Tobit posted a pic of the mystery board in this thread here: http://www.hardforum.com/showthread.php?p=1035127047
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
One pic says it all...
- Thread starter Tobit
- Start date
- Status
NF200
Only way to get SLI support probably.
Supposedly Intel and Supermicro will be releasing new bios's for some of their 5520 boards with overclocking functionality, we'll have to see.
Last edited:
I used to have a P6T6 which has 1xNF200 and never had any issues with it. Sure, it was too hot to touch but never had any issues from it in a well ventilated case. It does add latency, I will give you that but not an issue for most of the stuff we do. I can't think of any other way to get 4-way SLI.
The only thing required for SLI support is a BIOS that allows it, as you can see simply by looking at any one of the many P55 or X58 motherboards that support SLI without using the NF200. The main reason EVGA is using them here is for marketing purposes.
Now that's some interesting news I haven't heard of. Where did you come across this compelling information?
Ah, I even had a typo there, should have been 5520 chipset. That's what I hear. I'll fix the original post.
Well technically it's only a bios that allows it, but that would require a licensing agreement between Intel and nVidia for the 5500/5520 chipset to get SLI support, so far, I haven't heard of that happening, have you?
That's still very interesting news and a departure from the norm for both manufacturers, Skulltrail notwithstanding.
I don't know the exact details regarding how nVidia deals with licensing SLI support to different motherboards so I can't comment on that.
Mr. Pedantic
[H]ard|Gawd
- Joined
- Sep 19, 2009
- Messages
- 1,707
Either you use an nForce chipset, or you buy SLI rights from NVidia (which was done with X58 and P55). Either way, manufacturers pay a certain amount to NVidia beforehand, plus a charge (of $5 or so, I think) per motherboard sold. As far as I know, SLI support comes in the form of a key which NVidia GPUs require to enable SLI; if the motherboard has SLI enabled, the key is in the BIOS, allowing the two (or more) GPUs to recognize each other. If the BIOS hasn't got the key, SLI doesn't work, and the cards don't play along.
ya i thought the same thing. nf200 just adds more lanes because the x58 chipset wouldn't have enough lanes to support all of those x16 slots.
- Joined
- Feb 6, 2006
- Messages
- 13,984
SLI is driver related.
nVidia usually allows SLI if you use an NF100 or NF200 chip, that's they way it was originally going to work with X58, but they relented and licensed it for the chipset. I have a 5000X board from dell that uses an NF100 riser to allow SLI. Intels D5400XS board (skulltrail) did the same thing to allow SLI. If Intel isn't going to license SLI for the 5520 chipset (it does have 42 PCIe lanes natively, 36 of them 2.0), they only way to get around it would be to use the NF200 chip(s).
I'm aware of that. However, I don't know if nVidia also requires some kind of agreement with Intel for each chipset, or if any motherboard manufacturer is able to go ahead and license SLI support without needing a prior deal between Intel and nVidia for that chipset.
Well maybe EVGA will clear that up for us, I still see no reason for the NF200 chip unless it's there to enable SLI. SM already has 7 slot boards using dual tylersburg-36D (5520) chipsets, EVGA could have gone that same route, but it looks like they decided to use NF200 chips instead.
NF200 doesn't "add lanes". Its nothing more than a glorified PCI-E switch. Tylersburg is limited to 36 total PCI-E lanes, period. NF200 essentially takes a PCI-E x16 link, and splits it in two. Your cards each get an x16 connection and link, but your total bandwidth out will never exceed the input.
In simple terms, think of it like a GbE switch. Usually, you have a single uplink, and a bunch of systems connected to the switch. Each system gets a GbE connection, but you can never get full bandwidth on all systems, at the same time. Oh..and like all switches, NF200 adds a bit of latency.
Honestly, if you're going SLI, you'll probably get better performance from a motherboard that has a pair of x16 slots running at x8 electrical. Less latency, and more consistancy.
In simple terms, think of it like a GbE switch. Usually, you have a single uplink, and a bunch of systems connected to the switch. Each system gets a GbE connection, but you can never get full bandwidth on all systems, at the same time. Oh..and like all switches, NF200 adds a bit of latency.
Honestly, if you're going SLI, you'll probably get better performance from a motherboard that has a pair of x16 slots running at x8 electrical. Less latency, and more consistancy.
Last I heard, octo-core will be LGA1567-only for the next generation. LGA1366 will stay at hex-core. The next-gen platform (which will be a new socket, and new chipset, of course,) may see octo-core on the single/dual socket platform, but your X58/5520 board will not see eight cores per socket. (What, 12 cores/24 threads isn't enough for 'ya?
Oh, wait, it's not enough for me, either... )(And after tomorrow, I won't be able to talk about it. Ah, the joy of NDAs that haven't been signed yet.....)
