Hi all, I have wanted to write this review for some time. But with a certain piece of news that hit my wife and I very hard, our bathroom leaking into our hall and my main PC deciding to throw up "out of memory" errors constantly it had dropped down the list of priorities. Not to worry though, a complete system reinstall and a bathroom on order means I now have the time to get back into what I love doing - testing and messing around with hardware. Now just before I get into the nitty gritty I should point out I am not a professional reviewer, I do this sort of thing in my own time because I enjoy it. Intro As some of you may or may not know, I have been running multiples screens for gaming for a while now and I must say it is the best £450 or so I have spent in a long while. Unfortunately whilst multiple screens are fantastic from a gaming and productivity point of view, they do have a rather large GPU power (read: grunt) requirement. Over the last few months I have been playing with, and testing, high end single GPU cards from both Nvidia and AMD. My only caveat was they had to power three monitors from a single card so obviously this discounted the GTX4xx and 5xx from Nvidia. Overall I have been pleasantly surprised with both the ease of use and the performance on offer. Up until now I have only dabbled with single GPU cards as I wanted to avoid any micro stutter or driver based quirks that can be exagerrated by throwing multi screen gaming into the mix. Unfortunately even the best cards that Nvidia and AMD offer can be somewhat lacking when it comes to pushing nearly 7 megapixels, hence I have had to make the jump to SLI to garner the performance I require. My results below should hopefully pad out this reasoning a little further. Anyway, enough of the back story, lets get onto the interesting stuff; the cards! The cards and system setup The actual cards I ended up deciding on gave me quite a headache. I think I spend a good couple of weeks going backwards and forwards over exactly what I wanted to get and how much I was willing to spend. In the end I had it narrowed down to three choices: 1) Simply buy a second reference GTX680, cost £420 2) Sell my 680 for £350 and buy two non reference GTX670, cost £300 (£650-£350) 3) Sell my 680 for £350 and buy two non reference HD7970, cost £400 (£750-£350) Ultimately after trawling through reviews and balancing all of the factors I decided that a pair of GX670's would offer the best compromise of performance and cost. I ended up going for two Asus GTX670 Direct CUII 2GB cards from Ebuyer. I managed to get both for £660 posted (Saturday delivery). For your viewing pleasure here are a few pictures of the cards in question: Box art: Interior Box: Very well packaged: The card(s): Compared to the reference GTX670 the Direct CU is a vast improvement. The GPU itself is cooled by three chunky heatpipes (5 if you factor that two of them continue through the "core" and onto the other side) and two 92mm PWM fans. The VRM area is cooled by a black heatsink, which in turn receives direct airflow from the fan closest to the ouputs'. The "Mega" box of contents: I think Ihave been spoilt by the HD7970 box contents as quite frankly this is rubbish. The 680 was even worse mind! Card Porn: (GTX680 / GTX670 / GTX670) Mmmmm Nom Nom Nom: As you can see from the images above the Direct CUII cards are basically the same length and height as a reference GTX680. The inclusion of a backplate also makes the card feel and look alot better. The outputs are pure reference design as are the power connectors - 2 x 6 pin PCI-E. From reading reviews the PCB looks to be a complete custom job which could cause issues for those who wish to go down the watercooling route. Finally a shot of the cards in my system (Case - Define R3): System Setup For all tests unless otherwise stated this is the system setup: This details the main system components, graphics cards tested and drivers used. It also covers the "Stock" vs "OC" clocks for each card. Please note that the overclocked results for both the 680 and 670 are based on the max boost clocks. Due to the nature of the boost clock the actual overclock can and does vary from game to game. In addition and unless otherwise stated, the resolution used for testing is 3620x1920 being 3240x1920 with bezel correction. Benchmarks BF3 Test scenario - 5 minute run through of Operation Swordbreaker starting the moment you leave the APC right up to clearing the car park of insurgents. Note - I actually play this scenario, it isn't a time demo or anything like that. Settings - 3620x1920, All Ultra, No MSAA or motion blur Settings - 3620x1920, All Ultra, mixed MSAA, no motion blur Right out of the gate I must say that Surround & Eyefinity performance in BF3 is generally very good. GTX670SLI takes that to a whole new level by managing close to 60fps average with 2x MSAA thrown into the mix. Add an overclock and you have over 60fps average. Drop AA completely and you can be pushing over 70fps average at Ultra settings. I must admit though, whilst these framerates are fantastic a single GTX680 or HD7970 can happily play BF3 with a mixture of high and Ultra settings at the same resolution. As I found before with a single GTX680, 2GB of VRAM can be a limitation at this resolution. 