Gadgetory

hey everyone today we're doing an SLI benchmark on gtx 1070 so we just got our second 1070 in the shop this is an MSI gaming X and we've already got the founders edition aka reference 1070 so we're putting these in sli today for a benchmark and before getting to that this coverage is brought to you by enter max and I knew it look max a giant open-loop liquid cooling solution so the SLI setup here this is not an ideal setup and that's for a few reasons one these cards are not identical you normally want identical cards just kind of ideally to keep things simple but that's not what we have that's not what we can get because this is still a brand new card and right now a mix of two is the best I've got so we're going to do that anyway because it still works it works just fine really as a user the downside is that you're generally going to be kind of locked to your slowest device that would be the founders Edition card in this case so we'll lose some of the overclocking pre overclock benefit of the MSI version but we're going to go ahead and use them anyway so I'm just going to put this in the top slot it is the best card after all and if we decide to do any overclocking later it's just going to be easier if I have that there especially if we remove the founder's Edition one so we've got that in there it's good enough let's go ahead and connect this one to the lower slot and now for SLI the game has changed so as you've probably seen in our review of the 1070 the 1080 it's not the same anymore as it used to be where you can just use one cable between these feet at the top now there needs to be a high bandwidth bridge for Pascal which we don't have a high bandwidth bridge but we've got these LED bridges they're called and by Nvidia spec an LED bridge is just as good for our purposes the only place a high bandwidth bridge does better is in 5k and surrounded you're not testing today so we're going to go ahead and use the LED bridge again effectively the same as the high bandwidth bridge for our purposes this is a high bandwidth bridge higher clock rate and we've we've encountered a problem already so here's the here's the issue because the reference card has a shorter PCB and because I have no flexible bridge that's that's obviously not going to work I don't know if we can we're not going to be able to force that to work so that's no good unless we start hacking apart the bridge or something like that that's no good so want we got a couple options here one of them is and I will explain this I'm not going to do it but I'll explain it just so you all know we've got these two ribbon cables these are not high bandwidth even though one is branded as ro G it's not special it's just like this one branded by Nvidia so what one thing that people have asked is can you just do this do two ribbon bridges equal one high bandwidth bridge or LED bridge and the answer is no so these are actually regardless of the fact that there are two they are operating on a 400 megahertz clock it's they oscillate at 400 megahertz and these bridges are 650 mega Hertz that's a good deal different and having two doesn't make up that you get some gain but you don't get it all back and we'll show that in a future benchmark so I really don't want to do this this isn't going to be a representative of the best-case scenario and I don't know that we have the ideal situation for this but one thing we can do I think I have one somewhere here we go so I've got these riser cables some people are probably laughing at this already but we have these from sis for use in small form-factor pcs and riser cables are jet all they do is sort of elevate the PCIe slot and they're used very regularly in small form-factor pcs that the ax system integrator is build and that's normally just show off the faceplate of the card but we're going to use it for a more practical reason we're going to connect this here I I'm sure some of you will cringe at this but it will work connect that there is nothing conductive on the bono this is all plastic and ribbon so I'm not worried about that we can contact the board I connect this card to it and now our video card is taller it's a bit floppy but it's taller so let's go ahead and give us I think this we have to figure out which one of these will well line up I think this one will work this is actually a four-way bridge for Maxwell but it's still high bandwidth and as long as we line up the parts here with two headers then it'll it'll work well for Pascal so let's see if this will line up that looks promising okay so that's actually uh it's actually lined up I'm going to be let me needlessly neurotic here and get it so the MSI logos face up and so that the extra header at the top is bleeding over the north of the top card just because I want it out of the way in case we're doing something over here so they're connected this is reasonably stable fans like me spinning crazy fast so it's not going to spin loose or anything like that I am I'm happy with that for testing I would not say this is representative of what a user would do because I don't don't think first of all we can't get a screw in there unless we like an inch long screw secondly that's not going to work in a case but it will work for testing and so here's the thing we're using a ribbon a riser cable now PCIe riser cables we've tested this have about a 1% measurable but imperceptible impact on framerate performance