Showing results for 
Search instead for 
Did you mean: 

Drivers & Software

Why is R9 Fury X performing so poorly on Forza Horizon 4?


GPU R9 Fury X at +5% OC. max Power Limit (+50%)

CPU i7-4790k at +15% OC
Motherboard Asus Z97 DUAL / Deluxe.
Windows 10 64bit 18.03

Adrenalin 18.10.1

Monitor Resolution. 1080p 75Hz FreeSync.
Benchmark result on R9 Fury X.


And running at 4K 60Hz monitor at Ultra, pushing the CPU harder with +20% OC and the R9 Fury X at a +7.5% OC and +50% power limit


There is no DX12 MultiGPU support for this title, I have checked. It will not run in DX12 MultiGPU with a pair of R9 FuryX or R9 FuryX/Nano combination.

So I cannot even get increased FPS that way. Why is DX12 MultiGPU not implemented? I am pretty well convinced a pair of RX Vega 64 Red Dragons would beat an RTX2080Ti in this title if they were run in MultiGPU for example.

Normally at 1080p, an R9 Fury X would perform slightly above an RX580 8GB and it would normally extend that lead at 4K, provided required data can be supplied to the GPU by the 4GB of HBM quickly enough i.e. "there is enough VRAM".

Looking up a performance review for the game here:

Forza Horizon 4 Benchmarked - TechSpot

The RX 580 is performing much better than an R9 Fury X and the RX Vega 64 is perfoming unusually well.

Question is why?

Is it ...

(A). AMD don't bother optimizing the game for R9 Fury X gpus any more and they don't care about them.
(B). The 4GB of HBM is choking performance (I will accept that at argument at 4k, but I think it should be enough at 1080p).
(C). There is some particular feature or improvement in the RX580/GCN 4.0 architecture that is particulary better than R9 Fury X.
(D). The game was coded primarily for XBOX One X  Console and that GPU (Scorpio) is mostly Polaris GCN4.0 + "an RX580".
(E). RX580/Vega have better DX12 implementation.
(F). Vega GPU's make use of FP16 (I find this very unlikely since XBOX One X  does not support FP16). I think using FP16 on the game would have been announced though.
(G) there is a bug in the R9 Fury X driver.
(H). All of the above.

23 Replies

Play the actual game, not the benchmark.

Also DDU -> 18.10.2.

(1). RE: Play the actual game, not the benchmark.

Here come the attack dogs eh?
The benchmark is a good representation of the game performance.
AMD seem to think so: AMD boasts that RX Vega 64 comfortably overtakes GTX 1080 in Forza Horizon 4 benchmark | TechRadar
I have tested the benchmark runniing at 4K 2K and 1080p and I compared it to the actual game performance I see.

In a few scenes in the game the average 1080p perfomance does seem to go a somewhat higher than the benchmark scores, and at 4K the FPS I see is a few FPS lower with the same OC on CPU and GPU. However I feel that overall it is a good represtation of the performance I see.
In addition the benchmark is an easily repeatable test case, not reliant on variations in user input ot user generated scripts to provide test input to the game.

(2). Why will 18.10.2 make any difference. Sure I can take a look but no mention of fix for poor performance on R9 Fury X.

Fixed issues.

  • Some Vulkan™ API games may experience a crash on game launch.
  • On multi GPU enabled system configurations Assassin's Creed™ Odyssey may experience a random game exit when Adaptive Anti-Aliasing is enabled and the game is restarted.

The Driver was freshly upgraded to 18.10.1 recently and it's a new install of Windows 10 64bit.


Installed Adrenalin 18.10.2 No difference. 1080p result here:

The benchmark resulkts seem to indicate the CPU is fine.


Adrenalin 18.10.2 No difference. 4K result, everything OC'ed to the limit of stability comparison here:


I think i7-4790K at 20% OC, 4core 8thread is still pretty representative of a good gaming CPU.

