Printable View
Was looking at the setup on the new 1156 chips from intel, and saw the interconnects total, 16X pcie 2 lanes and a DMI bus that runs at 2GB/s
So how does the EVGA board with the NF200 chip manage tri 8x lanes?
Technically the DMI bus would max out at 4x, and that bandwidth is shared with hard drives, LAN, audio, USB and PCIE 1x slots.
Anyone have any idea?
it uses the pci-e controller (nf200) on the cpus lanes. i would recommend against getting an nf200 1156 since at that point u are at the cost ($) of the 1366 x58 but u have lost the latency advantage so the x58 and triple channel ram should be better by a good margin and you have more IO from the cpu with the 1366 and x58 than the 1156 since under sli or xfire the nf200 isnt making some sort of magical extra IO since the gpus need essentially the same IO from the cpu, but it will add IO from card to card but a native 16x16 would have the same card to card but with better pci-e to cpu IO
There is a good 3 way SLI comparison between X58 and P55 w/ nf200 here: http://www.evga.com/forums/tm.asp?m=100901346
If it uses the CPUs PCIE lanes, than triSLI/CF would actually only be running 5.33GB/s bidi? With 8x near maxed with a 4890/4870x2 I would think this would be trouble.
BTW, I have no intention of going socket 1156 now, or anytime in the immediate future. Was actually cheaper to go 1366 (with a fresh from RMA x58 platinum that is)
If this is the case, than EVGA should really put a disclaimer on their motherboard, TRISLI for 250 or earlier.
hmm no idea what cards he is running there. The images are not loading in that thread. and it was never mentioned. read the thread and all I can tell is he was NOT running 295s lolQuote:
Originally Posted by resident1509
Also a link in that thread points to a bittech article showing that there is a slight bottle necking on the p55 even in dual gpu situations. However they only ran the bench at a mesely 1680x1050, would be interesting to see if it was more or less of a problem at a more standard 1920x1080/1920x1200 res
Of course they were also using 280s with gmem speeds over the 1GHz barrier that causes 8x pcie 2.0 to be an issue (if what I read is correct)
Yes i dont believe that the total impact of the 'generic' x16 lanes wil be felt until the next gen cards are here. But three years ago i remember reading widely that two 8800 GT in SLI on a x8x8 take a 3-5 percent performance hit....no we are how much farther along with gpu's and there isnt more of an impact?? I believe there is and there are some tests being done that verify it, but its just not to a very large extent, yet. My feeling, however, is that as the gpus mature you will see this becoming a major issue on this chipset for those running sli....hell even if you put a raid card in one of the slots, it will knock your SINGLE gpu down to x8!!
minimum framerrates is where it is rearing its head, you spend alot of time there in gaming...Quote:
Then we get to multi-GPU scenarios, there is a surprisingly clear difference in favour of the Core i7 system, whose two full x16 PCI-Express links consistently provide a far higher minimum frame rate in almost every case. This is interesting because it actually highlights a subtle difference between PCI-Express 2.0 x8 and x16 that we had previously thought wouldn't be an issue - clearly the available bandwidth in very intense scenarios (where minimums occur) begins to run out on an x8 lane.
http://www.bit-tech.net/hardware/cpu...-performance/6Quote:
With DirectX 11 bringing more complex game environments and extra calculation data such as GPU accelerated physics and AI, the PCI-Express bus could become a system bottleneck by the end of PCI-Express 2.0's life.
and i agree, those tests were run at a low res!
Well, considering they are 3 Way SLI results I would hope so :rofl:Quote:
Originally Posted by Neuromancer
Images show 285s btw.
Quote:
Originally Posted by Computurd
Yup, minimum frame rate is what it is all about. I read that PCIE bandwidth was related to vmem speed. IE 1000MHz quadpumped vmem maxes out 8x pcie 2.0 (exactly half the bidi bandwidth) I do not know the veracity of this statement, it makes sense though. I just have this funny thing about not believing everything I read. Would explain why results are similar, but there is a slight differnce. You do not spend the entire time gaming maxing out the vmem, I would surmise.
Hey, I am an ATi guy, cut me some slack :ROTF:Quote:
Originally Posted by resident1509
This chipset is nice, but definitely not for those looking to do any type of dual pcie solutions. There are boards out with nf200 on them, but they only run at x16 x8, even though it is being marketed as dual x16 SLOTS!! that is misleading.
i think there may be a little bit of embellishment going on with the branding of this chipset.
