Gadgetory

clock for clock we've heard your interest in a fury X versus Vega Fe GPU comparison this benchmark looks at performance from multiple angles including synthetic tests gaming tests and spec view production tests all with a Vega Frontier Edition locked to the same clocks as our fury X this is mostly an academic exercise but could yield some interesting discoveries that help understand Vega Epis performance better particularly in workloads with geometrically complex scenes before that this coverage is brought to you by the core g21 enclosure from thermal take a $70 case with two 4 millimeter thick tempered glass side panels mesh ventilation in the front for breathability a rarity in cases these days and a power supply shroud with top mounted SSD sled learn more at the link in the description below so this test was very highly requested by the viewers and the readers on the website if you have more suggestions we are still following up on a lot of them but obviously we can't keep up with with the requests of potentially hundreds of users so we're doing our best but this one today looks at clock for clock performance that means that we are down clocking the Vega EFI card and we're doing so down to 1050 megahertz which is the out of box speed of this one the fury acts from 2015 and we do have a couple of caveats here so differences in architecture are more or less impossible to control for in this kind of test environment differences in memory are difficult to control for but not impossible and then there are plenty of other potential differences as well but ultimately we get something that is sort of an IPC test scare quotes around IPC because it's not exactly what it is but you get the idea with memory we still have a difference in bandwidth of 512 gigabytes per second on the fury X versus 483 gigabytes per second on the Vega f/e card that is partly because AMD is using two stacks of HBM on this one and four on the fury x so you end up with a 4096 memory bit bus versus half of that on the Vega efi card the difference being made up by basically Vega at these memory clock which is 945 megahertz out of box versus 500 megahertz on the fury X and this also is something saw in the Walkman tuning where there actually some leftover bugs we think from the fury acts in the drivers being used by Vega epi because Vega Effy's clocks for memory seem to down clock the 500 megahertz when you change really anything and wat man but it does work when you use something like afterburner or tri xx with other issues introduced at that point so the point of saying that is that this testing today 1050 megahertz clock for clock testing should help us unveil a few things one we might better understand the potential difference in driver performance and see if buggy drivers on Vega Fe could be a cause of some of the performance concerns that are being seen by the community and two are we able to see some of these smaller parts of vega's architecture in play when we control for clock speeds and voltage for example the small primitive discard that Vega has which the fury X does not that could be seen when we're at clock for clock because we eliminate one of those essential variables that said even if you were to control for everything that we have tuning sliders for so clock down to 1050 memory down to whatever it needs to be to be 512 and 512 or 484 and 484 if you control for all of those things there are still a whole lot of things that we don't know about one of which is that as you remove potential limitations on the GPU from one aspect of the architecture like maybe memory bandwidth or maybe the primitives discarding tool as you remove those limitations you encounter new limitations and we have no way of knowing what those are without basically being an D at that point so doing our best here to give you an idea now first has to get into its firestrike we have some gaming tests as well and we're focusing only on Vega and fury X today so there will not be comparative charts if you want comparative charts vs. and video or other AMD products check the review for Vega EFI that's where you can find those all tests are at ten fifty megahertz 1250 millivolts for the voltage and then for the power target we set as a plus 50% so there's no down clocking and we set the fan on Vega Fe to be unbearably loud that way it's not going to get hot and throttle itself so those were controlled for pretty well let's start with fire strike with fire strike providing a synthetic baseline and then later moving on to gain benchmarks we scripted 3dmark to execute five times on each configuration as there's quite some variance in 3d mark test test and the solids were close between each run the results are averaged with fire strike ultra the stock radeon vega frontier edition card operated a graphics core of 4906 points that's our baseline r9 fury x with 10 50 megahertz average a graphics score of 39 74 0.7 after several runs as well with a range of 21 points across all runs the Vega EFI card at 10 50 megahertz averaged the graphics score of 38 89 with a points range of 14 across all runs that's the largest versus the smallest the difference is at 2.2 percent favoring and these are nine fury x over the Vega f/e this could be a difference of memory bandwidth at this point but it's hard to say the exact difference we are confident however that the difference is outside of test variance and error there's about a two percent advantage on P reacts under our