so what does that mean here well when I say 1% I'm talking about I tested out on Maxwell card and it was a single card this is Pascal first of all Pascal has new sli rules and it's also 2 devices so I'm not a hundred percent sure how big the impact will be but I would expect still somewhere around the 1 percent area we can still get really good and accurate results from this configuration just know that they might be slightly improved if we were to do it with cards that are the same height what you want to do anyway but in order to test the impact of this cable I would need two devices of the same height anyway and I would need the high bandwidth bridges but we have some friends at other organized organizations who are helping us test that so let's go ahead and throw this on the bench we've got it set up more or less I'll do the power once that's connected we'll run some benchmarks and come back with the results alright so we've got the results in ran all of the tests for 1070 SLI versus 1070 standard including a 1080 and a bunch of other cards let's go through doom ashes of singularity Mirror's Edge catalyst the division and gta5 I ran a lot more tests than our present here I had help from some of our technicians with these so check the article links in the description below for all of the games tested but let's just start with a couple of the heavy hitters here talk about power and then we'll see what the results look like so on the screen now is the test bench we're using as I just said more information in the article below for testing methodology and results we use two different gtx 1070 s for this test because that's all we have available in these early days the card one as you saw it was the msi 1070 gaming X the other was a reference card we ran the SLI configuration at the reference clock for these tests giving us a clock versus clock comparison for a single 1070 versus two and we also tested the 1080 at its founders Edition clocks in all the tests the only instance where we tested the prio seed msi 1070 standalone meaning one of them at the pre OC value was in Mirror's Edge catalyst because brand-new and we're transitioning to the AIB partner cards for our benchmarks let's get into it with doom first doom does not have good SLI scaling at this time SLI is not officially supported in any respectable regard and currently forces AFR to to achieve it function without any real gains at 4k doom is intensive on all devices the 1080 Fe operates at 50 1.7 ish FPS average and that holds an imperceptible but measurable lead over the 1070 SLI configuration at 49 FPS average that's a 5.3 percent gap that's outside of our sort of exam I expected 1% difference from the riser but still very small a single gtx 1070 performs slower than dual SL the sli GTX 1070 s by 15.3 8% so it's really not good scaling and would be poor value laughable value for owners of 1070 sli but 4k is also intensive at 1440p sli shows in a more exaggerated fashion its 1% low and 0.1% low frame time performance which is less consistent than single cards the SLI configuration over a single card making it completely pointless for doom at this time its software and NVIDIA are supposed to be adding a better SLI support in the near future and volcán but it's not here yet 1080p is just as bad with a 3.4% delta between averages with the single 1070 versus dual 10 70's and doom is just like some other recent games we've tested including Just Cause 3 Assassin's Creed syndicate and some others where SLI just doesn't offer any value at this time ashes of singularity doesn't improve the narrative using the DirectX 12 API the SLI GTX 10 70s fall to the bottom of all charts and ashes of singularity showing dismal frame times like 42 point 5 4 milliseconds average at 1080 high and DX 12 against other Nvidia devices even the all the single cards this is a major downturn and is a result of SLI it would actually better to disable a slide for ashes of singularity and dx12 leaving the second card to rot in the case doing nothing the x11 performance is it more reasonable though and frame times are closer to a single 1070 which pushes 20.4 6 millisecond frame times averaged vs. 23.0 2 milliseconds for SLI scaling is around 5% for 1080p high so that's really not good and definitely not worth it you're gaining 5% for an extra 380 to $450 investment but at least it's better than the dx12 dip which was terrible this is just one game though so you really can't take much away from this other than the fact that sli 1070 s don't work well in ashes of singularity for DX 12 and that's a very specific use case it does not mean that they're bad in all DX 12 games or in all games just ashes of seeing with dx12 let's move on to the division the division shows an improvement over the last few tests for K post scaling of twenty nine point two seven percent over a single 1070 F II card and the SLI ten 70s we're on eight point seven six percent faster than a single GTX 1080 the one percent and 0.1 percent lows are worse though and represent poor frame consistency as tends to happen with SLI 1440p shows better performance for the dual GPU setup the ten 70's in sli push one hundred eighteen point three FPS and hold a substantial thirty-two point seven six percent lead over the ten eighty as the ten eighty now the ten seventy one percent and 0.1 percent lows it might not be as high as they should be but are still superior to single