67 replies to this topic

[FullMetalDad]

I sent off a support ticket and PGI got back to me promptly (thanks!). They said my AMD 8350 was indeed CPU bottlenecked (I provided the dxdiag.txt and a screen shot). I have the Black Vishera clocked at 4 GHz. It's unlocked so I can try overclocking, though I would have thought this processor was more than capable. I checked power settings (thanks siLve00) saw nothing out of the ordinary. I'll have to experiment this weekend and run some benchmarks.

[Nemesis Duck]

Not to be an ass but most of your statements on performance are meaningless without indicating your resolution. The resolution is how many pixels the GPU needs to process per frame, and as you can imagine, the performance goes down as the pixel count increases. There's a huge difference between 30fps @ 2560x1600 vs 30fps @ 1920x1080, most GPUs will cave at higher resolutions.

To communicate your hardware performance, you should list:

1. CPU: Brand, Model, Clockspeed
2. GPU: Brand, Model, Clockspeed
3. Video settings
4. Resolution

OP: Try overclocking the CPU to 4GHz. For me there was noticable difference going from 4GHz to 4.4GHz with my AMD 8350.

[Celtic Warrior]

Miner, on 24 October 2014 - 01:47 PM, said:

I sent off a support ticket and PGI got back to me promptly (thanks!). They said my AMD 8350 was indeed CPU bottlenecked (I provided the dxdiag.txt and a screen shot). I have the Black Vishera clocked at 4 GHz. It's unlocked so I can try overclocking, though I would have thought this processor was more than capable. I checked power settings (thanks siLve00) saw nothing out of the ordinary. I'll have to experiment this weekend and run some benchmarks.

That's really odd because we have the same chip and my systems flies through MWO. Have you unlocked all your cores?

[Kaptain]

MandyB, on 21 October 2014 - 05:47 PM, said:

Wow im running a Q6600 cpu (old i know) and a gtx 580 at max graphics and usually get 60+ fps, ive got links to my twitch stream if anyone wants proof. I'll check back again on this thread im interested to see what could be causing those issues you guys are having.

Overclocked? Resolution? I'm running a QX9650 @ 4.0 and I can't do 60+ and thats with a GTX770...

[JimboFBX]

As an experiment I messed with my settings:

sys_limit_phys_thread_count = 0
sys_main_CPU = 0
sys_physics_CPU = 1
sys_TaskThread0_CPU = 2
sys_TaskThread1_CPU = 3
sys_TaskThread2_CPU = 2
sys_TaskThread3_CPU = 0
sys_TaskThread4_CPU = 3
sys_TaskThread5_CPU = 1
sys_job_system_enable = 1
sys_job_system_max_worker = 0
sys_MaxFPS = 60
p_num_threads = 4
ca_thread = 1
ca_thread0Affinity = 2
ca_thread1Affinity = 3

This balanced out the CPU load pretty well according to task manager. Of course, total usage was still about 45 - 60% and no core maxes out. Yet, the frame rate generally stayed about the same, although it might had been a little better. This doesn't imply it's not CPU bound, as a single thread can jump from core to core but have it's max speed restricted to the speed of a single core.

The fact that testing ground with 12 mechs gives you 60 fps and a game with 12 mechs, even without action, gives you 25 fps at times seems to lean towards the multiplayer logic. I paid attention to the framerate and it seems like it doesn't get better until there's only 1 mech on a team left.

Unfortunately, some of the settings are determined by the server so I couldn't adjust several things that would affect CPU consumption to see if they may be a culprit.

Edited by JimboFBX, 25 October 2014 - 12:52 AM.

[Martin Oberhofer]

Its quiet simple

in most Multiplayer games you are forced to have the workload on one thread, you can outsource some parte of the calculation in a secondary on, but not everything.
That means CPU bottleneck is happening when the "Main Tread" occupies 1 Core, maybe there are 3-16 other threads doing side stuff (sound effects as an example) but the game is limited by the Main one. So it would be easy to split this one in two.
Well not so fast, you can split a thread if the 2 half is indenpendet from the first half. If you need the result of calculation A to start calculationB you cannot split them so easy (I am simplifying things here I know but its for an easier understanding).

For that reason Multicore CPUs are not that great for games as for other workloads. If you apply a filter to a image, you can split that image by the number of cores avaliable and each one can run fulltime -> here is where multicores work.

Imagine MWO being one "Main-Thread" and 3 independent Side-Threads
Lets say the Main-Thread needs 5GHZ to produce constant 100fps
each side thread 1.5GHZ for constant 100fps

If you would have a 5GHZ Dualcore, you would be fine and able to enjoy constant 100fps

If you now "upgrade" to a 3GHZ Quadcore, your fps would drop to 60fps even if you are now having 20% more raw calculation power.


And to make things worse for the AMD users, thier CPUs has not the IPC(instruction per cylce) as thier Intel Counterparts.
Meaning 2,4GHZ on a Core i3 are roughly the same as 2,8GHZ on a old Phenom
and even worse - almost 3GHZ on a Bulldozer which is more powerfull in a multicore enviroment (picture example of above)
but on the Single-Thread front it looses even out to it precessor.

