Gadgetory


All Cool Mind-blowing Gadgets You Love in One Place

CrossFire RX 480 vs. GTX 1070, GTX 1080 (FPS, Power, Thermals)

2016-07-09
and he made a lot of fanfare about crossfire with the RX 480 is mainly pointing to higher performance on to four 80s which would cost 200 times two to four hundred bucks or 480 bucks depending and saying that was better than it's single gtx 1080 which is of course a higher cost product 650 to 700 dollars or more as we all know with the current product availability so that was the fanfare at the Macau event and later at Computex with the RX for 80s Andy used the game ashes of singularity to benchmark that's the only one that we saw in sort of a live benchmarking so we bought our own second R X 480 paired them together we've got benchmarks for you here I did more than FPS we did thermal testing noise testing and power testing and we've also got the usual FPS benchmarks and you can find all of these linked in the description below in the article if you want full detail for things that might not be covered in complete depth here so first off when does crossfire work well and why crossfire mostly works in games where friends are not interdependent on one another so games where you have really heavy post FX post-processing and temporal effects effects that span from frame a to frame be like temporal analysis of samples of pixels for things like anti-aliasing things like that those types of effects and game environments do not necessarily work well with multi-gpu setups whether it's crossfire or SLI and this is assuming sort of an Lda or maybe even some MDA types of games so the reason for that is because the frames are interdependent so with AF our alternate frame rendering gp1 renders a frame GPU to renders a frame they put them together they output it that's your that's your frame rate so this is where with post-processing you kind of lose some of that performance and you can actually have negative scaling and we'll show that here so this table you're looking at is our bench with all the hardware that we use for these tests you can find that again link to the description below with the fold charts first we'll talk about thermals so thermals for multi-gpu configurations can kind of go through the roof with tight spaces blower fans help a bit by nature being more exposed to the air and Andy's got a cutout in the back of their cards that further helps with some breathability we tested crossfire with our cards two slots apart or one full usable slot gapped and checked on thermals peak temperatures and things like that this first chart is a zoomed out look at the frequency of GP 1 GP 2 and the temperature of GP 1 which is the hottest of the pair and this is an absolute temperature so it's not the delta value that we're gonna show in equilibrium charts note that the clock rate fluctuates when exceeding actually the fluctuations themselves exceed a range of 100 megahertz in some cases so we see a pretty big swing and that can cause spurious frame output and low frame times when the cards are under more duress and to offset this just like with the Nvidia cards and the throttles it's clock rate temporarily for a heavy thermal reduction so this process it's actually pretty normal between a.m. the end and video and repeats ad infinitum until you either stop gaming or things have smoothed out but let's zoom in to make it a little more clear what's going on so we've distorted the scale here and for good reason the temperature scale is now 70 to 90 and that helps see the spikes with greater resolution time scales been restricted to 500 to 800 seconds rather than the fold 0 to a 100 notice that every temperature reduction is accompanied by a brief frequency reduction and the two are directly related in their function in this way if you look at them in the charts the second card that seems to rounded a lower frequency throughout the test and isn't matching the frequency of the first GPU necessarily and if we look at our charts from the single arcs for a tea review we see that similar frequency and thermal values are being hit but the our X 480 does a breathe better overall and that's obvious because it's not being blocked by a second card so here's a chart looking at the quote unquote equilibrium temperature values basically average peak thermals for a single card all the other cards and the crossfire rx for a TS power draw of course is a very important metric when looking at multiple GPUs because even though you're increasing your potential for performance you're also drawing a lot more power so this shows total system power draw in volt amps or apparent power consumption and that means it's for the complete system not just the GPU and what we're seeing is nearly four hundred and thirty volts amps for the crossfire our X 480 configuration and as versus 240 on a single our X 480 and then of course if you look at the overclock power drop for the single our X 480 we would draw way more power we applied that to crossfire so you have to take that into account if you're planning to do crossfire with AIB partner cards that are pre overclocked a single gtx 1070 sits at 263 volt amps and a single GT X 1080 sits at 280 point six three volt am just for a comparison moving on to noise we gained a couple DB across the board with the crossfire configuration we're at 38 point 4 DB idle versus a single rx for ATS 37.3 DB outputs 41 to DB Auto versus 39.4 DB for the single card and then 44.9 at 50% vs. 42.8 also at 50% for the single