Gadgetory

everyone I'm Steve from gamers Nexus Donna and today we're reviewing the r9 fury x the new AMD card they just launched the fury non X version which is an air cooled version of this card and that's priced at about five hundred sixty dollars and it's rumored to have spurred and Vidya to drop the 980ti price by 20 bucks to six hundred and thirty dollars I haven't seen that take hold yet and retail outlets but it is supposed to be forthcoming so for the fury X this is $650 it's a liquid-cooled VG card so it uses the new fiji architecture and perhaps more importantly it's got HP m memory or rather redundant high bandwidth memory so it uses HP m rather than gddr5 and this is a big step in the new direction for GPUs this is something that Nvidia will be doing in the future as well as it drastically improves the memory bandwidth available for the card to work with so for this benchmark we look at the frame rate performance as usual we look at frame times and the average FPS look at the thermals and also at power in terms of what the power isn't looked into as in-depth in this video as it is in the full review so for full information on the power draw the thermals frame rate all of that hit the link in the description below for the feejee architecture drill down and the fury x review but let's get started with a overview of the new feature our Kotetsu by AMD and see if this promised return to form is actually able to compete in today's modern market it's been several years since a and these released a new architecture they've been on 28 nanometers since 2011 so this is potentially a big deal for them and it's important to really address all the architectural and new memory bandwidth increases and and all that sort of stuff in the depth that it deserves so andy's fiji architecture is brand-new for the fury x it's featured on the fury also and this is something that wasn't shown on the new 300 series video cards that we also reviewed those in fact were refreshes of the previous architecture they were more similar to Tonga in many ways than Fiji and they weren't true rebadged but their refresh is in that discussed in a previous video today we're talking about Fiji it's still 28 nanometers and it's probably one of the last 28 nanometer chips that will be produced by AMD and II really can't go any bigger than they already have the GPU dye size is 596 millimeters squared which is pretty darn big for a GPU and to put that in perspective Nvidia as a GM 200 max while chip is 601 millimeters squared so there's really not a lot of room for growth here and that limitation is imposed because of lithography issues where the stepper scanner systems that are used to create and fabricate these dyes these silicon wafers and dyes that are cut from them are limited by the reticle which is called the reticle limit we won't see larger GPUs if at all we definitely won't see them until after we move to a smaller fab process for PG&E is sticking with its GCN architecture that's graphics core next it's been around for quite a while now but they're using GCN at one point to effectively it's not really officially called that but that's what it is so fiji has a new version of GCN and there are a few improvements to GCN 1.2 some of them come from the 300 series that we already talked about those are not on a new architecture but they did make some improvements to GCN that were carried over here and one of the main ones that you'll hear about is called delta color compression this is similar to what we talked about with Maxwell where the GTX 980 introduced what Nvidia calls third-generation Delta color compression and these doing something similar here so the idea is that Delta color compression looks at the frames temporally meaning over time so you have frame N and frame n plus 1 and in this example on screen frame n is the one you're seen now where you've got the grid car and you see all the color is just fine frame n plus 1 is the next one because we're looking at a tempura Lee and you see all of the pink sort of pixels shaded in this helps indicate where colors have changed and what this is doing is telling the GPU hey these colors have only changed so much within the value we can just take the Delta of the change rather than fetching and redrawing based on an absolute value so this process reduces the bandwidth consumption of the memory by about 40 percent and the compression ratio here is about 8 to 1 so you can see that the tiles are compressed 8 to 1 similar to the 300 series Fiji improves the power efficiency of the fury X card so Andy has made great efforts to reduce the thermal and power envelope of their GPU they've done this in a few ways for Fiji the first main one is that APU technologies have been sort of lifted from a pea use and imported into Fiji more or less so a couple of those would include dynamic voltage which just means that the voltage modulates itself and regulates itself based upon load that helps reduce thermals and it helps reduce power drawn another item moved from ApS to the GPU is the clock rate and its ability to sort of fluctuate dynamically as well so it modulates the clock rate based upon load but of course the most immediate thing on everyone's mind is the CLC the closed-loop liquid cooler this is manufactured by Cooler Master I've got a great big whole page I located to this thing and the articles so sort of click on that and just read that if you want to learn more but the very short most important items of the CLC are primarily that it helps in reducing thermals which does a couple of things for power as well the main item here is that sticking of CLC on the card will help reduce the power leakage from transistors so as you heat components up on a device like a video card as you heat up the transistors there's a much greater chance of power leakage meaning energy wasted a couple of other changes to the thermal envelope include the way that the shader array is handled so the shader array will throttle when it's not in demand and basically it's similar to the core