GTX 970 SLI Review (Follow Up): Scores and Overclocking Journey

Now, I didn't post up scores in my last post. I had to redo it after an unfortunate thing happened to my RAM. One of my Corsair Vengeance Sticks Died so I'm down to 8 Gigs. In the process I swapped the card positions around and instead of MSI being the Top Card, the Palit returned its position as the top card for Temperature Control, as I found that with the MSI at the top, the temps become quite unbearable. As soon as I took it to the bottom, the Idle Temperatures of the MSI 4GD5T Armor Edition GTX 970 dropped Drastically by 25-30 Degrees C to around 31-34 Idle. I was able to get a much more thermal consistency which lead to more consistent overclocks and increased headroom.

Now I did quite a bit of overclocking tweaking and some of the legends you see are GTX 970 OC, GTX 970 SLI OC 1 and OC2

So how I tweaked them as follows:

GTX 970 OC: +222 Core Clock, 100 Memory.
GTX 970 SLI OC 1: 180 Core Clock, 294 MHz Memory, GDDR5 is at 7.6 GHz
GTX 970 SLI OC 2: 180 Core Clock, 394 MHz Memory, GDDR5 is at 7.8 GHz. - Final Stable Overclock

7.8GHz on the GDDR5! Fully Stable. That's very nice on SLI Cards!
I did have a GTX 780 in occasional scores, just to compare how much of a performance difference I get since that card.

Configuration of my Test System:

Core i7 2600K at 4432MHz
8GB RAM at 1648 MHz
128GB OCZ Vertex 4
2TB Western Digital Caviar Green
500GB Western Digital Caviar Blue
Corsair P8P67M-Pro
Corsair H70
1000W Corsair RM1000
Fractal Design Define R3
GPU 1: Palit GTX 970 "Vanilla"
GPU 2: MSI GTX 970 4GD5T "Armor Edition"

Swapped my GPU Positions around. Now Palit is the Top Card, and MSI is the Bottom Card

SOFTWARE BENCHMARK SCORES
3D Mark Firestrike: Standard, Extreme, Ultra
Catzilla ALL Benchmark

Seeing about 50-65% increase scaling all around. Not a good proportion but you can't get everything. It does however respond extremely well once I start tweaking. I am real chuffed to get 16,000 on FS Standard after overclocking to its current speeds.  I only had tested Ultra on SLI GTX 970s as prior to installing the SLI Config, the Ultra Benchmark mode hasn't been released yet. Still the scores would suck if I were to test FireStrike Ultra on a single card so I didn't bother. The 4788 score which I did post previously came as a surprise, after the first overclock I hit, but since then even with all the tweaks I can't get to beat that score.

*Update! I hit a stable 4896 after dialling my OC Memory back a bit! It was a good score nonetheless, thought you should know!

METRO LAST LIGHT

1080 HD Maxed, No SSAA

1080 HD Maxed with SSAA

A Very demanding benchmark indeed. I was initially having problems benching Super Sampling a single Palit Card, it crashed if I applied an OC. When I did OC the SLI setup, I could launch this without a hitch, so yeah no OC Single Card results. The score gap widens as I tune my GPUs. Notice at stock, the GTX 970 already beats the already OC-ed Zotac 780 I used to have. There wasn't good scaling from stock GTX 970 to GTX 970 SLI,  Very likely due to drivers. But tweaking to OC 2 Clocks nets some noticeable performance gains, especially Minimum Framerates.

BIOSHOCK INFINITE:
Ultra DX11, DOF

I didn't have stock GTX 970 SLIs as I completely forgot about them. However, that said, adding a second card really bumps Maximum framerates through the roof. It does respond well to the hefty overclocks I added in any case.

VALLEY BENCHMARK

I did show scores on 5760 x 1080 on Valley and I did say that having that on 3 screens looks absolutely stunning. However I didn't have scores for a single card config. On 1920 x 1080p, the GTX 970 OC had been benchmarked way before. So for a more apples to apples comparison. Apologies, no stock scores!

CONCLUSION

Now Overclocking was an absolute dream, a score of 16,143 on Firestrike Standard and 22,211 on Catzilla 1080p is the sort of results that I really should have! Again, 2 GTX 970s really is a sweet configuration for its price. So if you have the time, spend 2-3 days tweaking here and there. Now the scores will be better if I had a better CPU as it's currently the bottleneck atm. 1080p overkill? Depending on the game, no. Not so much, I thought this will be overkill, but I pulled my statement.

I did try 8GHz on the Memory but it started to artifact by then. So for now I dialled my clocks a bit back to 7.6GHZ on Memory and +175MHz on the Core just to get more stability.

We're talking in total 3328 Maxwell CUDA Cores, whose shader performance is equivalent to 5293 Cuda Cores according to Game-Debate, and near 500GB/s of Memory bandwidth. The bandwidth though is not optimal, seeing that this is on a PCI-E 2.0 based board. That said, hopefully I don't have to upgrade my Video Cards for at least another 3-4 years down the road, when that happens I won't sell it but I'll delegate it to something else once the future architectures are worth diving into.

That said, one 970 is good. Two is even better for the price. A quick update when the GTX 960 was released, I made a point why the 970s are so successful at LinusTechTip's Facebook post of the GTX 960 SLI video. The 970 has been an ideal value for 1440p gaming thanks to double bus width and double memory buffer and 2 970s in SLI are the cheapest gate to playable 4K gaming at near Maximum settings. Gaming currently at 1080p is overkill but that's all I got for now.