Hey everyone long time reader first time poster. I have to say the articles on this website are great.
Currently I'm on a E6600 with an ATI HD3870 . Needless to say I'm in real need of an upgrade to play BF3.
I really want to go all out on my next build and create a fully immersible gaming experience in my garage using three HD 3D projectors with eyefinity or nvidia's equivalent.
More than likely I will be using the new keplar cards when they come out and from my understanding I will need at least 3 GPU's in order to get decent frame rates for the 3 displays.
From what I understand the 2500k and 2600k cpus only have 16 pcie lanes on die so ive been waiting for the sandy bridge e which reportedly has 40 pcie 3.0 lanes (unofficially) which will allow me to do a 3 way sli with no bottlenecks.
Am I right in assuming that the 3930k is the right cpu for this project or is there a better alternative.
I would only get the 3930 if funds are virtually unlimited boards with Nvidia’s NF200 bridge will allow you to use 3 cards in SLi without bottlenecks and I don't think the new cards will be much faster in PCIE 3 than in 2. So I would get a 2500K and overclock it like crazy. But if cash is not a problem and you want to keep this as long as your old system then the 3930 will be a bit better.
Just get a mobo like the Asus Maximus Extreme-Z which has an NF200 chip which allows 16/16/8/8 for quad SLI/CF.
Also, depends on your resolution but most projectors that I know of are at best 1920x1080. I'm not quite sure what you mean, but I guess it's either use 2 projectors for 3D or use 6 projectors for surround 3D? Is that it?
If it's surround 3D you're talking about at 5760x1080 then dual GTX 580 3GB should do the trick nicely. Kepler is probably going to be a lot faster with the die shrink, so I'd think two of the top Kepler cards will perform even nicer. 3 way SLI does help remove any micro stuttering though so if you have that kind of cash then go for it.
Anyway, my point is, CPU-wise a 2500k or 2600k OCed to 4.5ghz will be suitable for this gaming rig if you pair it with an NF200 enabled motherboard. Otherwise, sure, go with the SB-e CPUs. They're more expensive but they're pretty damn nice.
^ Wow, wish I had the OP's budget for a gaming rig .
Anyway, since money appears to be no object, then yeah get a 3930K and oc it to 4GHz with a good water cooling kit. One of the reviews on the 3960X showed it to scale best with tri-SLI compared to other CPUs, plus the price will probably drop close to Intel's tray price of $550 once the novelty wears off and decent supply available.
I kind of don't see the point of getting a sandy bridge 2500k coupled with an NF200. From what I read the NF200 is a little bit of a gimmick. It decreases performance because it adds another component in between PCI-E and CPU that just adds latency.
Essentially all it does is compress the 32 lanes to 16. Not a very elegant solution at least on paper.
From some benchmarks it looks like running cards at 8x is nearly the same at running them at 16x.
In addition I don't see the point of going to regular sandy with ivy is just around the corner. The prices are still the same.
In regards to the new cards not being much faster in PCIE 3 I think this maybe true. From what I read GPU's have a hard time using PCIE 2.0 to the limit and PCIE 3.0 is only really going to benefit PCIE SSD cards.
However the new PCIE 3.0 effectively doubles the bandwidth of 2.0. So my thoughts turn to Ivy bridge and its 16x pcie 3.0. So I could run a card at 4x PCIE 3.0 and it would be the same as 8x PCIE 2.0. speeds. Ideally I would love ivy bridge to have at least 32 PCIE 3.0 lanes as I also want to throw in a PCIE ssd but that doesn't look like its going to happen.
Do you guys think Ivy Bridge could handle this project with minimal performance loss?
There is also another CPU im looking at the I7 3820 “partially unlocked.” which has 40 PCIE 3.0 lanes. Intel lets you set a multiplier six bins higher than the highest Turbo Boost clock rate: 3.9 GHz. The result is a maximum of 4.5 GHz using ratios exclusively.
Would this be any good?
The rig I want to build with 3 1920x1080 projectors looks like the setup in below link (starts around 5.50 mark). Like having 3 monitors but without the bezels.
Essentially I'm no millionaire but I have job and feel that I'm really enjoying my gaming at the moment. I just want to be able to play BF3 on ultra settings with a system that takes full advantage of the GPUS's in SLI. Of course it would always be better to spend less.
NF200 is a very elegant solution. Mutliple GPUs need the exact same data to be fed to them - they each have identical RAM usage, they just communicate via the bridge for flip-floping rendered frames. The point is, the NF200 splits the incoming signal, which is exactly what your GPUs need. It might add a tiny bit of latency but I doubt it could even be measured.
Anyway, since you're looking to the future then it's a moot point. I think waiting for Ivy Bridge is a good idea.
8x PCIe 2.0 does reduce performance. Last review I saw on it was here at Tom's, using 5870s. They lost roughly 3-5% performance. Considering that we now have 6950s, 6970s, 6990, 480, 570, 590 all as faster cards, you're looking at more performance loss. The flip side is that if you're pushing as many FPS as you need, that loss only matters in benchmarks which is to say - not at all.
As for the resolution, there's a lot of reviews on it. With current hardware, a pair of GTX 580s could run BF3 at ultra with 5760x1080. For a little wiggle room maybe go with Tri SLI 570s - also will reduce any potential micro stutter. But again, you're planning to wait for Kepler so it doesn't really matter right now. Who knows what Kepler will bring?