clock if the GPU detects that the shader units are unloaded it will modulate how much power is assigned to those shader units and effectively gate the power provided to them the shift to HBM further grants inherent power efficiency gains it's definitely a noteworthy amount but we'll have a full article and video on that in the future that's too long to go into here but HBM is the biggest architectural shift in GPU computing in quite some time it is clear that this is the way the industry is going in the future including Nvidia the question of course is does this actually matter right now is it enough to achieve parity to match the performance of the 980ti which is the most immediate competitor some other important architecture points before we dive into the framerate performance here Fiji's sim D uses a 64 asset wide wavefront which provides really a massive amount of plane potential for pixel and vertex shaders so this is something where Nvidia's warps which is basically their name for a collection of threads their warp is 32 assets wide right now and this is part of why we see Nvidia performing so well with gaming they've really fine-tuned their architecture to optimize for gaming tasks and AMD does really very impressively well with computes their compute is higher than the competing Nvidia device but they are a little more versatile so they're not as specialized in gaming and that's where we get this performance disparity where Andy will perform very well for OpenCL where you've got complex data structures and for Nvidia it's just really driven for gaming so that's that's where we see that disparity so more on all of this in the article we talked about whether the feejee is Rob's pound or triangle bound at different resolutions and things like that HBM is a future topic let's dive straight into some of the benchmarks here next we will talk about thermals and gaming performance for discussion on the power draw please see the article link in the description below we have two different thermal charts the thermal chart here shows average thermals and you can see that the 980ti hybrids at about 20 2.91 Celsius the reference 980ti on air from Nvidia's 61 Celsius and then sapphires fairly impressive cooler for the 300 series at 45 with a fury acts at 41 not very impressive for the the fury x4 thermals it is much better than Andy has been in the past so I will certainly grant them that but for a liquid cooler I think they could do a little bit better even with the cooling of RAM and VRM this next chart shows the thermals over time and we only have the 980ti and the fury acts on here so what you're looking at in the light blue is the r9 fury X you can see that it takes longer to sort of reach equilibrium and when it does reach equilibrium which is about 16 minutes or so into the the chart here you see that it's still warmer than the 980ti hybrid but much cooler than the 980ti reference so AMD is in a zone where we don't feel like we're thermally compromising our system by sticking their GPU into it whereas the 290x which was running at 95 Celsius in many cases began to be a bit of a concern because that warms up everything around it so this is a it is a step in the right direction and that is a good thing so this is where things get a little bit interesting we did much more statistical analysis on our game benchmarks than we normally do and that's all in the article if you're curious for this video we're only going to recap a few of the games will very briefly recap The Witcher 3 far cry 4 grid Autosport and shadow of Mordor and if you want the rest it's in the article The Witcher 3 illustrates the fury X is pretty confusing results very well what you're looking at right now is the 4k result and we're sitting at 56 for the average FPS with really a quite dismal 1% low and 0.1% low and as we move down into lower resolutions you will see that the performance gap the Delta between the 980ti and the fury X actually widens so for the fury accident crossfire the 4k performance is 56 fps worth pretty bad 1% 11.1 percent these single Fury axes at 31 FPS average with the 980ti 35 FPS average so 980 is in the lead we go down to 1440 and the gap widens now we go from 50 FPS on the fury X to 58 on the 980ti so our performance disparity has has grown quite a bit in fact it has nearly doubled it's gone from a 12% gap at 4k to a 24% gap at 1080p so why is that happening well this becomes a question of whether the fury X is limited by its ROPS its raster operation units or if it is limited because of geometry and from some analysis it is my understanding that the fury X is limited by its geometry performance this is something we know Andy is not particularly good at tessellation they're not really particularly good at complex geometric elements and games they're great at shaders and that's sort of anything else they fall short so that's what's happening here and we're getting throttled by geometry performance this is shown further in grid Autosport in grid you'll see that interestingly the performance actually becomes somewhat capped so at 4k the highest performing AMD device is the crossfire fury X configuration at 108 fps I should first note that grid often exhibits some sort of abnormal framerate performance versus other games so this is one thing to note when you're looking at the chart at 1440 notice how the whole chart has shifted in favor of Nvidia and where does the fury X crossfire sits it sits at almost exactly the same fps as in the 4k chart so what's going on here look at 1080 now we're hitting CPU limitations or something else in the system because the my ATI is heightened X 90 I hybrid all of them are effectively the same fps across the board so we're hitting some other bottleneck in the system why then is the fury X not also at the limitation at the 128th range it's still Cydia now 109 this is really quite confusing we saw this in some other games as well in the article you'll see those but at the end of the day it's some sort of limitation in the pipeline that is not present on the