Part 1 - Initial Scaling Review Part 2 - BF3 Online and Memory Data - Page 1, Post #18
Part 3 - BF3 Online High vs. Ultra and Stock vs. Overclocked - Page 5, Post #128
Well my three EVGA GTX 670 FTW cards arrived via UPS last Thursday. After a weekend which included moving into a new house and some testing of these cards I've got some information I'd like to share with you guys. I won't waste any time as we've got a decent bit to get through.
Three little bundles of joy!
Out with the old, in with the new!
Test Setup My current gaming rig is in my signature with more info on the Mods Rigs link, but here are the basics:
CPU: Intel Core i7-2600K at 4.7 Ghz on water
Motherboard: Asus Maximus IV Extreme-Z
Memory: 16 GB Corsair Vengeance at 1866 Mhz CL9
Power Supply: Seasonic Gold Series X-1250
Storage: 240 GB OCZ Vertex 3
Monitors: 3x Samsung 2333T up to 5760x1080
GPU(s): GTX 670 FTW, GTX 670 FTW SLI, GTX 670 FTW Tri-SLI (301.42 Drivers)
PCI-E: As you will see in the screenshots, my PCI-E settings are as follows -
Single Card: PCI-E 2.0 x16
I didn't grab a picture of the single card setup, but as you can imagine it was used in the top slot
SLI: Top Card - PCI-E 2.0 x8
Bottom Card - PCI-E 2.0 x16
Tri-SLI: Top Card- PCI-E 2.0 x8
Middle Card - PCI-E 2.0 x8
Bottom Card - PCI-E 2.0 x16
I'm not really sure why my bottom card is the one that defaults to x16 in either of the SLI configurations, but that's what it did. I should also note that all of my testing is done offline so far. The cable and internet man didn't get here until last night, so most of the testing was done by the time I got connection at my new place.
A little shot of my work area in my new basement! My new build uses the Xigmatek Elysium case with a 420mm rad mounted in the top and the 240 you see mounted in the front. These cards will be going on water soon. See my signature or Mods Rigs page for more details on the loop.
Test Methods For testing these cards I used some regular benchmarks as well as in-game play similar to the method [H]ard|OCP uses. I ran each test with a single card, SLI, and Tri-SLI and recorded the results. For the games, I played through a certain section of the game while recording my FPS with FRAPS and repeated the same section for each card configuration. I recorded 5 minutes of FPS in hopes that that would be long enough to get at a good average. For the benchmarks I simply ran them and recorded the results. For all tests I have summarized results into graphs that I'll get to in a little bit. For now, I'll go through each test individually.
3DMark 11 For this benchmark I just used the standard Performance benchmark. There really isn't a lot else to say about this one. I took screenshots of the results as well as the graphs.
Heaven 3.0 I ran Heaven at both 1920x1080 and 5760x1080 using standard settings with the addition of Normal Tesselation and 4x Anti-Aliasing. I tried to record screenshots, but for some reason the screen showed up blank when I tried to take the screenshots of 1920x1080 with SLI and Tri-SLI. I just ended up recording the results by hand on those. I do have screenshots for the rest of the configurations though.
Battlefield 3 For BF3 I ran all tests at 5760x1080. I ran one set on Ultra Settings with the exception of using 2xMSAA and another set on Ultra Settings to compare the difference between 2xMSAA and 4xMSAA. I ran these tests during the Ambush in the Car Park section of Operation Swordbreaker. It's close to the beginning of the game, but it was easily repeatable and had a good combination of a lot of gunfire and explosions.
Skyrim I tested Skyrim at 5760x1080 on Ultra Settings. I also modded the SkyrimPrefs.ini file to take VSync off. For these tests I ran around Whiterun for about the first 40 seconds and then exited into Skyrim and walked toward Riverwood along the paths. Along the way with each configuration I ran into the same 3 enemies, so I was happy with that as far as the as the ability to replicate goes.
Crysis 2 Crysis 2 was also ran at 5760x1080 on Ultra Settings. I did have the DX11 pack and High Res Texture pack installed, so this was full blown graphics. I had not yet played Crysis 2 on this build, so I ran the benchmark near the very beginning of the game when you walk out and get introduced to the binoculars and then later stealth mode for the very first time. Probably not the best place to bench the game as there weren't that many enemies, but I didn't feel like taking the time to find the perfect spot.
Results Now for the fun part
.
My Cards I just wanted to talk about my particular cards for a second. I have one card that is clearly better than the others which boosts to 1229 Mhz out of the box, the second that will boost to 1189 out of the box, and the third which will spend about half its time at 1176 and the other half at 1189. I can't say too much about the temps of these cards as far as factory fan profile goes. I have a fairly aggressive custom fan profile set in Precision X. Since these cards will throttle at 70C, I have basically drawn a straight line from 50C, 40% fan to 70C, 80% fan in the fan profile editor to try to hold temps at or under 70. Even in Tri-SLI with them all sandwiched together, I saw a max of about 71C on the top two cards with what I would guess to be about 80% fan speed. When it was either a single card or SLI with a space between them, they had no problem running about 63 or 64C on roughly 65% fan speed which is very quiet. I did not worry about recording temperature data because these cards are going to find their way under water soon. I also don't do any overclocking in this little review because I'm waiting for water for that.
