Introduction
Hello there everybody, I thought I would do something a little different for this People’s article and conduct an objective analysis from the consumer standpoint of the AMD and nvidia commitment to the people that really keep them in business, you, and analyse how each companies respective drivers, features and hardware compare and perform. The rivalry between AMD Radeon and Nvidia GeForce is undoubtedly a contentious one with both sides fiercely split, this article intends to look at both sides with an unbiased and objective eye picking the bones of both and seeing what each vendor’s strengths and weaknesses really are. Let’s see if we can perhaps for the first time ever come up with an analysis both red and green fans can agree upon, even if begrudgingly.
DirectX 11 or DirectX 12, So what’s the big deal?
DirectX is an Application Programming Interface (API) which is in itself a collection of smaller applications and is particularly relevant in game programming, DirectX 11 was first revealed at Gamefest 2008 and made substantial strides over previous DirectX iterations including features like DirectCompute, aka GPGPU, and improved CPU multi-threading for multi-core based systems, that multi-threading and DirectCompute is important so be sure to remember them going forward in this article.
DirectX 12 was announced at GDC (the Game Developers Conference) in 2014 the main enhancements it has over DirectX 11 is to hand over the ability for programmers to implement Crossfire or SLI themselves rather than rely on vendors (AMD/Nvidia) however such an ability adds substantial effort, cost, and development time so is gradually making multi-GPU setups fall even further out of favour. Also added is advanced low level programming to reduce driver overhead (going forward this will be important to remember as well) which allows for developers to utilise their own command lists and buffers on the GPU to give more efficient use of resources through parallel computation which in short is the ability to carry out multiple calculations and / or the execution of processes simultaneously with the overall main goal of DX12 to offer console levels of efficiency on other platforms including tablets and PCs.
Now you’re all clued in with that brief crash course overview on the main differences and highlights of DX11 compared to DX12 which serve as an important premise to this article it’s time we started delving deeper down the rabbit hole.
The RX580 and GTX980
For this article the two cards used will be an RX580 and GTX980, two cards separated by around two and a half years, the RX580 being the junior, on paper both cards are very evenly matched with the elder statesman GTX980 being built on the 28nm node with 16 Compute Units (CUs) and the RX580 being built on the much newer 14nm node packing 36 CUs. Let’s have a couple shots of GPU-Z for each card.
Both cards from a purely hardware level are similar with the GTX980 having the ROP and pixel fillrate advantage while the RX580 has the TMU, shader, and small bandwidth advantage. As most of us should know the GTX980 is essentially a slightly meatier GTX1060 6GB (sans the small amount of extra vRAM) so the RX580 and GTX980 certainly aren’t mismatched as some might think or try to claim, add to that the two and a half year advantage the RX580 has and the cards should be very evenly matched indeed.
For the sake of parity from a clock frequency perspective so the green side can’t bemoan the lower clock frequencies of the GTX980 in testing I also vBIOS modified the card to match the RX580 stock clocks as closely as possible and will be including those results as well.
It is important to point out at this juncture that you can never do apples to apples comparisons of an AMD and Nvidia card, completely different architectures, so a certain level of extrapolation is always required to see the broader picture this is why the clock matched GTX980 results are included for additional perspective only.
Test System
We should all be familiar with the test system now but here we go again...
CPU: AMD Zen 1700 @ 3.9GHz
Mainboard: MSI X470 Gaming Pro Carbon
RAM: 2x8GB G.Skill Ripjaws V 3200MHz 15-15-15-35 @ 3466MHz CL16
GPU: 4GB GTX980, RX580 8GB
Storage: 250GB Hynix SL301 SATA SSD, WD 120GB M.2 SATA SSD (OS drive), 2TB Seagate Barracuda
Opticals: 24x Lite-On iHAS324 DVD-RW, 16x HP BH40N Blu-Ray
Sound: Xonar DX 7.1, Realtek ALC1220
PSU: EVGA 1000w Supernova G2
OS: Windows 10 Pro x64 (latest ISO) and all updates
Case: NZXT Phantom 530
Drivers:
AMD Chipset Drivers v18.10.1810
GeForce 399.24, 417.71, Adrenaline 18.10.2, 19.1.1
Realtek Nahimic HD Audio Driver v6.0.1.8581
Xonar C-Media 1825 Driver
Intel l211AT Driver v23.1
Comparing ReLive to GeForce Experience
First let’s examine AMD ReLive against Nvidia GeForce Experience, aka GFE. Both offer very similar functionality so let’s see which comes out the better of the two.
