MSI GTX 970 Gaming4G OC'd

[BobCharlie]

I wanted to see how close to a 980 I could get with my 970, so spent the past hour testing. Used Passmark's Performance 3D test with it's generic settings, which aren't extremely intensive as others, but gives quick A to B results in a wide category so you can at least see if tweaks are helping or not. Goal wasn't a stress test, but to see how close it could go. I only stopped as I didn't want to go beyond +18mv, and figured further improvements would be minor w/o voltage bumps. Very last two tests ONLY had a voltage bump (+12mv to +18mv), but were otherwise identical settings and that alone saw an insane 8.1% drop in difference to 11.9% between the 980 (based on Passmark's 980 benchmark) and 970 in DX11 with a crazy 0.8% to 0.1% with Direct Compute, and a jump from 1578mhz to 1591mhz. Being 0.9 away from the 1.6ghz mark was close enough for me. Took photos of each test as proof.

This thing can seriously OC and scale nicely while doing it! Must have one of the good OC'rs finally. DX9 related tests barely fluctuated for some ODD reason, but it really doesn't matter as my old GTX 550ti could max those and the 970 is 2-3 times stronger with DX9. DX 10, 11, & Direct Compute improved GREATLY with almost every increase. 2D however, the GT 630 literally wiped the floor with this and even the 980 in so many tests it's embarrassing. I dunno if nvidia decided to sacrifice 2D performance in favor of expanding 3D capabilities, or if someone is seriously asleep in the coding department for the drivers, as even my old GTX 550ti killed both of these cards with 2D performance most of the time which is perplexing and sad given these monsters should be able to at least MATCH 5 to 7+ year old, now "low-end" cards, let alone get trounced by them. I never hit a wall here other than an early drop at +500 memory, meaning there's better numbers to be had. I wasn't comfortable going with higher voltages but figure there was plenty to still be picked up for someone more daring, though I might revisit this in the future and see how it holds up under demanding games, but I'll wait until I get a more powerful PSU. Looking at the last example, I was within 10% of DX10, 11.9% of DX11, and 0.1% Direct compute with a 980. That's utterly amazing given it can go further. Basically within 10% of a 980 on modern games with ability to Direct Compute at the same level is incredible. Temps never went higher than 36c with fans 100%, so could definitely push further. I do also have a secondary fan blowing across the heat exchange pipe on the video (from the side) to keep air circulating as the pipes get quite warm. This has dropped idle temps as much as 6c when main fans aren't running.

Passmark's Performance 3D test
Baseline: Stock settings all around.

3D Graphics Mark
970=7728
980=10341 (33.8%)
DX9 Simple
970=535
980=799 (49.3%)
DX9 Complex
970=121.4
980=174.8 (43.9%)
DX10
970=132.2
980=164 (24%)
DX11
970=178.9
980=222.1 (24.2%)
Direct Compute
970=4132
980=4848 (17.3%)



+205 core & +450 mem @12mv

3D Graphics Mark
970=7965
980=10341 (29.8%)
DX9 Simple
970=540
980=799 (47.8%)
DX9 Complex
970=117.8
980=174.8 (48.3%)
DX10
970=146
980=164 (12%)
DX11
970=189.2
980=222.1 (17.4%)
Direct Compute
970=4695
980=4848 (3.3%)



+240 core & +425 mem @12mv

3D Graphics Mark
970=7957
980=10341 (30%)
DX9 Simple
970=535
980=799 (49.3%)
DX9 Complex
970=117.6
980=174.8 (48.6%)
DX10
970=147.1
980=164 (11.5%)
DX11
970=187.9
980=222.1 (18.2%)
Direct Compute
970=4774
980=4848 (1.6%)



+245 core & +450 mem @12mv

3D Graphics Mark
970=8083
980=10341 (27.9%)
DX9 Simple
970=537
980=799 (48.6%)
DX9 Complex
970=121.7
980=174.8 (43.6%)
DX10
970=148.2
980=164 (10.6%)
DX11
970=188.7
980=222.1 (17.7%)
Direct Compute
970=4786
980=4848 (1.3%)



+ 250 core & + 460 mem @12mv (1578mhz)

