AMD Adrenalin Driver Analysis

https://i.imgur.com/PM1Eup8.png



Introduction

Today I thought it's time for a bit of a change of pace I figured it was about time the people took a look at the performance evolution of the Adrenalin drivers spanning from 2017 through to present driver releases to plot what sort of changes have taken place. There is a reputation around the Radeon Technologies Group that has been bestowed upon them giving the perception of Radeon hardware aging like fine wine yet there is extremely limited real world evidence of this, and some would say none at all. The purpose here today is to try to give credence to this reputation, or debunk it. So without further ado let?s get things underway.



Test System

There's been a few changes to the test system today the system memory has been highly optimised, the CPU is clocked a little higher than usual and the GPU has been overclocked and GPU memory timings I have optimised with my own custom timings, no one click optimisation here. The aim is to remove every limiting factor possible to see best case scenarios allowing both AMD hardware and software to put their best foot forward for the ensuing tests.

CPU: AMD Zen 1700 @ 3.9GHz
Mainboard: MSI X470 Gaming Pro Carbon
RAM: 2x8GB G.Skill Ripjaws V 3200MHz @ 3466MHz CL16
GPU: Sapphire Nitro+ RX580 8GB @ 1.425GHz, 2.2GHz (8800 effective), custom optimised timings
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 1709
Case: NZXT Phantom 530

Drivers:
AMD Chipset Drivers v18.10.0830
Realtek Audio Driver v6.0.1.8619
Xonar C-Media 1825 Driver



The Adrenalin Drivers

Let's begin taking a look at the drivers we will be testing with a brief summary of performance based improvements and major bug fixes each driver claims to bring. There have of course in addition been an abundance of smaller driver bug fixes, in the interest of trying to maintain a level of consistency you'll notice I've tested the final driver release of each month which typically end in .3. There are some exceptions to this but not many, hotfix drivers will not be included under the simple fact they only fixed very specific things thus it is extremely unlikely anything significant happened with said drivers other than the hotfixes.

Adrenalin 17.9.3
Multi GPU support enabled for Warhammer II

This driver will be our baseline from which all other drivers will be assessed by.

Adrenalin 18.2.3
Up to 39% better performance on RX580 8GB cards in Sea of theives
Up to 20% faster performance on RX580 8GB in FFXII Zodiac Age
Up to 29% better performance on Vega64 in Sea of Theives
Up to 13% better performance on Vega56 in FFXII Zodiac Age

Adrenalin 18.12.3
No major notes

Adrenalin 19.1.2
Up to 7% better performance in Anthem @1080p on the RX580

Adrenalin 19.2.3
Up to 10% average performance gains with mobile Vega Graphics
Up to 17% average performance gains in eSports titles with mobile Vega graphics
Up to 3% performance gains with Vega 64 in Dirt Rally 2

Adrenalin 19.3.3
Support for DX12 on Windows 7

Adrenalin 19.4.3
No major notes

Adrenalin 19.5.2
No major notes

Adrenalin 19.6.3
Fixed YouTube login for ReLive

Adrenalin 19.7.3
Up to 13% better performance in Wolfenstein: Youngblood

Adrenalin 19.8.1
Some system configurations may experience colour corruption after install of Radeon Software when running Windows? 10 May 2019 update. Issue is resolved in the latest Windows Update 18362.267 (KB4505903)

Adrenalin 19.8.2
Up to 10% improvement in Control running DX11

Adrenalin 19.9.1
Up to 8% improvement in Gears 5

Testing


All tested games are tested at 1080p using the highest quality preset where presets are unavailable maximum settings are used. Radeon Global Settings are all at default except for Tessellation which is manually set to 6x. We want to measure legitimate driver level optimisations and not allow the Radeon Technologies Group division to cheat their way through these tests with cheap tricks like simply automatically detecting a game .EXE and reducing the default level of Tessellation a game might use.

Due to some benchmarks inaccurately measuring minimum frame rates (recording at instances such as scene changes and the like) average and maximum results only are used. I could have used RTSS but for results that are as accurate and as easily reproduced as possible if you the reader are curious and want to do some testing of your own I elected to use only what the benchmarks themselves recorded and not outside software.



https://i.imgur.com/EfnoLX6.jpg

The first test is Alien Isolation, being an AMD Gaming Evolved title it?s a good example of AMDs level of dedication to ensuring games in this initiative at least get the very best optimisation.

https://i.imgur.com/3DXO8TR.png


There isn't a whole lot to say here performance is respectable but nothing to write home about. It looks like even earlier drivers than what is tested here could have had some driver optimisation for Alien Isolation if you want to take the level of consistency with maximum frame rates and do a little extrapolation based on the 17.9.3 result but that would be pure conjecture.