Using both ReLive and GFE the first thing to stick out is the main landing page ReLive is immediately more functional and to the point with a vastly superior layout with everything neatly categorised in separate tabs as well as also being neatly integrated in to Radeon Settings. The other major difference with ReLive is the remarkable amount of control over recording quality down to .5 adjustments and even the option of letting you choose between AVC and HVEC encoding.
Navigation with ReLive is also quite refined, often immediately taking you to where you need to be although it is rather irritating having to click between several menus to change options in ReLive and the performance monitoring options these should really just all be combined in one tab for more seamless navigation. This is the only area ReLive isn’t particularly refined though everything else is quite seamless. I can’t say I’m a particular fan of all of the options that are enabled by default though the upgrade advisor, banner advertisements, and notifications should all be disabled out of the gate by default. Pestering and intrusive notifications isn’t cool. We want quiet and beautiful, like the Normandy SR1, not the nagging Normandy SR2.
Compared to ReLive GFE is substantially more bloated, clunky and intrusive from the moment you begin to use it.
Moving to GFE it will immediately stand out being substantially more bold than ReLive, which isn’t necessarily bad, but bloated and intrusive certainly is and unfortunately GFE is blind (or ignorant) to those lines which should never be crossed. GFE also lacks the seamless integration of ReLive needing to be launched separately from the nvidia CP which feels very clunky and outdated, square pegs, round holes, if you will. The nvidia CP itself looks like something that has been ripped right out of Window XP and suffers from terrible lag when changing options, it is really rather awful and unpleasant to use and in modern times just smacks of laziness from nvidia. It is more than about time nvidia fully overhauled their software to integrate GFE with their CP, but let’s keep GFE optional, nvidia.
Back to GFE it also demands on an intrusive login from the very outset before you can do a single thing with it which is not only entirely unnecessary but also immediately imposes a feeling nvidia are data mining you which is a rather unpleasant feeling giving at the very least the impression of an invasion of privacy. Why do you need a login just to use what is essentially just capture software? It clearly feels like nvidia aren’t just content with their pound of flesh for the price of the GPU they want to pick at the carcass afterwards, find the wallet, and see if they can shake any change loose too.
Moving on to looking at ease of navigation the overlay and save location options for GFE in particular is cumbersome, popping up with a UI that engulfs the entire screen and navigating to a new directory to save screenshots or videos to is also very clunky feeling. Yes, that is a Cyberpunk 2077 wallpaper in the background.
Finally, GFE is also loaded with self-promotional content every time you go to the GFE Drivers tab, you can hide this self-promotional ad nonsense but it is still there lurking behind the curtain and ultimately adds even more unwelcome bloat the way it is done feels like you are using one of those terrible free to use applications that are loaded with ads, awfully distracting, unnecessary, and utterly irrelevant to GFEs purpose making it feel more like a virus not a piece of capture and streaming software.
GFEs landing page, it doesn’t offer much but it’s a little more convenient than having to launch individual software platforms to get to your games.
Ending on a more positive note for GFE it does have one ability ReLive does not (ReLive did however so I beleive this is just a bug with the tested drivers), GFE allows you to launch any installed games directly from it however this does still come with its own caveat, you see by default GFE likes to “optimise” any games it finds completely undoing any graphical options you have already set to every detected game you can’t select to have just one game done, GFE will just take it upon itself to do them all making you have to redo your settings in every game. I get this option may be useful for inexperienced users but for the majority of us it is just irritating there is surely a better way to do this, say initially offering to optimise any detected games which the user can then either accept or decline then from that point onward simply having an option where if the user right-clicks on a game icon within GFE you can select to have it optimised.
To be clear on the term “optimised”, GFE won’t go and automatically make custom performance based configuration changes to settings files it is just going to raise or lower the standard in-game graphical settings for what it thinks is an acceptable frame rate so... it’s not much good anyway. So much for trying to end on a positive note for GFE, sorry, I really tried.
