Benchmark review: Final Fantasy XV for Windows PC (Updated)

#5524041 Posted on: 02/28/2018 05:32 PM

---

Good so see a article on the games performance. Should make for a interesting read.
(And the benchmark makes comparisons pretty easy since it's a scripted run that is mostly identical in how it plays out.)

#5524044 Posted on: 02/28/2018 05:36 PM

---

Good so see a article on the games performance. Should make for a interesting read.
(And the benchmark makes comparisons pretty easy since it's a scripted run that is mostly identical in how it plays out.)

These are NOT results done with the benchmark released a while ago!, this is a manually played run in the first chapter of the game (demo). Thus this is based on real gameplay and thus in-game measurements. Check the FCAT video in this article for the measurement run to replicate if you'd like to.

#5524045 Posted on: 02/28/2018 05:39 PM

---

Looks like GTX 1060 is ~14% faster on average vs. RX580 at 1080p.

#5524051 Posted on: 02/28/2018 05:44 PM

---

"This is an NVIDIA gameworks and this NVIDIA optimized title"

Is it really just optimized when your company is actually creating the black box methods for how the game actually renders stuff? How would one determine something isnt working right when only Nvidia knows? I bet its always working as intended /s

Back in the day it was a big deal when companies made game specific driver optimizations rather than something that optimized for the whole engine. We moved from that to working with the game developers to make sure hardware works correctly. Each hardware vendor could do this since the only barrier was the developer.

Take that to the next level with gameworks actually locking out the other vendor. I guess that in a nutshell is the whole gameworks argument.

#5524053 Posted on: 02/28/2018 05:51 PM

---

A difficult discussion, you can see that the perf hit with gameworks on/off is roughly equal for both parties. FF XV definitely seems to like GeForce cards a notch better though. BTW as mentioned in the article, I tested with the gamework features disabled, to create a more fair and balanced test field for both brands.

#5524054 Posted on: 02/28/2018 05:56 PM

---

Should get better. Initially, the benchmark ran like complete crap for me, not a stuttering issue.
With the newest driver, I gained nearly 1000 points 1080p standard quality.
I just hope this isnt one of those gameworks games where AMD is locked out of the game engine.
Plus I wonder if a 100mb read/write mechanical drive is enough for this game. I have 4 SSDs, 1 for OS, 240GB, 480GB, and a 512 GB. The 3 game drive are full, so maybe I will clear 100GB off the 240GB. Its got 57GB free.

#5524057 Posted on: 02/28/2018 06:03 PM

---

"This is an NVIDIA gameworks and this NVIDIA optimized title"

Is it really just optimized when your company is actually creating the black box methods for how the game actually renders stuff? How would one determine something isnt working right when only Nvidia knows? I bet its always working as intended /s

Back in the day it was a big deal when companies made game specific driver optimizations rather than something that optimized for the whole engine. We moved from that to working with the game developers to make sure hardware works correctly. Each hardware vendor could do this since the only barrier was the developer.

Take that to the next level with gameworks actually locking out the other vendor. I guess that in a nutshell is the whole gameworks argument.


Source code for the following gameworks libraries is available here: https://developer.nvidia.com/gameworks-source-github

AnselSDK (EULA)
Flow (EULA)
FleX (EULA)
Blast (EULA)
Vulkan/OpenGL Samples (public)
NvCloth (EULA)
Blast (EULA)
HairWorks ( EULA )
HBAO+ ( EULA )
FaceWorks (public)
PhysX SDK ( EULA )
Volumetric Lighting ( EULA )
D3D Samples (public)

Nothing in the EULA/Terms signing up prevents AMD from seeing the source of the GW Libraries that are available. The bottom line is that the vast majority of Nvidia's libraries make use of tessellation and AMD's hardware is notoriously bad at tessellation levels above 16x.

And before you or someone else posts about the levels of tessellation being unnecessary:

http://imgur.com/a/VorPz
http://abload.de/img/amdreducedtessellatioclrrn.jpg - AMD "Optimized" Tessellation
http://abload.de/img/fulltessellation2jrki.jpg - Default Tessellation

x64 tessellation is also necessary in most forms of volumetric lighting.
Another example is GodRays where forcing lower levels of tessellation on AMD results in artifacting around dense regions of geometry.

Pictures of HairWorks are not a good example of 'over tessellation' because the high level of tessellation is to ensure simulation accuracy.

