Gadgetory

a new direct x12 benchmark allows explicit multi-gpu meaning that we can finally pair AMD and NVIDIA GPUs in the same system together or you can even pair non same cards from the same manufacturer so if you want to do a 970 to 960 for some reason you could theoretically do that with DirectX 12 in games which support explicit multi-gpu functionality and the new ashes of singularity benchmark is as far as we're aware the first game to support officially explicit multi-gpu and that's what we're benchmarking here today we're looking exclusively at the feature of multi-gpu for things like a 390 x with a 970 a bit odd but should be interesting 970 s paired with one another in normal SLI versus that set up and then a couple of other things thrown in there as well as you'll see throughout the video now we're also doing a DirectX 11 vs. DirectX 12 benchmark which will look only at the DX 11 verse TX 12 performance so we can see where do AMD and NVIDIA have their strengths and weaknesses and that will be in a separate video coming pretty much immediately after this one but for today we're looking at explicit multi-gpu when we say the word explicit for multi-gpu we mean that the developers have low-level access to the hardware and are able to better sort tasks so that vram and other resources never go unused this is contrary to DirectX 11 where the API overhead obfuscates low-level access and restricts developers in ways which have limited game graphics especially on multi-gpu platforms including the standard SLI and crossfire setups with explicit multi-gpu support the developers are able to use all the vram available from multiple devices something that historically has not been possible and they're even able to allow mixed card or mixed a brand configurations before diving deeper though it's worth pointing out a few key items here one mixed brand configurations for GPUs should likely be exceedingly rare so it's not necessarily a real-world use case or benchmark in that regard you wouldn't intentionally build a new PC and just decide to buy a fury ax and a 980 eye to use together it's kind of a weird thing to do so it'd be the type of thing that happens if you somehow ended up with a spare GPU from another vendor as great or something like that number two for this to build a PC intentionally in this way would actually be risky as the games would explicitly you have to support this kind of combination and they need to be dx12 enabled of course which is not that prevalent right now and it's unlikely that mixed brand configurations would often be tested so bugs and poor performance are probable in the future for dx12 games as for same brand configurations similar challenges apply but the use case is more likely and that a user may end up with for instance a gtx 960 and want to move to a 970 or 970 you wanna move to a 980 TI and with dx12 you could actually keep the older card the lower end one and run them both at the same time just put the more powerful one in the top slot in the event you have to disable the second one for dx11 games or whatever and this works for AMD as well if you want to do a 390 X to a fury or something like that the same development support flaw exists but the use case is exceedingly more likely oxide games built ashes of singularity on their nitrous engine which they call a fourth generation engine and this takes a job based approach to task management within games so instead of doing something like cry engine does currently anyway which is basically spawn a thread for each different task one thread for rendering one for physics one for game logic one for AI or sound or whatever instead of taking that approach the job based approach instead just assigns tasks as they come in to all the threads as would make sense for whatever type of CPU you're using or API or whatever so in the case of nitrous and the new engine from oxide the DirectX 12 support basically uses this job based approach to send out all the jobs asynchronously to whatever device should be processing that particular job and by asynchronously I mean that basically it's sort of parallel processing within parallel processing so on a GPU you're already a parallel processing unit that's what a GPU does it can process multiple things simultaneously and that's why they're so powerful for things like graphics now with asynchronous processing the bottleneck the traditional bottleneck of sort of queue blocking within the GPU is eliminated and that's because even though a GPU might be a parallel processor it still builds up this queue things that indirect x11 anyway needs to execute in a certain order in order to get to the next task and that is removed with asynchronous processing and AMD uses their ACPs for this and that's been a particular talking point for Andy over the past few months as you've seen in some of our videos let's look at some of our initial limited benchmarks so we've run a few dozen tests over the past two days on ashes of singularity and many of these will be published in a separate dx11 versity x12 article and video and for now we're just looking at the one question which is what happens when you combine video cards of different make or model and different brands so what happens when we do these cross Brand configurations for example one of our tests was a 390 X with a 970 and we're here to look at that performance because the first time we've been able to actually do that in a game in a real world scenario because ashes of singularity supports MDA and Lda approaches to multi graphics we thought it'd be fun to test what happens when we use this MDA approach which is multi display adapter that means you eliminate the bridge and you in this case mix an D with Nvidia we used an SLI bridge for a normal GTX 970 s an SLI without going cross ground across model or anything like that and you can see all of those initial results here and video devices currently show almost no difference between DX 11 and DX 12 within ashes of singularity and that's something we currently suspect is attributable to a lack of driver support looking strictly at dx12 explicit multi-gpu functionality running a gtx 970 and an r9 390x in conjunction yields a pretty significant gain over just a standalone 390 X or even a standalone GTX 970 this is because the game is able to access features on both cards and utilize them more fully creating a sort of Frankenstein's monster version of the more familiar SLI and crossfire setups the average FPS for our 390 X plus 970 config at 1080p and high settings with two tab MSAA you may know as 2x MSAA ran 96.5 FPS on an impressive average frame time of just 10 point 3 6 milliseconds very consistent throughout the test and consistent enough that there's little visible tearing for the most part now with a single 970 or even SLI 1970s we did see pretty substantial tearing at times and that ties back into the driver issues when you're running an Nvidia device as a primary or SLI setup for the GPU the 390x stand-alone pushed sixty five point seven MPs average which is a thirty seven percent difference against the weird combo configuration we ran and a single GTX 970 runs at fifty point six five FPS presently about twenty six percent slower than the 390x and 62 percent slower than the combo config SLI GTX 970 s were unimpressive at this time running only forty one point five percent faster than a single GTX 970 and markedly slower than our unlikely and the Nvidia combo what you're looking at now aren't our charts but they're actually charts generated by ashes of singularity now these are specifically for the DX 11 bench that we ran the DX 12 bench contains even more metrics and charts and the purpose of showing these on the screen is really just to demonstrate to you how truly overwhelming the data provided by ashes of singularity is it's a massive amount of data at times it felt more like data dredging rather than data analysis and there's a lot of crunching we still have to do for additional content for ashes over the next few days and for DX 12 and 11 and all that so do stay tuned for that as for now looking at our own charts and data for explicit multi-gpu functionality we can definitively state that cross manufacturer GPU support is functional seems reasonably well supported by the nitrous game engine and is definitely an interesting item to look at for performance in the future now is it something you'll use as an end user is it an actual use case in the real world I don't really think so but it's certainly cool it's interesting and for those off cases where it is a use case where it will happen in real computers then well it looks like it's supported this is all very interesting though AMD presently holds a very clear advantage with DX 12 in this specific game I have not published any other game tests just yet but do stay tuned for those because that could be where it gets a bit more interesting between two manufacturers for now though for ashes of singularity specifically and these definitely got a lead in terms of the dx12 performance gains over dx11 they've really shown their AC e ability here they're asynchronous compute engine is getting to stretch its legs of it finally with this better API and that means they're taking pretty big strides forward from their dx11 performance which is actually pretty pretty darn bad so the dx12 gap versus dx11 that delta for AMD is quite large whereas nvidia we're seen for the most part identical performance between 11 and 12 and that's really just a lack of probably driver support right now we'd think for this particular game and we'll try and stay on top of that as things change but for the immediate future we are looking at a DX 12 with DX load and benchmark check back for that and as far as the asynchronous processing on multi GPUs that are not the same brand manufacturer it works it's kind of cool but what do you do with that information really other than its neat so that's all for this time if you like this type of coverage as always hit the patreon link official video hit that article in the description below if you want to read more about this benchmark to see more charts thank you for watching I'll see you all next time