Benchmarks
For benchmarking I have chosen a selection of games that should represent an even and fair assessment for both red and green teams as well as some tests that by their very nature should be impartial. This selection includes 6 DX11 tests, 3 DX12 tests, and 1 OpenCL test. All tests have been run at 1920*1080 as this is the most common gaming resolution I’d suspect for the RX580 and GTX980 in modern titles with the maximum selectable graphics preset, in the case of the test not having any presets then all graphical options are set to maximum, and lastly for synthetic tests they are run at their default settings.
We’ll start with DX11 \ 12 synthetic and API overhead tests. Both cards had their power limits increased to maximum to ensure no GPU throttling and a “level” playing field.
Testing both cards at their factory frequencies provides the baseline, turning in a slim advantage for the RX580 but in truth the difference shown in this baseline test is not something you will notice that much in actual games.
Looking at the clock matched GTX980 next to the RX580 and observing the combined Fire Strike results is far more interesting telling the apparent story of the nvidia drivers having quite substantially better single and multi-threaded driver level API overhead optimisations compared to AMD which in short means better optimisations for more CPU dependent resolutions like 1080p and 1440p. It’ll be interesting to see how this pans out in further tests and practice.
Looking at the API overhead test results, which measure the theoretical maximum amount of draw calls the driver can make for various APIs, shows Vulkan performance being about the same between AMD and nvidia while AMD take a good DX12 theoretical lead in to the real world benchmarks. Nvidia on the other hand take a substantial theoretical DX11 single and multi-thread lead in to the real world benchmarks the trend started with the previous synthetic Fire Strike combined test is carrying through to these API results as well which while only synthetic does help add credence to what has been observed in these early stages. The nvidia drivers in theoretical testing workloads simply hang, draw, and quarter the AMD drivers, with deficits of roughly 37% for single thread and an astronomical 115% in multi-thread workloads. It is difficult to say from this basic test what is holding the AMD driver back it could be dynamic or static batching, or a combination of both. My instincts tell me it is a dynamic batching problem however as between the two approaches this is the one that incurs a CPU overhead penalty.
With drivers from both nvidia and AMD tested that are several months apart these results are not an anomaly this is how the drivers are optimised, which is to say the AMD drivers have theoretical lacklustre single thread performance with multi-threaded capabilities being non-existent getting utterly annihilated by the nvidia drivers, which I’ll state again because it is an unbelievable margin, by 115%. This could be a colossal bottleneck for AMDs hardware due to the drivers especially for DX11 titles at 1080p. This result for AMD is quite the surprise for a company that so loudly and proudly bangs the drum on multi-threading capabilities and the importance of it in the modern era. These results certainly warrant observing further.
DX11 Driver Benchmarks
FarCry Primal: Built on the Dunia engine which is a modified CryEngine it is not the most multi-threaded of engines utilising around 2 cores at most making it a good test geared more toward single thread workloads.
Shadow of the Tomb Raider: Built using the Foundation engine and with render paths for both DX11 and 12, being a title ported from Xbox and Playstation 4 the title should be well optimised for AMD hardware and will be pulling double duty for both DX11 and DX12 testing.
Deus Ex: The Deus Ex engine is based upon a heavily modified Glacier 2 engine, known as the Dawn Engine, and even includes a physics engine based on TressFX. The game is also an AMD Gaming Evolved title on PC and has both DX11 and DX12 render paths so just like Shadow of the Tomb Raider will be pulling double duty.
Metro Last Light Redux: The 4A Engine is capable of working on PC, Xbox 360 and the Playstation 3 and so favours nvidia hardware the engine is well known and documented for being heavily multi-threaded as such it will prove to be a superb test of driver multi-threading efficiency.
Alien Isolation: Built by Creative Assembly not many details are known about the CATHODE Engine however with Alien Isolation being released for PC, PS4, PS3, Xbox 360 and the Xbox One it is quite safe to assume the engine is well optimised for both AMD and nvidia hardware, on PC the game is also an AMD Gaming Evolved title. If you are a survival horror fan or just a fan of the Alien franchise you owe it to yourself to own this game, it is the best Alien game ever made bar none and the second best Alien film behind only Aliens 2. If you don’t already own this game, then shame on you, pick it up it is ridiculously cheap to buy now.
These games represent a solid mix of game engines for AMD and nvidia hardware, I said there would be no favouritism being played here, and I meant it, all scenarios will be looked at here not just the ones that paint a nicer picture for one side or the other. Let’s get down to the benchmarks.