Nvidia cards far cry 4 exhibits similar behavior to The Witcher 3 where we see the 980ti reference performing marginally behind the fury acts at 4k and the sli and crossfire configurations sort of tying in their average performance that I need T is Li winds out barely and 1% low performance but it's basically tied with if he reacts in crossfire when we drop to 1440 you see that the 980ti reference card jumps ahead by 15% so again we see a limitation in the pipeline probably some sort of geometry or something like that now we're looking at shadow of mordor at 4k and as with many of the other games in the charts we see the 980ti hybrid which is a non reference card meaning it's more similar to what you would buy in the market from one of the manufacturers because it's overclocked to 11 140 mega Hertz from the thousand megahertz base and that's basically every card on the market will be about that so that's our performing the fury acts by quite a bit really it's 50 verses 42 fps and moving to 1440 and 1080 you will once again see that the fury X loses its gap in terms of Delta between performance with the Nvidia devices GTA was interesting because we were able to exceed the 4 gigabyte hpm limitation on the fury X but check the article for information on that for their analysis let's very briefly look at overclocking here you're looking at the overclocking results table for the fury X and the next two will be the 980ti hybrid and 980ti respectively for the fury X we were only able to increase the clock rate by 60 megahertz to a maximum clock of 1110 megahertz it's a pretty small jump and the memory offset was 60 megahertz for a total of 560 megahertz and this was validated with GPU Z with a 264 and as bench so it was so it was applied to the cards this is about a 5% overclock the r9 300 series we were able to get about an 8% overclock neither of these are particularly impressive the 980ti we got in excess of a 20 to 30% overclock on average and that's that's pretty massive it has actually a very big impact on performance as you'll see in these overclock benchmark charts meanwhile the fury X C is almost zero gain it's really in my opinion not even worth overclocking I don't know if I am the plans to release voltage tuning utilities for the pure ax but as it stands now there's really no point to overclocking the feejee card it just doesn't produce any noteworthy FPS gains so concluding for the fury X this has been a pretty difficult card to review because it's all new architecture it's dot HBM which is brand new to any video card really that's in the consumer world and Andy has made the right steps for improving their performance and achieving hopefully parity with NVIDIA they've got a very powerful architecture that has huge compute performance it's got a massive shader array and those are good things for Andy it's certainly much better in terms of power efficiency over the 290x so the the fury X for power efficiency is only about 26 watts more than the 980ti and yeah it's still more but at this point it's not forcing you to buy a higher wattage power supply so that is a step in the right direction for sure and the thermal performance could use tuning I think that the tubing on the CLC is probably a bit of a concern for installation but overall thermal performance is greatly improved over the 290 X as well and other previous flagship so Andy is moving the right way they hear the criticism and they're responding to it unfortunately there are a few problems here in terms of performance for gamers so the main one is Andy still suffers from poor frame times in a lot of games shadow of Mordor is a very good example if you look at those charts the 18 FPS 1% and 0.1% lows are pretty bad and those are the numbers that impact fluidity of the experience so when you're playing a game and you see a sudden spike is from the 1% and 0.1% metrics where average FPS really is sort of irrelevant to that point because your gameplay has been interrupted so Andy still falls short there this could theoretically be resolved with drivers because PG and HB m are both very powerful items and the 4 gigabyte limitation of memory isn't really prevalent in any games we've tested except for GTA 5 at maximum settings and at that point it's producing unplayable frame rates anyway on all the cards and these drivers support has been pretty disappointing they've only released one wickel certified driver since December and that was the one for the launch of this video card so they've had a couple of beta is the GTA beta driver and witcher 3 beta driver come to mind but they're certainly behind on the driver and software optimization front and that's a big problem for andy especially they want to compete with and video which is really pretty prolific in terms of game industry support and optimization so andy struggling with the framerate output on games where it should in theory from raw hardware be able to produce a much better framerate and given their track record for drivers i'm not sure whether we're gonna see the steps taken that andy needs to do and until that point in time the 980ti is just it's a better buy right now in terms of performance in terms of frame times and if the rumored price drop is true then that's certainly a factor as well at overclocks much better so you get even more performance and if the price or performance disparity closes between the 980ti and the fury x then it's possible though the fury x will become a good buy because it is a promising card but it needs to be lower in price or higher in performance so that's not where it is right now I would be buying the 9d DTI at the $650 price point range I haven't looked at the fury non axe the air-cooled one that's much cheaper I from initial benchmarks would believe that that's actually a pretty competitive card with a 980 not necessarily the 980 TI so that's where we stand on this check the article for the full analysis it's 5,000 words or something and has a lot of charts that you don't see here in this video and I will see you all next time