19 replies to this topic

[p8ragon]

Hey everyone, I recently picked up an ASUS 7970 DC ii to replace a pair of faulty GTX 570s. I overclocked the video cards, and I've been able to play all graphically intensive games (IE natural selection 2, Crysis warhead, sleeping dogs etc) BUT mwo. After ~10 minutes the game will freeze up, and my video card will revert to 2d setting clocks. Anyone else have this problem? (it was prevalent in my gtx 570 as well)

[Catamount]

Could your GPU just be overheating? MWO isn't GPU-intensive in the sense that it takes a really powerful GPU to run it well, but it is in the sense that it produces a lot of heat in one's system.

[Cid F]

I use a HD7950@925/1500MHz, stock is 800/1250MHz and i have no problems

[MyCatTequila]

I use an old 6670 at the moment and it cooks it like a piece of rare steak!

[Gremlich Johns]

My current video card, a GTX760 4GB, has an Arctic cooling Accelero Extreme III. Currently at idle, it's 30 deg C at an ambient temp of 23 deg C. My OC'd FX-4100 has a closed loop cooler - currently at 32 deg C at typing speed. Neither have gone over 52 deg C under gaming loads.

Get an after market cooler if you are overclocking.

[The Gunman]

I'm running Triple Crossfired 7970s OC'ed from 950/1425 to 1000/1450.

I initially had heat issues because of poor clearance between the cards. 70c/85c/95+c that caused awful micro stuttering because one on the cards would overheat and throttle down.

Once the clearance issue was resolved I'm now getting 60c/65c/70c.

What temps are you getting durring play? Are you certain that the cards' fans are in fact spinning? Do you have sufficient case ventilation?

[Sunbasker]

Just buy some Cool Shots and consider upgrading to the double heatsi-- oh, right.

Different games will tax your card differently, even on max settings. It's possible that MWO isn't as well-optimized as some of your other games, and is therefore running the card constantly at 100%, whereas other games might lower the usage a bit down to what they actually need.

Protips:
-Use Vsync. It tears less often and usually saves tons of power and heat, because it limits your framerate to what your monitor will actually render.
-Listen to your card. If it's persistently doing this after X minutes of playing, it's overheating after X minutes of playing. Decrease the clocks a bit so that it doesn't. It's exactly like caring for your mech's heat, it has no choice but to cut back or face serious thermal damage when it reaches its limit. CPUs will also do this if they get too hot, either auto-downclock or totally shut down to protect themselves.

Just be careful and respect the fact that overclocking is, by nature, pushing the hardware beyond what it was designed for. You can often squeeze some extra juice out of it without damage, but when it complains, listen to it before you break it.

[Narcissistic Martyr]

Unless you're rocking a sandy bridge e system don't overclock your GPU for mwo for 1080p as this game is extremely CPU bound.

[1Sascha]

Overclocking for the sake of overclocking aside:

Considering reviews with "real"/synthetic benchmarks *and* my personal experience, I'd say GPU OCing isn't going to do a whole lot for any game. Why run your GFX-card hot when all you'll get is around 5 FPS more in actual games?

I'd rather spend the $$s for a fancy aftermarket GPU-cooler on getting a better video-card with a nice cooling solution already built in. Plus: You really don't need *that* much GFX-power to run the game at high/stable FPS. My new GTX660 does just fine on HD with everything maxed out (getting ~70 FPS online and never dip into yellow/red zone) and it "only" cost me around €170 - which I consider a middle of the road kinda price.


S.

[Flapdrol]

You cant just set the speed higher and expect it to be stable. I'd recommend testing with unique heaven, that's a good stresstest and can loop infinitely. If it's unstable and still cool you could increase the voltage a bit (msi afterburner can do this on many cards)

Anyway, overclocking the graphics card wont do much for mwo because in a fight you'll be cpu bound. If you want to overclock something overclock the cpu first.

[ArmageddonKnight]

Unigine Valley, Unigine heaven, 3DMark, 3Dmark 11. These are all good programs ot use when testing ur GPU.
Make sure ot also use a few game benchmarks. Metro LL benchmark, and Crysis benchmark are good.
And then ofc try a few games out.

I cant say how well a OC affects MWO since i use a GTX 780 Classified + OC and its totaly OTT for MWO, i run max settings and even downsample it at 2400x1350 just to reduce alaising.
How much FPS a OC gives is totaly dependant on the level of OC, if u can get around 200mhz OC on the core of the 7970 u can maybe get 10-15% better FPS. Add on some memory OC and u can maybe at 1 or 2 FPS more ontop. Memory OC's dont affect games as much as they do bench scores.

