I have a single MSI GTX 970 Gamign4G and in bench tests, I can get within 11% of a GTX 980 in DX11, 10% in DX10, and direct compute test get within 0.1 % of a GTX 980. It took a +250 core, +460 mem, +18mv, K-boost ON. Fans at 100% temps were averaging around 64c. I also left "Power Target' at 100. DX9 games are WAY behind a GTX 980. I used to run a GTX 550ti that could always max everything in DX9 games. On my benching, the GTX 970 is something like 200-300% more powerful with DX9 games than a GTX 550ti. After OC'ing the 970, was within 40% of the GTX 980 in DX9 simple and complex tests in bench.
I believe the 970 can go higher core with maybe a drop in mem if needed, but my PSU isn't strong enough to support the amperage demand of the system. I'm ordering a new EVGA G2 1000w PSU, and will resume testing. Have a thread in the OC forum if interested in previous results (header is MSI GTX 970). Was also using an OLD EVGA PX software to OC, so might get better results with most recent update which I d/l yesterday, as they made a bunch of fixes to K-boost along with improvements to voltage settings, so I suspect better scores will be had.
Point is, I'm within 10%, DX10, 11% DX11, and direct compute is basically a tie with a GTX 980. I paid $320 (and got a free Witcher 3 game) but it jumped to $350 a day or so later. For nearly $200 LESS, I'm within 11% of a GTX 980 which I think I can get within 5-9% with more aggressive OCing as each +001 to core was chiseling the % down until PSU hit a wall. You just can't beat that value. I used "Performance Test 8.0" by Passmark to monitor changes with each adjustment as it tests DX9 Simple/Complex, DX10, DX11, and Direct Compute in a few minutes, then compared the results against 5 other cards with the 980 being top card in comparisons. Once watermark results were established, I ran it in Heaven and got a 3174 with a max 254.6 FPS, minimum frames were 27.4, though there's a bug when the test first starts that seems to drop the minimum immediately even though the test is still initializing as FPS immediately jump to 150+ but the 27 was already logged, so I suspect score would be higher if that wasn't the case, and average FPS were 126. Either way, stock everything with no OC nor K-boost enabled saw a 2791, min frames of 9.9 and max of 233, with average FPS being 110. Remember, I'm at the limit of my PSU, actually over the limit by nearly 10 amps on the 12v+ rail, so not even gaming until new PSU arrives, or I'll drop power target to 60% and underclock. Running stock everything and only K-boost, saw score go from 2791 to 2828, and min. frames jump from the lazy to respond 9.1 to 27.1. That was with NO OC. Just using K-boost which locks it at boosted clock rate of 1328 mhz in my case.
Anyways, it's pretty clear the GTX 970 is a beastly card if you a buy a better, non-reference variant and at minimum, run K-boost. For $320, you can run pretty much anything out now. But a 2nd and you are around $640 and performance jump is significant (run K-boost minimum). Or, you can save a $100 and by a single 980. Even if the 980 can get a 15% improvement with OC, I don't see the value compared to SLI'd GTX 970. Although, both cards are horrible at 2D tasks. Somehow a GT 630 walks all over them in the 2D world. DX9 shouldn't matter as like I said earlier, the 970 is double to to triple the performance of a 550ti which I used to own, and the 550 could easily max DX9 games (for the most part).
Also, whichever card/cards you get, get the best Gold or even better Platinum PSU you can afford. Look at the 12v+ rail amperage. If it says "up to, "MAX" or "peak", that is NOT what it'll do continuous. My current PSU is a 4 year old Thermaltake 600w. It's actual continuous rated on unit is 32amp 12v+. Looked on Thermaltake's site and they questionably changed the 32 amp cont. to 50 amp MAX. Very bad. That means cont. is 64% different than max. More efficient "Gold' or "Platinum" PSU might not be that far off, but even at a 64% difference between cont. amperage (what it'll do consistently) and "max" (what it can do in short boosts like power up), that means a 850w rated at 70.8 amp MAX, is putting out 45.312 amp cont. That's enough for a single card, but not SLI. I dunno if a 90% efficient PSU directly translates to less spread between max and cont. or not, but be VERY careful in the PSU you run. Better to have too much PSU than not enough, as it'll pull more current trying to stabilize the voltage, which is bad. People that buy a psu solely on rated wattage are making a mistake. The amperage output is what matters. I've seen people on Newegg and elsewhere say "It worked great for a week, then died, RMA'd it, new unit worked for a week then dies, I'll never buy another one these, went with an EVGA platinum PSU and all is well". The real reason it died is it didn't put out amperage, he smoked it, then smoked 2nd one, then by chance he picked one that DID supply enough cont. amperage.
Choose PSU carefully. Most only show MAX ratings in spec section, despite claiming to be "continuous" in the descriptions section. At very least, take "MAX' rating and multiply by .64 or 64% to show a ballpark. Better PSU like gold or platinum, can assume it's a little higher. If in doubt, contact PSU company directly and ask for continuous values on all 3 voltage rails i.e. 3v, 5v, 12,v as what's listed on most is MAX, meaning MAX. Not what it'll do all the time. Same thing as amplifiers and RMS wattage. That matters. "Peak or max" don't really matter and is a last resort before dying number.