4xMSAA is completely unplayable on GTX670 SLI just as it was on a single 680. Overall though it is more of a technical issue as I found 4xMSAA didn't really add anything to the experience over and above 2xMSAA at my resolution. Crysis 2 Test Scenario - 5 minute play through of the "battery park" level. Settings - Extreme Preset with DX11 and HD Texture pack. Both a single GTX680 (overclocked or not) and a single HD7970 (sadly not shown on the graphs) were borderline playable in this game. Realistically I had to drop the settings down to "Very High" to get over 40fps average on either card. GTX670SLI on the other hand just slices through it like a hot knife through butter. Once overclocked, you are looking at framerates north of 60fps average and minimums nearly doubling a single GTX680. I havn't forced any MSAA as the game engine seems to add some form of MSAA automatically. Jaggies were minimal during my playthrough. Crysis 1 Test Scenario - 5 minute play through of the "contact" section near the beginning of the game. Settings - CCC Level 6 Custom config with HD Texture packs This is the main game that drove my SLI purchasing decision. I wanted to play through it again in all its glory but hit a stumbling block when my GTX680 couldn't even manage playable framerates. Technically the game was playable with a number of settings dropped, but when you want to see how good it can look where is the fun in that? GTX670 SLI offers up "playable" framerates in general but I can't help but think that Tri SLI is really a requirement here. Unfortunately Tri SLI would require a complete system rebuild which, thanks to our bathroom replacement, isn't going to happen anytime soon. MSAA is a total no go at these settings even on dual 670's overclocked. VRAM isn't the issue but GPU grunt is. I really couldn't have sat through playing it on a single GPU with AA as it would have been a slideshow. One oddity I have noticed with this game is that even at my resolution I appear to be CPU limited. Whilst not a problem in itself it is proving difficult to pin down the exact cause as the stutter I experience - which I attribute to CPU limitation - occurs randomly, even at the same section of the game. I need to do more investigation into this before concluding anything. Just Cause 2 Test Scenario - Dark Tower built in benchmark Settings - Everything maxed with 4xAA & 16xAF This addition (along with Batman below) is just a bit of fun. GTX670 SLI absolutely slices through the benchmark without a single hesitation. Note - I used the built in benchmark here as the open nature of the game means benchmarking a repeatable yet gameplay representative section is very difficult. Batman: AA Test Scenario - Built in benchmark Settings - Everything maxed with 4xAA & 16xAF (No Physx) Again this is just a bit of fun. I don't own Arkyam City so this had to do. Not much to say here. GTX670SLI is total overkill even at this resolution. Unigine Heaven 2.5 Test Scenario - Standard Benchmark Settings - Default benchmark settings Now I don't normally run synthetic benchmarks (I find them ultimately poinless when it comes to performance comparisons), but I have included this as I was using it for stability testing and though "ah what the hell". I have only run Heaven 2.5 instead of the newer 3.0 as the 680 and HD7970 were run in 2.5. Heaven, along with Crysis 1 shows that even two top end cards overclocked can sometime be lacking at Surround / Eyefinity resolutions. Indeed as with Crysis 1 MSAA is a total no go even on 670 SLI overclocked. VRAM Testing This graph is relatively straight forward but I know VRAM usage is a big talking point at the moment. All told, I ran into VRAM limitation only once during testing and that was in BF3. Even then it wasn't a game killer. Oddly enough the one game I thought I would have problems with; Crysis, had no issues at all with VRAM. Even upping MSAA to 4x only increased reported VRAM usage to 1.75GB. Now don't get me wrong here, at my resolution there will be games out there that will require the settings to be dropped to ensure VRAM isn't an issue. I don't own a single one of them but they do exist (Skyrim with alot of mods for example). Some may ask the question; 'Why didn't you go for 4GB GTX670's?', the answer was simple; price. A pair of GTX670's would have costed another £150 yet offered me no realworld benefits for the games I play. Looking to the future though I will definately be looking for more than 2GB (and even 3GB) on my next card purchase. Card Temps & Noise I don't own any real testing equipment so this is more a subjective take on all the cards I have tested to date in Surround / Eyefinity. Temperatures on my two cards are as follows: Idle: Card 1 - 30 Card 2 - 28 Load: Card 1 - 80 Card 2 - 75 The above readings were from my BF3 test scenario and per the auto fan profile in Afterburner. In terms of noise the Direct CUII coolers are for want of a better word; silent. Even at 55% fan speed I can't here them over the case fans in my Define R3 case. Conclusion Excellent performance, Silent coolers and a reasonable price. That is basically my summary of Asus Direct CUII GTX670's in SLI. To be honest there isn't alot more to add. In single card configuration you are basically getting GTX680 performance at a noticably lower price. In the end I am glad I went for GTX670 SLI for two main reasons: 1) Saving of ~ £120 over GTX680 SLI / GTX690 yet nigh on identical performance 2) Avoided possible Eyefinity / Crossfire issues which seem to be quite prelevant. Anyway I have waffled on long enough, hopefully this has been an informative and interesting review. Thanks for reading
Great review, but how come you haven't used 4xMSAA? Are you hitting VRAM limits? Is that why the single GTX 680 2GB would not run?
