Jump to content

Cpu Or Gpu?


109 replies to this topic

#81 Fire and Salt

    Member

  • PipPipPipPipPipPipPip
  • 526 posts
  • LocationFlorida

Posted 17 October 2014 - 05:17 PM

The way people talk about this game, I was certain that one core of my processor would be at 100%, and three would be sitting there slacking off...



But:
Posted Image

Also note MSI afterburner keeping tabs on the EVGA superclocked ACX2.0 GTX980.
This was mid game (task manager and MSI were on monitor #2)
I was getting 65 fps.

The spikes in GPU usage were from heavy particle effects... getting slammed by LRMs.




The question is... if my GPU was at less than 50%, and my CPU was about 50%.... why wasn't I getting 120 FPS instead of 65?



FWIW, I get 250FPS-180FPS in testing grounds, and in game I get 120FPS-60FPS, except for maybe very short drops below that, which is irrelevant, because if I am getting less than 60 FPS, there is so much smoke on the screen, I can't see anyways.





Note: 60FPS is fine, since my monitor is only 60hz anyways (for now)
I'm just curious - where is the bottleneck?




Settings:
Motion Blur - OFF
All other settings: Ultra High
Post AA
1920x1080
V-Sync- OFF




Some other things that have me curious:
My processor running at 3.85 GHZ. I guess turbo is engaged all the time... probably because I have a liquid cooler....

Also, why is my core clock for my GPU at 1430mhz?
The specs say is turbo speed is 1367. (with the factory overclock)
I did not overclock it (at least not intentionally) and MSI afterburner is showing the clock offset is 0mhz from stock. http://www.evga.com/...=04g-p4-2983-kr

The memory clock rate for the GPU is almost exactly the spec. I am at 7012 effective and the EVGA card is specified at 7010.

Edited by Fire and Salt, 17 October 2014 - 05:26 PM.


#82 Flapdrol

    Member

  • PipPipPipPipPipPipPipPip
  • The 1 Percent
  • The 1 Percent
  • 1,986 posts

Posted 18 October 2014 - 05:49 AM

View PostFire and Salt, on 17 October 2014 - 05:17 PM, said:

The way people talk about this game, I was certain that one core of my processor would be at 100%, and three would be sitting there slacking off...

The question is... if my GPU was at less than 50%, and my CPU was about 50%.... why wasn't I getting 120 FPS instead of 65?

Note: 60FPS is fine, since my monitor is only 60hz anyways (for now)
I'm just curious - where is the bottleneck?

Some other things that have me curious:
My processor running at 3.85 GHZ. I guess turbo is engaged all the time... probably because I have a liquid cooler....

Also, why is my core clock for my GPU at 1430mhz?
The specs say is turbo speed is 1367.

The reason you don't see 1 core on max is because it's continually being transferred from one core to the other, too fast to see in taskmanager. Your cpu is absolutely the bottleneck, the reason you don't see one core near max is because the thermal management hides it.

To have the cpu turbo up near max is not surprising, it should always do this if the temperature and powerconsumption allows it. Mwo+taskmanager+afterburner only loads your cpu to 60%, so clockspeed can stay high, temps won't be a problem either.

The gpu does the same, as long as powerconsumption and temperature allows it it'll up the voltage and the clocks, to get better performance. Some games will not stress a particular part of the gpu fully, so the power saved there is converted into higher performance, in some games the turbo will be higher than others. When it comes to mwo you're cpu limited, so the gpu clocks can go all the way up, doesn't get you higher fps, but it'll have the image on screen a bit more quickly.

#83 ninjitsu

    Member

  • PipPipPipPipPipPip
  • FP Veteran - Beta 2
  • FP Veteran - Beta 2
  • 402 posts

Posted 18 October 2014 - 10:00 AM

You've got a lot of headroom to OC that chip. I can get my 4670k up to 4.5 on air easy. Should be able to go a little farther with a 4690k

#84 Sprouticus

    Member

  • PipPipPipPipPipPipPipPipPip
  • Bridesmaid
  • 2,781 posts
  • LocationChicago, Il, USA

Posted 18 October 2014 - 12:06 PM

So I got the 8370, got installed (default cooling, no OC).