Beginning with Deus Ex there is really no difference between cards or drivers the game runs identically accounting for the usual benchmark variation and small margin of variance I automatically allow. That is not to say I don’t think further optimisations in both nvidia and AMD drivers could not be done, but both cards are each other’s equal here.
There are definitely some questions to answer though the game isn’t exactly bleeding edge graphics the most frame killing graphical option appears to be volumetric lighting cleaving off about 15FPS from both cards when set to maximum other graphical options had a minimal impact on both cards often with many options having settings that can only be described as feature creep rather than having a noticeable visually beneficial impact. Perplexingly the higher clock frequency GTX980 results do not improve frame rate at all which is extremely peculiar suggesting some kind of artificial limitation.
As we move on to Shadow of the Tomb Raider results swing notably favourably to the GTX980 with even the baseline 980 results comfortably thwarting the RX580 with an impressive additional 14% on minimum frame rates. There is certainly more AMD RTG could and should be doing to improve minimum frame rates here and from a title that has you would think been designed from the ground up to inherently favour AMD hardware given the console heritage someone is definitely not pulling their weight here and I’m not inclined to look in Crystal Dynamics direction either, AMD RTG. These results also show the results observed from the synthetic tests are seemingly having an impact in actual games, it’s not just theoretical anymore people. These results do strongly hint at a drivel level CPU API overhead issue which is going to particularly impact performance at 1080p.
FarCry Primal steps up to the plate next and the RX580 is mediocre at best in this title with even the baseline GTX980 managing to just pip the RX580, a result I’d normally chalk off as a margin of error outcome but the GTX980 results were repeatable consistently across several runs. The clock matched GTX980 results only pile on the pain for the RX580, really. There is once again an argument to be made that the inferior driver level CPU API overhead with the AMD drivers is playing a role here.
Moving to Alien Isolation and Metro Last Light Redux measured frame rate had to change slightly to include maximum instead of minimum frame rates due to the minimum frame rates not being representative for either the GTX980 or RX580 of what a typical minimum frame rate actually is in these titles due to the benchmarks recording the minimum frame rate at scene changes.
Looking at Alien Isolation performance for both cards is solid but nevertheless this result for AMD must be a bit embarrassing as the game is meant to be an AMD Gaming Evolved title yet the RX580 is usurped by even the baseline GTX980. Not a particularly glowing endorsement of the AMD Gaming Evolved initiative. Again an argument can be made that the inferior CPU API overhead with AMD RTGs drivers in DX11 titles is playing a role here as there is simply no questioning that Creative Assembly have optimised this title to the fullest, an assertion backed up by the sheer amount of platforms the title has been designed for, and how well it runs on all of them.
Moving to the final DX11 test, Metro Last Light Redux shows decidedly unremarkable performance for the RX580 and a resounding thumping for the card with the clock matched GTX980 results. CPU utilisation in Metro LL also uses vastly more of the CPU than any other tested title so CPU core scaling from the 4A Engine is not an issue here. What is very concerning is that the 4A engine has not changed that much for Metro Exodus it would be extraordinarily detrimental to AMD RTG to continue to ignore the continued downward spiralling performance they have in the Metro titles which are deceptively popular.
For those that like to keep tallies on this sort of thing that’s one draw for AMD RTG with the RX580, and five losses. If this were (european, proper) football that would be grounds to be looking at the manager and beginning to wonder if he’s the right man for the job as that would be 17 points dropped out of an available 18. It is time to move on to the DX12 tests.
DX12 Driver Benchmarks
Real world DX12 games are thinner on the ground than I would like, or ones with built-in benchmarks anyway, but let’s look at the DX12 results there are starting with Deus Ex. The results here do not change from the DX11 ones, at all. Have we just caught Eidos Montreal doing something naughty? It certainly looks to be that Deux Ex is not using a native DX12 code path and is just slinking all of those DX11 calls in to a wrapper that translates them for DX12, if correct this serves as an interesting insight in to just how damaging for AMD RTG, and DX12 as a whole, simply using an API wrapper to convert those DX11 calls can be. Even more interesting is the fact that with the clock matched GTX980 frame rate does not budge still which adds even more mystery to the Deus Ex results but does add to the suspicions whereby there is an artificial limiting factor somewhere.
