My sli msi gtx 980 ti +zotac 980ti refernce =bad results!?

[saeedkunna]

HELLO
I bought the msi gtx 980ti gaming and i was going to get another one but beofre i did i got a gift and it was zotac 980ti reference model so i canceled the seconed msi , i put the zotac under water since i already have a waterblock (it was for titan x ) and it was realy good i could easy overclock it to 1380 with stock voltage i didnt try to go highter also the msi perofrmed well in OC mode it reached 1370 with very good temps , now the problem when i put those tow in sli i get bad results in benchmarks and bad scaling in games gpu usage goes down on both GPUs and when i try to overcolck the zotac when they are in sli i get crashes unless i go down like +80 on gpu clock and +100 on memory?
i have tried diffrent drivers(353.30 and 353.38) but still the same my temps are good the watercooled zotac never goes higher than 40 and the msi max is 72 .
my setup is
windows 8.1
asus rampage v extrem x99
intel i7 5820 @stock water cooled
16 gb ddr4 gskill
psu evga supernova 1300 watt
samsung 256 ssd
2TB wd x3=6TB
the zotack is in pcie slot #1
and the msi is in #3
i use msi afterburner
i game @4k with every thing maxed
here is some results and games photos :

heaven bench



http://cdn.top4top.net/d_77194872f39.bmp

3dmark

http://up.top4top.net/







some games
http://up.top4top.net/





http://up.top4top.net/






http://up.top4top.net/

 

[Pandamonia]

50% scaling on the second card is all you can expect.

Bad move SLI 980ti.

 

[railven]

I wonder if Nvidia really abandon SLI optimization with their announcement of Nvidia Link.

Woof at that scaling. I reran my card last night at another suicide run. Scored 20889 on graphics. Was able to hit +305/+310 +20mv stable. 100% fan can never broke 60C.

Nvidia got some 'splaining to do!

 

[saeedkunna]

I wonder if Nvidia really abandon SLI optimization with their announcement of Nvidia Link.

Woof at that scaling. I reran my card last night at another suicide run. Scored 20889 on graphics. Was able to hit +305/+310 +20mv stable. 100% fan can never broke 60C.

Nvidia got some 'splaining to do!

which driver do you use?

 

[railven]

which driver do you use?

I'm on Win10 so using 353.50 which currently appears to be specifically for Win10.

Prior to that, I was using 353.38 without issue.

 

[guskline]

OP: Did you try OCing your cpu? It appears you have the 5820 at stock. 3.3 turbo to 3.6. Try setting it to a set 4Ghz OC and see if that helps. Also, it appears your gpus have different stock settings for core clock. I wonder if that is messing with your results? Do you have any other PCI e devices that could be eating up lanes. The 5820 has 28 PCIE lanes while the 5930/5960x have 40.

 

[saeedkunna]

1-i overclocked the cpu to 4Ghz still i get very bad gpu usage from both gpus (in the witcher 3 and Valley Benchmark 1.0)

2-i also set both pcie to gene 3

3-i have only the 2 gpu's so one is runing @16x and the other is 8x

 

[Termie]

A couple thoughts on this. First of all, people are talking about SLI scaling here, but unless you put up numbers with one card, there's absolutely no reason to start talking about scaling issues. I frankly don't think that's the issue. A few times GPU usage goes down, but often it's quite high in your screenshots. It's NEVER going to be 99% in SLI.

That doesn't mean there's not an issue, though, as your system only does about 10% better than mine scored with 780 Ti SLI in 3dMark, and it should be 40-50% better. Here's my score for reference - just focus on the Graphics Score which is the apples-to-apples comparison: http://www.3dmark.com/fs/3861805

Third, your cards are not running at reference speeds - they're actually running below the reference boost. 1164MHz isn't slow by any means, but reference boost is 1202MHz, at least on my 980 Ti. So that is a bit unusual, especially if these are overclocked models. I can't quite tell from the OP, but maybe you intentionally downclocked them due to crashes.

Honestly, the very first thing you should do at this point is pull one of the cards out and test a single card on its own. It sounds like the Zotac might be giving you problems.

 

[n0x1ous]

did we rule out vsync being on?

 

[saeedkunna]

A couple thoughts on this. First of all, people are talking about SLI scaling here, but unless you put up numbers with one card, there's absolutely no reason to start talking about scaling issues. I frankly don't think that's the issue. A few times GPU usage goes down, but often it's quite high in your screenshots. It's NEVER going to be 99% in SLI.

That doesn't mean there's not an issue, though, as your system only does about 10% better than mine scored with 780 Ti SLI in 3dMark, and it should be 40-50% better. Here's my score for reference - just focus on the Graphics Score which is the apples-to-apples comparison: http://www.3dmark.com/fs/3861805

Third, your cards are not running at reference speeds - they're actually running below the reference boost. 1164MHz isn't slow by any means, but reference boost is 1202MHz, at least on my 980 Ti. So that is a bit unusual, especially if these are overclocked models. I can't quite tell from the OP, but maybe you intentionally downclocked them due to crashes.