Based on the available evidence you certainly can't say AMD RTG ever optimised for Alien Isolation, they clearly never have, for a game in their own Gaming Evolved initiative no less. At best this first test shows a steady state over time which is fine, but it does look like driver polish is lacking given the almost static results across the board. That bottle of fine wine hasn't been opened yet.



https://i.imgur.com/pNxxNbH.jpg

Following on from Alien Isolation we move to Metro Last Light Redux, the 4A engine is well known for being heavily multi-threaded as well as looking absolutely stunning even today. Primarily this will be an excellent test of batching and threading capabilities.

https://i.imgur.com/058CUz4.png


Performance here is actually fairly good you are made to work for it though without highly optimising the test system 90FPS average was a struggle to obtain being 87-89FPS. Optimising the system netted another 7-10FPS which is the difference between rather unremarkable given the age of the game and quite good. There's a tangible indication that AMD RTG just might be starting to optimise their driver more if the fairly consistent maximum frame rate results are any indication. You certainly can't pin any hopes on such finite observations with only an increase in maximum frame rates but it is still a slight whisper on the breeze of the wind of change that performance for DX11 titles could be on the rise in the near future with driver updates.

Ultimately you should look at these results and see another steady state scenario, not fine wine, which is not necessarily bad but as yet there's nothing to give AMD that "fine wine" moniker their GPUs have been associated with, not in my book.



https://i.imgur.com/6R9GRMq.jpg

For the final DX11 real world test we are going to break out Final Fantasy 15, looking at as many scenarios as possible is imperative in building a complete picture and not a blinkered one. To build such a picture you must look at both titles hardware does well in and not so well in. The latest version of the FF benchmark will be being used with all settings at maximum, all nvidia options are disabled with the exception of hairworks because if you aren't at least using this you might as well play the game on console.

https://i.imgur.com/mN5KSpr.png


Once again performance here is a steady state with the oldest tested driver being among the best performing, the Adrenalin 19.7.3 drivers and onward do take a fairly sharp fall in this test though for some reason; it is the difference between a "Fairly High" and "Standard" result. There is also a notable amounts of hitching / micro stutter / whatever you care to call it present in newer drivers that is not there in the older tested drivers.

Most puzzling of all is that AMD were meant to add driver optimisation for FF15 yet there is zero evidence of them ever doing what they claimed they would, and let's be clear here; the oldest tested RTG driver was released September 2017, some 11 months after Final Fantasy 15s release date RTG had more than ample time to incorporate these optimisations at that point.



https://i.imgur.com/LgxUQxK.jpg

Our first DX12 test sees the enjoyable, if somewhat abruptly ending, Deus Ex Mankind Divided come in to play, as always the maximum quality preset is used.

https://i.imgur.com/E57l2fi.png


Average and maximum frame rates are rock solid across all tested drivers, no complaints here as Deus Ex isn?t made on the most optimised of engines but what is nice to see (and something I commented I would like to see prior to holding this article back to look at more newly released driver sets as they come out) is some improvement to frame rates with newer drivers 19.8.2 managed, it's not much, but given how extremely consistent these results have been you can take the observed improvement here as a legitimate one, even if it is just for the maximum frame rate. Sadly the newest set of drivers sets things back again and it's as you were.



https://i.imgur.com/FASVbqp.jpg

We now come to Shadow of the Tomb Raider, a game Radeons typically do well in likely due to the games console heritage. Once again the maximum quality preset has been used.

https://i.imgur.com/2ohGxAk.png


This was unexpected, there's actually a performance improvement that can be plotted I wouldn't blame any of you for nodding off while looking at these charts things have been so uneventful. At least we know RTG have done something performance related with their drivers over the last two years, even if that improvement only really happened recently.



https://i.imgur.com/ZF4rgL8.jpg

The final DX12 test brings Ashes of the Singularity to the party, the maximum "crazy" preset was used for testing.

https://i.imgur.com/jw88bqo.png


There's no surprises to be had with Ashes either, performance across all tested drivers is consistent although I had expected the RX580 with the optimisations it has combined with the system optimisations to muster around the 40FPS mark at least. Something is slightly amiss here.




Performance Overview

https://i.imgur.com/ue3BWEs.png


To put all of the tests in to a more easy to follow graph above is a summary of performance optimisations, in short Tomb Raider sees a notable improvement with Adrenalin 19.7.3 onward of almost 10% which is certainly worth mentioning and carries through to the latest drivers suggesting this is not just a happy accident. Based on the remainder results clearly performance optimisations have not been on the cards for a very long time which would certainly go a good distance to explaining the performance gap that has opened up between RTG and nvidia in a lot of titles between team Red and team Green GPUs in the same class, good hardware can't survive without good software and good software can't survive without good hardware. Such a long period without a focus on driver level performance optimisations does beg the ever more looming question; Just what are RTG doing over there? The glimmer of hope is the impressive Tomb Raider improvement hopefully this is a sign of things to come for the RTG drivers and AMD GPU owners. Outside of this one result there's nothing else to talk about which is a real shame after all the noise and blow-hard AMD has made over the Adrenalin(e) drivers. For all the fancy facade the drivers have undergone with the Radeon Settings underneath lays the same mediocre driver everyone used to criticise, at least in the tested games here.