test conditions and if you're curious this scanning would indicate that the stock clock the Vega frontier Edition is about 28% faster than the stock clock speery X and if you're curious what has hardware we're using or anything else about methodology as always you can check the link in the description below to find all that out in our 2015 review of the fury x we noted some interesting nonlinear scaling across lower resolutions going from 4k down to 1080 so let's next look at fire strikes extreme and normal 1080p benchmarks to see how that affects things and fire strike extreme which is basically 1440p the fury x again is roughly tied or marginally ahead of the Vega f/e 10 50 megahertz card the lead held by the fury X cannot be confidently declared as a consistent advantage test to test the variance is high enough at least here that we're able to call these results effectively identical in our original fury X review the card closed the gap on some competitors when getting closer to 4k resolution we're seeing some of that here where the fury x held a noticeable and measurable lead of 2.2% at 4k but no significant difference at 1440p let's see if that continues at 1080p and fire strikes at normal benchmark 1080p the Vega FPS star card a graphics score of 21,000 355 for our baseline the rni in fury X operated a score of 16 5-3 1.7 the range of 85 points and Vega f-e at 10 50 megahertz operated a score of 16 7 4 9.3 points for the range of 37 points the tables have now flipped the fury X is behind by 13% where it had previously tied or held a lead over Vega at E at 10 50 megahertz let's take a closer look at the individual tests that comprise bio strike to better understand performance behavior so here's the important stuff to know fire strike has two graphics tasks within it will produce two numbers graphics fps 1 and graphics fps - and those are combined with all the other aspects of fire strike to create your total your combined on your graphics scores the graphics scores being what we just need at the moment ago for the FPS scores we get a better look at where the differences emerge between these two cards because the two tests gt1 graphics test 1 is a tessellation heavy test it has a high poly count it is very focused on a lot of quads a lot of primitives a lot of Polly's basically geometric complexity is what you're looking at with gt2 you're primarily looking at a heavier compute workload and that also could correlate with a higher stress on memory bandwidth so we can see a potential difference here being at 512 and for 84 but for the rest of it for the first test GT 1 there's a potential advantage for Vega because of its newer geometric handling so the geometry pipeline has changed and has the small primitive disk Carter that we saw on Polaris and detailed when the RX 400-series launched last year around this time looking at the results for FPS 1 and FPS 2 in the fire strike 1080p test with fps one being GT 1 the geometrically complex test we see that Vega is notably ahead in GT 1 roughly 4.5 percent and about 1.2 percent behind in GT 2 the other tests show the same trend but with a lessened difference at higher resolutions the firestrike 4k attacks for example which is fires like Ultra posted Vega Fe in the lead consistently at around 20 dot for 1 FPS for gt1 but behind the consistently at around 14.4 for FPS for gt2 this behavior is mirrored to 1080p but the difference is lessened as GPU load increases so it becomes a lot less significant at the higher resolutions and at fury X starts to make up ground at the higher resolutions which is what we saw in our initial review of it two years ago this is interesting data for Vega and P reacts and it's the first step to better understanding Vegas performance in our next test coming up because we now know that potentially in the more heavily tessellated or geometrically complex scenes there's a possibility that Vega FPE comes out ahead of the fury X at least clock for clocks being the main focus here because stock it will pretty much always be ahead but Clarke for Clarke is what we're looking at so that's the first step now we don't have a 100% for sure answer as to the reason for the performance we're seen in those fire strike results but we've got a pretty good educated guess hypothesis that it's basically geometry choline and tessellation choline and primitive discarding happening on the vega card with GT 1 and then GT 2 you're potentially seeing either a memory bandwidth difference or just some other architectural difference that allows the Fury acts to be at least marginally ahead of the Vega efi card but GT 1 is where we see the more significant difference and it's reasonable to assume and make an educated guess that that's because of the primitive discard architectural differences which do not exist on the periods moving into gaming and covering production last let's resurrect a benchmark from another era we're looking at Metro last light which gives us granular control over tessellation specifically so we'll have a line plot showing trend of performance in a moment but we're starting with the stateful benchmarks with 4k and very high quality settings supported by high tessellation the r9 fury X operates an average FPS of 43 lows at 33 and 27.7 fps 0.1% and the Vega epi 10 50 megahertz card the down clock 1 sustains a 39.3 FPS average with comparable time that lows to the average this places the fury X about 9% ahead at 1440p with the same settings