As for MWo's CPu dependancy, i cant say from personal experiance as my CPU doesnt have any issue with MWo, however having seen other forum posts to do with performance issues the trend tends ot be weak CPU = low performance even if u have a good GPU. So yea OC cpu first for MWO UNLESS u know for a fact ur GPU is ur limiting factor.

[1Sascha]

Quote

Unigine Valley, Unigine heaven, 3DMark, 3Dmark 11. These are all good programs ot use when testing ur GPU.
Make sure ot also use a few game benchmarks. Metro LL benchmark, and Crysis benchmark are good.
And then ofc try a few games out.

Not really an expert here, but I'm not *quite* sure how reliable all those stability tests are when what you want is RL-stability.

I fiddled around with my new system for about a week or so until I found a stable (CPU) OC. I did tons of testing to ensure stability. Used Aida64, Passmark, 3DMark, PCMark and Unigine. The funny thing was that even after hours and hours of running said tests without problems, I'd sometimes still get BSD-ed when I jumped into MWONL (or other games). Only since I slightly reduced my OC (4.1 on cores 0/1, 4.2 on 2/3@1.25 VCore - didn't want to push voltage any higher than this) did I find stability with this CPU.


S.

[Catamount]

1Sascha, on 13 September 2013 - 01:04 AM, said:

Not really an expert here, but I'm not *quite* sure how reliable all those stability tests are when what you want is RL-stability.

I fiddled around with my new system for about a week or so until I found a stable (CPU) OC. I did tons of testing to ensure stability. Used Aida64, Passmark, 3DMark, PCMark and Unigine. The funny thing was that even after hours and hours of running said tests without problems, I'd sometimes still get BSD-ed when I jumped into MWONL (or other games). Only since I slightly reduced my OC (4.1 on cores 0/1, 4.2 on 2/3@1.25 VCore - didn't want to push voltage any higher than this) did I find stability with this CPU.


S.

A stress test is a stress test. It doesn't matter what's stressing your hardware. The trick is to get something that actually represents a harder load on your system than the software you'll be using.

The likes of Passmark and 3Dmark are NOT CPU stress tests; you want Prime95 for that. You will be hard pressed to find a CPU that's "Prime stable" across a multi-hour test that isn't stable in any game out there

For the CPUs, 3DMark is usually a reasonable test, but only if used correctly. A combination of 11 and the new 3DMark are really needed for a decent test, imo (Firestrike is a pretty good quick and dirty stress test, though, if one that's not long enough). For real stability, however, what you want is MSI Combustor or, even better, Furmark. That software is so stressful, that it can damage hardware if care isn't take to have proper temperature and good power; it's rare, but Furmark does kill GPUs.

Edited by Catamount, 13 September 2013 - 09:44 AM.

[ArmageddonKnight]

To test CPU OC's u should use Prime95, IntelBurnTest, OCCT, and AIDA 64.

I personaly use Prime95 the most and run 12 hour stress tests to fully make sure my CPU OC is ok. Prime 95 runs in 15 minute increments for each of its different size CPu stress tests, hence why it take so long to do a full run. infact iirc a 100% ful lrun is somthing like 16hours, but most i do is 12. usualy if it passes the 4 hour mark you are usualy ok.

The programs i mentioned in my other post are mainly used in GPU OC's, hence why u may have missed a CPU OC's instability, you were using the wrong programs and stressing the GPU instead of the CPU.

Edited by ArmageddonKnight, 13 September 2013 - 09:48 AM.

[Smokeyjedi]

ArmageddonKnight, on 13 September 2013 - 09:46 AM, said:

To test CPU OC's u should use Prime95, IntelBurnTest, OCCT, and AIDA 64.

I personaly use Prime95 the most and run 12 hour stress tests to fully make sure my CPU OC is ok. Prime 95 runs in 15 minute increments for each of its different size CPu stress tests, hence why it take so long to do a full run. infact iirc a 100% ful lrun is somthing like 16hours, but most i do is 12. usualy if it passes the 4 hour mark you are usualy ok.

The programs i mentioned in my other post are mainly used in GPU OC's, hence why u may have missed a CPU OC's instability, you were using the wrong programs and stressing the GPU instead of the CPU.

Don't forget Furmark for dedicated GPU flogging almost as fat and hefty as MWO itself....LOLZ great to run with prime 95 in tandem love dem 8cores......

