Amd Cpus?
#41
Posted 02 July 2014 - 02:44 PM
#42
Posted 02 July 2014 - 03:04 PM
stick to intel cpu
and
nvidia gpu
if your sole purpose to play this game only.
Edited by King Arthur IV, 03 July 2014 - 08:33 AM.
#45
Posted 03 July 2014 - 08:49 AM
The Nvidia cards might run marginally faster, but seem to have more issues. And I've at least seen suggestions of semi-working Crossfire implementations, and few/no working SLI fixes.
Back to the OP...
I'd suggest an FX8350 with a decent Liquid Cooler, going up to 8 more more GB RAM, and switching to 64bit Windows(this is a freebie...your windows license is good for both 32 and 64bit versions). Maybe also go with at least an R9-270X.
For MWO the FX-6xxx is a side grade at best, the FX-4xxx is probably a downgrade. Best bet, without breaking the bank is to go FX-8350, and overclock the beejebus out of it.
Or, be like me. Save your pennies and wait for Haswell-E in September.
Edited by darqsyde, 03 July 2014 - 09:00 AM.
#46
Posted 03 July 2014 - 10:13 AM
darqsyde, on 03 July 2014 - 08:49 AM, said:
The Nvidia cards might run marginally faster, but seem to have more issues. And I've at least seen suggestions of semi-working Crossfire implementations, and few/no working SLI fixes.
Back to the OP...
Well, to add insult to injury, Nvidia cards are slower at basically every single pricepoint, until you reach utterly ridiculous levels of spending and just run out of AMD GPUs to compare, not because there's anything inherently wrong with Nvidia, but because Kepler is fundamentally a failed GPU architecture. It costs over twice as much to make Kepler do what Volcanic Islands or Southern Islands will do, so they lop off effectively all of the GPGPU capability and then skimp badly on the VRAM to make it still-but-not-quite-as-badly inferior, and you get a GK104 (or if you want the full-flavor variety with all the VRAM, a very expensive GK104).
Drivers are good from both companies these days of course, but yes, Nvidia cards do have more issues with MWO in particular. You can examine the forums at any point and see that whenever a GPU problem is encountered, it's an Nvidia GPU 5-6 times as often as an AMD GPU. That ratio has held steady almost since closed beta, though I haven't bothered looking all that recently.
Edited by Catamount, 03 July 2014 - 10:15 AM.
#47
Posted 04 July 2014 - 02:15 AM
SLDF DeathlyEyes, on 02 July 2014 - 02:44 PM, said:
Hey bud, I am wondering how your getting this low FPS!? Obviously @5Ghz you've OC'd a bit, so I imagine your NB / FSB are high enough and your Ram is running at speed. Are you on a Std HDD or SSD?
Anyhow you can always PM me if you want to go through things. I run 8350@4.7Ghz and Maxed detail, AAoff, 1920x1080, Vsync on and net 60-40Fps.
#48
Posted 04 July 2014 - 09:53 PM
WarGruf, on 04 July 2014 - 02:15 AM, said:
Hey bud, I am wondering how your getting this low FPS!? Obviously @5Ghz you've OC'd a bit, so I imagine your NB / FSB are high enough and your Ram is running at speed. Are you on a Std HDD or SSD?
Anyhow you can always PM me if you want to go through things. I run 8350@4.7Ghz and Maxed detail, AAoff, 1920x1080, Vsync on and net 60-40Fps.
indeed
FSB @256X19=4861mz
NB@2560
HT@2560 with
2046mhzDDR3 will land you much higher settings and framerates.....specially if you are experiencing thermal throttle to a 7X multiplier when MWO thrashes your cores like a pissed off chimpanzee thrashes tourists who falls into their cage @ the zoo.......Use core temp to see real time breakdown of core action./
#49
Posted 05 July 2014 - 10:53 AM
SLDF DeathlyEyes, on 02 July 2014 - 02:44 PM, said:
WarGruf, on 04 July 2014 - 02:15 AM, said:
Hey bud, I am wondering how your getting this low FPS!? Obviously @5Ghz you've OC'd a bit, so I imagine your NB / FSB are high enough and your Ram is running at speed. Are you on a Std HDD or SSD?
