cancel
Showing results for 
Search instead for 
Did you mean: 

Drivers & Software

hitbm47
Forerunner

Proof AMD is either deliberately hogging pre-Navi performance or very negligent.

Hi community, my system:

Windows 10 64-bit latest build,
i7 870 (OC'ed to 3.9Ghz stable),
16GB DDR3 RAM (Clocked at 1866Mhz),
MSI Radeon RX 480 Gaming X 8G,

I am providing screenshot proof that AMD is still performing absolutely unacceptable on the Polaris/Vega kernal in Unreal Engine 3 (DirectX9 version) games, but it seems the lost optimization has been reintegrated into the Navi kernal and optimizations, when I have been reporting the issue since 2017.

Furthermore, I have been noticing my GPU is hitting 100% in some game titles with 70-80watts, whereas in the past it has only ever hit 100% utilization at 110-125watts and this is impacting game performance by 15-20FPS (in games such as Kane & Lync 2, WRC9, Alien Rage, GTAV, etc.). To continue, this is even with great temperatures in my system (RX 480 staying below 76C at 1303Mhz which is the MSI default clockspeed).

I am not going to name the author of the drivers, but out of my frustration of AMD not fixing or lack of attention to five-year long driver issues, I decided to give the Digitally signed modded drivers a try and herewith follows the screenshots at 1440p:

Screenshot of Polaris/Vega kernal dipping to 50FPS-30FPS or lower (depending on CPU OC) since 17.7.2 - 22.5.2:

AMD_Polaris_kernal_hogging_1_unreal_engine_3.jpg

 

Screenshot of Navi kernal over Polaris/Vega providing expected performance as it used to in 17.7.1 or earlier at 70FPS or more:

AMD_Polaris_kernal_hogging_2_unreal_engine_3.jpg

 

So as you can see, a massive difference and how it should've been and how it was in 17.7.1. I find this to be very bad business practice or negligence from AMD, and I really hope that they are planning to port to optimizations to pre-Navi cards as well, including those new DirectX11 optimizations that does not work on Polaris yet without these modded drivers. The only thing holding me back from permanently using these modded drivers with the Navi kernal is some artefacts in Epic Games login and one game thus far.

I have already proven that it still works well on Nvidia, now I have proven that the lost optimizations have secretly been ported to Navi. There is no more excuses now AMD.

Please pass this through to the dev team @dorisyan , @xhuang , @dipak , and please don't close this thread ray_AMD.

Kind regards to other affected users.

0 Likes
6 Replies
Chester1994
Challenger

I don´t really think they are doing it on purpose. Sadly, making drivers use efficiently the hundreds of thousands of apps out there must be difficult. Even more when we talk about a game engine released in 2006 (In my opinion the most important game engine ever realesed in addition to Source).

However, I´m interested in those modded drivers you are talking about. Can you provide a link? Thank you!

@Chester1994thank you for your response, the point is they are taking these type of things up much too lightly and it carries over to other scenarios. For example, those DirectX11 optimizations should've been present 5-6years ago already and now they are only bringing them to Navi.

Furthermore, I have gained some stability and FPS in Assassin's Creed Origins as well by using them, I can tell you they are the most popular AMD modded drivers on youtube at the moment, do you mind if I send the link to you privately? I don't want AMD taking them down, because these people/person are actually trying to make a difference.

EDIT: @Chester1994 , I tried to private message them to you, but the website states that you do not have PM enabled.

EDIT #2: I give in, hopefully AMD does not make an issue of this, but here is the installation guide for the drivers, it should give you an idea of how to install them: https://youtu.be/PZSWJxkPZA0

EDIT #3: Also, it is not just a handful of games, it is like 50-60% of them on average that have unstable performance in AMD Official drivers. I really like the hardware itself, but the performance is in my opinion unjustifiable, especially if your in a 3rd world country where the prices are even more expensive and you saved up money like me to buy the RX 480 in 2016 for ~300USD.

Kind regards

Thank you for the link, I´ll check it out only out of curiosity. I have an RX 580 4GB.

You can also enable hypermemory in that driver setup to use some of the RAM as VRAM.

hitbm47
Forerunner

I have now also determined that AMD is much more bottlenecked or inefficient in their driver about PCIe lane modes than Nvidia.
For example, removing my GTX 1060 3GB from my motherboard resulted in giving x16 PCIe lanes to the RX 480 and giving a 5-10FPS boost in WRC9 in affected areas, whereas the GTX 1060 was unaffected to using x8 PCIe lanes and could keep the same frame-rate as the RX 480 needed in x16 lanes.

This is a major issue for people that are even maybe using Crossfire for older titles, since if they are playing a game that can't support Crossfire, they might be severely affected in some scenarios by AMD's way of handling PCIe lanes and not actually a CPU bottleneck. To add to this issue, AMD released modern-ish chipsets such as found in the MSI B450 Tomahawk where the second PCIe slot runs at only x4 PCIe Version 2 (not even Version 3) on AM4, whereas the 990FXA AM3+ platform did x16 PCIe 2 for both the primary and secondary slots.

Furthermore, AMD seems to be affected by some mismanagement of communication between VRAM and other components of the RX 480, here you can see a screenshot of Kane & Lynch 2 being severely underutilized even at x16 PCIe, whereas the GTX 1060 3GB could do 70FPS+ in x8 PCIe V2 mode:

kane_and_lync_2_PCIe_bottleneck.png

EDIT: Now to think of it, AMD is probably hogging performance in scenarios such as this as well to promote SAM (Smart Access Memory). Just 4-6weeks ago when I tested this exact Kane & Lynch scenario, the RX 480 was able to keep 60FPS+ in this scenario at x8 PCIe V2 speed, and now look at the performance on official 22.5.2 drivers.

 

Any AMD employees have some input to give?

Kind regards

0 Likes

The entire encoder/decoder has been broken for the 6000 series for more than a year and their drivers suck ass for vr related content, i wouldn't get any hope on having this fixed or any attention from devs at this point.