WEBVTT 00:00:07.200 --> 00:00:16.320 Now, the purpose of this video is fairly simple. Because of its extremely 00:00:12.719 --> 00:00:18.480 scarce availability, the GTX 690, you 00:00:16.320 --> 00:00:21.439 know, isn't as popular as people would like for it to be. There's lots of 00:00:19.840 --> 00:00:26.800 people out there, I know this for a fact, who want to buy a GTX 690, who 00:00:24.320 --> 00:00:33.920 want the ultimate performance, but cannot find one in stock. anywhere. So, 00:00:30.960 --> 00:00:39.600 I have one here today and I want to treat this video as sort of a um a what 00:00:37.520 --> 00:00:43.760 can you do if you want ultimate performance and can't get a GTX 690 sort 00:00:41.920 --> 00:00:47.440 of video. So, it has come to my attention that depending on which 00:00:45.120 --> 00:00:52.559 retailer you look at, NCIX not being the one that I'm referring to here, a GTX 00:00:49.520 --> 00:00:56.800 690 can go for as much as $1,200, which 00:00:52.559 --> 00:00:58.879 is almost exactly the price of three GTX 00:00:56.800 --> 00:01:06.479 670s. So, this is going to be a head-to-head comparison in my most 00:01:00.640 --> 00:01:10.000 recent um in my most recent 00:01:06.479 --> 00:01:14.560 suite of benchmarks between 3-way SLI 00:01:10.000 --> 00:01:17.040 GTX 670s and a single GTX 690. I think 00:01:14.560 --> 00:01:20.720 we all probably know which one's likely to win at this point, but in case you 00:01:18.799 --> 00:01:23.840 don't, that is the purpose of doing these kinds of tests. So, I don't know 00:01:22.000 --> 00:01:28.000 if you can see my Fraps meter in the corner there. Look at that. It's like 00:01:25.439 --> 00:01:30.240 exactly where the light is making it so you can't see it. Whatever. I'll tell 00:01:29.040 --> 00:01:37.360 you what it says. So, this is Battlefield 3. All of the tests are going to be run at 1080p. Yes, I am 00:01:34.799 --> 00:01:41.280 aware that 1080p isn't necessarily I actually don't know which way I'm going 00:01:38.560 --> 00:01:48.399 anymore. I am aware that 1080p is not necessarily the ideal resolution for, 00:01:46.159 --> 00:01:53.680 you know, very high-end graphics card setups. However, I did notice that we do 00:01:52.000 --> 00:01:57.680 see scaling in some of the games that I've tested already where we're seeing 00:01:56.000 --> 00:02:03.040 more than double the performance in 3-way SLI as we're seeing with a single 00:02:00.880 --> 00:02:06.799 card. So, that is a very positive indicator. I'm using the brand new 00:02:04.799 --> 00:02:12.400 driver from NVIDIA released on May 22nd. So, this is their R300 driver. This is a 00:02:09.039 --> 00:02:14.319 WLE driver that supports uh the GTX 670 00:02:12.400 --> 00:02:19.040 and the GTX 690. You can see I'm getting aboutund 00:02:16.160 --> 00:02:22.400 195 FPS in Battlefield 3. This is all on ultra settings, so I'm going to be 00:02:20.080 --> 00:02:28.239 maxing out every game, uh, The Witcher 2, Battlefield 3, The Elder Scrolls, 00:02:25.120 --> 00:02:30.000 Skyrim, and Crisis 2, and bringing you 00:02:28.239 --> 00:02:36.480 guys the numbers head-to-head between these two very ultimaty solutions. Now, 00:02:33.680 --> 00:02:41.120 I had really wanted to test 00:02:38.560 --> 00:02:45.760 four-way SLI, but I actually didn't realize that the 670 does not support 00:02:43.200 --> 00:02:48.400 four-way SLI. So, one of these cards is not on the bench, and we're just going 00:02:47.120 --> 00:02:53.519 with these three. So, I have my reference card, my Galaxy card, and my 00:02:51.280 --> 00:02:58.800 other more different Galaxy card. The one that looks like this with the 00:02:54.640 --> 00:03:01.200 glowing logo. Pretty sweet. 00:02:58.800 --> 00:03:05.599 So, I'm finished my testing with the GTX 690. So, in case you guys were wondering 00:03:03.280 --> 00:03:11.280 what a GTX 690 looks like on the test bench, it's a little something like 00:03:07.760 --> 00:03:12.640 that. Looks pretty cool. Let's turn off 00:03:11.280 --> 00:03:18.319 the lights. You can see the glowing GeForce logo here, which glows very 00:03:16.000 --> 00:03:22.159 strongly and looks very cool as well. One's pretty toasty. 