For sure the code can be optimized to distribute the workload better - but still a MP code would never benefit more from a higer core count than from simple single thread performance.

[FullMetalDad]

Agreed. Good point regarding resolution. I'm only running 1680x1050 (native), so I'm not really pushing too many pixels. All cores unlocked, but I need to run some bench tests on how each core is being hit. As I mentioned, I do have everything set to Very High (which is what Nvidia "Game Experience" suggests, solely based on video card). Anti-aliasing from POSTAA to none made no difference as expected if vid card is not the bottleneck. As I've mentioned I would think an AMD 8350 at 4 GHz should be able to handle the CPU load (I have Black but not overclocking presently). Everything I've read regarding parking cores to free up bandwidth (8 cores but only 4 memory channels) has suggested it doesn't really buy much, but I will play around this weekend.

Thanks for all the input.

[JimboFBX]

Does DX9 have a higher CPU overhead than DX11? I feel like I'm getting tangibly better fps (hitting 60 fps much more frequently) with DX11 and most settings at high. This is with the CPU affinity tweaking I did above.

[Nemesis Duck]

In theory, DX11 should be more efficient (faster) than DX9 is, but it depends on if the code takes advantage of the situation.

DX11 introduces new features to allow more 3D funtionality to be offloaded from the CPU to the GPU, freeing the CPU to focus on non-graphics related tasks. IE. networking, sound, AI

In some cases but not always, DX11 also CAN have more efficient versions of DX9 function calls. Things you were already doing with the GPU in DX9 now execute faster in DX11.

In my experience, at the settings I use, DX11 and DX9 are about the same.

[Summon3r]

Miner, on 25 October 2014 - 05:58 AM, said:

Agreed. Good point regarding resolution. I'm only running 1680x1050 (native), so I'm not really pushing too many pixels. All cores unlocked, but I need to run some bench tests on how each core is being hit. As I mentioned, I do have everything set to Very High (which is what Nvidia "Game Experience" suggests, solely based on video card). Anti-aliasing from POSTAA to none made no difference as expected if vid card is not the bottleneck. As I've mentioned I would think an AMD 8350 at 4 GHz should be able to handle the CPU load (I have Black but not overclocking presently). Everything I've read regarding parking cores to free up bandwidth (8 cores but only 4 memory channels) has suggested it doesn't really buy much, but I will play around this weekend.

Thanks for all the input.

you should definitely consider OC to 4.4+ your cpu will do it easy and MWO will love you for it.... baring that set your particles/environment/object detail to high .... particles are murdersome on even the high end intel cpu's that are OC'd.

[Aratan Aenor]

I've had no problems with MWO and CPU core loads. My stock Phenom II 1055T runs between 50% and 65% on all 6 cores while playing.

I run a smooth 60 fps with v-sync, DX11, max settings and MSAA enabled. My GPU load runs between 20% and 80% depending on the amount of action on screen.

Tried CrossFire with same settings, and got between 25 and 40 fps with a few drops to 15, GPU1 load maxed out at 40%. GPU2 load bounced between 0% and 5%. (Multi-GPU support please) CPU usage remained 50% - 65% across all cores.

CPU: AMD Phenom II 1055T @ 2.8 GHz
RAM: 8GB DDR3-2133 9-10-9-24
GPUs: 2x HD 7970 3GB 1000/1375 (Gigabyte WindForce3 OC Edition)

[Chrithu]

The problem according to a dev post I can't find right now is that the way they are rendering mechs for animation purposes causes too many draw calls for low and mid end CPU/GPU setups to handle.

OCing my CPU helps to some extend. Unparking cores didn't change a thing.

I don't even want to know how many new players this particular issue has caused to immediately uninstall again.

As a F2P MMO you need the casual masses to survive and they usually don't spend loads of money on a high end gaming rig.

[JimboFBX]

Exactly. I'm pretty sure my 5 year old cpu is still more powerful than most laptop cpus unless you go alienware or something

[FullMetalDad]

Quote

you should definitely consider OC to 4.4+ your cpu will do it easy and MWO will love you for it.... baring that set your particles/environment/object detail to high .... particles are murdersome on even the high end intel cpu's that are OC'd.

Tried this past weekend. Changing the multiplier from default 20 (4 GHz) to 21 (4.2 GHz) made the system unstable playing MWO - kept crashing to desktop. I even had to run the MWO repair tool as some files got corrupt. I tried setting several different core affinities (0,2,4,6 vs 1,2,5,7), then only over clocking those but got generally worse results: seemed to better with all cores up. I was running a custom CPU OC profile for MWOclient, so I might try some general overclocking of entire system. I have the "AMD Overdrive tool" that came with the unlocked Black edition so their is a fair amount of tweaking that can be done. I have OC'd before and would think going from 4.0 to 4.4 (10%) would not be too stressful - maybe a bad chip? I can run some stability tests.

I also will try reducing the graphics settings:I had read particles in particular was a killer. Thanks Summon3r for your input.

Edited by Miner, 27 October 2014 - 05:29 AM.

