Ha, well I "think" I figured out my variance in scores.
Over the past few days I've since reloaded my system from scratch and just finished another battery of 3dMark tests (over 100).
If someone wants to help confirm these tests and prove my insanity, please do.
Bottom line, it looks like OC software (Afterburner, Precision X) is affecting high bandwidth with SLI HB.
It doesn't start to become noticeable in 3dMark until you start running Fire Strike Extreme and very obvious in Ultra.
I started testing other benchmarks such as UNiGiNE's Heaven and do see the hit in performance but don't have enough scores yet to show the variance.
Cards: x2 EVGA GTX1080 Founders Edition
All 3dMark scores are Graphics only with Graphics Test 1,2
Demo, Physics, Combined disabled.
All tests ran 6 times throwing out the lowest score and averaged.
Tested without/with OC Software, Single Card, Single Bridge, Nvidia SLI HB Bridge
OC Software: Precision X 6.0.1, Afterburner 4.30beta4
Fire StrikeSingle Card/SLI Disabled: 21673 (+/- 29)
Single SLI Bridge: 34964 (+/- 55)
SLI HB - No OC Software: 39336 (+/- 110)
SLI HB - OC Software: 39249 (+/- 95)
Fire Strike ExtremeSingle Card/SLI Disabled: 10259 (+/- 20)
Single SLI Bridge: 18321 (+/- 32)
SLI HB - No OC Software: 19249 (+/- 70)
SLI HB - OC Software: 18962 (+/- 80)
Fire Strike UltraSingle Card/SLI Disabled: 5104 (+/- 11)
Single SLI Bridge: 9415 (+/- 23)
SLI HB - No OC Software: 9564 (+/- 33)
SLI HB - OC Software: 9282 (+/- 36)
Observations:
1. Ultra scores are consistently worse with OC Software loaded at defaults than a single SLI Bridge.
2. No change with/without OC software loaded with single card/SLI disabled.
3. No change with/without OC software loaded with single bridge.
4. In each test run per configuration the variance is less than 1%.
5. Able to reproduce these scores and variances over a few days now and they are still less than 1% per configuration.
6. Rebooting -or- Exiting the OC software, disable SLI, enable SLI consistently gets me back into the "without OC Software" scores.
7. I tested my dual flex bridge config again and as everyone else is reporting, no difference with the Nvidia SLI HB bridge.
8. Same results with 368.25 or 368.39 drivers.
If the OC software is setting different defaults than a stock card settings I'd think it would show up in the single card or single bridge tests.
My concern is with SLI HB if we can't even get default settings in OC software to benchmark the same as stock card settings isn't that affecting our overall OC benchmarks?
Update: If you want to try to test this yourself use Precision X. While Afterburner is showing a hit in scores Precision X has been worse and more obvious.
post edited by Signal64 - 2016/06/30 12:34:48