1) I currently run 5760x1080 in 2D surround vision with dual GTX 570's. I can tell you that you do NOT want GTX 570's, any amount of them, for surround gaming unless you're budget limited, or already own one 570 and are looking at buying a second, etc. You will run into VRAM (GPU memory) bottlenecks with huge resolutions long before you'll run into actual processing power bottlenecks.
For example, when I'm running GTA 4 my cards run between 60-80% load but barely playable fps (~30-35). This tells me that my memory is full (I imagine because the map is huge and detailed) before the dual 570 chips are fully loaded. Also, Battlefield 3 is suppose to be a huge game in terms of VRAM memory requirements. I would NOT run Battlefield 3 on any amount of 570's at 5760 x 1080 unless you're going to buy the 2.5 GB versions. I wouldn't go with any less than 3GB 580's. Two is probably fine, three if you want to eliminate micro stutter.
I should add that I play all my games at 5760x1080 with my dual 570's. Games that memory is a problem I just have to turn things down. Nothing I can't deal with given that I already owned one 570 and wasn't going through the trouble of selling it an buying 580's when I made the move to 3 monitor gaming. I also wasn't very keen on the idea of dropping 1200 on cards when 1200 will buy a WHOLE lot more card in 6 months. But if you're buying new and have the budget, buy as much VRAM as possible.
2) I'm in a very similar position in that, due to external factors, I'm in a really strong position to build a new PC within the next couple of months, but am really struggling with what platform. I've been waiting on SB-E to see what it is all about thinking it might be the next natural step after my current Nehalem platform. But I have to say there are some things that disappoint me.
The main thing that disappoints me is their overclocking ability. It sounds like these things take huge voltages (aka heat) to reach very high clocks (4.2+ Ghz). I'm in a position where part of the time I could use 12 threads. However, MOST of the time I probably can't use more than the standard 4c/8t. I would say 30% of the time at most I can utilize 12 threads.
So here is my dilemma. Whatever system gets built is getting built on a very quality custom water cooling loop. Do I buy a 3930K estimating that I'll probably be able to clock it to a maximum safe 24/7 clock of 4.2-4.4 and give up performance that 70% of the time that I can only use 8 threads? Or do I buy a 2600K/2700K estimating that I'll probably be able to run it safely around 4.7-4.9 and give up big performance that 30% of the time I can use 12 threads in exchange for greater performance the other 70%?
Here's why waiting on Ivy is probably not an option for me. Like I said, if I'm doing it it needs to be done relatively soon. If I made the commitment to wait for Ivy, we're probably not looking at an actual release date until April-May next year. Once that is released, I'm not going to be keen on buying one of the first models. It will be a new lithography process and who knows what the first chips will be like. I've always kind of been a fan of waiting for things to mature. For example, when I bought my i7 920 I got the second stepping (D0) rather than the original (C0). I couldn't be happier that I did. I got a great chip that OC'ed to 4.0 Ghz with very low voltage and temps compared to the average chip. Had a bought an early version I wouldn't have gotten near as good of a chip. So, then I'm waiting for the second stepping of Ivy, which will be when? August? Out of the question.
That begs the question, do I wait for C2 stepping of SB-E? Hopefully they'll be much better overclockers in terms of voltage required.
One other thing that bugs me about SB-E is the X79 chipset is really just the P67 with more PCI lanes and quad channel memory. That sucks. Really not impressed with the chipset. I'd say my disappointment is much more directed at X79 than SB-E itself. One review I read (don't remember which one) summed it up very nicely, X79 has given us everything we don't want and nothing we do want. I'm pretty "meh" to quad channel memory. Where is the USB 3.0? Where is the Sata support? You're buying a $1000 chip and can only connect 6 Sata devices and actually 2 at current gen speeds (Sata III)? Give me a fricken' break. I've got at least 4 Sata III drives I'd like to connect, and that hasn't touched a media drive yet.
Needless to say, I'm pretty torn. I'm honestly leaning towards Sandy Bridge since it's so mature I know I would get a good chip with a developed chipset. Then I will probably pedal the mobo and chip about this time next year when Ivy is good and mature if it ends up being something to get excited about.
For me, what it comes down to is, will there be a C2 stepping of Sandy Bridge? And will there be a more developed chipset in a few months much like there was P67 before Z68? If both answers are yes, then I may be willing to wait, yet again, in hopes of SB-E becoming everything I want it to be much like it was the second round of Nehalem before I pulled the trigger.
I should also add that CPU/Platform is not all I'm waiting on. One piece I'm pretty much sold on is the Seasonic X-1250 power supply that won't be available in the market until next month, and I also really want to see what the new Intel SSD 520's are all about. I have no clue when they'll be out. There was a rumor they'd be released with SB-E, but that obviously didn't happen.
Waiting for computer parts is a dangerous game, and honestly one you can't really win at.
I personally would go with the 3930k over the sandy bridge equivalents if I had the funds. The fact that the pci lanes are physically built into the system instead of on a 3rd party controller is more attractive to me. Also some games are starting to utilize more then 4 cores so the extra 2 cores might provide more longevity and life to the system. Depending on the game though I would consider turning off hyper threading and running off the 6 physical cores because of the possible performance loss.
And I totally agree with thebski on waiting for computer parts, I decided to wait for bulldozer and was thoroughly disappointed with the performance levels, and for its low performance level the incredibly high power consumption.
I do like the 40 PCI-E lanes built in to the chip. What would be really nice is if it had PCI-E, USB 3.0, more Sata, etc. all built in that way an Asus Rampage IV Extreme didn't cost $450 and use a ton of power by the time they get done adding all the 3rd party controllers to it.