[Nemesis Duck]

Same here. I had to add some vcore (1.39v) to make mine stable at 4.4GHz. You should really be OCing/disabling cores/adjusting voltage in the BIOS, it's always been the better way. It might getting better with UEFI. While you're in the BIOS, disable power saving stuff, like C6 state, but especially Cool'n Quiet, which will lower your CPU speed to save power and sometimes execute code at lower speeds, which is exactly what you don't want when gaming.

Make sure your RAM is clocked to it's spec and not just set to AUTO. If both your BIOS and your RAM support a memory profile, load that as it's the best you can hope for. Once you establish a baseline, and read a little about memory timings on the internet, try lowering some values. The saying is AMD's like lower latencies while Intel's like high bandwidths, so your AMD will benefit going from 9-9-9-30 to 8-8-8-24 more than 1600 to 1866. (unless you have an APU!)

[Summon3r]

Miner, on 27 October 2014 - 05:28 AM, said:

Tried this past weekend. Changing the multiplier from default 20 (4 GHz) to 21 (4.2 GHz) made the system unstable playing MWO - kept crashing to desktop. I even had to run the MWO repair tool as some files got corrupt. I tried setting several different core affinities (0,2,4,6 vs 1,2,5,7), then only over clocking those but got generally worse results: seemed to better with all cores up. I was running a custom CPU OC profile for MWOclient, so I might try some general overclocking of entire system. I have the "AMD Overdrive tool" that came with the unlocked Black edition so their is a fair amount of tweaking that can be done. I have OC'd before and would think going from 4.0 to 4.4 (10%) would not be too stressful - maybe a bad chip? I can run some stability tests.

I also will try reducing the graphics settings:I had read particles in particular was a killer. Thanks Summon3r for your input.

OC through your BIOS, up muliplier and up vcore slightly long as u have a decent cooler you should easily hit 4.4 on that chip

[FullMetalDad]

Quote

Same here. I had to add some vcore (1.39v) to make mine stable at 4.4GHz. You should really beOCing/disabling cores/adjusting voltage in the BIOS, it's always been the better way. It might getting better with UEFI. While you're in the BIOS, disable power saving stuff, like C6 state, but especially Cool'n Quiet, which will lower your CPU speed to save power and sometimes execute code at lower speeds, which is exactly what you don't want when gaming.

Make sure your RAM is clocked to it's spec and not just set to AUTO. If both your BIOS and your RAM support a memory profile, load that as it's the best you can hope for. Once you establish a baseline, and read a little about memory timings on the internet, try lowering some values. The saying is AMD's like lower latencies while Intel's like high bandwidths, so your AMD will benefit going from 9-9-9-30 to 8-8-8-24 more than 1600 to 1866. (unless you have an APU!)

Quote

OC through your BIOS, up muliplier and up vcore slightly long as u have a decent cooler you should easily hit 4.4 on that chip

Thanks for the tips. I tried a "standard" OC config thru BIOS on my ASRock MB that put the clock at 215 MHz (4.3 GHz core speed), but again MWO crashed. I'll have to see which parameters this factory default OC changed. Interestingly, changing all video settings to LOW in MWO only increased performance by about 10 FPS versus VERY HIGH - still had dips in 30s and normally 40s in match. I did up the VCore using the AMD tool while adjusting the multiplier, but no joy. Maybe in BIOS.

Appreciate the comments Assmodeus and Summon3r. Thanks again.

[Summon3r]

Miner, on 28 October 2014 - 04:59 AM, said:

Thanks for the tips. I tried a "standard" OC config thru BIOS on my ASRock MB that put the clock at 215 MHz (4.3 GHz core speed), but again MWO crashed. I'll have to see which parameters this factory default OC changed. Interestingly, changing all video settings to LOW in MWO only increased performance by about 10 FPS versus VERY HIGH - still had dips in 30s and normally 40s in match. I did up the VCore using the AMD tool while adjusting the multiplier, but no joy. Maybe in BIOS.

Appreciate the comments Assmodeus and Summon3r. Thanks again.

wait a sec you arent OC'ing your BCLK (base clock) are you? that would be very tricky and likely unstable... also for a visual step by step youtube your mobo + cpu and oc'ing and you should get a very helpful video!

[FullMetalDad]

Quote

wait a sec you arent OC'ing your BCLK (base clock) are you? that would be very tricky and likely unstable... also for a visual step by step youtube your mobo + cpu and oc'ing and you should get a very helpful video!

In BIOS I set CPU core frequency individually from 200 MHz to 215 MHz (or any 5 MHz increment). Clock multiplier 20 (4.3 GHz CPU clock). The PCIe reference speed is set separately and maintained at 100 MHz. Alternatively, I can set the clock multiplier in steps of 0.5. Memory settings can also be fine tuned, though I usually leave these to the default, as well as the tweaking the core voltage. My MB is an ASRock that comes with OC configured into the BIOS (you can even choose to boot OC or not at the POST screen). I just need to play around a bit to find a stable configuration. The YouTube suggestion is a good one: someone may have already gone through the tweaking process. Thanks!

[JimboFBX]

It just occurred to me...

I have a 64-bit processor

This is a 32-bit program

Couldn't this potentially run more efficiently as a 64-bit executable?