I understand the above benchmark link is running their R9 Fury X with the best performing and in reality available gaming CPU from Intel,  the Core i7-8700K  6 Core 12 thread.
The FPS numbers are better with the i7-8700K,  but even so, I think the R9 Fury X should be running much faster than this.

FYI the Fury X is performing fine on other gaming titles and also it has a Push Pull fan configuration on the Radiatior and both fans are running maxed out so I can't do much more to keep it cool and fast. Fan noise is pretty loud when I run the above benchmarks.


Filed one of these: AMD Issue Reporting Form


Just submitted one of these:
Online Service Request | AMD

Thank you for submitting your Service Request

To ensure that our response will not be blocked by spam filters please add and to your safe senders list."


Done all I can now.

I think R9 Fury X is underperforming in Forza Horizon 4 for some reason and I think it should perform better than this.
They are AMD previous flagship GPU. Putting in some effort to at least investigate it could show AMD is not simply dumping the older generation R9 Fury X card.
That might make people feel more comfortable investing in RX Vega 64/56 GPU's at this point in time, before Navi launches.

I think it would be good to see if DX12 MultiGPU could be implemented on the newer Forza Series games running on XBOX-One-X.
The Forza 4 games are clearly already well optimized for AMD cards from RX 580 -> RX Vega 64.
DX12 MultiGPU support might be possible and it might show similar impressive performance to other titles such as Battlefield 1, Sniper Elite 4, and others.
However I think  that Playground game developers reported MultiGPU support will not be coming to Forza 4 Horizon for PC.


Video recording of Benchmark and gameplay at 4K here:

Forza 4 Horizon 4K Ultra Benchmark Results. - YouTube



1st) The perf for 4k is fine, what more do you want actually?

2nd) The game is fresh released, wait for perfomance updates from the DEVS.

Not everything is a driver problem, like come on.


RE: Demo?...... yes in that video, and your point is ?.......

RE: 1st. No it is not fine at all. The benchmark run from GPU cold looks o.k. Performance drops off later. The R9 FuryX is only running on par with an RX580 4GB/8GB.

See the reply I was posting. The FPS drops down to 20 in the video I produced, in the playable part of the Demo. It was dropping down as low as 16 FPS. 

RE: 2nd. The game is fresh released, wait for perfomance updates from the DEVS
I don't see that happening somehow, and, I think the drivers do have quite a bit to do with GPU performance in game.
I think the drivers have only been optimised for RX 580, Vega 56/64 in this case. 


I looked at the CPU use on i7-4790K running at +15% OC with R9 Fury X running at max fan speed (~ 3000 r.p.m) , + 50% power, +5% OC, default (it's locked by Adrenalin) HBM frequency.  The R9 FuryX also has an additional high performance 120mm fan in pull configuration running at max fan speed the other side of the 120mm radiator. The R9 Fury X is a Sapphire Card, is in mint condition, and has been tested thoroughly running on other Games/Benchmarks where it performs very well indeed.

Overall the Total CPU % Use is only 40% during benchmark or gameplay.
The CPU load seems distributed evenly across all 4 core 8 thread in this case.

I see ~ 29 FPS in the benchmark result.
The Benchmark in the link (​) shows 44FPS at 4K.

The difference in FPS Performance seems to be close to the core/thread count ratio of the Core i7-8700K (Intel® Core™ i7-8700K Processor (12M Cache, up to 4.70 GHz) Product Specifications ) clocked at 5 GHz with 32GB of DDR4-3400 memory
My result with i7 -4790K(Intel® Core™ i7-4790K Processor (8M Cache, up to 4.40 GHz) Product Specifications )  Clocked at 4.6GHz  with with 32GB of DDR3-1600.

i.e. 12/8 * 29 FPS =  43.5 FPS which, rounded up = 44 FPS.

