Benchmark vs true in game performance

[SonOfCrusader76]

OK, so just now I've been doing some benchmarks.

Current game settings average at a lovely 50fps.
DX 11 high - a perfectly fine 35fps.

The thing is, no way in hell am I getting these average fps in game - what's going on?

And for what it's worth, CPU DX9 is 11fps on average.

I don't get it - I thought a benchmark was supposed to be indicative of in game performance?

[Crazyeyesreaper]

it is lol when no melee is happening,

CPU bench = minimum FPS you should get if your GPU is decent, aka CPU is bottleneck

if minimum FPS is lower than your CPU bench FPS then the GPU is the bottleneck.

[talljoe]

OK, so just now I've been doing some benchmarks.

Current game settings average at a lovely 50fps.
DX 11 high - a perfectly fine 35fps.

The thing is, no way in hell am I getting these average fps in game - what's going on?

And for what it's worth, CPU DX9 is 11fps on average.

I don't get it - I thought a benchmark was supposed to be indicative of in game performance?

This game is insanely CPU bottlenecked. There's a benchmark chart out there (it has appeared in this forum) from one of the review sites that tested CPU performance in S2. With ultra unit detail and ultra shadows most cpu's will dip into the teens or low 20s for min FPS (while zoomed in).

I think an OC'd 3770k managed to break 30fps, but most others fell far short. I know with my i7 950 at 3.7ghz that I dip into the low 20s when zooming in on a big melee fight. (While my gtx 680 is sitting there doing 60%...)

[SonOfCrusader76]

it is lol when no melee is happening,

CPU bench = minimum FPS you should get if your GPU is decent, aka CPU is bottleneck

if minimum FPS is lower than your CPU bench FPS then the GPU is the bottleneck.

OK, thanks for that. Definitely a CPU bottleneck then - or more specifically - a CPU that's not having all its cores utilised.

[Crazyeyesreaper]

yea pretty much.

[A Barbarian]

The only thing I want to hear about Rome II is that it has proper multi core support. Almost everything else like units, AI, sound, etc. can be modded in, but multi core support has to be planned- and implemented as low as possible in the game engine itself. That's there CA failed miserably so far.

Also, there are no CPUs on the horizon which would magically double- or triple the current single core performance.

[irishron]

The only thing I want to hear about Rome II is that it has proper multi core support. Almost everything else like units, AI, sound, etc. can be modded in, but multi core support has to be planned- and implemented as low as possible in the game engine itself. That's there CA failed miserably so far.

Also, there are no CPUs on the horizon which would magically double- or triple the current single core performance.

I won't consider buying the game wothout it.

[SonOfCrusader76]

Should it really be this hard to optimise the engine? And if so, will 'heavily modifying' the engine enable CA to optimise it more easily?

The only time I get low FPS is when the camera is looking at a lot of units - and it gets really bad when lots are clumped together. Apart from AA/AF, what is the best setting to lower to improve the situation?

[talljoe]

Son,

General: AF isn't a taxing option and should be set at 16x. With your 1gb vram card you should run MLAA (or none) and probably no higher than medium textures. AA and textures use up alot of vram, and at ultra settings my S2 uses over 1.4gb of vram.

In terms of the huge FPS dips that happen when lots of units/zoomed in: it appears that 2 settings really affect the CPU and cause bottleneck: unit detail and shadows. Those two settings are the ones that rely on CPU and cause even todays fastest, most OC'd CPUs to dip into the low 20s and teens when zoomed in.

This is, of course, completely insane. In my book it makes Total War the most taxing game in history. There's that bit tech chart floating around showing the CPU test (at only DX9!) returning less than 20 fps on an OC'd 2600K. The french site also shows the CPU test in DX11... which has CPUs like the OC'd 3770k returning 18fps.

This is why some players, myself included, won't get Rome II unless something is done about this. I paid alot for a fast gaming rig so I could have a smooth and beautiful total war gaming experience. Silly me! I mean, of course I should have known that my i7 950 at 3.7ghz and GTX680 2GB would net me 19fps mins while big battles are happening!

[SonOfCrusader76]

Thanks, Joe - I'll do some tweaking.

[Crazyeyesreaper]

minimum i see is 34 in the CPU bench these days, the DX11 and DX9 dont do a damn thing for the CPU bench or how well it runs however it does improve FPS in game for the most part if running similar settings across the the shader models do to better optmized shader code but I digress soft shadows hits the CPU hard but for the most part you wont see huge dips untill in battle do to the 1 on 1 fighting and the animation system both using 1 core and both getting lumped onto a single core.

the big benefit for Shogun 2 is IPC or instructions per clock in that situation Sandybridge and Ivy Bridge excel compared to previous CPUs

example say the Core 2 Quad is 100% i7 Nehalem is 115% Sandybridge is 130% Ivy Bridge is 145% in terms of Shogun 2 these numbers are being pulled out of my ass but they are for reference, each generation Intel has improved on IPC and thus at the same clock speeds performance improves from 1 generation to the next,

Where as for AMD users have noticed its the opposite

Phenom to Phenom II showed IPC improvements but mostly L3 cache and clock speed made the difference,
Phenom II to Bulldozer IPC performance dropped thus Shogun 2 didnt improve much at all any benefit seen is clock speed nothing more.

overall its rather sad situation for Shogun 2 and Creative Assembly but then again they are the ONLY company pushing these kinds of games with large troop numbers youll notice most games use a single sprite or a unit composed of a block of soldiers instead in that move in formation etc or they just plain have far less numbers. Theirs a reason for this its not easy to code for or optimize but considering how long CA has been doing this i would expect far better work out of them. considering i can manage the same size battles across all their game engines for the most part.