Anyhow you can always PM me if you want to go through things. I run 8350@4.7Ghz and Maxed detail, AAoff, 1920x1080, Vsync on and net 60-40Fps.
It's probably how MWO is utilizing the CPU. It's only quad-threaded and if it's using only two modules, it's only getting two FPUs, and MWO is very FPU-heavy. You might as well be giving it an FX4320 at 5ghz at that point, or an 8320 at below stock clocks.
Edited by Catamount, 05 July 2014 - 10:56 AM.
#50
Posted 05 July 2014 - 01:03 PM
Catamount, on 05 July 2014 - 10:53 AM, said:
12 named threads, all seemingly requiring floating-point access, assignable to six cores, plus ~ 4 more variables able to spawn up to 12 other threads each …
#51
Posted 05 July 2014 - 01:23 PM
It may have numerous threads, but clearly not numerous threads of especial significance.
#52
Posted 05 July 2014 - 01:38 PM
If ca_thread0Affinity isn't happy, will you notice a difference between a X4 or X6 Phenoms, and what tweaks did it take to paint yourself into that corner?
#53
Posted 05 July 2014 - 09:49 PM
Still, are you saying that it's solely because something like an 8350 lacks the horsepower to feed ca_thread0Affinity? I'm not sure that would explain why MWO takes poorly to anything with less than four cores, but is indifferent to anything with more. An FX4350 can process a single thread as quickly as an 8350, so if that's the entire bottleneck in and of itself, why does the 4350 flounder so badly by comparison?
And if MWO has more than four threads that actually matter, why doesn't something like a 3770k seem to in any way outperform a 3570k? Why doesn't a 3930k absolutely scream by comparison to both?
Edited by Catamount, 05 July 2014 - 10:07 PM.
#54
Posted 06 July 2014 - 04:52 PM
Catamount, on 05 July 2014 - 10:53 AM, said:
Catamount, on 05 July 2014 - 09:49 PM, said:
My current i7-990X, with the 4288MHz overclock:
The previous i7-920, which ran at 2.8GHz all the time thanks to swell cooling:
Both had a fps limit of 70 to go with my 1800 x 1440 x 70Hz CRT. True Color, of course.
That's ca_thread0Affinity on the last core of both systems, with "sys_budget_soundCPU = 0" no-longer "preserving" the last 15 percentage points of any given core to run them thar RealTek cardz. Note the scale difference(s) from one core to the next.
I've got
r_WaterUpdateThread = 0 sys_TaskThread2_CPU = 4 sys_TaskThread4_CPU = 2 sys_TaskThread0_CPU = 4 sys_TaskThread1_CPU = 0 sys_TaskThread3_CPU = 2 sys_TaskThread5_CPU = 1in my user.cfg, but the other five are their stock "locations":
;ca_thread0Affinity = type: int current: 5 ;ca_thread1Affinity = type: int current: 3 ;sys_main_CPU = type: int current: 0 ;sys_physics_CPU = type: int current: 1 ;sys_streaming_CPU = type: int current: 1 ;r_WaterUpdateThread = type: int current: 5 ;sys_TaskThread0_CPU = type: int current: 3 ;sys_TaskThread1_CPU = type: int current: 5 ;sys_TaskThread2_CPU = type: int current: 4 ;sys_TaskThread3_CPU = type: int current: 3 ;sys_TaskThread4_CPU = type: int current: 2 ;sys_TaskThread5_CPU = type: int current: 1for reference.
Teh "Once Size Fits All" nature of that list tasks me greatly; That ca_thread0Affinity stays on the last core isn't what I would expect from looking at that list. It does seem CryTek thought Main, TaskThread2, and TaskThread4 were all special, 'cause they got cores all to themselves, but why? You'd figure Main is doing the net code, but never seems too loaded; Task(s)2 & 4 do seem to carry a load, but I have no rhyme or reason for any of these "sys_TaskThreadX_CPU".
It uses 6 cores, and can use Hyper-Threading on a quad; But the load imbalance in the CA subsystem requires such a beefy IPC/ Overclock, it becomes practical to just use an I5.