GPU configurations so we're still good shape overall against a single gtx 1070 the SLI configuration holds a fifty point three seven percent lead an average FPS 1080p scales about thirty eight point three percent over a single 1070 with the average FPS at one fifty seven in its best-case scenario 50 percent scaling really isn't bad for the ten 70s and begins to show some real value but let's see how they do in Mirror's Edge catalyst and GTA 5 this game just came out you can view our separate GPU benchmark video on the title to learn more about the results overall and pitfalls and things like that general intensiveness of the settings note that we did run the GTX 970 gaming X for this test rather than the founders edition so we're comparing against a pre overclocked single GPU here not the founders Edition at 1080p hyper the gtx 1070 sli tops the bench with 105 point six seven FPS average and retains fairly strong 1% and 0.1% lows despite being imperceptibly lower than the gtx 1080 the SLI configuration scales twenty three point six eight percent over a single gtx 1070 which is eighty three point three FPS putting it within three FPS of the single gtx 1080 or about a delta of 3.5% 4k high shows a gap of four fps between the 1080 and 1070 sli and a gap of seventeen f s between the 1070 and 1070 SLI the 1080 and 1070 SLI are separated by seven point five five percent but the one percent and 0.1 percent low metrics are much higher on the single 1080 showing a twenty two point three percent Delta for one percent lows and about the same for 0.1 percent lows this is a fairly constant trend for SLI and crossfire and versus a single 1017 on SLI we're seeing Delta values of thirty six point five six percent for the average FPS with the 1% in 0.1 percent Lowe is marginally improving note again that this single gtx 1070 here is the gaming x so the gap is shrunken versus a reference 1070 with a lower clock rate at 1440p we see a delta of 8.4 percent with the 10 70s and sli versus the 1070 single card but remember that the settings have changed from high at 4k to ultra on 1440p which shifts some of the load in the pipeline to be more intensive after rasterization because there's so much more post-processing going on the 0.1% lows are a fair bit reduced for SLI versus the non sli set up posting forty eight point six seven FPS zero point one percent low vs. 63 FPS 0.1% lows 1080p ultra was pretty uninteresting you can see those results briefly on screen now but check the article for more detail on that GTA 5 at 4k it shows 55.8% scaling versus a single gtx 1070 in the average FPS department so that's finally a major improvement 1% in 0.1 percent lows are also improved in that land above 30fps this makes 4k at very high ultra sort of mixed settings completely playable on sli gtx 1070 s whereas a single 1070 sees enough dips in performance in 1% and 0.1% lows to necessitate a very slight settings reduction versus a gtx 1080 the sli 1070 s at 4k show an improvement of 34 point oh seven percent so even against the $700 F II version the 1080 we're getting a 34 percent performance gain with effectively reference clocks on SLI 10 70s that's pretty reasonable value anyway we hit a CPU bottleneck at 1080p with the GTX 1080 and SLI 10 70s and effectively the the 1 GTX 970 all landing at around the same 120 to 125 FPS limits so you can't really use these results because of that bottleneck so this is just a cross-section of our test we also did black ops 3 shadow of mordor and Metro last light and all those are in the article below very quickly black ops 3 did show some pretty insane scaling in a good way in some of the tests including at least 50% gains and in some cases more than that so it really just depends on the game you're testing how good the gains are others like the division or rather Mirror's Edge catalyst Doom results are kind of mixed in some cases or just terrible depending on the km so it all depends but let's talk quickly about the power and then get into the conclusion here just to quickly make a note of power draw you obviously do consume more power by going with an SLI setup versus a single GPU setup that's an immediate drawback but can be overlooked by some users so it's really up to you whether or not you fall into that group where it's something you can overlook that's light n 70's post a full system draw of over 300 watts when drawing the when running the graphics test to 3dmark benchmark and you can see the 1070 single card and 1080 single card power draw metrics also in this chart with the 1080 doing a bit better in terms of power draw overall but it's weaker in some gaming so just depends on what you're after so that pretty much concludes our testing of our next Franken build which is this really crazy riser card and it's not that crazy but I guess a better word would be precarious or haphazard or a lot of money waiting to be in a trashcan if it goes if the bridge becomes loose but we tested it it went pretty well I wouldn't recommend building it this way but let's talk about if it's worth doing at all in terms of just SLI 1070 there is a single 1070 first of all SLI and crossfire have always been a hard sell that's because of there's a lot of complexity so the risk is obviously a game comes out it's not optimized by the developer or by the drive is to run SLI or crossfire in such an instance you basically end up with one functional card and then a brick