Moving to Shadow of the Tomb Raider the results show a much more expected outcome with the title showing quite a healthy frame rate increase with DX12 and the RX580 even picking up top place for a change against the baseline and clock matched GTX980 results. There is evidence of AMDs potential to be dominant in DX12 for only the second time here the first being way back in the synthetic tests, there are even signs that there have been performance improvements between the two tested AMD driver sets which is a first for any of the games tested for the red team. While things do look more positive for AMD here it cannot be considered anything more than a conciliation prize.
In football terms; "We’ve played eight games now, how we doing gaffer?"
"We’ve drawn 2, won 1, and lost 5. Our success ratio is 25%. We're in relegation form right now."
"Err... can you put me up for sale gaffer? I want a move to a different team."
Luxmark 3.1 Results
The last test for the cards will see them pitted in Luxmark 3.1.
This test was run with the driver set to graphics and then compute but it did not improve the results for either card the superior amount of compute units on the RX580 sees it outperform the GTX980 but considering the extra amount of CUs the RX580 has you can’t help but wonder why the RX580 hasn’t scored better, with 125% more CUs than the GTX980 and only scoring at best around 23% better is highly disproportionate. Fortunately I can provide you with a strong answer to this question in brief the memory on the RX580 is horribly optimised, atrocious in fact, a case of extraordinarily poor stock memory timings leads to substantially below par Luxmark results, to the tune of around 20% (about 3000 points). The GTX980 on the other hand shows to be doing remarkably well with only it’s 16CUs but unsurprisingly in this test there is very little more to be garnered when clock matching the 980 with the RX580.
Frame Pacing
For the final part of this driver analysis it is time to take a look at how the 417.71 and 19.1.1 drivers are performing for frame pacing. Here is where this article becomes somewhat more interactive as you see I’ve always felt there is much that gets lost in translation with a simple frame pacing chart as what a chart can show is often in reality imperceptible to the human eye so instead I decided to try something a little different, I decided to record a benchmark run of some of the tested titles then upload these videos, completely untouched (raw, if you will), for you to (hopefully) more accurately experience frame pacing with the drivers for yourselves. If this approach for frame pacing proves popular I’ll expand it in future. Of course for the best experience it is recommended to go full screen.
Alien Isolation
AMD 19.1.1
nvidia 417.71
Metro Last Light Redux
AMD 19.1.1
nvidia 417.71
I’ll let you all judge for yourselves about if you can see a difference because once something is pointed out, everybody “sees” it. Suffice to say without specifically pointing anything out I think frame pacing regardless of card is as good as one another there’s not enough difference for me to say categorically that one side does frame pacing better than the other with each side perhaps being a little smoother than the other in different areas.
Driver Performance Summary
For this driver performance summary we’ll compare the RX580 relative performance to the substantially lower clocked baseline GTX980 and the clock matched GTX980 results.
These results are with the 19.1.1 and 417.71 drivers, let’s begin dissecting this. Looking at the RX580 results relative to the baseline GTX980 you can see there is very little in it except for Tomb Raider which see the RX580 take a battering to the tune of a 14% deficit. I chose to include minimum instead of average results for Tomb Raider for point of being thorough and not letting anything get swept under the rug, minimum frame rate is arguably more important than averages so for these reasons I see no harm or foul play in including them as I’m sure some will try to claim. The Alien Isolation results are also a fair distance off the pace of the competition by 6.5%, despite still solid numbers in this title overall. The Deus Ex results did not show enough of a difference to outright say one card is faster than the other with run to run variance being in the region of 3FPS or so for both cards so I decided the fairest thing to do here is to call it a draw which is why all results here are marked equal. There are some areas of quite serious concern which I would be remiss if I did not remind you only become apparent with the above more detailed analysis and not this at a glance overview. Nobody could argue either if you wanted to conclude the RX580 and the baseline GTX980 perform well within themselves but both for different reasons, the RX580 because of what appears to be a combination of CPU API overhead driver level capabilities and woeful memory timings while the GTX980 simply because the sample used here is clocked far below the true potential it has.
Moving us on nicely to if you compare the RX580 to the GTX980 with adjusted clocks, which I would suspect most people are going to for a more apples to apples comparison (though I’d advise against it these results are here for additional perspective only) we can see that the GTX980 finally manages to stretch those legs and in doing so completely leaves the RX580 in the dust. How does it manage to do this? Better architecture? That is definitely a possibility, or is it far superior driver optimisation, particularly for DX11. That is the more likely of the two, I’d say.