For laughs:
This should be the last time I tested on the Pentium G620/ GTX650Ti HTPC, in the training grounds, in DX11
sys_spec = 0 sys_spec_full = 4 cl_fov = 79 d3d10_TripleBuffering = 1 d3d11_TripleBuffering = 1 d3d9_TripleBuffering = 1 r_DrawNearFOV = 79 r_GeomInstancing = 1 r_multiGPU = 0 r_ShadersPreactivate = 1 r_silhouettePOM = 0 r_stereodevice = 0 r_TexturesStreaming = 1 r_TexturesStreamingSync = 0 r_UsePOM = 0 r_vSync = 0 sys_budget_soundCPU = 0 sys_budget_streamingthroughput = 15662080 sys_budget_sysmem = 2048 sys_budget_videomem = 1024 sys_job_system_enable = 1 sys_job_system_max_worker = 0 s_NumLoadingThreadsToUse = 2 r_ShadersAsyncMaxThreads = 2 sys_spec_Water = 1 sys_streaming_memory_budget = 2048 ca_thread = 0 r_MultiThreaded = 1 e_ParticlesThread = 0 ;sys_budget_frametime = 22 sys_MaxFPS = 45 sys_budget_fps = 45 ca_thread0Affinity = 0 ca_thread1Affinity = 0 r_WaterUpdateThread = 0 sys_main_CPU = 0 sys_physics_CPU = 0 sys_streaming_CPU = 0 sys_TaskThread0_CPU = 0 sys_TaskThread1_CPU = 0 sys_TaskThread2_CPU = 0 sys_TaskThread3_CPU = 0 sys_TaskThread4_CPU = 0 sys_TaskThread5_CPU = 0 es_MaxPhysDistInvisible = 1 e_ObjectLayersActivationPhysics = 1 e_OnDemandPhysics = 0 p_cull_distance = 1 p_num_threads = 2 sys_limit_phys_thread_count = 0 sys_physics = 1 sys_spec_Physics = 1 e_particles = 1 e_ParticlesCullAgainstViewFrustum = 1 e_ParticlesGI = 0 e_ParticlesLights = 0 e_ParticlesPoolSize = 0 e_ParticlesPreload = 1 e_ParticlesShadows = 0 r_UseGSParticles = 1 r_UseParticlesGlow = 0 r_UseSoftParticles = 0 q_ShaderWater = 0 r_fogShadowsWater = 0is the high-light reel for it's user.cfg; should be nothing higher then High, and no Glow.
So: I believe
ca_thread0Affinity = 0 ca_thread1Affinity = 0 r_WaterUpdateThread = 0 sys_main_CPU = 0 sys_physics_CPU = 0 sys_streaming_CPU = 0 sys_TaskThread0_CPU = 0 sys_TaskThread1_CPU = 0 sys_TaskThread2_CPU = 0 sys_TaskThread3_CPU = 0 sys_TaskThread4_CPU = 0 sys_TaskThread5_CPU = 0punts thread sorting to the OS;
ca_thread = 0 r_MultiThreaded = 1 e_ParticlesThread = 0pretty much tells the game you are "only a duel-core, so extra threads are just overhead," and the fps limit(s) unload both the CPU and the GPU greatly from doing frames it would blow the deadline on anyways. (The
Note how it looks like "sys_budget_soundCPU = 0" does nothing on a duel-core: It's really applied per thread.
Moving this box to DX11 required a different bunch of settings then DX9, and still the loads are higher …
Every other week, I get a yen to build one of them thar FX boxes, if only to see which thread(s) truly need access to the math-co …
Edited by Goose, 08 July 2014 - 12:59 PM.
#55
Posted 08 July 2014 - 03:12 PM
just upped the cpu from the fx 8320 and added the SSD and the 3rd card! This is her 1st stroll in the BIG WORLD!