that does nothing you disable crossfire SLI through the control panel for your GPU vendor turns it off doesn't do anything so obviously as a consumer that feels pretty bad because you've got in this case 380 to $450 depending on the price the card doing nothing and your build and that could have been put towards something like a single 1080 of course and you'd actually be spending less money so that's the risk with SLI now the gain is when there are gains they're pretty good and they can put you in theory beyond what a single 1080 can do which we've tested and you've seen in some instances so that's the that's the risk with SLI lower frame rates and poor frame times are also an issue where you'll see these 1% in 0.1 percent low metrics dipping below a single 10 70 even though the average is higher and depending on how they dip if it's a really hard hit it introduces something called micro stutter or really just this sort of really nasty judder when you're playing and that's not something you want that's actually worse even though you've got a higher average it plays worse with the micro stutter so it makes sense again to disable a slice that's one of the other issues here AFR introduces some of these just general complexities having multiple devices introduced to some of these and a lot of games now are also moving to interdependent frames which means that the frames depend on one another when they're being rendered so that's either temporal anti-aliasing or temporal effects or post processing all the stuff that happens at the end of the GPU pipeline after geometry and rasterization that's all at the end and when they're interdependent it's worse for SLI that's why Nvidia is moving away from bass lab at least in terms of 3-way 4-way SLI if not all Isolite altogether in the future but for now they've moved away from three-way and four-way SLI and they require an enthusiast key to unlock it and that's why it's because of both these frames now the way they're being rendered because the new API is DirectX 12 in bulk and it changes the way games address multiple GPUs still there are some benefits so in black ops 3 we saw huge gains where if you even had a 200 Hertz monitor which is absurd you could actually fulfill that full 200 Hertz with 200 frames per second and you'd actually be saturating it depending on your graphic settings of course 1080p you'd be fine 1440p I think we're at 180 FPS which is tremendous so a lot of frames for a game like black ops 3 so it there are benefits and black ops 3 is one of those games that's actually very almost surprisingly well optimized for all vendors so there's there's good things we saw 50% scaling and beyond in games like GTA 5 sometimes in the division that's useful for 144 Hertz gaming or just to make 4k playable because GTA 5 very high in ultra settings 4k is not really perfectly playable on even at 1080 and the 10 70s in SLI step past that and make it now pretty playable more than 70 FPS in some instances so that's good too and then 210 70s can Kotwal will cost more than 1 1080 so that's one thing to note here 110 80 you buy cheap one 600 bucks buy middle-of-the-road one 650 bucks let's just let's put these up sort of the low-end let's say you buy these for $400 each that's $800 so spending 200 to 150 more for this setup so you want to get 200 150 more performance out of it and that doesn't always happen even at its best though we're still we're seeing deltas of around 30% versus a single 1080 so that's decent but whether that's worth it is up to you and it depends on a game too so it can cancel out depending on if you're playing a lot of different types of games where 1070 just doesn't even work in some instances like doom then you kind of lose out multi-gpu it's still a hassle it's still a big bet to make you sort of have to look at our data on a game by game basis decide for yourself if you like it or not and hopefully the data is not for you to make that decision now I don't normally personally give an opinion on what I would do with if I were in your situation trying to buy cards but I normally do with SLI and crossfire and that's because it's kind of a weird genre of buying devices that is hard to figure out on your own so here's what I would do I would not buy 210 70s personally and I've said this I think in every SLI or crossfire test we've done I would still buy a single card that's because I play a lot of different types of games when I do play games and I don't want to deal with the issue of it just not working which is what happens with something like doom or some of the other games we've tested so Ash is a singularity is another good example so I would still be buying a single card maybe a 1080 if I were spending that much money or waiting and seeing what else comes out of course the four ATS are coming out they're talking big about Crossfire's will try and validate that see if the claims are reasonable and if they are and if crossfire does better than SLI well maybe that'd be a reasonable option but otherwise I'd be looking at a 1080 rather than 210 70s save some money you get good performance sure in some games you're 30% slower but it's still so good that's kind of who cares so that's how I look at it but as always all the informations on the website link below for you to work out yourself thank you for watching page fouling to post roll video subscribe for more content and I'll see you all next time