Cpu Or Gpu?
#41
Posted 15 October 2014 - 02:56 AM
#42
Posted 15 October 2014 - 04:39 AM
Flapdrol, on 14 October 2014 - 11:43 PM, said:
You are cpu limited, don't get a new gpu.
Yup. CPU for sure. The graphs don't lie. If your GPU isn't pegging the 100% mark for most of the match, your CPU can't feed it well enough.
#43
Posted 15 October 2014 - 05:02 AM
I ran 2 more games after posting that pic last night and the GPU was pinned about half the time for both games. Not sure if that makes a difference in the recommendations.
I am guessing if I update the GPU, I will quickly become CPU bound. The CPU seems to be very active. What should I expect the CPU to look like if I was completely CPU bound? Would it pin the CPU at 100% or just get to the 70-80% mark regularly?
Thanks again for all the help guys. I know this seems like a lot of work, but MWO is really the only game I play right now so doing a full system upgrade doesnt have a ton of value for me. The game runs ok, I would just like it to perform better under specific stresses that seem to kill performance. In particular smoke/particle effects seem to kill it, even with the settings reduced for those. I have seen my framerate drop into the high teens when I am smoking and walking backwards.
'Anyways, thanks again for all the feedback.
SpdDmnAdam: What does clear that override file do to help utilize the GPU better? (If I can get the GPU to pin more regularly, then that might tip me towaards getting a new vid card)
#44
Posted 15 October 2014 - 05:14 AM
Quote
It would most likely not pin the CPU at 100%. CPU usage is not a good way to measure a CPU bottleneck in games. It hasn't been since the days of single-core chips. This is because if one core is holding back instruction execution on the others, your CPU will be bottlenecking, but might display low usage, because, to take your CPU as an example, three modules and a core and a half would be sitting around twiddling their thumbs while one module is in a desperate dead sprint to keep up.
To measure CPU bottlenecks in games, there are really only three things you can do:
1.) Examine GPU usage and infer based on whether it hits 100% or not (also not a good method imo, but probably okay if you're feeling lazy; it's really asking for extraneous variables)
2.) Vary low and high settings and see how it impacts your min and average fps (good method, no guarantee against spurious data since you're still indirectly inferring, but it's a solid inference)
3.) Overclock your CPU and see what happens/ try another CPU (best method; inference about the bottleneck directly follows from results, with essentially no possibility of other factors slipping in and screwing up results/conclusion)
^-theoretically you could also underclock a bit; if you're CPU bound your performance should decrease linearly with CPU clock since there's no headroom. But who the hell wants to do that?
Edited by Catamount, 15 October 2014 - 05:18 AM.
#45
Posted 15 October 2014 - 05:17 AM
Sprouticus, on 15 October 2014 - 05:02 AM, said:
Mwo doesn't scale well to all cores, so it'll never be able to load the cpu to 100%. Therefore:
xWiredx, on 15 October 2014 - 04:39 AM, said:
I'd just look at GPU use and FPS (that's on the graph too right?), if gpu use and fps go down at the same time you're CPU limited when the fps is lowest, and it would make most sense to upgrade the CPU
#46
Posted 15 October 2014 - 05:20 AM
It's not even Piledriver, and we all know what a bundle of fun first gen/Bulldozer was.
Edited by Catamount, 15 October 2014 - 05:21 AM.
#47
Posted 15 October 2014 - 06:30 AM
Catamount, on 15 October 2014 - 05:20 AM, said:
It's not even Piledriver, and we all know what a bundle of fun first gen/Bulldozer was.
Im half tempted to do both, but the cost would be enough that I could be halfway to a full upgrade.
#48
Posted 15 October 2014 - 06:41 AM
Flapdrol, on 15 October 2014 - 05:17 AM, said:
Mwo doesn't scale well to all cores, so it'll never be able to load the cpu to 100%. Therefore:
I'd just look at GPU use and FPS (that's on the graph too right?), if gpu use and fps go down at the same time you're CPU limited when the fps is lowest, and it would make most sense to upgrade the CPU
Yea, the FPs and GPU seem to be in sync most of the time. The GPU pins itself at times, and the framerate flattens itself out but it is usually in the high 40's (per the graph) which is acceptable for me playing the game. Mind you I was used to low 30's until they did the performance passes last year so high 40's is fine.
Which means I am CPU bound like you guys said. I am guessing bumping myself to an 8xxx series CPU will make my GPU bound most of the time, as was said earlier, I can then manage framerate with settings in game.
Any thoughts on a fan if I get an 8370?
#49
Posted 15 October 2014 - 06:52 AM
Make sure to check the CPU support list for your motherbored and flash your bios before going to an 8370. It's newer and some older motherboards may not support it.
Edited by ninjitsu, 15 October 2014 - 06:54 AM.