card and 60 7.4 DB at a hundred percent which is very loud versus 60 5.6 DB for the single card also at 100% so although the DB increase looks like maybe one or two decibels in a lot of cases and of course this is logarithmic so it's not just a subtraction addition thing although that is the case there's a pretty annoying whiny noise that does bring itself out a bit more with crossfire at the hires fan rpms and there's also a slight vibration that's caused from turbulence between the two cards it's not that annoying if it's within a case and you're under 50% fan speed you're probably fine but past that point it does get to be a bit annoying ok so enough with the power noise and thermals now we're gonna get into the game benchmarks these are not all of our benchmarks you can find the rest again in the link that's in the description below Mordor shadow of Mordor at 1080p ultra we see a thirty eight point seven percent increase over the single our X 480 the 0.1% lows are a little on the low end but overall acceptable and this pairing is three point nine percent faster than a gtx 1070 the GTX 1080 is eleven point seven percent faster than the crossfire Rx for ATS moving the 1440p we see the poor 0.1% lope introduced some noticeable stutters and framedrops but we've improved our averages overall and see a scaling of 72% over the single our X 480 performance improvement is about twelve point four percent versus the gtx 1070 gaming X or eleven FPS average if you want that number and that said the GTX 970 gaming act sustains a better 0.1 percent low performance and this is generally a trend with multi-gpu configurations versus single 4k pushes the cards closer together with minimal average FPS gaps differentiating the GTX 1080 and the RX 4 80s the gtx 1070 runs at 50 1.3 FPS average but keeps significantly better 0.1% low frame times than the RX 4 80s in crossfire GTA v was retested using these 16 not 7.1 drivers that results stuttering the rest of these games were tested with 16,000 6.2 at 1080p at the very high and ultra setting combination we use saw that rx for 80s and crossfire performing at 106 FPS average with slightly lower 0.1% lows than the single our X 480 configuration they're still positive scaling on the averages since we've seen the single 8 gigabyte card runs at 85 point 3 FPS and that's a scaling improvement of 19.5% so not impressive but definitely present and 0.1% low values are negatively scaled with the single card performing nearly 20% better than the 2 R X 480 is strictly in 0.1% low values at 4k with GTA 5 we're seeing the crossfire rx for 80s pull ahead in the charts and achieved parity with the stack of GT X 1080 GPUs the crossfire Rx for 80s are outpaced only by the AIB partner gtx 1088 from msi and the sli gtx 1070 s reference cards may be out of black ops were clearly hitting a throttle point at 1080p in the current newest Call of Duty game for now anyway with our top few cards hitting a wall around 218 FPS there's obviously a throttle point so black ops 3 remains one of the better optimized games on the bench and also has a history of posting better and the performance than in some other titles the crossfire cards tie with the SLI GTX 10 70s at 1080p creating a scaling of 39% over a single rx 488 gigabyte card and that a 130 2.3 FPS average for the r x 480 as opposed to 218 on the crossfire cards at 1440p we see the sli gtx 1070 s leading the crossfire our X 480 is by 3.5% crossfire 1% and 0.1% low performance is acceptable though is clearly lower overall than the gtx 1070 and GT X 1080 cards despite maintaining a strong average the crossfire configuration shows an increase of 52% over the single R X 48 gigabyte card moving on to 4k this is actually playable on crossfire for ATS though the 0.1% lows do dip low enough to produce visible stutter when plane it's not ideal but the average is positioned between the GTX 1080 game and X and the sli gtx 1070 s average FPS performance is effectively doubled over the single rx 488 gigabyte card but 0.1% low performance remains comparable Mirror's Edge catalyst uses a lot of post-processing which is often interdependent on frames and that's not great for scalability on multi-gpu configurations and we're actually seeing negative scallion and Mirror's Edge catalyst with a single arts 480 at 1080p Ultra performing 35% faster than the crossfire cards so this is an instance where you would definitely want to disable crossfire to play the game at 4k high we're seen 25 FPS average for dual Rx for 80s that's in crossfire vs. 33 FPS average for a single card so that's again negative scaling granted the game is not really playable on either device at 4k high but the scaling is still negative so it does illustrate that and that's not what you want when you buy a second video card we also tested hyper and 1440p and you can find more information on that below ashes of singularities this games interesting this one is optimized for MDA so we used multi display adapter for this through DX 12 and of course the X 11 does not support that quite the same way as DX 12 does it's not optimized the same way Asha's is where am these our x4 ATS should it be expected to perform well considering they're bombastic comparison versus at 1080 looking purely at frame rates we're seeing the same poor dx11 performance as we've observed on all of AMD's currently tested cards on our bench including the fury X r9 390x and our X 480 but that subsides as we switch over dx12 this inflates and these scaling