For AMD to have drivers that appear to be this poorly multi-threaded in DX11 titles still especially considering the amount of time AMD has had to address the problem for a lack of a more diplomatic way of putting it is criminal. This problem is only set to get worse if AMD do not properly address the issue as there are many titles both current and upcoming which still predominantly use DX11 and most of said titles, if they offer a DX12 path, will likely just be DX11 calls in an API wrapper that translates them for DX12 thus not being a native DX12 code path which will completely negate any advantages of DX12 which makes both AMD RTG and DX12 look bad to the less informed. At this point I’ll stress once again a certain level of extrapolation is required here due to the completely different architectures of the cards and factors that just will not be known until titles release but these performance deficits range from a crack to the size of Valles Marineris. AMD only have themselves to blame here and must resolve what looks to be apparent performance bottlenecking driver issues particularly for engines that make heavy use of multi-threading.
Moving to the DX12 results there is nothing that requires much talking commenting on the Deus Ex and Tomb Raider results would only have me repeating what we have all already observed and discussed but let’s give credit where it is due AMD RTG for the games that do use a true DX12 code path have some fairly good performance on their hands I’m sure they are chomping at the bit to try and showcase more.
Driver Stability & Bugs
Neither the AMD or nvidia drivers had any serious bugs, a few quirks here and there, such as the Radeon Settings menu not displaying properly if an OC failed after the driver reset and some titles experiencing their colours being inverted on recordings, the latter of which is now fixed in the latest drivers, and let's not forget the apparent bug with Radeon Settings whereby games would not launch when they should. I did find however that the AMD drivers had a lot more minor visual artefacts particularly in Tomb Raider and Metro LL in the guise of fog and smoke effects glitching in varying but similar manners some of which can be evidenced in the Metro LL frame pacing video. For nvidia the only bug I noticed is what now feels like has existed from the dawn of time, the display output if using a HDMI cable being set to RGB limited rather than YCbCr 4:4:4, I didn’t notice any visual bugs, minor or major, in any of the tested titles.
Putting the drivers head to head was also an excellent chance to put the AMD RTG claim of superior stability to the test, I tested frame rate, frame pacing, and the drivers stability in general as I could not find the exact criteria defining this superior stability proclaimed. Things were close, frame pacing was very similar as was frame rate stability, in terms of overall driver stability and quality though that has to be given to nvidia as their driver did not require a system restart to be able to use the control panel properly if an overclock failed after the driver reset itself nor did the nvidia drivers have visual glitches in tests with smoke or fog effects.
In short, AMD do not have the better drivers, not as far as overall stability and optimisation goes, not by a long shot.
Closing thoughts & Impressions
This article has been a “through the window” look at both AMD and Nvidia as such these findings can be taken as a glimpse but a glimpse with enough substance and evidence to add a reasonable level of credibility to the findings. With that said let’s get this conclusion underway.
AMD RTG are widely accepted as being a company where their GPUs get better with age, this sounds fantastic at first, but let’s just think about that for a moment. What this really means is that for an inordinate amount of time, whatever the circumstances may be, AMD GPU drivers remain un-optimised which means you as the consumer suffer, how? Possibly more bugs and lower frame rates and that is just the start of it, which leads us to the fundamental nature of the AMD drivers, specifically what would based on all empirical evidence be in part, and potentially a sizeable one, down to what appears to be terrible multi-threaded driver level capabilities, I’ll remind people here that AMD do love to talk about the importance of multiple cores and simultaneous workloads for their CPUs so to see a driver from the very same company, and a graphics driver at that, so single threaded (minded?) causing what appears to be a rather large throughput bottleneck in multiple circumstances in this day and age is an aberration. Some prime examples would be with the 3DMark API overhead test, Luxmark and Metro results, and what is now the eagerly awaited, if somewhat controversial, Metro Exodus release using the very same but updated DX11 4A engine you’d think these reasons would be enough incentive for AMD to fix their poor driver threading capabilities given the 4A engines well documented scalability with multi-threading and the onslaught of new game engines that will benefit quite substantially from good threading optimisation, but apparently not. We have a vicious cycle here, for what AMD does well in one hand is taken away by the other either through incompetence or laziness, in any DX12 title that merely uses an API wrapper to convert DX11 calls to DX12 rather than having a dedicated DX12 code path will again see those DX11 vulnerabilities raise their head for AMD RTG, bringing us back to that throughput bottleneck there is evidence of. Rinse and repeat ad infinitum for every DX12 title that simply converts DX11 calls to DX12 through a wrapper. I’m sure those more knowledgeable will also be thinking right now how fruitless and ironic the endeavour all seems to have been to birth Mantle, and in turn DX12 essentially, only to have the same driver threading problem bite AMD RTG in the arse all because they did not want to tackle the problem head on, rebuild the driver and be done with it. Now, some might argue “oh but cost”, to those people I would pose this question; Do you really think creating a whole new API in Mantle was cheaper, or easier, than addressing the driver shortcomings and rebuilding it? Also do not forget that a great many AMD RTG cards are predominantly intended for 1080p or 1440p use so there’s another reason added to the mounting pile to address driver level inadequacies.
