TWC Total War Warhammer benchmark thread - last updated: 26.12.2016

[A Barbarian]

Unfortunately Afterburner does not work with DX12 as there is no overlay and also the F12 shortcut does not work then for screenshots. What is the tool you use for monitoring and screenshots?

I haven't tested under Win10+DX12 yet so I still use Afterburner.
OBS apparently works with DX12 but I don't know much about it's features:
https://obsproject.com/

[yutterh]

Hahahah this bench mark is garbage. Like ridiculous garbage. The freakin cavalry run back into the empire unite and are facing the wrong way and a messed up formation. The battle lasts loke 3 seconds hahahaha I don't understand how it is so garbage hahaha

[alQamar]

Hahahah this bench mark is garbage. Like ridiculous garbage. The freakin cavalry run back into the empire unite and are facing the wrong way and a messed up formation. The battle lasts loke 3 seconds hahahaha I don't understand how it is so garbage hahaha

Dude you should have a look at R2 benchmark there are many flaws with unit collision wrong pathfinding clipping lod popping but anyway it's all we got served. I agree the Attila scene had some benefits over Warhammer.

Gesendet von meinem SM-G900F mit Tapatalk

[yutterh]

Rome 2 was far superior as a tool and the clipping and parh finding was not ridiculous to me. Trust me I ran that bench mark a lot.

[alQamar]

Rome 2 was far superior as a tool and the clipping and parh finding was not ridiculous to me. Trust me I ran that bench mark a lot.

If you insist I could upload a video on YouTube again highlighting all of these. Please trust me I've spend more than 60 hours in this tool for my extensive works here.
Most of the issues got introduced with patch 4.

---
Looking forward to it Lestat

If you all agree we should post the following:

Screenshots of all tests maxed out through low preset in dx11 and dx12 (if possible) + cpuz log

Any further ideas?

Gesendet von meinem SM-G900F mit Tapatalk

[yutterh]

If you insist I could upload a video on YouTube again highlighting all of these. Please trust me I've spend more than 60 hours in this tool for my extensive works here.
Most of the issues got introduced with patch 4.

---
Looking forward to it Lestat

If you all agree we should post the following:

Screenshots of all tests maxed out through low preset in dx11 and dx12 (if possible) + cpuz log

Any further ideas?

Gesendet von meinem SM-G900F mit Tapatalk

Sorry, I meant I have seen it a lot and the overall bugs of clipping and pathfinding don't really take away from the whole point of the tool. Which is too benchmark the different settings of the game. Rome plays for a extended amount of time, has a close up vegetation, shows a lot of units marching and quite a bit of actions. As well as various shadows and fx. This just goes so quickly with barely anything happen. I was just laughing at how obvious all the flaws were. Im sure the cavalry was suppord to do a counter charge and not fall back behind their own lines. Which is why the screen follows then dor. A little while. Just saying this benchmark is quite inferior to any benchmark we had before it, by a mile.

[LestaT]

Benchmark tool should be in the graphics options like in the past

Gesendet von meinem SM-G900F mit Tapatalk

Yeah. Found it. Have to disable back the intel beta thing and verify game cache.

Initial test base on my usual setting is 25fps. Will do proper benchmark tonight.

[SPARTAN VI]

DX12 performance was a massive dissappointment with my GTX 970. I'm guessing they're using async compute? On the campaign map alone, first 20 turns in with the Vampire Counts, my frame is normally in the 80s under DX11, but dropped to the 60s in DX12 @ 1080P and everything maxed except screen space reflections = off and AA = MLAA.

[alQamar]

So my guessing about a repetition the Tomb Raider disaster has come true. For the same reason I hesitate to clearly recommend the 1070 and 1080 they seem not to be future compatible otherwise as Barb said next to code, also driver optimization is another issue

So again should we bother to call for benchmarks yet or wait or completely give up

Gesendet von meinem SM-G900F mit Tapatalk

[A Barbarian]

If optimized correctly, the DX12 version should have the same speed as the DX11 version, even without Asynch Compute. Because DX11 has no Asynch Compute and it is also just an optional component in DX12 (u can execute some tasks in parallel).

Therefore, the conclusion has to be that the current combination of CA's DX12 render path for NVIDIA cards and NVIDIAs DX12 driver is not as optimized as the DX11 versions.

U have to keep in mind that in DX12 the engine developer have to implement more mechanisms themselves which were automated in DX11 before (e.g. texture allocation). The ofc, allows for more flexibility for the engine developer and if done right also for a more efficent render engine implementation but it also leaves less room for driver optimizations.