[M4NTiC0R3X]

Catamount, on 08 September 2013 - 11:20 AM, said:

Could your GPU just be overheating? MWO isn't GPU-intensive in the sense that it takes a really powerful GPU to run it well, but it is in the sense that it produces a lot of heat in one's system.

This is true, the heat.. ever since 12v12 I realize my computer.... it's going to die. From overheating, but it's four years old it's about that time anyway. Hello Black Friday!

[1Sascha]

Quote

The likes of Passmark and 3Dmark are NOT CPU stress tests; you want Prime95 for that.

That's what I thought until I did some reading up on Prime and Haswells. According to multiple sources, you shouldn't use Prime on Haswells since it's not certified for the new CPUs. All the OC-related articles/videos I found recommended Aida64's torture test instead - which I used.

*shrug*

As long as my system keeps running stable, I'm a happy camper. I just found it odd that I could successfully and extensively stress test the system with all sorts of benchmarks/torture tests and then still get BSD-ed out of a game after 30 - 60 mins.

S.

[ArmageddonKnight]

Haswell is a ***** to get good clocks out of anyway becouse they are so hot. granted they run faster than ivy and sandy at the same clocks but getting 4.5ghz on a haswell requirs a h100 at the very least in 90% of casses, even then you gotta hope u get a above average CPU to keep voltage down.

IMO the transitions from Sandy to Ivy to Haswell has netted next to 0 performance gains in the real world. Reason i say that is, if u get a normal everyday air cooler like the CM 212Evo, or even a high end Aircooler like a NH-D14 or SilverArrow, the Sandy will get higher clocks than the Ivy and the Ivy will get higher clocks than the Haswell, all of which negate any clock for clock performance gain (~5%) each one once had over the previose.

NH-D14 level air cooler on 4 core 8 thread i7's.
2700k: 4.8 - 5.0ghz
3770k: 4.5 - 4.7ghz
4770k: 4.3 - 4.5ghz
UNLESS u delid Ivy and haswell. if you Delid u 'could' drasticaly improve temps thus OC's IF ur CPU stil has OC headroom (i.e more volts still help) But deliding isnt for every1. The only upside to Ivy and Haswell i see is that they can OC whislt using fast RAM 2400mhz+. Sandy doesnt tend to like going much higher than 2133mhz.(atleast i know thats the case for the LGA 2011 chips)

Edited by ArmageddonKnight, 14 September 2013 - 06:27 AM.

[Chavette]

Dont torture your machine with this game, wait untill dx11 so it has some half decent optimization.

[Catamount]

ArmageddonKnight, on 14 September 2013 - 06:24 AM, said:

Haswell is a ***** to get good clocks out of anyway becouse they are so hot. granted they run faster than ivy and sandy at the same clocks but getting 4.5ghz on a haswell requirs a h100 at the very least in 90% of casses, even then you gotta hope u get a above average CPU to keep voltage down.

IMO the transitions from Sandy to Ivy to Haswell has netted next to 0 performance gains in the real world. Reason i say that is, if u get a normal everyday air cooler like the CM 212Evo, or even a high end Aircooler like a NH-D14 or SilverArrow, the Sandy will get higher clocks than the Ivy and the Ivy will get higher clocks than the Haswell, all of which negate any clock for clock performance gain (~5%) each one once had over the previose.

NH-D14 level air cooler on 4 core 8 thread i7's.
2700k: 4.8 - 5.0ghz
3770k: 4.5 - 4.7ghz
4770k: 4.3 - 4.5ghz
UNLESS u delid Ivy and haswell. if you Delid u 'could' drasticaly improve temps thus OC's IF ur CPU stil has OC headroom (i.e more volts still help) But deliding isnt for every1. The only upside to Ivy and Haswell i see is that they can OC whislt using fast RAM 2400mhz+. Sandy doesnt tend to like going much higher than 2133mhz.(atleast i know thats the case for the LGA 2011 chips)

Absolutely positively correct

There has been effectively zero practical performance gain from SB to Haswell, even at stock clocks! 5-10% incremental differences amount to nothing that would ever be noticed in the real world. In fact, you'd be hard-pressed to find a gaming difference between CPUs with anything less than a 50% difference.

Add in overclocking differences, and you'd be hard-pressed to find any real advantage for Haswell, or its forseeable upgrades. I'm not saying Haswell is bad; it's just that Intel's offerings haven't progressed much, and have almost regressed if you consider OCing. I'd buy Haswell right now, and bought IB last year, but it's pretty much what you'd get if you bought Sandy Bridge :/

Edited by Catamount, 16 September 2013 - 05:57 PM.