More positively while AMDs graphics driver itself would appear to leave a lot to be desired for DX11 titles mercifully due to many factors being eliminated with DX12 AMD GPUs do fair better in circumstances
where a true DX12 code path, such as Tomb Raider, and titles that use the Vulkan API such as Doom 2016 are used. AMD have solid, unquestionably superb performance in these cases combine that with the simply stellar ReLive with everything neatly organised and implemented with no nasty in your face intrusive “login” Radeon Settings is certainly superior to anything nvidia can currently offer and AMD do love to try and bring any new features to as many of their GPUs as they can through their driver updates which if nothing else does earn better long time staying power and also allows existing owners of older AMD GPUs to trial new features out, and if they like them, perhaps invest in newer, faster AMD GPUs.
Taking a look at AMD RTG hardware It is easy to forget that the RX580 is hailed for the 1440p performance it offers and is considered a “happy surprise” but let’s reflect on that, is there anyone out there that really believes the RTG engineers accidentally over engineered a card intended for 1080p gaming? Of course they didn’t, they knew very well what they were creating and why. The brutal truth is that AMD are aware (or the RTG division at least is) how poorly their drivers are optimised for multi-threading so they had to try and brute force it by over engineering the hardware which I’m going to say just does not work any more and is even more glaringly obvious with the RX590, a massive clock boost on the GPU yet it can’t pull away from the RX580, particularly factory overclocked ones, by more than a few frames in most cases.
Historically I don’t think too many people will argue that AMD reference PCB designs are superior to nvidias meaning if you intend a long haul out of the GPU to really eek your money’s worth out of it with prices often cheaper than their rivals AMD GPUs also signal an easy option for most people, cheaper, therefore save money, albeit usually at the cost of performance and / or additional power.
With that all said however it is no excuse to brush the aforementioned software issues under the rug, my final words to you, dear AMD RTG, is imagine if your software engineers actually dedicated the majority of their time to addressing underpinning critical flaws with your driver and less time on social media, then, you’d have something to truly be proud of. Get it done and give everybody the true experience from their GPU they deserve. What is necessary, is never unwise or a waste of resources or time.
Now, dear nvidia, it’s your turn.
You certainly can’t complain in terms of performance nvidias drivers reveal there is evidence that the substantially superior single and multi-threaded optimisation gives a notable FPS advantage in 1080p DX11 gaming for the GTX980 it would otherwise be incapable of. Nvidia do get their drivers right more often than not being slick as greased lightning and well optimised across the board, AMD could certainly learn a thing or two particularly when it comes to CPU API overhead optimisation. However, this takes us to the caveat of the nvidia drivers, when they work well, they work really well, and when they aren’t working well, it may as well be a complete collapse nvidia cards almost exclusively live and die by the level of optimisation a series of cards gets before nvidia decide it is time to move on something which can be evidenced on 900 series cards with the 400 series driver releases up to 417.71 where frame rate in Tomb Raider prior to 417.71 within the 400 series driver releases nose dive by 7FPS on minimum frame rates and 10FPS on average frame rates, you might as well have replaced the GTX980 with a GTX950 such was the feeling. You could argue however that the advantage is that you get drivers well optimised out of the gate for the newest hardware and you get to enjoy your nvidia card from the very start mostly, this would be a fair argument to make I have always experienced far fewer optimisation and performance problems with nvidia drivers, AMD RTG just don’t measure up or even come close in this area.