I am now 100% GPU bound. Which is awesome. I am getting about 15 fps more 30-55, with some 20-30 hits in specific situations (smoke on Frozen seems to be the worst) but it is rare. very high settings for all except particle (low) and shadows(low). No AA


Thanks for all your help guys, if you really want to know what the max for a 7770 is, I think I may have a good example machine :)

Edited by Sprouticus, 18 October 2014 - 12:07 PM.


#85 Goose

    Member

  • PipPipPipPipPipPipPipPipPip
  • Civil Servant
  • Civil Servant
  • 3,463 posts
  • Twitch: Link
  • LocationThat flattop, up the well, overhead

Posted 18 October 2014 - 01:35 PM

View PostFlapdrol, on 18 October 2014 - 05:49 AM, said:

The reason you don't see 1 core on max is because it's continually being transferred from one core to the other, too fast to see in taskmanager.

Not heard that one before; In fact, I thought Cryengine placed it's twelve named threads specifically to avoid the OS/ firmware/ whats'it from doing just that …

It would be nice if Teh Devs would open up some of the diagnostic commands, wouldn't it? Posted Image

#86 ninjitsu

    Member

  • PipPipPipPipPipPip
  • FP Veteran - Beta 2
  • FP Veteran - Beta 2
  • 402 posts

Posted 18 October 2014 - 02:14 PM

The game doesn't hit my cores evenly. Core #4 takes the brunt of it on my machine Posted Image

#87 Flapdrol

    Member

  • PipPipPipPipPipPipPipPip
  • The 1 Percent
  • The 1 Percent
  • 1,986 posts

Posted 18 October 2014 - 02:34 PM

I don't know what makes windows do or not do it. But if I run prime95, and only run a single worker it'll show 50% on each of my 2 cores.

#88 Goose

    Member

  • PipPipPipPipPipPipPipPipPip
  • Civil Servant
  • Civil Servant
  • 3,463 posts
  • Twitch: Link
  • LocationThat flattop, up the well, overhead

Posted 18 October 2014 - 02:40 PM

Posted Image There's a test to run …

#89 Pet Dude

    Member

  • PipPipPip
  • The Howl
  • The Howl
  • 90 posts
  • LocationUS West

Posted 18 October 2014 - 02:52 PM

I wonder if network performance has anything to do with. I always wondered how much data you are recieving and sending could cause invalid metrics on CPU load and GPU load. Plus I know I get lag when I max out my bandwidth. Is the testing grounds tied to an internet connection?

#90 Goose

    Member

  • PipPipPipPipPipPipPipPipPip
  • Civil Servant
  • Civil Servant
  • 3,463 posts
  • Twitch: Link
  • LocationThat flattop, up the well, overhead

Posted 18 October 2014 - 03:13 PM

Yeah: I remember when I thought "talking to the sever" was a bottleneck, too …

You might want to spend a few hours with DNSBench, after a quick visit with TCPOptimizer …

#91 Pet Dude

    Member

  • PipPipPip
  • The Howl
  • The Howl
  • 90 posts
  • LocationUS West

Posted 19 October 2014 - 02:39 PM

Well I just got my FX-9370 and R9 290 Sapphire Tri-X running so lets see how it is.

BTW if you looking for a new GPU MSI Gamer R9 290 is $250 at newegg. You are not beating that price for performance ever. Comes with 3 free games and a free gaming headset. That value is insane. Like $450 for just $250.

Edited by Pet Dude, 19 October 2014 - 02:50 PM.


#92 Fire and Salt

    Member

  • PipPipPipPipPipPipPip
  • 526 posts
  • LocationFlorida

Posted 20 October 2014 - 09:47 AM

Yea I know that the GPU and CPU can both go into turbo if the temps are low enough etc...
I just find it odd that my CPU goes to exactly 3.85 GHz as specified, while the GPU goes a little over... Not much in terms of % anyways so whatever...


I bought the CPU with the intention of overclocking it...

Will be interesting to see if I get a 20% FPS boost from a 20% overclock. My GPU certainly doesn't seem to be a bottleneck...
Not really in a rush to do so, because my monitor is only 60hz anyways...




So here's a question:
If I get a monitor with say.... 2.25x-4x the amount of pixels... Will my framerate stay basically the same? If I'm CPU bound, adding more pixels to draw shouldn't make a difference, right?