Continuing on with gameplay in the  Demo (that's shown in the Video) The FPS dips down as low as ~ 20 FPS at times.
I saw it go as low as 16 FPS with further testing and longer gameplay duration as the GPU heats up.

I consider 29- 30 FPS ~  playable in Forza 4 Horizon at 4K Ultra. However as the FPS drops into 20-25 FPS the gameplay is too laggy, the car does not seem to respond well enough to control input.

The visuals produced by the R9 Fury X are of very high quality overall. I have no complaints on that front. No texture flashing or dropping is seen (which might indicate not enough VRAM Capacity or Bandwidth or HBM over-temperature). However when the FPS does drop into 25-20 territory then I start to find it causes eyestrain/headaches and the game is unplayable, even though it looks great.

Total CPU Utilization rises to 100% when loading the game.

It gets as high as 60-70% sometimes when running pre-rendered demo video.
Overall though, total CPU Utilization it stays at ~ 40% when running 4K Ultra and achieving ~ 29 FPS average.


Here is a performance comparision of the two CPU's:

the ~ increase in core ration is also ~ close to the increase in effective speed of the CPU in single thread so it's difficult to tell why the benchmark system performs better exactly, also it has faster DDR4 RAM which may help. I looked at running i7_4790K with 4Cores hyperthreading off and also 2 cores 2 threads and the benchmark performance did drop but only by about 1-2FPS, giving 27/28 FPS. The gameplay FPS did drop a bit more than that, however the i7-4790K with either 2core 2 thread or 4 core, hyperthreading off simply loads the remaining available cores/threads more. Understandable since total CPU Utilization was only 40% during gaming with 4 cores 4 threads available.


Looking again at:

The RX Vega 56 is hitting       64 FPS average.
The R9 Fury X is only hitting  44 FPS average, same as an RX580 4GB and 1 FPS slower than an RX 580 8GB.

That seems a very large 44.5 % performance improvement difference between the RX Vega 56 and an R9 Fury X, even in a DX12 title.

Sure the base clock speed is a bit higher:

1,156 MHzvs1,050 MHz

And the Turbo Clock speed is higher:

1,471 MHzvs1,050 MHz

But then the Vega 56 is basically a chopped down Fiji and if the game is well oprtimized in DX12 then that should matter.
I don't think a reference Vega 56 will sit at turbo speed the entire time in this game.

I think a generous performance boost figure would be 25%.

An average figure would be more like 13.6% according to this:

So that's why I question the performance difference and I think the R9 Fury X driver is not performing well on Forza 4 Horizon.

If you translate those figures back to my R9 Fury X on i7-4790KI I would expect to see the following.

Based on a generous +25% performance difference I would expect to be able to run at an average:         34.1 FPS not 29 FPS.
Based on a typical performance improvement of +13.6% I would expect to be able to run at an average:  37.6 FPS not 29 FPS.

Would that additional performance matter?  - yes, since the game is ~ playable at 4K Ultra with 30 FPS average, with brief dips down to ~ 25 FPS.
An additional 5-8.5 FPS average increase would definitely help.

Alternatively if AMD cant improve the driver performance on the R9 Fury X  I ask the question again.
Why is this title not running on DX12 MultiGPU?

It is clearly very well optimized for RX Vega 64/56/and RX580.
Sure running a pair of AIB RX Vega 64 requires a 1500W PSU so that's a bit extreme and the RX Vega 64 runs at over 60FPS at 4K. 
A pair of RX 580's in crossfire would only burn ~ 2x the power of a GTX 1080, and might scale well in DX12 MultiGPU so could benefit.


Is the game using a lot of video memory? That could well explain the perfomrance difference as the Fury X only has 4GB which may not be enough in this game at 4K resolution.