numbers substantially given the poor dx11 performance so they're scaling does look much better than perhaps the actual perceivable output is but it does help that crossfire rx for IDs out match a single gtx 1070 an average FPS at 1080 high and the two cards nearly achieve parity to 480 as that is with the single GT X 1080 single cards and against a single R X 480 the crossfire configuration runs 36.8% faster 4k high post the crossfire rx for 80s at 59.2 FPS average for DX 12 or between the two gtx 1080 cards and we land above the gtx 1070 gaming X and crossfire than maintains a 41% speed increase over a single our X 480 frame times are also important to look at the average frame latency at 1080p high has the RX 40s and crossfire performing between the GTX 1080 and gtx 1070 and well above this single our X 480 when looking at the X 12 is strictly the X 11 doesn't really look good with the crossfire setup but then again it doesn't look good for any of AMD's cards on this bench and it stretches am these scaling metrics given the poor dx11 performance here's a percent change chart to show scaling from dx11 to DX 12 between these cards and architectures the Talos principle offers both the x11 and Vulkan API benchmarks with dx11 we're seeing clear scaling between the RX for 80s and crossfire where's is the single card option 1080p ultra post an increase of 45.7% for a comparison duel 10 70s posted a performance improvement of 17.4% which that's of course not that exciting but the story changes as we look to Vulcan crossfire for ATS actually posed lower Vulcan performance than the single card then the single rx 480 of 71 point 6 FPS crossfire vs. 74 point 4 FPS for a single art X 480 that makes the single card 3.8 percent faster at 1080p moving to 1440p we see similar results with the single 480 actually about 2 FPS faster than duel for ATS and vulcan using dx11 instead we see 150 3.6 FPS with crossfire vs. 70 9.8 honest card so obviously the API is not favorable towards multi GPUs at this time and that is more on the game side so API matters and that remains true all the way through 4k which is playable on DX Alvin with crossfire but totally unplayable when and volcán the division is another game that shows negative at crossfire scaling at 1080p high our tests are showing a performance difference of more than 10 FPS average between the single card and dual cards made worse by dismal 0.1% low values and that makes the single our X 480 13.8% faster than averages and 43.6% faster than 0.1% lows and then the crossfire configuration 1440p is again similar with the RX for a DS and crossfire dragon behind the single card setup or 56 FPS average versus 50 1.3 FPS average and there's an even bigger gap in the lows as we've seen in the last few tests so that leads us to the question of whether or not this is actually worth it with crossfire cards and my answer is the same as it's been for SLI x 70's and sli everything else we've ever tested for the most part SLI 9 70s we did as well so this is is it's true for Nvidia and for AMD I would presently advise against going with two cards unless you have a very specific game or use case where you know that these cards will be leveraged properly so if you're playing one game where you know the scaling is good you playing that game a lot and it seems worth it in terms of value of spending money on two cards versus more money on one card then yeah I guess maybe consider it but otherwise first sort of a more general use case unless you're taking into account production tasks rendering what-have-you those things we're not looking at here but for a general use case I would say a single card is still better than dual cards and that goes for SLI and it goes for crossfire so I'm making the statement statement for both options here for ATS they have negative scaling in some games which is clear in Vulcan testing of course it's incomplete right now but that is a game and it's something you can play a Talos principle so negative scaling there we're seeing negative scaling in the division we're seeing good scaling and ashes of course you try to expect that they brag on that point and then we see negative scaling and mirrors and we see some spurious performance in terms of the 0.1% low values and other games and that's normal for multi-gpu that's where we see the frame latency czar kind of stretching out between frame outputs because of the way crossfire SLI work so overall not necessarily something I recommend unless again you have a specific use case and you want to do it but power is higher of course then a single card whether that's a single 1070 or even a single 480 if you'd prefer to drop your performance overall and then thermals are higher noise levels are higher and performance is hit-and-miss depending on the game so that's the crossfire test these were both 8 gigabyte cards the results would be different with 4 gigabyte cards we have a 4 gigabyte where's a gigabyte single arts for ID test online you can see that on the channel and that shows where sometimes that u gigabyte takes more sense but sometimes it doesn't matter so as always thanks for watching patron like the post roll video if you want to help us out directly with furthering these projects subscribe so you don't miss any content and I'll see you all next time
We are a participant in the Amazon Services LLC Associates Program, an affiliate advertising program designed to provide a means for us to earn fees by linking to Amazon.com and affiliated sites.