(I'm gonna buy a higher resolution screen eventually if it slows down too much I'll just get rid of AA. Matters less with smaller pixels.)




#93 Peter2k

    Member

  • PipPipPipPipPipPipPipPipPip
  • Elite Founder
  • Elite Founder
  • 2,032 posts
  • LocationGermany

Posted 20 October 2014 - 10:03 AM

View PostFire and Salt, on 20 October 2014 - 09:47 AM, said:

Yea I know that the GPU and CPU can both go into turbo if the temps are low enough etc...
I just find it odd that my CPU goes to exactly 3.85 GHz as specified, while the GPU goes a little over... Not much in terms of % anyways so whatever...


I bought the CPU with the intention of overclocking it...

Will be interesting to see if I get a 20% FPS boost from a 20% overclock. My GPU certainly doesn't seem to be a bottleneck...
Not really in a rush to do so, because my monitor is only 60hz anyways...




So here's a question:
If I get a monitor with say.... 2.25x-4x the amount of pixels... Will my framerate stay basically the same? If I'm CPU bound, adding more pixels to draw shouldn't make a difference, right?

(I'm gonna buy a higher resolution screen eventually if it slows down too much I'll just get rid of AA. Matters less with smaller pixels.)




With a gtx980? More pixels? You're fps are gonna stay the same I'm sure
Secondly I use evga's pixel overdrive to go from 60Hz to 75 (don't want to push my LCD further, but the difference is nice)
Thirdly
This game is heavily cpu bound, most graphics cards would be bored out of they're mind.
Though I do have to say that something must be limiting this game when it comes to actual online play, because if you deactivate vsync and head into a training map, u can see the fps skyrocket. With the same amount of detail going on.
Just no network stuff going on.

Will be interesting to see how Star Citizen can handle all the players.
Same engine.promised mantle support, should reduce things for CPU, will see when and if it makes a difference.

Edited by Peter2k, 20 October 2014 - 10:04 AM.


#94 Pet Dude

    Member

  • PipPipPip
  • The Howl
  • The Howl
  • 90 posts
  • LocationUS West

Posted 20 October 2014 - 10:41 AM

I think across the board the game itself would be considered inefficient. Seems to not use resources up to their fullest potential. GPU, CPU, bandwidth. It leans much more towards CPU use though. I'm not sure how many people are running the game on anything more than 1080p on 1 monitor. I have the capabilities to run multi monitors and higher resolutions but that's a lot of desk space, bezels, and cost for monitors. At single monitor below 4k resolutions even an FX-8350 works fine. Still pulling an adequate K/D ratio.

#95 Goose

    Member

  • PipPipPipPipPipPipPipPipPip
  • Civil Servant
  • Civil Servant
  • 3,463 posts
  • Twitch: Link
  • LocationThat flattop, up the well, overhead

Posted 20 October 2014 - 10:48 AM

View PostPeter2k, on 20 October 2014 - 10:03 AM, said:

Though I do have to say that something must be limiting this game when it comes to actual online play, because if you deactivate vsync and head into a training map, u can see the fps skyrocket. With the same amount of detail going on. Just no network stuff going on.

That's eight other, non-firing, unmoving targets you are looking at, Hoss, instead of 23 belligerents …

#96 Peter2k

    Member

  • PipPipPipPipPipPipPipPipPip
  • Elite Founder
  • Elite Founder
  • 2,032 posts
  • LocationGermany

Posted 22 October 2014 - 06:43 AM

View PostGoose, on 20 October 2014 - 10:48 AM, said:

That's eight other, non-firing, unmoving targets you are looking at, Hoss, instead of 23 belligerents …


I heard that argument before ;)

that's not actually a challenge for a modern PC though, is it?
graphics aren't that good anyway, there isn't even some calculation going on for AI, or advanced physics like in other games

how do ALL the other games handle more than one moving target on screen at the same time, u know?

Maybe the game is just "waiting" for the other players; where they move, what they do, shoot, ...
that may be a good reason why my CPU is kinda underused when playing this game
hmm, if we had 24 players from US or Canada going in a private match, having pings below 50
would the fps be better for everyone?

sometimes I'd really like to know, cuz EVERY other games seems to perform better.