Therefore it's not really a surprise that it takes a little longer for developers to surpass the performance level which was achieved before by the highly optimized DX11 NVIDIA drivers. AMD DX11 drivers on the other hand always lagged behind the NVIDIA counterparts in terms of performance which is why one reason for some large gains on the AMD side is simply that their much lower level of optimization was far easier to match or exceed

[alQamar]

What I don't get is that the new on die VRAM with amazing bandwidth on paper doesn't help AMD to beat Nvidias GDDR5 models with a narrow VRAM bandwidth of 256 bit or less

Gesendet von meinem SM-G900F mit Tapatalk

[A Barbarian]

VRAM bandwidth, while not unimportant usually doesn't have such a big impact on performance. Only when u go really high on resolution, 4K or higher, or high MSAA it starts to impact performance more. NVIDIA also seems to have the better texture compression. The move from AMD to use 4GB of HBM while commendable might not have been the greatest idea, because HBM 1 is limited to 4GB VRAM and at resolutions when VRAM speed would be more important, the cards start to run out of memory.

Btw, the 1080 is using the new GDDR5X which is slower than HBM from the Fury but faster than GDDR5.

Said that, NVIDIA was pretty clever to limit the 1080 to a 256 bit memory bus because that's what's probably holding it back being a really good 4K GPU. When the 1080TI appears it will certainly most likely have a 384 bit memory interface being the first 'real 4K GPU'.

[alQamar]

I think so. They preserve selling points very good. Both of them.

Gesendet von meinem SM-G900F mit Tapatalk

[A Barbarian]

Here is a very good review of the DX12 benchmark. It's finally a review which shows exactly that the CPU bottleneck still exists. In their extreme test scene they get 11-16 fps on their i7-4930K CPU (GPU doesn't really matter since u can see that it is only utilized by 32% during that scene):

http://www.dsogaming.com/news/report...dias-hardware/

This is the kind of scene I wish would be part of the benchmark. Or at least a somewhat similar scene:

Spoiler Alert, click show to read: 

[alQamar]

I've updated the thread with my own benchmarks now, refined the rules topic, and also included benchmark of a high performance gaming PC that I've built and benchmarked for TWC member revan.be.

[SPARTAN VI]

Intel Core i7 6700, 4.0 GHz, ASRock Fatal1ty Z170 Gaming-ITX/AC, 16GB DDR4-3200 MHz CL 16-16-16-36@1.35V Dual Channel, MSI GTX 1070 Gaming, Windows 10 Home 64-bit (10586.545), resolution 1920x1080, DX11, no SLI

Max Settings:
min FPS: 50
max FPS: 59
avg FPS: 54.7

Ultra preset:
min FPS: 97
max FPS: 109
avg FPS: 104.5

High preset:
min FPS: 107
max FPS: 151
avg FPS: 141.8

Medium preset:
min FPS: 97
max FPS: 166
avg FPS: 158.5

Low Preset:
min FPS: 177
max FPS: 199
avg FPS: 189.9

Intel Core i7 6700, 4.0 GHz, ASRock Fatal1ty Z170 Gaming-ITX/AC, 16GB DDR4-3200 MHz CL 16-16-16-36@1.35V Dual Channel, MSI GTX 1070 Gaming, Windows 10 Home 64-bit (10586.545), resolution 1920x1080, DX12, no SLI

Max Settings:
min FPS: 44
max FPS: 52
avg FPS: 48.2

Ultra preset:
min FPS: 81
max FPS: 103
avg FPS: 94.9

High preset:
min FPS: 90
max FPS: 135
avg FPS: 106.6

Medium preset:
min FPS: 114
max FPS: 151
avg FPS: 127.6

Low Preset:
min FPS: 151
max FPS: 194
avg FPS: 185.1

CPUZ_SPARTANVI.txt

I'll add my results from my old i5 3570K + GTX 970 machine when I have time.

[SPARTAN VI]

Intel Core i5 3570K 4.4 GHz, ASRock Z77 Extreme 4, 16GB DDR3-1600MHz CL 11-11-11-30@1.35V Dual Channel, Asus GTX 970 STRIX, Windows 10 Home 64-bit (14393.105), resolution 1920x1080, DX11, no SLI

Max Settings:
min FPS: 32
max FPS: 42
avg FPS: 36.4

Ultra preset:
min FPS: 63
max FPS: 77
avg FPS: 72.2

High preset:
min FPS: 68
max FPS: 105
avg FPS: 91.8

Medium preset:
min FPS: 99
max FPS: 123
avg FPS: 112.1

Low Preset:
min FPS: 42
max FPS: 145
avg FPS: 134.0

CPUZ_SPARTANVI_i5.txt

[alQamar]

Many thanks for your efforts! I will incorporate them.

Gesendet von meinem SM-G900F mit Tapatalk

[juanjux]

Hi,


Thanks all for the benchmark data, it's very useful. I've a request for the ones of you with a GTX1080, could you use Nvidia's DSR to render the game at 4K and post the results? I'm interested in switching to a 1080 and a 4K TV soon (currently playing with a 970 at 1080) and I would like to know the kind of performance that I can expect.

[alQamar]

Hi everyone I plan to update the results section and the drivers section, that has seen several updates across the board. Just in case anyone has already done benchmarks but not posted them, let me know.

Gesendet von meinem SM-G900F mit Tapatalk