Honestly, the very first thing you should do at this point is pull one of the cards out and test a single card on its own. It sounds like the Zotac might be giving you problems.

no the zotac supose to boost 1076 mine reach 1164 without over clocking the msi is 1279 MHz so the msi is automaticly downclocking to match the reference zotac, here is tests for every card both scored very well

zotac:



msi gaming 6g:

did we rule out vsync being on?

if you mean in games setting i've tried to lower turn off everything i still get bad gpuusage and bad fps .

 

[n0x1ous]

If you had vsync on the gpu usage wouldnt be maxed if you were hitting monitor refresh rate regardless of in game settings. Make sure you don't have vsync set globally or something in the nvidia control panel

 

[Ketchup]

Hmm. Berryracer was noticing a similar issue, so I wonder if there is something to this, that is possibly driver related.

OP, if you look at the scores for all 3dmark tests, do you see that one score is substantially different between single card and SLI?

 

[Termie]

That Zotac model is honestly running under reference speeds, but otherwise, the single-card 3DMark scores are in the right range.

This is not a Vsync issue - Heaven is at 150fps.

Do we know of anyone that has 980 Ti SLI running correctly? It's so new that not many have tried it. The scaling is definitely bad, but the GPU usage doesn't explain it.

Let me give you all a concrete example from the OP's testing: 3dMark FS Graphics Test two scales from about 70fps to 110fps going from single-card to SLI. That's 55% scaling. Very bad for a GPU-limited test. But his GPU Usage is 97-99%. Thus, it's not a GPU usage problem. This is not at the hardware level, it's at the software level, and therefore it looks like drivers.

 

[saeedkunna]

nop all single results are good and very close the problem with sli .

 

[saeedkunna]

That Zotac model is honestly running under reference speeds, but otherwise, the single-card 3DMark scores are in the right range.

This is not a Vsync issue - Heaven is at 150fps.

Do we know of anyone that has 980 Ti SLI running correctly? It's so new that not many have tried it. The scaling is definitely bad, but the GPU usage doesn't explain it.

Let me give you all a concrete example from the OP's testing: 3dMark FS Graphics Test two scales from about 70fps to 110fps going from single-card to SLI. That's 55% scaling. Very bad for a GPU-limited test. But his GPU Usage is 97-99%. Thus, it's not a GPU usage problem. This is not at the hardware level, it's at the software level, and therefore it looks like drivers.

i've tried both 353.30 and 353.38 both give me same results but i agree it good be driver related

is there any chance that becouse one card is reference (the zotac) and the other one is custome pcb (the msi gaming 6g) they are not working very well in sli ?

 

[RaulF]

Try reseating the sli bridge.

Wish I could help more, I went through something similar about 2 years ago.

That's when I figure how much better crossfire was in my experience. SLI was always glitchy for me.

 

[saeedkunna]

i reseated the bridge and i did a littel bit over clock +80 to core and +100 to memorey
3dmark FS still not good http://www.3dmark.com/3dm/7830907?



heaven benchmarks
@1080



@4k

 

[moonbogg]

You are running firestrike standard. I hit a wall with standard. The difference between reference 980ti's and 980ti's at 1515mhz is about 400 points. That's it. A wall is hit because the res it just too low I think. You need to do firestrike extreme. Your standard score looks right for a mild overclock. I remember getting about 21k with a mild overclock.

My reference 980ti score was right around 20k, just like you. When overclocked all the way to 1515mhz, 2300+ is what I get as shown below. That's a small difference. Do firestrike extreme and compare it with mine below if you want. Specs are in sig.

 

[Termie]

Moonbogg - did you ever try running one card in FS Extreme? Just curious if you can confirm that good scaling occurs, and thus there is no driver issue.

I'm in the middle of moving some parts around, or else I'd try FS Extreme with my 980 Ti right now. Might be able to do this later today.

 

[saeedkunna]

this is at @1242 i tried to go to 1350 but it crashes @ end of the test

 

[moonbogg]

I'll run a single card tonight on firestrike. The OP's last score of 13k looks pretty normal I think for the clock speed he was at. My 15700 score was at 1515mhz.

 

[n0x1ous]

Just run firestrike ultra - then no worries on gpu's not getting used?

my single 980ti @ 1400 gets around 4500 on that.

 

[moonbogg]

I didn't have to time to analyze, but I did a quick run with my standard OC of +225 core and +300 mem.

 

[Azix]

Who do I need to know to get gifted a 980ti? Tell me

 

[Berryracer]

Hmm. Berryracer was noticing a similar issue, so I wonder if there is something to this, that is possibly driver related.

OP, if you look at the scores for all 3dmark tests, do you see that one score is substantially different between single card and SLI?

indeed, in Ice Storm 1.2 I get around 150K score but if I enable SLI that score drops down by 50%

I have a strong feeling it's a driver issue since we both are experiencing this with the 980 Ti.

Furthermore, please note that the 353.xx drivers all have horrible Throttling issues

See the rants here:

Welcome to the nVIDIA Throttling Zone