If zanzabar is right, and the NF200 chip connects via the CPUs PCIE controller, than it is outright falsehood. Cant get 24 lanes out of 16 no matter how you divide it. (Hmm maybe they mean PCIE 1.0 lanes??? I guess that Math would work :p)Quote:
Originally Posted by Computurd
Seriously though, despite being nVidia only, EVGA is a great company, and I really doubt they would do something so... false.
Really want to know how this is pulled off, since from Intel's own architecture marketing, there is no way it can do more than effective 18 or 20 lanes (dunno if the DMI bus is 2GB/s bidi or not)
Perhaps, since the board is aimed at enthusiasts and they OC... perhaps upping the bclock, also increases the speed of the DMI bus and allows them to skate that line a little closer.
Bah pure speculation.
the dmi bus is slow and is the equivalent to the other inel chipsets buss from the NB to the SB u cant do much with it
If they are using the DMI lanes they are in real trouble of misleading the consumer outright, you cant pull slow lanes out and use them as a quasi-patch to claim x16.....i think that they should come forward with at least some sort of 'vague' explanation of how they did this....at least so that people dont think that they are just claiming x16, but actually have a REAL, true working x16 functionality. I have a sneaky feeling about this one.....Quote:
Finally, some have noted that the DMI connection between the CPU and chipset could be a potential bottleneck for high bandwidth accessories and devices on Lynnfield systems. At only 2 GB/s, that DMI connection could be swallowed up simply by saturating the 8 PCIe 1.0 lanes on the P55 chipset, not to mention the 6 SATA channels, 14 USB connections, etc. If motherboard companies are serious about adding things like SATA 6G support to these boards we need to be sure a bandwidth solution is worked out first!
There's already a noticeable impact when running 2 4870x2's on P55 according to anand, with rumours of the 5000 series being 20-30% higher performance I'd definitely go X58
thats comparing a real 16x16 and not an 8x8 with an emulator to look like 16x16, your missing the point of the thread for the impact of the nf200. i know that there was one on the classified e790 (x58 with nf200) and e760 (x58 without)Quote:
Originally Posted by Roch
http://www.anandtech.com/mb/showdoc.aspx?i=3519&p=4
it has no impact and lowers ocability in an sli/CF setting, if u had a gpu and a raid card then it would help since u wouldent have the hard load all of the time on the gpu and raid card so it could do something but with all gpus in sli/xfire they need the same data in the vram and have almost identical IO on the pci-e buss so it wont help at all just adds a little latency
SO you are saying that the nf200 can limit the ocability? I dont understand??
yah, the more chips and the more parts the less u can oc and they get hot. and IMO for gaming u would not want it, if u have a gpgpu farm or want to use a raid card and have slot limitations then it would have a use but for the 1156 i dont get it since it raises the cost to the point of the 1366Quote:
Originally Posted by Computurd
The NF200 is a little more than just a PCIe switch, it has some bandwidth saving features. The VGAs can exchange data thru it at 16x without having to go all the way back to the chipset.
http://images.bit-tech.net/content_i...i/nf200-1t.jpghttp://images.bit-tech.net/content_i...i/nf200-2t.jpg
http://www.bit-tech.net/hardware/mot...rce_780i_sli/1
Ultimately these features make up to the added latency of having it in the first place. So in the worst case it's useless but since you can have a 3 way SLI (or Crossfire), it does it's job on Lynnfield.
On a X58 board it's pointless, unless you just use it as a PCIe switch to have more slots (like on the P6T6 WS) or to fool the drivers in thinking that 4x GTX285 are 2x GTX295 (like on the Classified SLI 4x).
I don't think it will limit oveclockability, specially with Lynnfield. The NF200 might work pretty well with the PCIe clocked up to 150MHz.
The problem is the P55, the DMI channel is PCIe based, overclocking the PCIe bus may corrupt data.
But like zanzabar said, it's pointless to trow $200 on a P55 board just to get 3 way SLI/Crossfire, you better go X58.
Thanks for the explanation Doompc. Forgot all about this thread, and was just commenting on something similar over at OCF. Had to check for updates, and I am glad I did :up:
cause that is how you roll! :ROTF:Quote:
Originally Posted by Neuromancer