Gadgetory

hey everyone today we're talking about the gtx 960 and for gigabyte form so the gtx 960 shipped a while ago we have a full review on the channel and on the website and i've described the architecture i've described the memory subsystem done the benchmarks today we're looking at the 4 gigabyte version this is an EVGA gtx at 960 Super SC it's a super clocked card ships with higher clocks than the reference design and it is 4 gigabytes equipped so the stock model ships with 2 gigabytes of memory and that's what I have in this Asus Strix 960 which is the one I tested the 4 gigabyte model against and this is a 2 gigabyte card so that is more similar to what shipped a few months ago and what we're looking at with the 4 gigabyte card is strictly one thing here it's will 4 gigabytes of video memory actually get utilized will it benefit you as a user versus the 2 gigabyte 960 model that is already out for $210 the 4 gigabyte model this EVGA one specifically is available at about 240 250 depending on where you look so it's a price jump of about $30 and we need to know whether that actually gets you anything in gaming historically for gigabyte cards have not necessarily done a whole lot versus 2 2 gigabyte counterparts and this is because of all kinds of bottlenecks elsewhere in the architecture and with the 960 the concerns are still present so we have a 128-bit memory interface on the gtx 960 and even the GK 106 predecessor to the GM 206 which is what the 960 uses had a larger memory interface it had 192 bits however Nvidia EVGA everyone will be very quick to tell you that the Maxwell architecture is well optimised for memory it actually uses a lot of memory compression technology including Delta color compression which compresses about 18% of the memory transactions occurring within the chip and the memory so this ultimately produces better performance on the GM 206 versus the GK 106 despite having a smaller memory interface so it's not all just a numbers game there's all kind a compression going on there too if you look at all the other technological updates in the Maxwell architecture we end up with about a 25% bandwidth increase and memory capabilities and transactions so overall it is better than the predecessor despite the numbers being slightly lower still 128 bits even with the 112 is gigabytes per second memory bandwidth of the 960 has the GM 206 has we're not 100% sure on paper whether a four gigabyte card will benefit the user and so that's what we tested in these tests in order to make sure we were actually utilizing the memory what we did was we first measured the video memory consumption of each of the major titles and I'm only showing a few of them here on this chart but we did test all of them and they're all linked in the description below the whole article delves into all this very deeply including the memory architecture and games like Assassin's Creed unity far cry 4 and battlefield hardline somewhat unsurprisingly use the most video memory and in theory should get the most direct benefit from my 4 gigabyte card because they're actually exceeding 2 gigabytes so one would think that because they're exceeding 2 gigabytes then a 4 gigabyte card will allow greater frame rates this isn't always the case and we'll go into that in a second here let's talk about what video memory actually does before looking at the benchmark numbers video memory on a video card is gddr5 memory and it is significantly faster than ddr3 system memory it is also located physically on the PCB where the GPU is located so it doesn't have the transfer over such great distances as we would with ddr3 it also limits some of the interfaces and bosses that we have to travel through to communicate between the GPU the rest of the card subsystems and the memory so for this reason we want to store as much of the games information on the video card memory as possible some of the things that gets stored in the video memory would include textures which if you download mods for Skyrim you might have 4k resolution textures those are definitely being stored in the video memory up to the Pasi limits and this allows the video card to more rapidly transact fetch that memory that's containing the textures and display it on the screen without too much of a frame drop or noticeable jitter in the frame performance we also store things like shadow maps and normal maps and specular maps all these different mapping elements that are used artistically in the game to display information with a certain effect those are all stored in the video memory and then they're accessed effectively every frame so that's pretty regularly now what happens when we run out of memory on the video card well what ends up going on is the system memory and the video card memory must swap with each other they have to start swapping data and we store only the most frequently accessed files in the video memory or the currently accessed files in the video memory everything else is the system memory during the swapping we're forcing the transaction to occur through a PCIe bus and that adds latency that adds physical travel time to access the system memory so now we have added latency just to go through the bus to get to the system memory come back and we're communicating with a CPU during all this and the GPU and it adds a lot of overhead so you want to avoid that wherever possible and you really the only good way to do that is to improve the memory subsystem of the architecture and to improve the capacity of the memory on the video card which is the one we've done here with the GTX 960 have four gigabytes when the video card begins cashing out on the memory you'll start seeing things like very large and noticeable frame drops and gameplay so you might see a sudden minimum of 9 FPS your game sort of just freezes almost out of nowhere and that's when it's fetching dumping and swapping data most likely it could be a million other things as well but if we're talking strictly on the video card side that's sort of what it would be and then we have to