For the next test I will be taking a selection of the tested drivers and putting them through 3DMarks API overhead test as a means to measure threading and batching capabilities.

https://i.imgur.com/Ub556dS.png

As we can see from the drivers selected for this test multi-threading and / or batching seemingly remains a problem in DX11 for AMD and the Radeon suggesting a poorly optimised driver stack, worth noting is that DX11 performance is slightly better with the oldest driver tested compared to the newer drivers but all results absolutely pale in comparison to the GeForce driver results which I added here to put things in to perspective.

How much these theoretical results hamstring RTG hardware performance in real world DX11 gaming is something that will only really be known if RTG ever address this issue they seem to have, at worst it is costing gamers performance in the real world, at best it is a synthetic anomaly that requires addressing either through the driver or on the part of Futuremark but as things currently stand with so many question marks it is certainly enough to put a lot of people off buying Radeon as there?s an awful lot of DX11 titles out there.

I'm going to be extremely critical of AMD RTG here as you can go all the way back to 2015 and the R9 290X where Radeon hardware exhibited this exact same issue. I know there's people out there who will get on the defensive here for me saying this but before you do you should see this SOURCE. To reiterate again as I have before this API overhead test is not to measure a GPUs performance in any given API but rather how efficiently draw calls are handled / executed. The Radeon results look bad in the same way the GeForce DX12 results look bad in reality the real world performance at worst in terms of percentages is most probably going to be double digits so not earth shattering but enough to sway a purchasing decision especially if you already own a large library of DX11 titles and thus something AMD RTG should consider an important issue to at the very least to look in to.

The final test I'll be doing for this driver analysis will be with Tessmark which while being more of a hardware test none the less will be interesting to run to measure Tessellation performance and seeing how it stacks up between a few different generations of cards.

https://i.imgur.com/yCuaKvr.png

A nice little surprise here from the RX580, fully optimised and overclocked the card manages to beat out a GTX980 by a healthy margin, if only the specifications of an RX series card translated in actual games more often, something which both developers and the RTG driver division must share blame equally for.


Conclusion

Where to start with this peoples conclusion, I suppose we have to start with how despite what AMD promised of implementing driver optimisations for Final Fantasy 15 no driver tested here today, and that is 13 of them, exhibited any such optimisation but rather a regression with newer drivers. If I'm to be generous I could say the driver optimisations for FF15 have been broken since 19.7.3. No tested driver showed any optimisation efforts on RTGs behalf for Alien Isolation either, which while performing decently anyway if there is more optimisation that can be done to improve performance it should - it's that simple, particularly as nvidia with competing same in class GPUs perform better in a title that is meant to be one on AMD RTGs home turf thwarting an OC'd and optimised RX580 which can only be considered somewhat of an embarrassment for AMD RTG. The bottom line as I highlighted earlier is quite straightforward here; AMD RTGs performance in DX11 titles needs work, no question about that, and for more than just because they need to play catch up but because a shortfall like this will cost them sales.

More positively performance in most the tested titles did remain consistent while Tomb Raider showed remarkable improvement over recent driver releases, pleasing stuff indeed. Tessellation performance also proved to be very strong with a fully optimised and OC?d RX580 on paper at least. RTGs real problem, as it always has been, is their hardware being underutilised and never fulfilling the true potential it has it is about time AMD RTG shakes this monkey off their back.

The ultimate conclusion and the one this analysis set out to prove or disprove is does RTGs GPUs age like a fine wine? Well, based on the evidence here no, no they absolutely do not it is a title that has been bestowed without merit, and as it is here, evidence to support the claim.

-------------------------------------------------------------------------------

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 other work you can do so by Clicking Here.

I have been using AMD gpus since X1900XT crossfire days.... Here as of late, seriously considering going Nvidia 2070 super direction... This, and the lack of any real performance out of the AMD team really leaves me wondering if "navi" is going to be the "savior" AMD thinks it will be, along with the lackluster drivers being dropped on end users....

Honestly the 5700XT doesn't impress me due to the myriad of problems the 5700 series are having. If I were looking for a bang for buck GPU now my pick would be a Vega56 or GTX980Ti, overclocked the card will reach 1070 levels at least which makes it a potent 1440p card still while a V56 OC'd will at least reach V64 performance and do it while consuming less power. Optimised and overclocked RX580s would be another consideration but are obviously weaker. I can't say any new GPUs are particularly impressive either $/?400+ for a 2060 Super or 5700XT for GPUs that are meant to be midrange cards is absolutely abysmal value. We have nvidia to blame for those prices though they just opened the door for AMD to charge more. Theres no such thing as a sweet spot $/?250-300 GPU you can buy from new right now, and I seriously wonder if anyone will ever be able to get those types of GPUs again. The used GPU market is about to become a lot hotter.