Hi, thanks for responding. HBM: The 4GB Question - The AMD Radeon R9 Fury X Review: Aiming For the Top

Knowing exactly how much video memory is used/needed by the GPU in a game/application is difficult to know unless I would be able to run a detailed debug/analisys tool as a the programmer/code performance analyst. It's a matter of can the correct input data be delivered to the GPU quickly enough to be processed, as far as I understand it, not simply how much is stored directly in VRAM on the GPU. Some games / applications will simply load as much data into VRAM from System RAM as possible,  texture files etc even if they are needed or not, and fill the VRAM to just under 3GB, 4GB or 8GB, or whatever. OK it's on the VRAM and ready to be used if needed, but, as long as the application (& drivers) are written to manage what data is required, and make sure it gets into the Vram quickly enough from System RAM, the application/game should run well enough. There is an article discussing the 4GB of HBM on R9 Fury X here: HBM: The 4GB Question - The AMD Radeon R9 Fury X Review: Aiming For the Top

I have been surprised at how well the R9 FuryX/Nanos can run at true 4K or 4K Virtual Super Resolution settings on numerous games. I regularly run BF1 at 4K Ultra in DX12 MultiGPU. That runs fine mostly 50-60 FPS on an R9 FuryX/Nano runinng at +5% OC with i7-4790K clocked at +15% (~ 4.6GHz on core 1and 2 cores and 4.5GHz when running 3, 4 cores.
In addition I have tested and run things like Middle-Earth: Shadow of Mordor 6GB Ultra HD texture pack which is supposed to require 6GB of VRAM minimum.
It runs fine on a Sapphire HD 7970OC 6GB  or an R9 FuryX or R9 Nano with 4GB of HBM, benchmarking / running at  same FPS as without use of that texture pack.
Wheras a R9 280x (a refresh version of the HD7970 OC, not a rebrand with 3GB of VRAM)  or a GTX780Ti with 3GB of Vram simply cannot handle it and runs at very low, unusable FPS.

I took a quick look yesterday to see how my oldest Nvidia GPU I still use regularly, an Asus  GTX780Ti 3GB OC Direct CUII + i7-4770K clocked at 4.6GHz on all 4 cores ran  with an Nvidia Driver from ~ end December 2017 (I will update the exact details later - the machine is busy running rendering) behaved with Forza4 Horizon. The Nvidia drivers used are not specifically optimized for the game. I was using the Nvidia equivalent of Virtual Super Resolution to render and run the benchmark at 4K on a '1080p 60Hz' panel). I  recored what happens/ how it performs with the 4K Ultra Benchmark, and they were quite good in terms of FPS compared to the R9 Fury X. The GTX 780Ti It was not as fast, but it was not that far behind, given the driver is unoptimized for Forza 4 Horizon, and the GPU was not overclocked at all. The benchmark runs and then before 33% progress a warning is flashed up on the screen regarding VRAM capacity. The benchmark crashes shortly afterwards.

I will upload that video running on a GTX780Ti with that old driver and point you to it if you wish.
I see no such VRAM capacity warning running with the R9 FuryX or Nano during the benchmark or running the game demo.

I could also see if there is a new Nvidia driver optimized for Forza 4 Horizon and install that and run that on a GTX780Ti.
I could look at overclocking the GTX780Ti and see if that improves it's FPS performance.
I could also remove a PCIeSSD from the third PCIex16 slot on that machine so I can run the secondary R9 Nano at PCIe3.0x8 speed so I give a fair comparison of those 2 air cooled GPU running with same i7-4770K CPU at same OC.
All of the above would take me time of course, but if it helps investigate and improve R9 Fury X / Nano performance then I will see what I can do.

Regarding VRAM capacity - please look at this 4K benchmark result I referred to above at the following link: Forza Horizon 4 Benchmarked - TechSpot

Those numbers are completely independent of me.