[A Barbarian]

the DX11 and DX9 dont do a damn thing for the CPU bench or how well it runs

You are right, DX11 by itself isn't slower. When using the same options it's usually faster than DX9, although some older cards might render the same options faster in DX9. It's the different options like shadow detail which can hit the CPU hard.

By default the CPU benchmark uses minimal graphical settings. But you can change the benchmark settings configuration file to use any other graphical option and also to switch on DX11. Then you can check out which graphical feature also taxes the CPU.

minimum i see is 34 in the CPU bench these days,

You are talking about average I guess, I find the minimum fps more interesting. It's listed in the benchmark log file and is usually about 33% lower than the average.

[SonOfCrusader76]

The annoying thing about playing in DX9 is the blood - it looks too bright and cartoony. Looks fine in DX11 though - but I'm guessing it might be worth a try lowering to DX9 just to see if it makes any difference. Then again, my card's not exactly 'old', is it? I hope not - I got it recently!

[A Barbarian]

@SonOfCrusader76
If you look at a way to get around the CPU bottleneck situation of battle closeups then DX9 will probably not help you much since it's much more GPU related.

[SonOfCrusader76]

@SonOfCrusader76
If you look at a way to get around the CPU bottleneck situation of battle closeups then DX9 will probably not help you much since it's much more GPU related.

OK, thanks for saving me some time. I reckon I might knock everything down to 'low' and see what that does. Then increase settings bit by bit until a good compromise is reached, because right now when the units all clump together, it can drop as low as 9fps!

In fact, past experience tells me that if I knock everything down to low besides the textures, then it won't look too bad. Let's go find out.

Thanks for your help, gents.

Edit: One other thing with this in mind: CA, please allow us to set different graphics options for the map and battlefield! Just because the battlefield runs like crap doesn't mean we should have to put up with an ugly map as well.

[Crazyeyesreaper]

You are right, DX11 by itself isn't slower. When using the same options it's usually faster than DX9, although some older cards might render the same options faster in DX9. It's the different options like shadow detail which can hit the CPU hard.

By default the CPU benchmark uses minimal graphical settings. But you can change the benchmark settings configuration file to use any other graphical option and also to switch on DX11. Then you can check out which graphical feature also taxes the CPU.


You are talking about average I guess, I find the minimum fps more interesting. It's listed in the benchmark log file and is usually about 33% lower than the average.

no my MINIMUM is 30-36 these days depending on how lucky i am when i run the bench if i am operating at full overclock or not etc

but on average of runs my minimum is 33, avg fps in the cpu benchmark is 38

so minimum 33 avg 38 is what i see 95% of the time seen it go as low as 30 and as high as 36 on the minimum, avg ive seen jump around from 36-40 depending on the run.

[A Barbarian]

ah, i thought you were talking in general, not only about your 3770K CPU

[Crazyeyesreaper]

yea my 2500k couldnt push like the 3770k but to gain that performance edge i have to deal with a CPU that hits 80+ degrees C compared to 2500k at 55 C but yea its possible to max Shogun 2 melee battles it just takes an Ivy Bridge system heavily overclocked.

but as for the CPU bench,

Resolution and Direct X have 0 impact by editing these settings all people are doing is shifting more load over to the GPU, instead of testing the CPU which is well what the CPU bench is for lol,

[A Barbarian]

turning on ultra shadows in the CPU benchmark configuration does add to the CPU workload as well, not just the GPU.


Here is a test i did a while ago about the CPU load for shadow rendering:

No Shadows: 30fps min, Ultra Shadows 20fps min. The fps are minimum values I encountered during the test scene.
Settings: All ultra (except shadows) + MLAA + 1680x1050 with a 7970@1000Mhz and a 2500k@4.2Ghz.

30fps no shadows

Spoiler Alert, click show to read: 

20fps ultra shadows

Spoiler Alert, click show to read: 

64% GPU usage during the 30 fps fight scene (CPU core 1 was 100% at the same time):



In comparison no fight: 75fps ultra shadows

Spoiler Alert, click show to read: 

[SonOfCrusader76]

Right - after some tweaking, I've got the game running at a good FPS at all times - but now there's another problem! The damn bodies are disappearing - and one of the very best things about TW battles is surveying the scene of sheer carnage after the battle has been going on a while. Which setting governs the bodies (and arrows) disappearing?

It seems the primary culprit of the low FPS before was the unit details - I suppose in order to get the bodies back, I have to turn that back up as well? Hence back to low FPS. Catch 22 in other words?

I'm hoping that when CA say they're going to try and keep Rome 2 close to Shogun 2's requirements, they mean they're going to optimise the damn CPU performance properly!