Hey xnikx,
 
Let me give you some of my thoughts ... first of all there's no reason you shouldn't be gunning to run Ultra on BF3 with 3x 4GB GTX 670's. I have been running Ultra without MSAA since this review and the performance is just incredible.
 
Since this review I have done a decent bit of swapping with my rig, so if you are looking for benchmarks on High to compare to the testing I have already done I won't be able to provide that for you. I realized 3x GTX 670 FTW was just a good bit of overkill, and I didn't really feel like it balanced very well with my platform. I had the opportunity to trade one of the GTX 670 FTW's for a 3930K from a member here and did so. Now my rig looks like my sig says, but I'm running a Rampage 4 Formula, 3930K, and SLI GTX 670 FTW's. So ... I can provide with you High benchmarks on that setup, but I don't have access to the 3-way 670 FTW's anymore.
 
Another thing, I feel your pain about the investment it takes to drive 3 monitors, and that's what I always tell people. Most people look at my computer and say wow, why would you spend that much on monitors? My response ... buying the monitors is the easy and cheap part (unless you're going 30"), but buying the GPU horsepower to drive them is the part that hurts and the part that you can be somewhat overlooked beforehand.
 
All that said, I don't think we will feel this pain too much longer. Here is what I mean ... just last generation (500 series) it basically took 3-way 580 3GB cards to comfortably run surround set ups ... basically one of the highest end setups possible. Here we are ... only one generation later ... and a single GTX 670 can run the same surround set up. We went from taking $1800 of GPU power to being possible with $400 in a single generation. You can see where I'm going with this. As of the 500 series, running surround was absolutely bleeding edge technology and it took the highest end set ups to do it. A year later, it's much more mainstream and easier to pull off. By the 800 series, even the midrange cards will be able to pull off 1080P surround pretty easily I bet, so relief is coming for you and I as the technology becomes more and more mainstream as the bleeding edge shifts to 4K displays and ... dare I say ... surround 4K.
 
What I would do if I were you ... start with 2 GTX 670's. These things are going to be so much faster than your 460's it will seriously blow your mind. I don't care how much you're expecting out of them, they will impress you. I would also bug Jacob (most effectively through Twitter) and try to find out when they are coming out with the GTX 670 FTW 4GB. He said a couple weeks ago that it would be soon, so it could be any time now. Those cards would be the absolute best price/performance.
 
I would buy 2 GTX 670 FTW 4GB when they are released, and if you feel you need a 3rd in a year then pick up a used one to complete your 3-way set up then. I see almost no way you install SLI GTX 670's in your system today and think "Wow, I need a 3rd card."
 
Getting the FTW model is even more important if you're ever going 3-way and staying on air. I wouldn't have wanted to leave my 3-way 670 FTW's on air very long, but I'm particular about my cards and their temps. I wouldn't even consider putting 3-way reference 670's in and leaving them on air just due to the cheaper cooling based on how I saw the FTW models perform. They were good enough, but I wouldn't have wanted any worse cooling in a 3-way sandwich.
 
Hopefully this helps a bit.

Let me just run some quick tests when I get home tonight. I will put together some quick graphs of Ultra minus MSAA, High, and both with a stock 3930K and a 4.8 GHz 3930K.
 
I don't have much (make that any) experience with Ivy Bridge, but my gut tells me you won't have to OC that CPU at all. You are buying a K model though, so it should nearly be illegal not to at least put a mild 4.0-4.2 OC on it just because it is so easy . Just kidding as what you do with your hardware is completely up to you.
 
I also think you will see you can stop worrying about High and bump your standards up to Ultra and tweak from there. We'll see what the FPS numbers say tonight, but I'm saying I average over 60 (somewhere near 70) with SLI 670 FTW 2GB on Ultra minus MSAA at 5760x1080. It is not uncommon at all to see the Logitech displaying 80-90 FPS.
 
I unfortunately have to do some work work at home tonight as well, so if I don't get results posted tonight don't freak out ... they are on their way.
 
As for temps, just going off of memory I think I remember the top two cards would top out around 72C in 3-way SLI. That's just going off of the top of my head, but it was definitely less than the 80C you are experiencing now. There just isn't much room for them to breathe in a 3-way sandwich.