baron we saw a 9% lead for the Fury accent moment ago at Fort a the fury X now runs a 78.3 FPS average against the 10 to 3 megahertz Vega fps 70 4.3 FPS average the gap has closed a bit the fury act now is reduced to a 5.4 percent lead rather than 9 percent finally with 1080p at very high and high tessellation the same settings we've used for the other two the fury X operates a 100 11.7 FPS average with a 1050 megahertz Vega variant at about 112 FPS average so they've more or less tied at this point now that we've gotten to 1080p let's plot a performance trend line with all the different tessellation settings starting with average FPS the average FPS is on the left and the lower horizontal axis is for the tessellation settings in Metro last light this is where it gets interesting here with average FPS plotted for 1080p we see that the theory axe retains a stronger lead with tessellation disabled on the far left but loses that lead as tessellation becomes more demanding ultimately giving way to a very slight bump at the end for Vega however that slight bump at the end is within our standard deviation so we can call that effectively tied at the higher tessellation settings if we can further increase tessellation settings in this benchmark we'd suspect this trend would continue normal tessellation starts to close the gap but not much and high tessellation is the real turning point for Vega standard deviation for these tests was 0.577 FPS basically 0.58 as for 0.1% low metrics these follow mostly the same progression as seen in this line graph high is the turning point once again with a slight advance for fury acts when tessellation is off and mostly tied towards the far right end of the plot moving to doom at 4k Ultra was Vulcan and async compute the stock Vega F II card operates in average FPS of 60 4.2 was a fury X at 58 at 4 FPS average and the Vega Fe 10 50 megahertz version at 57.5 FPS average doom isn't an intensive enough game to test at lower resolutions given its FPS cap so we'll save that for the next title with Ghost Recon wildlands at 4k first we're seeing the stock Vega F II card operate a frame rate of roughly 37 out seven okay average with lows at 34 1% and 33 at 7:01 1% that's our baseline the fury X at 10 50 megahertz operate 832 FPS average with lows at 39 and 25 this place is the fury X about 15% behind the Vega F II card though its low end performance weakens a bit at 10 50 megahertz the Vega FP card operates its average at 33 FPS marking at 5.3 percent behind the fury X or about 19.6% behind our baseline card which has a 1600 megahertz clock moving to 1440p the gap now minimizes just like we saw with fire strike if you recall those numbers of Vega FPS stock runs a baseline score of 62 FPS average as the card stretches its legs a bit at the lower resolutions we'll see this continue with 1080p one or 7.1% lows are maintained relatively close by the fury X operates at 50 FPS average with 0.1% local frame times consistently around 36.3 fps part of this may be game level optimization at this point on the developer's end as the fury ax is old enough to see waning support from developers more importantly the Vega EFI card at 10 50 megahertz scored effectively the same 49.7 FBS Everage vs again 50 with some of its test passes scoring that to the FPS mark and others at 49 so again basically the same this resolution equals performance of the cards and that we see a slight gain for Vega in the 0.1% loaded apartment let's look at 1080p to build some more resolution scaling data our baseline Vega fe card performed with an average FPS of 81 with this resolution the fury X runs 60 3.3 FPS average again effectively equal in performance given test to test variants those have also evened out in this test as VM consumption has gone down and load has lightened this particular set of data aligns very well with fire strike where we see performance equalized at resolutions lower than 4k the fury X always did perform better at 4k even in our original review years ago that seems to maintain at least somewhat true today we only ran for K for GTA 5 the numbers position the fury X at 43 FPS average and the F ecard at 10 megahertz at forty point seven FPS average this is outside of variance and is a measurable and repeatable result placing the clock for clock Vega F II card at five point three percent behind the fury act and IPC so to speak but can i - is of a singularity with directx12 high settings and a 4k resolution Vega Fe stock card establishes a baseline of 69 dot for FPS average the very acts at ten fifty megahertz at stock speed run 65.4 FPS average with low-end frame time performance slightly more behind and setting Vega Fe to ten fifty produces a fifty 7.3 fps results with lows behind the fury X this makes the Vega Fe ten fifty megahertz card about twelve percent slower than the fury X in this test and makes the fury X about five point eight percent slower than the Vega Fe stock card for our last gaming test were using Sniper Elite for at 4k with DirectX 12 and high settings with async compute enabled the Vega Fe stock card provides a baseline performance of 53.3 FPS average with the fury X 10 50 megahertz card at 50 1.5 FPS average so the fury X is still behind Vega when stock but matching clocks places at Vega at 51 FPS average so pretty close you see are effectively equal and are within our standard deviation for this test you cannot confidently declare that one is faster than the other when clocks are