BUT
CryEngine isn't exactly used enough to make a judgment on such things, the only other game with a heavy focus on MP is Star Citizen?
and that doesn't let you play with the ships u bought yet, let alone against someone human.
Maybe (that's a big maybe) they promised Mantle support because they know of some problems? Be it optimization or Engine limits.
I think Paul already posted they went with this engine because its free, can have eye candy, and hoping for heavy support from CryTek at the time

but maybe there is a reason why this engine isn't used as much as Unreal engine for instance

on a side note

Windows 10 fixed my performance problem with Particles (like Steam from overheating)
got a new install of 7 as well, still performance drops with steam n smoke
and seems to run somehow "smoother" all together

#97 Catamount

    Member

  • PipPipPipPipPipPipPipPipPip
  • LIEUTENANT, JUNIOR GRADE
  • 3,305 posts
  • LocationBoone, NC

Posted 22 October 2014 - 07:32 AM

Point of interest, SC does let you fly the ships you own (well, some of them, more being added), and against other players in both racing and dogfighting. Don't know if coop of any sort is in yet; haven't been in in a couple weeks because of computer and school stuffs :/

I have noticed the Windows 10 performance improvement also. I actually have some benchmarks around showing higher minimums, though averages didn't budge much vs W7. Despite my best intentions of keeping something else around for important work, 10 has already become my primary OS for everything, even schoolwork. Oh why, PGI, did you go with CE3... Yeah, sure, you have to pay for Unreal 3, but you'd have saved that much money in man hours making the damn thing work! :P

Edited by Catamount, 22 October 2014 - 07:33 AM.


#98 Flapdrol

    Member

  • PipPipPipPipPipPipPipPip
  • The 1 Percent
  • The 1 Percent
  • 1,986 posts

Posted 22 October 2014 - 08:09 AM

If you're overclocking the performance scales pretty well with cpu clockspeed. I don't see any reason to believe network performance has anything to do with it, also, lag gets you a bunch of rubberbanding, but not fps drops.

#99 ninjitsu

    Member

  • PipPipPipPipPipPip
  • FP Veteran - Beta 2
  • FP Veteran - Beta 2
  • 402 posts

Posted 22 October 2014 - 08:12 AM

On a slightly off topic subject, not every game performs better than MWO. World of Tanks looks like crap and has a hard time properly managing hardware. I think it's a symptom of Free to Play games.

#100 RadioKies

    Member

  • PipPipPipPipPipPip
  • The 1 Percent
  • The 1 Percent
  • 419 posts
  • LocationThe Netherlands

Posted 22 October 2014 - 09:07 AM

Keep in mind MWO has nothing of optimisation. I use hope they'll start optimising the game after CW phase 2 (aka the next phase).

MWO is verry CPU sensitive. I resently upgraded my graka and nothing else. My specs:
Phenom II 940x4 (@3.0ghz)
6gb ram
Asus GTS250 DK 1gb fancypants edition
WD1001FALS black edition 1gb HDD
1680*1050 monitor

I upgraded the graka from the GTS250 DK 1gb to a MSI N560GTX-Ti twinfrozr II 1gb. The card performs about 130% up to 190% better compared to the GTS250 when playing games like Sleeping Dogs, World in Conflict and Risen II. With MWO I have 0 (zero!) performance boost. Ok, maybe 3 to 5 fps more sometimes, but nothing noticeable. Using DirectX11 or 9 doesn't do anything for me.

I'm playing the game on low settings, 1680*1050 @ a fps of ~32. Meanwhile Shadows of Mordor is running 60+ fps (with drops to 34fps when way too much grass,rain and orcs are on screen) with way higher settings:
Posted Image
Even though Alien Isolation is mostly corridors that game runs on max settings at 60fps. Cryengine sucks.

tl'dr:
Get a powerfull CPU to run the game better and maybe get memory that's compatible with the CPU speed, MHZ and low Cas timings. Maybe it's because how AMD has the memory controller in the CPU, but my CPU performance (with benchmarking) is quite dependent on the speeds my memory is running on.





1 user(s) are reading this topic

0 members, 1 guests, 0 anonymous users