wait for it to catch up and refresh everything and we can play at a normal frame rate again but it's not always this simple because everything depends on how the game has been optimized how the drivers have been optimized how the video card was built how the architecture is engineered how the API like DirectX or mantle is interacting with all this in between so there's a million parts to this and it's not as simple as just memory but strictly for purposes of this benchmark we're only looking at memory to whatever extent is permissible games like Assassin's Creed far cry 4 are very poorly optimized so we can only get so far with those games and games like Metro last light and grid are extremely well optimized and that presents other problems as we discuss in the article linked below so it's a very interesting problem the benchmark and to sort of scientifically try to try to draw conclusions of system performance so how does this thing perform well in Assassin's Creed unity we see one of the biggest disparities in performance at the 1% and 0.1% lows and these lows are representative of your effective minimum for the 0.1% and your most noticeable frame drops for 1% when I say these percentages what I mean is with 1% low you're reflecting 1% of the lowest performing frames so rather than looking at an outlier like minimum which a lot of sites do we're looking at something that's more statistically replicable and noticeable to the user so at 1080p assassin's creed sees a giant jump from the minimums at the 4 gigabyte and 2 gigabyte cards but it's not persistent through all the other games with far cry 4 our gains are much less interesting the only really massive difference is at 1440p where we see the lows jump 52% using the 4 gigabyte card and that actually does become quite noticeable in gameplay it's a lot less jarring in this fashion but not all games even are like this so you have grid Autosport you see in this chart almost no difference whatsoever at any resolution for the performance of 4 gigabytes versus a 2 gigabyte card and specifically talking out this EVGA card we're testing they market the device as up being that are optimized for 4k gaming and better allowing higher textures and higher resolution gaming at 4k so 4k is sort of a buzzword for the card in terms of marketing speak and that's why I'm testing all these games with 4k resolution except for Assassin's Creed which is really far too heavy on the GPU to test at that resolution Metro last light is another title that sees effectively 0 from four gigabytes of video memory with the GTX 960 at all resolutions from 1080 to 4k and so in this instance it's not worth buying if you're just playing about your last slide we've got city skylines a new game that's actually really cool for 4k testing because it is the kind of game you would want to play at 4k and I say that because you when you zoom out you really can see a lot of the game at once you can fit a lot on the screen and it looks good at a high resolution but even so doesn't change the fact that there's no noticeable gain using a four gigabyte card versus a two gigabyte card and then finally we look at battlefield hardline a new battlefield title that uses massive amounts of video memory 4 gigabytes effectively in the test that I was running using a four gigabyte card so it was maxing out the memory and in battlefield hardline we do see pretty substantial differences between 4k resolutions with the cars I know this only has average FPS that's because we couldn't measure with fraps I'll explain in the article regardless big difference with 4k but still unplayable 4k even using the 4 gigabyte card that said if you were to tweak your settings and lower them from Ultra down to high or hybrid of medium high then you will see an advantage with the 4 gigabyte card that is actually playable 1080 and 1440 not quite as impressive and and we do see a bit less gains at these resolutions overall we can look at synthetic tests like 3d mark where there's almost no difference we can look at real world game tests like grid where there's almost no difference or assassin's creed where there's a huge difference and we're left with a very simple answer whether the 4 gigabyte card is worth it for you versus the 2 gigabyte card depends it entirely depends on what games you're playing and what you hope to do with the video card higher resolutions will see a more noticeable gain for different games and in that respect to be playing 1440 you should probably consider a 4 gigabyte solution whether or not it's the 960 you should be looking at something maybe higher than 2 gigabytes regardless 4k you're probably not going to play into regularly on the 960 it is possible with some games it's kind of possible with things like City skylines but really nothing like Assassin's Creed unity don't expect to play that at 4k regardless of the amount of memory on any 960 beyond this the 9 64 gigabyte model is $240 so it does enter some of the AMD price territory where there are very competitively priced cards available and it's not too far away from a 970 so you do need to consider those options but I will leave that part of the argument up to you it is up to you to decide whether you want to consider some of the other devices and we've reviewed almost all of them on the channel at this point ultimately at the end of the day for gigabyte 960 video cards do have a noticeable substantial advantage in framerate for some games that said other games either because they use less memory or they're better optimized will see effectively zero advantage from the memory increase so keep all these charts in mind when buying the card determine if it's right for you or not if you're still confused leave a comment on our forums we check those much more regularly than the YouTube comments please subscribe if you like this content sort of lets us know we can see how many of you subscribe and like each video so it lets us know if you like the format or if we should change it up and thanks for watching I will see you all next time you