It's like when we switched from DDR2 to DDR3!
800MHz is the switching frequency, where both are avalible at that speed.
Which is faster, neither, but DDR3 made you feel better!
I have a machine running DDR2 and the Intel rating for every chip that runs in the board Is 667MHz, except the extreame chips!
My board has DDR2-800 support.
What does that mean? When I bought it I got the lowest latency 667MHz I could find(800MHz cost 2 to 3 times as much) and overclock it, because I know the board will run 800MHz easy.
Everyone running DDR2 machines before DDR3 came out was overclocking into DDR3 speeds long before DDR3 made it into the public hands. Venders (Gateway,Dell,etc.) were making a killing selling "New DDR3" machines to everyone who didn't know any better, running at 800MHz. The memory bandwidth was worse or the same as older DDR2 machines but the potential was there. Same thing now with PCI-e 3.0!!!
Sandy bridge platforms have 20 PCI-e lanes at 2.0 bandwidth , normally you run 16 for the graphics card and 4 left to run I/O on the board.( add in SATA/USB/AUDIO/etc chips) When you run 2 graphics cards you need lanes to feed the second one, 4 isn't enough. So you have to devide the first 16 in half 8x/8x, and leave the other four to run your board. If current graphics card were using the full PCI-e 2.0 bandwidth there would be no point to add in a second card, as each would only get half the bandwidth avalible, you would be better off with one good card using all of it.
Sadly, current top end graphic cards just barly use above half of the 2.0 bandwidth, thats why you barly see any difference when you cut their badwidth in half and run them at 8x.(they couldn't use it all anyway) If they could use it all, than you wouldn't see any performance scaling when you added in the second card, because the first was using it all up!! The PCI-e 2.0 spec would become the bottleneck and nobody would have more than one graphics card.
Sandy bridge-E CPUs, is basically like having 2 sandy bridge CPUs on the same die. 2-banks of cores, 2-dual cannel memory controllers, and 2-20 lane PCI-e controllers(now 40 lanes or double the bandwith, which techniclly could be called 3.0)
3.0 specs call for a doubling of the bandwith. Sandy bridge-E does have double the band with, but it devided it out into more lanes, not adding it into doubling the current lanes bandwith.
When Intel made the SB-E chips there were no motherboards or devices that were 3.0. Intel can't claim to have 3.0 when it doesn't exsist yet, and no way to varify it. So no 3.0 claim anywhere in Intel's specs, and there probably won't be until Ivy bridge-E.
Now that SB-E is out and we got better graphics cards, we can push past 2.0 spec a few ways.
1st- we have 2-2.0 controllers on die, no singe card out can use up all of that bandwith of 16-2.0 lanes , so we can't do it with a single card..... 2 cards, maybe but it's more like the oveclocking DDR2 and calling it DDR3 thing, We theroretically could push past 2.0 spec, but probabably are not doing it with 2 modern cards each on 16-2.0 lanes(Aka 32 lanes of 2.0 could be called 3.0). If we were at 3.0 we would see no more scaling buy adding in another card, but we could be above 2.0 spec if each card pulled 50.1% of 2.0's avalible bandwidth, and that would be the first way to get into 3.0 spec, using all 2.0 equepment and never calling it 3.0 like Intel is doing.
2nd- way add more cards and it gets easier because each card has to use less bandwith.. with 3 cards each one only has to use 33.334% of 2.0's avalible bandwith, which the board would be running 16x/8x/8x. One full 16 lanes of 2.0 and 16 more devided in half , each with a half the bandwidth. We all know that only the highest end cards are even affected by dropping down to 8x and it's only a few percet drop in performance.(which means thet were only drawing a few percent above half of 2.0's specs in the first place) So if each card uses 50% of 2.0's avalible bandwidth, with 3 cards your 50% above 2.0 specs. (you can say thats 3.0, but we still did it with all 2.0 stuff)
With 4 cards it gets easier and we don't need as good of cards , they would only need to use 25.1% of 2.0's bandwidth, even though they will each have 50% avalible to them to them at 8x/8x/8x/8x
3rd- way we get 3.0 which I would call a true 3.0, Is having a BIOS update to take a CPU with 2-2.0 controllers and have then dump 32 lanes of bandwith out on 1-16x slot of a 3.0 ready board and have a card that could use it in a single card configuation. Only thing we are missing now is the graphics cards. The current 3.0 cards can't do it! If they could an easy way to test is stick one in a 1155 machine at 16x and run the benchmarks, add in another, if performance scales the first one wasn't even using 50.1%.
With the cards climbing in performace the way they are now the next generation of cards will be using more than 50% of 2.0. They may push close to 75% and will give great performance ganes to all of us with 1155 machines(goning from less than 50% to over 75% still on 2.0)
But what about people who are using 2 cards 75% x 2 = 150%. They would get a great boost with the first card using 75% of the 2.0 bandwidth, but when they add in the 2nd card, the motherboard would cut the bandwith in half and feed it out to each card and you would starve each card with 50% when they each want 75%.
This is what Ivy bridge is gong to fix for the 1155 guys. If we take 2.0's bandwidth and double it, (to 200%) and get them all in out hands , some just by putting 3.0 on the box, before the ultra hot next gen cards are avalible. You will get your new board and IB CPU and throw in your old 2.0 card that struggles to use 50% of 2.0's bandwith( 25% on your new 3.0 board) and won't notice a change(and blame the card), but you will be super happy with your 10% increase in cpu power and the thought of a big gold star with 3.0 on it stuck to the side of your case. Later when you upgrade to one of the next gen cards (next year)capable of 75% of 2.0's bandwidth(37.7% of 3.0) you will be reading about the peformance of people running 2(150% of 2.0/75% of 3.0).
If they didn't bring out 3.0 on 1155 and kept us on 2.0, you would be reading reviews about how awsome the new cards are if you run just one(75%) and how horrable they scale when trying to run two(trying to run150% chooked down to 100%)
But the problem is How do they get us to buy all new stuff that we don't need now and can't use, until we buy more stuff???..........put 3.0 Support on the box! wait a year than sell us more stuff to make it work.
But remember when they double Ivy bridge's 8x lanes up to full 16x bandwidth , Ivy bridge-E will have 2 controllers on die and we will start the above all over, but IB-E will be pushing 4.0 speeds at 8x/8x/8x/8x 3.0 lanes equivalent to currennt 16x/16x/16x/16x 2.0.