3DMark 11 Single Card:
SLI:
Tri-SLI:
Summary:
I don't know that I put very much stock into benchmarks like Heaven and 3DMark 11, but people like to see them so I ran them. I think what we see here is pretty good SLI scaling. When you look at the graphics score it's not quite linear but pretty close.
Heaven 3.0 Single Card 1920x1080:
Note: No visible screenshot for SLI or Tri-SLI 1920x1080, but results are in graph.
Single Card 5760x1080:
SLI 5760x1080:
Tri-SLI 5760x1080:
Summary:
We don't see scaling quite as good here as we did with the graphics score on 3DMark, and something is obviously wrong with this benchmark in surround resolutions and SLI configurations. It doesn't concern me too much because these benches don't mean too much to me, but this is what I found.
BF3 Single Card:
SLI:
Tri-SLI:
Summary:
For BF3, I am amazed at how well these cards ran it. I doubt you could play with 4xMSAA unless you can tolerate somewhat frequent stutters, but the game is phenomenal on 2xMSAA. Frankly, I can't really tell the difference in appearance on the screen anyways. I am really interested to see if disabling Aero makes 4xMSAA playable. That's testing that this testing has bred
. I have to say in this game that I was amazed at how smooth SLI and even single card were. Even when it would dip down to 27 or 28 fps with one card on 2xMSAA, it was the smoothest 27 or 28 fps I'd ever experienced. I honestly thought Tri-SLI wasn't as "smooth" as SLI even though the frame rates were higher. I think you can kind of see it in the FRAPS recordings too with the huge variance in frame rates, especially toward the end.
Skyrim FRAPS:
Summary:
You can easily play Skyrim on a single card at 5760x1080 from what I experienced. Overall, I wasn't too impressed the the Tri-SLI scaling of this game. Maybe as drivers mature more it will get better, but to be honest it's just not needed in this game. You will noticed that in each configuration the frames dropped to basically nothing around the 40 second mark, but this was just the transition between Whiterun and Skyrim. You could also feel little hitches in this game, with all configurations, which I perceived to be textures loading in to the memory when transitioning between parts of the world.
Crysis 2 FRAPS:
Summary:
This was a fun game to test because the benefits of SLI and Tri-SLI were very clear. The game was simply not playable with one card, was playable and decent with two cards and was pretty awesome on three cards. The differences between the three configurations were by far the most noticeable in this game. I won't complain about 2.5x Tri-SLI scaling.
Data Summary 
Scaling Summary 
My Thoughts I had a lot of fun testing these cards and putting together the data. Overall, I have to say I was a little disappointed with the SLI scaling, but that could be due to my Z68 platform and it could also be immature drivers. I would love to do further testing with a Z77 and/or X79 platform, but unfortunately money doesn't grow on trees. I was really curious to see how these cards scaled with the 2GB of memory, and I can't say that I was terribly surprised that they were simply stunning. SLI is really all you need for a very smooth game play experience, but there were advantages to Tri-SLI. If you look at the FRAPS graphs, Tri-SLI rarely dips below 60 FPS in any of the games with the exception of Crysis 2 where I thought Tri-SLI to be the most "necessary."
I know everyone wants to know about memory usage, but honestly I didn't bother trying to record it. Memory usage doesn't concern me near as much as memory
required, and I don't know how to get at that number. I will say that my memory usage in the three games hovered around 1900-2000 MB most of the time, but as we know most games will use more than they need if it's available. I think the 2GB of memory does limit you on 4xMSAA in BF3, but I can't say that it was limiting in anything else. I have a hunch that the Skyrim texture loading pause has more to do with the memory bandwith than the memory size, but I have no way of proving that. I also know Crysis 2 was much more GPU horsepower hungry than it was memory hungry because there was such a noticeable difference between all three configurations.
Overall, these cards are really insane. I'm tempted to return them and go with SLI 4GB 680's just because I don't know that I
need three of these things, and my head says that it's money better used in 4GB of memory considering how fast two of these things are. On the other hand, my heart says three of them looks way too cool in there and I just need to order water blocks so the overclocking and real fun can begin.
I do have more testing ideas. I want to redo the games with Aero turned off to see what that does for me. That should help identify any memory bottlenecks I may have had (
cough cough BF3 4xMSAA cough) . I also want to buy Max Payne 3 and test on it since I've heard some grumblings that it's really memory hungry. For now though, that's all I've got. I'm ready to actually game on these babies for a while. Let me know what you guys think!
post edited by thebski - 2012/08/09 08:39:36