#50
Posted 15 October 2014 - 07:42 AM
Also, the 8370 is still quite an expensive cpu, maybe it's more cost effective to switch to intel, at least for this game. If you want to stay with amd and the board supports it I'd say get an FX6350, doesn't break the bank, x3xx series has some fixes to the uncore or something that improves performance and the chip has good clocks plus enough cores for the few games that do scale well to them.
I run a cheap H81 board (only 2 ram slots, few usb/sata ports) with a pentium g3258 overclocked to 4.6GHz, on maximum settings I drop to something like 37, on low to 50, or stay above 60, depends on how big the brawl gets. if I look at your graph you drop to 20 fps most of the time, so I reckon it would be an upgrade even with a more conservative overclock. Of course you could try overclocking your current cpu as well, see what it can do.
If you don't want to overclock I'd recommend the cheapest i5, more expensive, but the pentium at stock is a bit underwhelming, or so I hear, wouldn't know myself as I only ran it stock when installing windows.
#51
Posted 15 October 2014 - 07:56 AM
Thanks again everyone, I will make sure to let you all know how it turns out!
#52
Posted 15 October 2014 - 08:03 AM
Flapdrol, on 15 October 2014 - 07:42 AM, said:
Also, the 8370 is still quite an expensive cpu, maybe it's more cost effective to switch to intel, at least for this game. If you want to stay with amd and the board supports it I'd say get an FX6350, doesn't break the bank, x3xx series has some fixes to the uncore or something that improves performance and the chip has good clocks plus enough cores for the few games that do scale well to them.
I run a cheap H81 board (only 2 ram slots, few usb/sata ports) with a pentium g3258 overclocked to 4.6GHz, on maximum settings I drop to something like 37, on low to 50, or stay above 60, depends on how big the brawl gets. if I look at your graph you drop to 20 fps most of the time, so I reckon it would be an upgrade even with a more conservative overclock. Of course you could try overclocking your current cpu as well, see what it can do.
If you don't want to overclock I'd recommend the cheapest i5, more expensive, but the pentium at stock is a bit underwhelming, or so I hear, wouldn't know myself as I only ran it stock when installing windows.
I thought abuot ther 6350, especially for the value, I will be GPU bound for sure now.
But for the cost of the 6350 + a good fan, I can get the 8370 and use the freakin stock fan (with better thermal compound) and still get better performance without OC.
#53
Posted 15 October 2014 - 08:19 AM
#54
Posted 15 October 2014 - 08:52 AM
Flapdrol, on 15 October 2014 - 08:19 AM, said:
Eh... we are essentially counting each module as one core for MWO since it really ties up the CPU and needs the full resources. Pretty sure the 8370 will outdo the 6350 (but I don't think we have anybody that's tested it for certain).
#55
Posted 15 October 2014 - 09:31 AM
#56
Posted 15 October 2014 - 09:40 AM
#57
Posted 15 October 2014 - 10:00 AM
Sprouticus, on 15 October 2014 - 09:40 AM, said:
I still don't know how people do that. Thermal compound is easy. A dot in the center about half the size of a pea is more than enough. When securing the heatsink, the pressure will spread it evenly as long as you didn't somehow create a huge air bubble in your dot. If you're going to reuse the heatsink, Arctic makes a kit to emulsify the stuff and it wipes off pretty nicely. There's a purify or whatever bottle to clean it after. Takes 5 minutes, no big messes. Use MX-4 for maximum happiness.
#58
Posted 15 October 2014 - 12:04 PM
xWiredx, on 15 October 2014 - 10:00 AM, said:
I still don't know how people do that. Thermal compound is easy. A dot in the center about half the size of a pea is more than enough. When securing the heatsink, the pressure will spread it evenly as long as you didn't somehow create a huge air bubble in your dot. If you're going to reuse the heatsink, Arctic makes a kit to emulsify the stuff and it wipes off pretty nicely. There's a purify or whatever bottle to clean it after. Takes 5 minutes, no big messes. Use MX-4 for maximum happiness.
Wait, i dont have to spread it at least SOME?? I have always worried I would not get even coverage and it would cause heat issues on the CPU...
#59
Posted 15 October 2014 - 12:10 PM
#60
Posted 15 October 2014 - 12:49 PM
Some very high viscosity thermal compounds like IC Diamond 7 do recommend twisting a bit after pressing down and before locking down the cooler to help the process out, at least as I recall, but most pastes aren't like that. Of course most pastes also dry out given enough time (I have an AS Ceramique 2 horror story there from not reapplying for 3 years), so I suppose there's small tradeoffs either way. The bottom line is that regardless of what you get, it's pretty hard to screw up application unless you either apply way too much (half the size of a .177 BB is good) or do something that introduces an air bubble, which you won't as long as you press the heatsink on and leave it there.
Edited by Catamount, 15 October 2014 - 12:49 PM.
1 user(s) are reading this topic
0 members, 1 guests, 0 anonymous users