It AMD wants to do well in GameWorks games it needs to figure out how to turn this around:
https://images.anandtech.com/graphs/graph11717/90104.png

Force AMD's hardware into a lower tessellation level completely eliminates the gap in performance. This idea of GameWorks artificially gimping AMD's performance because it's "a black box" is nonsense and always has been nonsense.. even when they didn't have the source. Nvidia had no trouble issuing a driver update for TressFX in the first Tomb Raider fixing it's performance before the source for TressFX was released. In fact the majority of the time neither Nvidia/AMD ever get a games source and yet they are both capable of optimizing for binaries and do so regularly.

#5524059 Posted on: 02/28/2018 06:07 PM

---

That's what I get for being in a hurry, so this is based on the demo build then alright.
Should cover almost everything then aside from perhaps VXAO and be pretty much on par with the final game aside from perhaps some tweaks and the later regions possibly having a more pronounced effect on the GPU or CPU load of things depending on if it's a arid area, forest or a town.

Good read, 580 is still keeping up on AMD's side though the 1080Ti is in a class of it's own as expected leaving the Vega behind by quite a margin, going to be interesting to see what new cards NVIDIA has planned if there's a consumer variant for 2018 and if there's any availability at all ha ha. (And price but it is what it is.)


Far Cry 5 as the next game performance test I guess, AMD managing to hold onto it this time unlike Far Cry 4 or Watch_Dogs 2 as a Vega GPU title and some tech.

And the full version of FF 15 on March 6th but I guess this initial area won't differ too much compared to the demo build here, game really loves VRAM if you let it have access to it from what I've read ha ha.

EDIT: Well I suppose AMD might be able to get something more out of the game from newer drivers, NVIDIA looks like they're already well optimized by this point but perhaps there's advantages to be had here too, multi-GPU should be possible by some method if not already supported for example.

And unless the entire AMD GPU lineup is under-performing Vega looks to be in a class above the 580 so it's not held back that much although it can't match the Pascal architecture on NVIDIA's side but that's been the case in most benchmarks so that's nothing new.
(Vega 64 isn't too far behind the 1080 though, not close enough to match it but not too behind to be completely outclassed either though that and then the Ti just dominates in most current games.)


Interesting to see, I mean this is a GameWorks game but all the effects can be disabled either in-game or via the launcher leaving mostly the driver side of things and what the game itself is geared towards. (Geometry or shader work or a bit of both.)
(Amazing how well Pascal held up really, will be fun to see what the next reveal will be and then Navi or what's next from AMD.)

EDIT: And I guess Vega and Polaris aren't too bad if they ever return to their regular price class (As if!) and well if there's any actual availability for these GPU's at all ha ha.
(Even NVIDIA is starting to feel the demand a bit now, and the pricing increase so yeah, fun times ahead but that's a separate problem entirely and already a well discussed one.)

#5524061 Posted on: 02/28/2018 06:13 PM

---

Should get better. Initially, the benchmark ran like complete crap for me, not a stuttering issue.
With the newest driver, I gained nearly 1000 points 1080p standard quality.
I just hope this isnt one of those gameworks games where AMD is locked out of the game engine.
Plus I wonder if a 100mb read/write mechanical drive is enough for this game. I have 4 SSDs, 1 for OS, 240GB, 480GB, and a 512 GB. The 3 game drive are full, so maybe I will clear 100GB off the 240GB. Its got 57GB free.
I had benchmark on fast SSD and demo on normal HDD and it ran the same, slight streaming stutter time to time when running across the map.

#5524063 Posted on: 02/28/2018 06:21 PM

---

Hilbert, theres some missing info in the chart on p8. I presume the processor core scaling chart colored bars are for resolutions. Its not mentioned so can be confusing.

#5524065 Posted on: 02/28/2018 06:30 PM

---

Imo the 980 Ti does 'weird' in the results. Only slightly above 980 and notably behind 1070.

Either the game is better optimized for Pascal (if that's even possible) or highest preset would need more VRAM maybe?

Note how 8 GB cards have 6+ GB usage even at FullHD.
If that was simply due to filling available VRAM then the usage should be closer to 8 GB perhaps.

1070 and above have 8 GB VRAM.

#5524068 Posted on: 02/28/2018 06:41 PM

---

Legend added, thanks for reporting.

#5524071 Posted on: 02/28/2018 06:43 PM

---

To say that Adrenalin 18.2.3 officially supports FFXV is an outright fallacy. AMD shall release an optimized driver upon the game's official release, rendering this article redundant.

#5524076 Posted on: 02/28/2018 06:57 PM

---

If they do, we'll update the article with the new driver set and see if it makes a difference. Would be good if they do.

#5524097 Posted on: 02/28/2018 07:45 PM

---

It did the same for me. I tried it on 960 EVO, 512 GB Sandisk, and 3TB 115mb read.