Yup, most benchmarks agree with that. At least the newer NF chips actually talk to the main chipset at PCIe 2.0 x16 now. The older ones were limited to PCIe 1.0. Which meant that you gained nothing by putting it on, except faster inter-card communication. (i.e. the speed between the two video cards was the full PCIe 2.0 x16; but then those two cards shared a single PCIe 1.0 x16 connection; which is the same speed that each card would have had under PCIe 2.0 x8. I'd rather have each card have 'half' speed, than have both cards share half speed with the system; even if they are full speed internal to each other.
And of course, doing four-card SLI you have issues with going between NF chips; whereas with a 16/8 switch, each card can talk directly to every other card at the same speed.
Yep..those NDA's do suck. However, just think of the Intel Tick/Tock product plan. Westmere (hex core) is the "tock" for dual socket. The followon will be what you are looking for..albeit with significant architectural changes that will require a new socket, and a new chipset. Note that there is an interesting feature that Core i5 has..that core i7 doesn't yet........ Also note that Larabee did get cancelled....
Of course, you don't have to wait quite that long. Octocore will be out shortly, but not for single socket....
I heard the exact same thing. Beckton (octal-core) will be S-1567 whereas Gulftown (hex-core) will be compatible with S-1366. Thus, current dual S-1366 boards such as this EVGA Classified iteration should be able to run Gulftown with perhaps no more than a BIOS flash, but Beckton will not be compatible even physically.
- Joined
- Feb 6, 2006
- Messages
- 13,984
i7's are only single socket capable... so it'd be kind of a waste.
Finally! A real use for the QPI Link!
Even Hex core would give you 12 real and 12 virtual cores. What could you do with 24 cores worth of processing anyway??
I'm glad to see the comeback of Dual CPU boards, seemed like it kinda dropped off with the core 2's
Just wonderful wretched excess...
Even Hex core would give you 12 real and 12 virtual cores. What could you do with 24 cores worth of processing anyway??
I'm glad to see the comeback of Dual CPU boards, seemed like it kinda dropped off with the core 2's
Just wonderful wretched excess...
- Joined
- May 21, 2008
- Messages
- 5,840
dual socket isnt going to be mainstream, I hope thats not what you mean. But yea, it would be nice to see a few more non server boards
Negative Decibel
2[H]4U
- Joined
- Dec 13, 2005
- Messages
- 3,588
That is sexy!
Now just need a case that can contain this monster that isn't butt ugly.
All the 10 slot 4Way Classified cases are god damned hideous.
http://www.evga.com/support/faq/afmmain.aspx?faqid=58719
All the 10 slot 4Way Classified cases are god damned hideous.
http://www.evga.com/support/faq/afmmain.aspx?faqid=58719
This is a total waste of money in my opinion. After all windows doesnt even utilize four CPU cores never mind 8 cores. Also why would you need 24GB of ram and a whole lot of GPU's. You would need to remortgage your house to be able to afford this thing.
Last edited:
- Joined
- Feb 6, 2006
- Messages
- 13,984
Windows utilizes four+ cores just fine.. games may not, but Windows does. And some people do need 24GB of ram.. and some just like to have that much space. Distributed computing projects can make use of multiple GPUS.
Negative Decibel
2[H]4U
- Joined
- Dec 13, 2005
- Messages
- 3,588
lol... thank you captain obvious! and yes. windows will utilize that many cores just fine. 24 GB of ram can be used, not by most people... and I'm sure it's not cheep, but that's not the point.
You don't spend much time in this part of the forum do you?
We could max out that board CPU and RAM wise and still make it cry.
This board or ones like it will probably become the base for the 100k+ ppd rigs into the future.
- Joined
- Apr 7, 2006
- Messages
- 10,477
Sniper_Merc
[H]ard|Gawd
- Joined
- Mar 25, 2004
- Messages
- 1,037
as the wise and powerfull Shaggy once said.......
ZOINKS!!!!!
ZOINKS!!!!!
Tigerbiten
[H]ard|DCer of the Month - February 2007/January 2
- Joined
- Sep 24, 2002
- Messages
- 5,028
My main box is built on a Skulltrails MoBo.
My dual E5355's, 8 cores, @3.4 Ghz are running at 100% load 24/7 folding.
I'm running out of free ram with only ~1.5Gb free, 8 Gb in system.
Going to 16 cores I'd need over 9Gb memory as a minimum amount.
I can only fit 3 vid cards in the system.
I'm only running 3x 9800GTX's, but an upgrade may be in order.
If I could get a new good overclocking dual cpu mobo for 55** series Xeon's then I'd buy it like a shot.
Luck .............
- Status