SLI on triple screen, not gaining any performance

Rigel

250RPM
Original poster
Premium
Nov 12, 2010
250
394
31
Adding a second 2080 just got me 1 (one) more average FPS in the AC benchmark. MSI Afterburner shows both cards load at 97-98%.
All the screens are plugged in to the same GPU through DisplayPorts. I've just found out this might be wrong since here are specific instructions on how to connect monitors to the GTX GPUs; no info for the RTX ones though.

I've tried a bunch of different combination in nvidia inspector. The SLI configuration is currently set on "span displays with surround" since I can't get AC to use all three screens on "maximize 3d performance" mode.

Is it just a matter of using the right display outputs? What other things should I check?
 

Billy Pilgrim

1000RPM
Mar 25, 2014
1,566
412
I'm no expert, but can say that AC performs well with SLI - I know that from experience (used to use SLI). Span displays eith surround: iswhat I used, so it's right, i think. But I didn't have all display cables connected to one card so make sure you re sure about that
 
  • Like
Reactions: Rigel

Jens Roos

500RPM
Premium
Feb 23, 2017
895
1,290
48
I have 1080ti in SLI at 7680x1440, what are your performance using the AC benchmark?

For reference I have an average of 185fps going back 1 step on some settings, all max around 135fps.

You are using an hb bridge to connect your cards?

edit, i have used 3dp on one card or split on both, no bigger difference.
 

Jens Roos

500RPM
Premium
Feb 23, 2017
895
1,290
48
maximize 3d performanc - that is used for maximize performance on one display only
 

Rigel

250RPM
Original poster
Premium
Nov 12, 2010
250
394
31
I have 1080ti in SLI at 7680x1440, what are your performance using the AC benchmark?

For reference I have an average of 185fps going back 1 step on some settings, all max around 135fps.

You are using an hb bridge to connect your cards?

edit, i have used 3dp on one card or split on both, no bigger difference.
Here's the benchmark with one card - with SLI it's pretty much identical.
I'm using the nvidia nvlink to connect the cards.

IMG_20190127_124612-882x766.jpg
 

Jens Roos

500RPM
Premium
Feb 23, 2017
895
1,290
48
Well, that's lots of pixels..

5760x1080 = 6 220 800
7680x1440 = 11 059 200
11520x2160 = 24 883 200

AC, little reduction
POINTS: 27137
FPS: AVG=185 MIN=59 MAX=264 VARIANCE=19 CPU=82%

AC, all max
POINTS: 20117
FPS: AVG=137 MIN=51 MAX=234 VARIANCE=16 CPU=64%

I would first test your GPU performance to make sure nothing basic is wrong.
Passmark is an lightweight test program
https://www.passmark.com/products/pt.htm

Here your can see other result:
https://www.videocardbenchmark.net/high_end_gpus.html

Here is my result:
upload_2019-1-27_18-5-4.png


In this test I have about 20% gain compared to a single card, depending how well SLI is utilized there is normal to have 20-50% practical gain using SLI over a single card.

You could also test lowering your resolution and see how it impact performance, at least for the GTX, v-sync is almost mandatory as otherwise stutter of different kind are common.
 

Rigel

250RPM
Original poster
Premium
Nov 12, 2010
250
394
31
So I had access to the PC the other day, tried connecting the monitors in a similar way to what the nvidia website showed for the 1080ti (two to the main card, one to the other) and it worked! Now it runs well at 4k resolution.
 
  • Like
Reactions: Billy Pilgrim

Jens Roos

500RPM
Premium
Feb 23, 2017
895
1,290
48
I think it's related to something else unless the RTX is very different.

I have tried
3-in-top-no-in-bottom
2-in-top-1-in-bottom
1-in-top-2-in-bottom

I could notice a difference in gpu heat of the 2 cards, after I water cooled the card I have the monitors all in the top card and gained 10fps in AC benchmark, but that I think i more related to the fact that the gpu is 50 degrees instead of 84 when loaded 98-99%.

But it's great it working for you now, I understand the frustration when spending all that hard earned money and it doesn't deliver. Happy racing in 4k :thumbsup: :)