MB - ASUS Crossfire V Formula Z 990FX
CPU - AMD FX-9590 8-c Blk Edt 4.7GHz
RAM - Corsair Dominator Platinum 32gig DDR3 2133mhz
[color=CPU/GPU Cooling - Aquacomputer Aquaduct 720xt Mk V External L/C
Case - CoolerMaster HAF X Full Tower (2x200mm, 4x160mm fans 1 and 2 intake/exhaust respectively)
SSDs - Samsung SSD 840 EVO 250GB
HDDs - W/D VelociRaptor WD5000hhtz 500gig at 10krpm
HDDs - W/D VelociRaptor WD1000dhtz 1tb at 10krpm
Graphic - 3x Crossfired AMD Radeon HD 7770's
Monitors - 3x22in Samsung S22c300
PS - Thermaltake Toughpower Grand TPG -1200M 1200w
Edited by Tho, 08 July 2014 - 03:16 PM.
#56
Posted 19 July 2014 - 12:10 PM
near 100% cpu load all the time.
my settings are very modest (nothing above medium, some on low), in a brawl I'm Always cpu limited.
haswell quads should easily outperform this pentium, but it seems to outperform my old i5-750 on 4.0, and it's dirt cheap (+ many H81 boards allow overclocking, stock cooler gets you pretty far).
The way I test is with msi afterbuner. I log cpu use, gpu use and framerate. In a brawl fps tanks, and so does gpu use, this means the cpu cant keep up. on my old quad I had roughly 50% cpu load in a brawl, so 50% of that cpu was wasted by the game not scaling. This Pentium is near full load all the time, outperforming the old quad with better perf/clock and higher clockspeed.
Settings might make a difference on scaling too, if you have high object detail there may be a lot of directx cpu overhead, I've set the pre rendered frames in nvidia control panel to the minimum (1). This means the cpu can only work on one frame at a time. Perhaps this is why it failed to load up my quad, but I like the low input lag it brings. "no scaling" and fullscreen are also important for low input lag.
#57
Posted 19 July 2014 - 12:54 PM
Catamount, on 05 July 2014 - 09:49 PM, said:
Still, are you saying that it's solely because something like an 8350 lacks the horsepower to feed ca_thread0Affinity? I'm not sure that would explain why MWO takes poorly to anything with less than four cores, but is indifferent to anything with more. An FX4350 can process a single thread as quickly as an 8350, so if that's the entire bottleneck in and of itself, why does the 4350 flounder so badly by comparison?
And if MWO has more than four threads that actually matter, why doesn't something like a 3770k seem to in any way outperform a 3570k? Why doesn't a 3930k absolutely scream by comparison to both?
Blew through the AMD FX part of the question, didn't I?
Them FX chips are caught in a crossfire: ca_thread0Affinity demeans a critical mass of Hz and IPC; Then, the seemingly vast majority of game threads demand access to the math-co of any given module, which is directly opposed to whole point of AMD's Module architecture: "Math co-processors aren't that important anymore."
So: We have a couple reports of ~30% boost by restricting MW:O to a 1:1 cores-to-math-co ratio, through two different means (shut cores off in the BIOS, on one hand, or restrict the game to "teh left" cores of a module, using stuff like Processor Lasso or Bill2's Process Manager); We don't yet have any reports of user.cfg settings that should do something close to that, in theory; This is important 'cause we've heard Teh Twelve Named Threads all want math-co access, but I can think of two or three settings that would spawn a whole bunch of "unnamed" threads, and we don't know if these can get by without floating point access.
The "critical mass" thing? I once saw someone on Tom's Hardware say AMD needs ~50% moar Hz to make up for the IPC gap: That means 6.2GHz, don't it?
I don't have the parts to do the tests I'd like to see, and no one else is trying; But it still comes down to "reduce a module to one core" for MW:O purposes …
Edited by Goose, 31 August 2014 - 11:04 AM.
#58
Posted 19 July 2014 - 05:05 PM
#59
Posted 20 July 2014 - 07:21 AM
Goose, on 19 July 2014 - 12:54 PM, said:
D'oh: http://mwomercs.com/...ost__p__3394675
I guess he was still getting his ~30% …
#60
Posted 20 July 2014 - 09:11 AM
At 6 cores, I never get above 100 degrees Fahrenheit, either.
Edit: typo, addendum.
Edited by Apocalypse Pryde, 20 July 2014 - 09:12 AM.
1 user(s) are reading this topic
0 members, 1 guests, 0 anonymous users