Please compare the 12GB GTX Titan X to the 6GB 980Ti to the RX 5808GB to the R9 FuryX 4GB to the RX 580 4GB ...etc.
If the game really needed say, 6GB of VRAM capaciity, and absolutely could not get data from system RAM into the GPU VRAM and presented for processing by the GPU unless the 6GB of data was sitting ready in the VRAM, then, I would expect those cards with 4GB of VRAM capacity to run at much lower speeds, and for the GTX1060 6GB to significantly outperform an RX5804GB card for example:


I save this now and add some more info next.



I cannot edit my previous response to add additional information so here is some other data regarding trying to work out if GPU VRAM Memory used (I don't think it is the problem) is causing poor performance of the R9 Fury X at 4K Ultra settings on Forza 4 Horizon Benchmark:

(A) The Radeon Overlay is not working for me on Forza 4 Horizon so I cannot see VRAM use.

(B) If I install latest version of ASUS ROG GPUz just so I can run the Monitor (I have set it up NOT to change any GPU Clocks / Overclocking settings) it can see and monitor everything such as GPU SCLK, HBM CLK, but the GPU Memory use does not seem to be monitored at all.
In any case, a couple of references about the Limitations of GPUz are that it reports VRAM requested by the application b ut not what is actually used/needed.

Here is another link regarding VRAM use and myths:

(2). The Myths Surrounding Graphics Card Memory - The Myths Of Graphics Card Performance: Debunked, Part ...

I have used GPUOpen RadeonGPU  Profiler before, so I thought I would have a quick look and see if I could run a profile on the DX12 Forza 4 Horizon Benchmark using the latest version of it and see if it would tell me anything useful about VRAM use. For anyone interested in that here is a link to more information on it:

The following video is useful:
How-To Use Radeon™ GPU Profiler - YouTube 

To summarize I tried to use the latest version of the Radeon Profile GUI to run a profile of Forza 4 Horizon on my PC & setup.
It fails to launch correctly.
I have run out of time to look at that any further.
I am uploading a video to show you what happens with it.
After I have done that I have no more time to look at this any further for today.




Here is a quick video showing my attempt to run latest version of AMD GPU Open Radeon GPU Profiler on Forza Horizon 4:
Quick look at running AMD GPU Open Radeon GPU Profiler on Forza 4 Horizon. - YouTube
It seems to start OK, then it seems to get confused and think more than 1 Forza 4 Horizon application is running.
I have tried multiple times, inluding log off log on / reboot the PC, same thing happens with the Radeon GUI Profiler.
That's about all the time I have to look at this for today.




RE: Is the game using a lot of video memory?

I do not think insufficient Video Memory is an issue on the R9 Fury X on te Forza 4 Horizon benchmark at 4K Ultra.

I have completed looking at the performance of an Nvidia GTX780Ti 3GB versus an R9 Nano 4GB HBM running on same i7-4770K CPU with same PCIe 3.0x8 bandwidth. An Nvidia ASUS GTX780Ti 3GB OC DCU II (Kepler GPU) can complete the benchmark fine, provided I do not run Shadowplay recorder at the same time. This indicates to me that 4GB of HBM on the R9 Fury X or R9 Nano should be more than enough. I can post the comparison results if you want. I have been too busy to do it yet. I do have that data. However I am now confident I know what insufficient video memory looks like on this Forza 4 Horizon benchmark.

Here is a video exaple of a single R9 Fury X running the Forza 4 Horizon benchmark at 1080p. In this example I deliberately preloaded the 4GB HBM Vram memory with data using a program with a low CPU overhead before running the Forza 4 Horizon benchmark. The 'halting'  load data into VRAM -> execute game, load data into Vram -> execute game  shown in this video is more like what I would expect to see if insufficient VRAM is the issue on the 4K Ultra benchmark that I pointed you to earlier. That benchmark runs smmoothly. I see no warnings about low video memory.