While nvidia might have the upper hand for driver optimisation and performance their GeForce Experience software is just horrific being clunky to use overall with an intrusive login requirement before you can use the software at all feeling like the only purpose of it being there is to data mine you for information to sell, self promotional ads, and an unwelcome, unnecessary, amount of bloat with the application make it feel like those free alternatives you have to pay money for if you want to get rid of all the irrelevant bluster. You then have to consider that nvidia are also rather unfriendly with their driver features often being locked to the latest generation, or maybe going to the previous generation if you are lucky. Some stark examples of this is with the recent move to enable Raytracing capabilities on some non RTX cards such as the 10 and 16 series but not the 900 series, why not nvidia? The 900 series has perfectly good cards well capable of 1440p still and knowing how creative some developers are no doubt the Compute Units on these cards will be put to good use for Raytracing just like Crytek recently demonstrated on Vega. Then there is the issue with Adaptive vsync support, only the RTX and 10 series get it, no love for the 900 series here either. The decision to not include the 900 series for these features just feels like a manoeuvre to try and force you in to buying an RTX card, not a very nice feeling at all.
Finally we get to nvidias reference PCB design, there’s not a lot to say about them except for purely and simply they are not as well engineered, never have been, often skimping on component quality. This is something nvidia have begun to address, but it’s not there yet and you have to question if you want something from a company that is pulling an Apple and beginning to compete with their own AIBs, a lack of competition is never good for the consumer.
Applicable to both sides is each manufacturer’s commitment to respecting their customer base and earning respect and trust, both are things that are earned by doing the right thing, not blindly given through some false sense of loyalty. Companies care not for any of that, just your cash, and they will do anything to get it including claiming things that are not wholly true. In this manner both sides have done good and bad things, what one may consider worse somebody else may not we are not here to get in to those semantics but rather which company has done the more consumer friendly thing with no hidden agenda, unbiased and objectively that would have to be AMD.
My final words then, no doubt this conclusion has not been what either side expected and I doubt I'm going to make any friends at AMD or nvidia for some of the things said in this article, or maybe they'll respect it, I doubt, but theres no harm is skeptical optimism. This article became an amalgam rather than something singularly focussed that I usually do and there are certainly areas I would have liked to flesh out with more detail but Renderdoc was actively blocked somehow by Deus Ex so I was unable to explore the anomalous results there further nor was I able to decompile AMDs driver to attempt to ascertain more information. I did however say I would be calling things exactly as the results came in and not pulling any of my punches, that is exactly what I have done. Neither side is better than the other here there are some rather glaring issues on both sides, acceptance of this whether you are red or green, being vocal and demanding better, is exactly how progress and improvement is made. I like to see progress over stagnation nor do I want to see the same mistakes being made time and again, how about you? The question right now then comes down to which side better suits your philosophy and / or gaming habits. Whichever side that is, is the lesser of two evils for you. Just like anything I write I expect you to do the same here, take all the presented information then make up your own minds independent informed thought is one of the few true freedoms we still have, use it, don’t be afraid of it, and don’t squander it.
------------------------------------------------------------------------------------------------------------------
If you liked this article consider throwing a little scratch my way via a paypal donation doing so will show your love and help me as a freelance writer write more articles and reviews. Sharing this article will also help. If you'd like to read any of my reviews you can do so by Clicking Here.
The 7970 and 390X will have to wait until I have time to dig the former out and try to locate the latter but I can give you the info for the 580 I’m using at the moment, it’s a sapphire Nitro+ card, Device instance is PCI\VEN_1002&DEV_67DF&SUBSYS_E3661DA2&REV_E7\4&1C3D25BB&0&0019, Hardware IDs are; PCI\VEN_1002&DEV_67DF&SUBSYS_E3661DA2&REV_E7, PCI\VEN_1002&DEV_67DF&SUBSYS_E3661DA2, PCI\VEN_1002&DEV_67DF&CC_030000, PCI\VEN_1002&DEV_67DF&CC_0300, Compatible Ids are; PCI\VEN_1002&DEV_67DF&REV_E7, PCI\VEN_1002&DEV_67DF, PCI\VEN_1002&CC_030000, PCI\VEN_1002&CC_0300, PCI\VEN_1002, PCI\CC_030000&DT_1, PCI\CC_030000, PCI\CC_0300&DT_1, PCI\CC_0300, and matching ID is; PCI\VEN_1002&DEV_67DF&REV_E7