matched clock doesn't seem to be making a linear difference here either and once we get another over fog in attempt then we wouldn't expect to see big gains from Jesse core clock modifier in this particular game regardless vega fe and fury ax with equal clock speeds are roughly equal in performance for sniper leap for we're moving on to spec view perf tests now which will provide some production level workloads for the charts and for some perspective outside of gaming the titan XP will not make an appearance here if you want to see how to tighten XP compares to vega fe they do trade blows in a couple of the tests then check the vega f ee review that we posted already here's what we've got with 3ds max the vega f ii card has a stock weighted performance of 140 9.3 FPS with the ten fifty megahertz version at a 120 1.5 FPS weighted average so positive scaling is posting a 23 percent gain with a fury acts at ninety two point seven FPS that place with the vega fe ten fifty megahertz card at 31 percent of the fury X at ten fifty megahertz much different than what we saw with gaming and even with fire strike moving to Katia the Vega EFI card stock operates 19% faster than a 10 50 megahertz card which in turn operates 55.4% fascism the same clocked fury X the energy test is next and gains about a two x performance of lead by moving to the Vega card from fury X o'clock four o'clock note this is not a power consumption test it's a specific test named energy or sometimes power you may see online in the spec you perv sweet Maya Post scaling of 40% for the FE card which seems to align with other dx11 tests as well the snx test is one of the most interesting though this test is generated from Siemens NX software with model sizes that are 7 million to 8.5 million vertices in complexity this is engineering software and the fury X gets eviscerated here and is multiplied nearly seven times over in performance by the clock for clock Vega F e card with this particular pro application Vega Fe appears to be showing its strength in vertex processing again with seven to 8.5 million vertices depending on which scene is being rendered in this particular engineering software this testing tells us a few things but be careful about drawing definitive conclusions here we certainly will not be because it's still kinda too soon to tell for example how this performance extrapolates to rx Vega which is what most of you and our core audience are interested in so just again disclaimer to be careful about extrapolating anything look at it as an academic exercise we can potentially learn some things about how this card performs but let's let's kind of be careful about how that translates other cards so specifically to Vega F e some of the differences we saw could be explained by partly a memory bandwidth difference it's not necessarily huge but there is a difference 512 44 that's part of it but not a big part the bigger part potentially is some of the geometry and primitive sculling performance that we see on Vega Fe in theory the in theory because speaking with build Zoid as well and a couple other folks in the industry we're not 100% positive that the geometry choline is working properly right now on this card with these drivers for the games at least so that stated that difference could change going forward we're not sure the other difference you could tend to be looking at a driver performance increase this has been something we've been saying since the review be very careful about how much you are expecting from a driver release in terms of performance gains especially with an unreleased product Rx Vega be careful what you're expecting but it is reasonable to expect some kind of performance gain with the driver update in the very least it would eliminate some of the bugs that we've seen with afterward for instance there are plenty of bugs that we've talked about in the past so there's not going to be any harm done from a driver update unless they make it worse that's that's only a good thing to look forward to for the most part but we don't know exactly how much of the lymphatic performance going forward so geometry : is part of it the memory performance is part of it in terms of bandwidth and then if you encounter a game where you're drawing more than 4 gigabytes on the memory then there's potentially going to be an advantage of the 16 gigabyte card but that's just it's not the most common thing and remember that if you use something like gpu-z to tell you the memory usage it's not actually the memory that's necessarily being utilized by the game or actively deployed it's the memory that's been requested by the application so it's kind of on standby that doesn't tell you what's actively utilized so keep that in mind as well but anyway cool academic exercise if you want to discuss it feel free to below this gives you a baseline for answering the question of what's the clock for clock performance because so many of you asked and it was an interesting thing to do so hopefully that helps with that we've got plenty more tests to do with this card we still have no no firm statements on what to expect for our X Vega other than we will be reviewing it wait for our review so as always thanks for watching you go to patreon.com/scishow and Isis tops out directly with these tests and our work or you can go to Store that gamers Nexus dot net to pick my shirt like this one or like the GN Graf logo shirt subscribe for more I'll see you all next time you