I use Core Temp and watch the usages across the cores on my Logitech G19. I don't have a way to log it, but you can watch it and get a good idea. I'm not sure if there is an on screen display program for it or not. If you do have an accessory display when you get your 600's you can just have the Core Temp gadget on the desktop or a number of different programs really.
 
You are hugely VRAM bottlenecked with those 1GB 460's, so I'm sure that's a lot of the stutter.
 
I doubt I'll get the results posted tonight, but I'll try to get the data logged and have it up tomorrow night.

Part 3
 
Ok, so I have the results from what I guess I'll call part 3 here. This testing was pretty straight forward. I set out to test BF3 online but mainly Ultra settings without MSAA vs. High settings which are automatically without MSAA. I used a pretty similar method to what I did in the first two parts. I used FRAPS to record the FPS for 5 minutes at a time hoping to catch a good average. I chose servers that were full with 64 players and all were on Caspian Border. There were 3 or 4 different servers used throughout the testing as servers would change levels and I wanted to keep it all on the same level. All servers chosen were under 50 ping. I tested three different settings, mostly for my own curiosity. Here's the rig currently:

3x Samsung 2333T at 5760x1080 60hz
Intel Core i7-3930K
Asus Rampage IV Formula
16 GB Corsair Vengeance
Seasonic X-1250
SLI EVGA GTX 670 FTW both running PCI-E 3.0 x16 - 304.79 Beta Drivers
 
I tested everything at stock settings, the 3930K overclocked to 4.8 GHz with GPU's stock, and finally the 3930K at 4.8 GHz with GPU1 running 1320/6202 and GPU2 running 1270/6202. Everything is on water, so temps are pretty low. I don't have a very good block mount on GPU1 so it will hit 47 overclocked while GPU2 will hit 42 overclocked. I got extremely lucky with my 3930K as it only takes 0.015V offset to run 4.8 GHz, or basically 1.36V, so it spends most of its time around 50-52C across the hottest cores in game while overclocked.
 
When I ran the first batch of benches I was shocked at how close High and Ultra were, so much so that I actually ran two more runs of each of the overclocked settings just to verify results. I did only run one run of the fully stock settings. As a result, the graph below shows 3 sections. The stock section is the average FPS for the one run that I did for stock settings while the two overclocked sections are the average of the average FPS for the 3 runs that I did. Basically they are a 15 minute average across a couple of different servers.



 
 

 
You can take from it what you will, but I just think it shows how difficult BF3 is to test objectively. On Run 2 the Ultra average was higher than the High average, but on both Run 2 and Run 3 they were only 1 frame apart.
 
To xnikx, hopefully this will show you that there is no reason not to run Ultra for you. The difference between High and Ultra is very minimal. I would say it is a much bigger deal if you are up against a VRAM bottleneck, but with 4GB you definitely won't be as I'm not with 2GB without MSAA. Also, as I suspected the average frames for my overclocked settings, which I usually play at, are right around 68-70 frames which is pretty incredible for 670 SLI in my opinion. I was a little surprised, but I did think overclocking the CPU helped. I think it helped more by feel than it shows in the benchmarks. I really should have done two more runs for the stock settings as well as I don't know how much I believe that Ultra average for the stock settings. Oh well.
 
Those are my thoughts and there is the data for you to pour over. If you have any questions about anything just ask!
 
Edit: Just for a shameless plug I thought I'd throw in a couple of pictures of the updated rig!
 


 

Sounds like a plan bud. You will have no problems with that setup. I am actually going to be selling these 670 FTW's to a couple friends in a month or so and swapping for some 680 Classified Hydro Coppers. I just have too much of a need to tweak and have to play with the EVBot and voltages.

Nah, I just want to play with voltages and overclock the Classified cards and bench them. I think I could get my 3930K to 5.0 GHz or maybe even beyond, so hopefully I could snag some pretty high scores for a 680 SLI configuration. I wasn't very impressed with 3-way scaling and it sounds like 4-way is even worse, so I could just never justify going with a full 4-way set up. Faster cards will always be out soon enough for me. I also wouldn't mind adjusting settings just to achieve the frames I'm gunning for. At 60hz even these 670's have no problem pulling that.
 