Sorry guys the full review should now be up. I hit reply instead of preview... DOH! I havn't run a single GTX670 through the grinder as the GTX680OC results are a perfect proxy. EDIT: The detail for this is now in the review but to answer you question specifically - 4xMSAA was completely unplayable on a single GTX680 and GTX670 SLI due to a lack of VRAM. It was technically playable on a HD7970 but the framerate was too low to actually be used. HD7970CF or GTX6xx 4GB should allow playable framerates with 4xMSAA but the benefits over 2xMSAA are minimal. EDIT 2 - Just realised you may mean across all of the games (where AA isn't listed). This is a quick summary: BF3 - VRAM limitation Crysis 2 - Appears to run AA automatically as jaggies were minimal. Crysis 1 - Unplayable with any form of AA due to a lack of GPU grunt Heaven - Unplayable with any form of AA due to a lack of GPU grunt The latter two would have been a complete slide shown on a single 680 / HD7970 with any form of AA. I will add a couple comments to this regard in my review.
Great read Xaser + rep However, I would like to point out that it is not simply a lack of VRAM that makes it unplayable, it is more to do with the lack of PCIE bandwidth on your motherboard with the 2500k. I think if you get an Ivybridge cpu and enable gen 3 PCIE you will have great improvements when you hit the VRAM limit
Assuming you are talking about BF3 then it definitely a VRAM limitation. The games craps out with a Direct X out of memory error right after memory reads over 2050MB in afterburner - exactly the same thing happened when running a single GTX680. I would have to test out an Ivy Bridge platform myself before making any judgements on a lack of PCI-e bandwidth.
Nice review Xaser , I enjoyed the read. Nice to see the PCIe bandwidth finally being saturated by these newer cards. Am I right in saying its running dual x8 on that platform? Just wondered if my i7 x58 setup with x16 has a bit more to go if I upgrade to two cards. + rep
Awesome work dude! Its good to see real world performance rather than pro reviewers all the time. Just gotta wait for my (1) directcuII to arrive. +Rep
I didn't test much when I had the 680 2GBs to see what difference 2.0 had over 3.0 personally but how can this video be explained if 2GB VRAM isn't enough for 3 screens? Edit: watching again, he has no msaa and ao so its hardly conclusive. Need moar testing! I can't really compare my set up as I don't have such a high res...
As you say it is very hard to draw any conclusions unless you run the tests yourself. There are so many factors involved it can be difficult to pin down the exact cause. One thing I should note however is that a 4 way comparison is going to highlight PCI-e bandwidth limitations far before a 2 way setup will. In the video Vega mentions that under 2.0 GPU utilisation was only in the 50's. During my own testing my GPU utilisation on both GPU's was 99% in every scenario bar Crysis 1. I would love to be able to test this out myself by getting a nice Z77 motherboard and i5-3570K but other things have taken funds away from computer upgrades at the moment. MSAA in BF3 makes a massive difference in the amount of VRAM used, especially at Surround / Eyefinity resolutions. In my case moving from no MSAA to 4xMSAA increases VRAM usage by well over 700MB. As you say in the video no MSAA or HBAO is used, I also note that terrain decoration / Effects was set to low therefore it isn't directly comparable. At the same settings I am only using around 1300-1400MB of VRAM. EDIT: Thanks for all the replies guys! I enjoy doing this sort of thing so it is good to see that you are enjoying reading it too. Ivoryspike - yes you are correct, The P67 MB I have runs both slots at PCI-e 2.0 x8 when both are in use. Your X58 platform has more theoretical PCI-e bandwidth available to multiple cards but you may end up being more CPU limited in certain scenarios reducing any potential benefit.
I'm surprised by the temperature on the bottom card. After extensive playing my single card rarely pushes 60 degrees at the moment. Also in a Fractal R3 with two front fans and one on the bottom of the chassis. Excellent article! Tempted by GTX 670 SLI but my single card is doing a sterling job by itself at the moment!
