3D11 secrets?

[aussie-revhead]

Hi guys, I am struggling to get decent scores from my GTX580/2600K combo, my best ever only once score was 6950 where mostly its in the low-mid 6000 range. I just did a run and scored 5900 with gpu at 920, while testing max clocks at various voltage levels I found my setup is very inconsistant, so what causes this? What tweaks/settings etc have you guys found for 3D11?

Any help would be greatly appreciated.

[Splave]

IMO there are none, fast gpu is it

[crossg]

3DM11 is pretty tough on the new cards. Is that score with Physx on or off? My best single score is here with Physx enabled. Used Nvidia inspector and upped the voltage to 1.138v. One notch higher with the core and it fails. I see a lot off people get better core speed with lower memory settings. I'm going to try the same settings with the Sandy at 54X and see if it increases at all. Posted the screenie in case you can't see the settings on line.

[aussie-revhead]

Come on guys, I have seen a wide variation of scores with similar hardwares and lesser gear beating better gear, what have you guys found to improve your scores? For example I know of one guy with a single GTx580 and a 2600K, his rig outscored my best by 300 points yet I tried up to 300Mhz more cpu clock, more gpu clock, tighter ram timings, looser rams clocked higher. I knocked out a better score with a 930 at 4400 than I did with a 2600K at near 5400 then he bumps his PB by 100 points ... something is going on here, is there any way to reduce tessellation in the benchmark like the AMD guys do? Can the old LOD tweak be implemented somehow (since AB is a reskin of rivatuner)?

[ZenEffect]

afik LOD will only work effectively with dx9 applications. LOD CAN be adjusted but you will sacrafice FPS as it has to be done through as "bug" in the driver that has to do with increasing fsaa and fxaa values iirc or was it fcsaa or mabye ftwomgbbq... w/e the point is that it didnt seem like a good idea once i looked into what it takes to adjust lod in dx10 and 11 applications.

[aussie-revhead]

No probs, thanks. I personally am chasing one guys score and I cant work out why he scores so much higher than me even when I outspec him in every respect..... anyway new rig coming together now so I will see how the 970 does.

Cheers.

[Q56_Monster]

None of the old school tweaks work with 11. CPU frequency doesn't matter either. As splave said, massively OC'd GPU FTW. SB....meh, overrated on 11. Low to mid 6000 is in the normal range for your core. Every 25mhz in core gives some nice gains. I haven't tested any bloomfields in 11, but looks like that shouldn't your problem. Some 920's at 4ghz have no problem hitting that score, so it really comes down to pushing your GPU....HARD

[chew*]

No probs, thanks. I personally am chasing one guys score and I cant work out why he scores so much higher than me even when I outspec him in every respect..... anyway new rig coming together now so I will see how the 970 does.

Cheers.

Sometimes screenshots do not tell you what frequencies a bench is run at........some guys downclock to look effecient.

Some are running diff clocks in each game test

Sometime the wrong cpu features enabled/disabled on a particular board can also impact scores.

[dinos22]

work the GPU clocks or volts when on LN2 in different tests and use SB platform if you are on single card

[aussie-revhead]

Guys what causes combined test to lock up on loading? Always for me the loading bar stops about an inch from the end and hangs a bit then jumps to the end, this may be normal but now with my current rig it often doesnt finish loading, it will sit for a while then system shuts down and resets. It was suggested to be a psu problem but its a new 850w so I doubt it, specially not while only loading. Also suggested was low vcore but Im running 1.325v for 4100 on my 970 so it seems high already right? Anyway I will try on SB again soon and see what happens.

I did manage P8139 with the new drivers and 970 at 4100 so Im sure there is more in it yet.

Cheers.

[SniperOZ]

here you go aussie this MAY fix your problem with combined test hang

http://www.3dmark.com/support/systeminfo-updates/

[aussie-revhead]

I think I may have found it, I changed alot of things last night to try to isolate this problem and the one thing that seemed to fix it was ram timings/clocks. With my ViperII's at 2000-8-8-8 I get hang most of the time even though it will pass sometimes and blend primes happily, with ram multi one notch slower or even at 2000-8-9-8 the problem seems to go away, I will test this further but thats how its looking now. I have a set of GT's to try too so maybe my E760 might like them better.

Afaik 3d11 wont run without the sys-info update anyway.

[aussie-revhead]

Guys, I have lost about 2-300 gpu points in 3D11 and Im not sure why, driver version is the same as previous, clocks are the same , yet Im down on all my scores, what can cause this?

[Hondacity]

use stretched..instead of center

[aussie-revhead]

Hi Rob, yeah I found this earlier when chasing low scores on my 1155 rig but this is a recent problem. It is stretched. I just found something though, I think my setup doesnt like my new Mushkin ram as I lowered the ram speed and got my score back. I always run 3D11 with BCLK up as high as I can as it helps physics and combined scores but while trying all sorts of things I set it to 1600 and gpu score came back. I got this at 5300 1032/4838 with ram at 1866-9-9-9:

P8504 http://3dmark.com/3dm11/1583687

Im disappointed with this cpu too, it takes 1.66v to pass at 5300, just failed at 5400/1.7v yet it can 1mb pi at 5650 with 1.63v ..... maybe both the cpu and gpu need some cold.

[Hondacity]

it needs to be cold....what is your cpu cooling method?

my cpu can do 5604 @ -35c

@ -30 max is only 59xx

[aussie-revhead]

The cpu and gpu are water cooled. I posted a WTB for pots for both and already got a reply, though it will be DICE only for now, its n00b friendly....

The 580 likes 1032 core as I can clock the mem up to its limit, at 1045 core I need mem at stock or the screen blanks out under load.

[Hondacity]

good to know pots are arriving..you'll love subzero

[aussie-revhead]

I tried DICE once on a 3520 I had, first time I saw a 5200Mhz screenshot for real, so it will be good to try it again, if only I didnt sell the pot so quicky .... your 580 will love the cold too Im sure.

I saw your latest 580 adventures too - amazing stuff.

I may have located a dual head phase unit too, but it couldnt hold sub-zero on a 3520 at 5ghz 1.47v so I dont think it will like my 2600K ......

[Poci]

For the sake of comparison with other end users benchies like 3d mark11 should just run at default p setting with default video driver settings.
Anyone can tweak the driver to get a higher score..... OCing on the other hand is the fun part.

[Hondacity]

oh really..tweak the drivers? how much increase in points?

and yeah go to hwbot.org

we run default p settings....

[Poci]

oh really..tweak the drivers? how much increase in points?

and yeah go to hwbot.org

we run default p settings....

Tweaking as in sacrificing image quality just for those extra points....

No way can anyone make comparisons if one guy has it tweaked one way and another the other. Totally different results.

Hence the need to leave drivers at factory defaults for benching.

Btw yes I have hwbot in favs..... for years.... its where I go to see what kind of OCs ppl are getting from certain batches of chips.....

[T_M]

Please go away

[Poci]

Please go away

May I ask what I have said that offended you so much? Or are you another benchie bragger?

[Hondacity]

lol