If these 670 FTW's are pulling 68, I would think 680 Classy's OC'ed to 1350-1400 could pull 75-80, but maybe I'm wrong.
 
I'd love 120hz monitors, but I'm not ready to make that investment yet. I'm thinking surround 1600 or 1440P is going to be my next monitor upgrade, and it's going to be a while before I'm ready to do that as that would be what, $4000+? That's not far off what I have in my whole set up currently, lol.

well, they just came this afternoon. i wasn't able to test them till now, they came an hour before i had to go to work.
 

 
I only played two games off bf3, one on seine  and the other on sharqi.
 
All settings maxed @ 5760x1080; ultra on all, 4xmsaa, aa post high, motion blur on, 16x af, hbao
 
Average fps was 41-46, lowest i seen was 33, highest was 57.
 
The game actually felt very playable since the frames were usually in the 40s, but could notice the low fps when it dropped lower than 38.
 
Temperatures were pretty good, top card never passed 72C but holy **** are these some loud cards. I really thought these would be quieter than my 460s.
 
It's still a little early to say  but these cards seem like a good buy.  However, i was having some weird issues where one of my cards gpu usage would drop from 98% to 2% while the other would remain at 98%.  This would cause major stuttering.  The problem seemed to fix itself, lets hope it stays that way.
 
Just wanted to give some quick input, ill post more after playing more with the cards.
 
Gigabyte x58 usb3
Intel i7 950 @ 3.68ghz
12gb Ripjaw 1333
Dual 4gb 680gtx sli
850w Corsair hx

Yea I've actually just acquired a 3rd 670 FTW again and am in the process of trying to tune it to perform better. I'm on the verge of a complete OS and driver reinstall because right now the performance is absolutely pathetic. I was seeing 30-40 FPS last night with 30-40% GPU usages in online BF3. Something is simply not right. I know the usages won't ever be pegged in the upper 90's like they are with 2-way SLI, but what I have now is just not going to work.
 
 

Great data, thanks alot !

thebski

Yea I've actually just acquired a 3rd 670 FTW again and am in the process of trying to tune it to perform better. I'm on the verge of a complete OS and driver reinstall because right now the performance is absolutely pathetic. I was seeing 30-40 FPS last night with 30-40% GPU usages in online BF3. Something is simply not right. I know the usages won't ever be pegged in the upper 90's like they are with 2-way SLI, but what I have now is just not going to work.

 
hey guys, been a while since ive posted.
 
now im looking to pickup a 3930k, asus extreme iv rampage, and a third 4gb 680 ftw. however seeing that people with similar systems to this are having the same issues, its got me kind of worried.
now thebski, what do you mean by your three cards will never be pegged in the 90s like 2-way sli?  is three way sli a good idea or should i just stick with two cards?
 
this definitively has to be a driver issue.  
using the 306.97 drivers i was actually getting great performance in bf3.  my gpu usage would be in the high 80s or 90s and fps would rarely drop below 60.
 
i upgraded to the 310s, which nvidia promised would boost surround and multi gpu performance.  for me, these drivers were terrible. many people on other forums i frequent have said their surround performance definitively increased.
 
@spyle, now the mesh quality thing made a huge difference for me on the 306.97 drivers.  with everything on ultra, msaa off, 16x aa, and mesh quality on low i was getting terrific performance.  when mesh quality was on ultra, performance went down.
with the 310 drivers, the complete opposite occurred, the higher i boosted all my settings, the better the performance.  that ment mesh quality on ultra gave me better performance.
 
people say this is a cpu bottleneck but bf3 perfoverlay shows that my cpu is not bottlenecking at all.  And if a 3930k cant cut it, what cpu can?
 
@boffinboy. i dont know the awnser to your question, but i would imagine you would be better off with just the two cards in sli.
 

if you are thinking about 7680x1600
 
then you'll need to step up to 3-way sli 670/680 minimum with 4gb models to boot, with heavily overclocked vram (again because of the resolution + AA that you'll most likely use)
 
and at 3+ sli, i would seriously start looking at pcie3.0 (for 2 cards and below, i wouldn't bother)