move from ddr2 to ddr3. The problem of high initial ddr3 cost will of course change much like it did with the transition to ddr2. By the time Nehelam is launched we will probably see not only higher performing mature DDR3 but also price parity.
Intel are going to be pushing the move to ddr3 harder than they did the move to ddr2, and if the volumes grow faster the price will drop quicker.
At the moment If I were building a system I would have course go with DDR2 simply because of the price difference for similar performance, but with the G33/G35 launch there will be a tangiable performance benefit of going to ddr3. I doubt the benefit would mitagate the additional cost, but as the costs drop and the performance benefit increases as ddr3 matures it will be hard to justify not going with ddr3.
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.