In this case where I preloaded the HBM memory with data you can see I get the the WARNING "Low Video Memory" notice that I see on the GTX 780Ti 3GB GPU running the benchmark: Deliberate limiting the available HBM Vram on R9 Fury X on Forza 4 Horizon benchmark. - YouTube



And on to another point ...
AMD Fury Series Voltages And HBM Overclocking Unlocked With Sapphire's Latest TriXX Version, v5.2.1
More specifically this:
AMD Fury X Memory Overclocking Unlocked In MSI Afterburner - HBM Overclocked And Benchmarked, GPU Vo...

If the Crimson ReLive / Adrenalin Driver did not lock HBM overclocking in MSI Afterburner or Sapphire Trixx since just before the RX Vega 56/64 launch,  I should be able to get addidional performance from the R9 Fury X, and those performance differences between RX Vega 56 and R9 Fury X would very likely be lower.


Final test I overclocked the CPU to + 20% OC on the i7- 4790K (all four cores/ 8 threads running at 4.8 GHz). I set the R9 Fury X to a GPU CLK overclock of +7.5%, I turned off ReLive. That gets me back to average FPS = 35 on the benchmark. The min FPS still dropped down to around 22. That is definitely an improvement. Turning off ReLive Recording improvement =  2FPS. The CPU overclock gives another ~ 2.5, and increasing the R9 Fury X OC to 7.5 gave another 1.5 FPS.

However the game froze at the start of the Summer Stage in the demo and benchmark. Radeon Settings crashed and reset.

So that's it I will have to look for a 4K FreeSync monitor with a low end Freesync range of 20FPS to run this game with a single R9 Fury X. 


Here is the response from AMD Support.

Dear Colesdav,

Your service request : SR #{ticketno:[8200839103]} has been reviewed and updated.

Response and Service Request History:

Thank you for the email.
I understand you are noticing low performance of Fury X in Forza Horizon 4 game compared to RX 580 graphic card.You have already tried clean installation of latest drivers and reported the issue as well.
I will forward this internally for further investigation. I request you to stick with the current setup as of now and wait for next driver release.
Thank you for contacting AMD.
In order to update this service request, please respond, leaving the service request reference intact.

Best regards,


AMD Global Customer Care


My response:

Thank you for responding. I appreciate it. I look forward to see the next AMD driver update. 

I will update the ticket and any more information I have regarding the Fury X performance on Forza 4 Horizon and any progress I make trying to improve the performance

every week from now.
Any detailed information will be updated here:

I will likely investigate if I can somehow overclock HBM on the R9 Fury X as that should help push the FPS performance higher.

The other issue that is interesting to me. Why is the FPS performance so much slower on an i7-4790K versus the i7-8700K.
I tested running i7-4790K with 4 cores hyperthreading off and 2 core 2 thread versus default 4 core 8 thread.
Reducing the number of cores or threads only resulted in ~ 1 FPS performance reduction in the benchmark.

Boosting all cores up to 4.8GHz from 4.6,4.6,4.5,4.5GHz on core 0,1,2,3 did give a surprising increase in FPS.
This was surprising to me given this is a DX12 title.

My assumption in general is DX12 driver overhead is lower than DX11 for AMD cards (seems true, total CPU utilization is only 40%, however ... the FPS IS very low).
The DX12 driver is multi-threaded (CPU workload does seem to split evenly between all 4 core / 8 threads based on a quick high level look at CPU process monitor).
However in this case a small 0.2-0.3 GHz increase in CPU Clock frequency on the i7-4790K CPU cores does seem to boost FPS by a significant amount.
Thanks again.


Here is a video showing how an ASUS Nvidia GTX 780Ti Direct CUII   performs at 4K on the benchmark, using up to date Nvidia Drivers.
Forza Horizon 4 GTX780Ti OC 3GB at 4K Ultra + GeForce ShadowPlay Video Memory Warning - YouTube

I will upload the benchmark result later.

It completes the benchmark fine, if I do not record with Nvidia GeForce Shadowplay.
Therefore 4GB of Vram on the R9 Fury X should be more than enough.