00:03:20.560 --> 00:03:30.000 Little bit of coil wine. Not too bad. Not as bad as uh my 680 or uh or any of 00:03:26.800 --> 00:03:31.680 my 670s, but not as good as 590, which 00:03:30.000 --> 00:03:38.000 had absolutely no coil wine. At least the one that uh that I have. Um 00:03:36.319 --> 00:03:42.720 overall, not that loud though. about pretty similar to GTX 590 in terms of 00:03:40.959 --> 00:03:48.319 the actual fan speed. Still got a game running on it right now. Definitely 00:03:45.120 --> 00:03:50.480 quieter than the 3-way GTX 670 solution 00:03:48.319 --> 00:03:57.439 and uh consumes less power, but remember it's two GPUs versus three GPUs. And I I 00:03:55.360 --> 00:03:59.920 see that that's five video cards, but I was only running three of them at a 00:03:58.640 --> 00:04:05.599 time. So, I'm just going to make up my little graphs and then we're good to go after I tell you the test bench, of 00:04:03.120 --> 00:04:12.239 course. So, this is a Radeon radio and radon nothing. uh Core i7 3930K at 4.4 00:04:09.920 --> 00:04:18.560 GHz with 16 gigs of Kingston HyperX RAM at DDR3600 00:04:14.720 --> 00:04:20.320 megahertz. I also have a Mushkin Kronos 00:04:18.560 --> 00:04:24.000 Deluxe drive in there and I've got my Gigabyte 00:04:21.840 --> 00:04:30.639 uh what is this thing? It's a GA something or other X79 00:04:26.560 --> 00:04:32.639 UD7. So, this is up to a 4-way SLI 00:04:30.639 --> 00:04:40.000 Crossfire X capable motherboard, and it's all powered by an OCZ ZX series 00:04:36.160 --> 00:04:42.479 1250 watt power supply. Easily capable 00:04:40.000 --> 00:04:46.880 of powering those three GPUs. Actually, you know what? It's uh 2:00 in the 00:04:45.440 --> 00:04:52.320 morning and I'm actually really tired, so I'm not going to make graphs, you guys. You'll just have to bear with me. 00:04:50.240 --> 00:04:58.160 So, here we go. Um, I actually have an upcoming video where I plan to do some 00:04:54.080 --> 00:04:59.840 GTX 670 versus Radeon 7970 scaling, 00:04:58.160 --> 00:05:03.840 which will be really cool. So, 1, two, and three-way configurations for both, 00:05:02.000 --> 00:05:08.960 seeing how they compare. But for this one, we're just looking at three times 00:05:05.440 --> 00:05:11.120 GTX 670s versus a single GTX 690. What 00:05:08.960 --> 00:05:16.240 is the better value proposition? So, you can see here in Crisis 2, the uh 3-way 00:05:14.240 --> 00:05:20.720 SLI configuration actually has a lower overall minimum, but a higher overall 00:05:19.199 --> 00:05:25.360 average, but not by much. We're talking a 5% performance advantage. In 00:05:23.199 --> 00:05:31.840 Battlefield 3, the 6 the 3x 670 solution just destroys 00:05:29.280 --> 00:05:35.840 the single 690. So, this is excellent scaling that we're seeing in this 00:05:33.199 --> 00:05:41.759 particular game. Um, I mean, we've got an additional 25 to 30% performance, 00:05:39.440 --> 00:05:46.720 probably about 25% performance from adding that additional GPU. So, that is 00:05:43.919 --> 00:05:49.520 a pretty good example of of of excellent scaling there. And you can see the 00:05:47.759 --> 00:05:56.240 minimum is even higher. That's more like 40%. Outstanding. Skyrim. Yeah, those 00:05:53.759 --> 00:06:00.400 look pretty similar to me within margin of error on this side. However, that 00:05:58.160 --> 00:06:07.680 minimum frame rate is very different. So, the GTX 690 was uh actually well the 00:06:05.120 --> 00:06:12.080 3-way SLI configuration was about 35% higher than the GTX 690 in terms of 00:06:09.759 --> 00:06:18.160 minimum FPS. And I did notice that when I was actually using it. Uh, Witcher 2 00:06:14.560 --> 00:06:21.039 is another one that uh, scaled okay in 00:06:18.160 --> 00:06:24.880 terms of averages with the 3 by GTX 670 configuration, but the minimum was a 00:06:23.360 --> 00:06:29.520 little bit lower. Well, quite a bit lower. But something to bear in mind about these minimums is sometimes when 00:06:27.680 --> 00:06:36.080 you're benchmarking, your frame rates will dip for sort of no reproducible or 00:06:33.360 --> 00:06:38.639 easily explainable reason. And uh, so there's a little bit of that to bear in 00:06:37.280 --> 00:06:44.000 mind as well. Personally, I know minimums are important to a lot of people, but personally, I put more 00:06:41.759 --> 00:06:48.639 weight into the averages, especially because of the way I'm collecting data 00:06:45.520 --> 00:06:50.720 with Fraps. Um, you can get a minimum 00:06:48.639 --> 00:06:54.080 that gets, you know, if both cards dip like this, you can get one of them that 00:06:52.400 --> 00:06:58.240 records at the very bottom of the dip because it happens to fall on that split 00:06:56.400 --> 00:07:00.880 second and one of them that falls here, but that doesn't mean the dip didn't 00:06:59.440 --> 00:07:06.080 happen. So, something to bear in mind about Fraps frame rate recording. So, 00:07:03.199 --> 00:07:13.840 thank you for checking out my 3-way GTX 670 versus GTX 690 performance review. I 00:07:11.039 --> 00:07:17.360 think that we can see that 38 by GTX 670 is a very viable option if you really do 00:07:15.520 --> 00:07:21.680 want the ultimate performance and you can't get your hands on one of these bad 00:07:19.520 --> 00:07:24.720 boys. Don't forget to subscribe to Linus Tech Tips.