3D Graphics Mark
970=7938
980=10341 (30.3%)
DX9 Simple
970=534
980=799 (49.4%)
DX9 Complex
970=117.5
980=174.8 (48.7%)
DX10
970=147.5
980=164 (11.2%)
DX11
970=185.1
980=222.1 (20%)
Direct Compute
970=4809
980=4848 (0.8%)




BEST SETTINGS of test
+ 250 core & + 460 mem @18mv (1591mhz)

3D Graphics Mark
970=8064
980=10341 (28.2%)
DX9 Simple
970=535
980=799 (49.3%)
DX9 Complex
970=117.6
980=174.8 (48.6%)
DX10
970=148.7
980=164 (10.3%)
DX11
970=198.4
980=222.1 (11.9%)
Direct Compute
970=4845
980=4848 (0.1%)


EDIT: If anyone wants to see the actual screen shots of the tests, say something here and I'll upload them later. Didn't realize you couldn't add attachments directly from local media.I'm too tired to mess with photobucket right now.

 

[BobCharlie]

Absolutely no interest in these results? Thought they were fairly impressive considering some people seem to sputter at +220 core while claiming a high+500 to+550 RAM. While RAM at +500 saw a decrease in numbers, essentially going backwards 1-2% when core was at +205 and +12mv, having mem in the low to mid 400's saw actual improvements.


Hopefully my results will help others OC'ing theirs for the 1st time. In this line of tests, it was clear bumping voltage wasn't needed. Also, if you hit early walls on your card, consider the fact that not all cards are created equal, just like not all CPU are equal. With CPU, like the AMD Athlon II series, they'd start out creating the TOTL CPU. Some had defective areas in a one core, but remaining 3 cores were OK. Rather than junk it, they locked out the faulty core out, and made it a 3 core CPU. Some had an issue where the 3rd core wasn't 100%, so it was deactivated in the BIOS, making it a 2 core, where it would run 100% perfectly. Then there was a huge explosion of curiosity when "unlocking" that "hidden" core got people excited because some were good enough to game on, and they got a "free" upgrade essentially.


As far as GPU go, not all are equal. I think the best ones from the mold that test the best, go to select companies, and from there, where they end up in their line-up is quantitative of how good they essentially, with the best ones receiving the higher OC standard, and in case of these GTX 970, getting the excellent heat sinks with the exchange heat pies that help dissipate heat much much better than others w/o them. Seriously, it maintains cool enough temps under low loads it can shut the fans off completely, which a is a little trippy as it's the first card I've owned able to do that. And will hover in the low 30c range. Run fans forced @ 100% and temps are around 23c. I do have about 4 fans for case venting not including an extra 2 fans doing a pull/push on both sides of the CPU cooler's radiator so have around 6 fans total, plus the 2 on the actual video card so 8 fans in all. I do have one of my smaller 3" fans custom mounted blowing on the side of the GPU, so it'll drop the heat pipe exchanger's temps when the main fans are off. This helps tremendously as the pipe is hot to to the touch at idle w/o the extra fan. With the fan, it's at near room temps. I saw nearly a -8c drop in idle temps from positioning the fan there. Had a second, smaller one on the pipe nipples but it lost a blade during adjustment.

Anyways, you can clearly see why spending that extra $20-30+ to go with the TOTL variant within the company's line up that come with a slight OC, is actually beneficial if plan on OC'ing it, as the odds are it tested best with the what the company was given from nvidia hence the confidence in OCing beyond the reference numbers and locking it in at theat rate. Plus they get more for it. Seems I got a really good OC'r


I'll try Heaven and post those results as I just downloaded it and will see how it goes there, as that's all anyone seems to care much about. I'll also upload the previous test screen shots with Passmark's testing as I took a pic with each test I mentioned in initial post and will add Heaven's too while at it (assuming it survive's Heaven). I was really tired the day I did the other tests and wasn't too interested in going up with voltages as was too tired/lazy to research other's voltages or what might be safe. I think I feel confident now to try a little a more to keep it stable and monitor max temps too. Thanks

 

[SteffzStarkz]

Nice work!

Do you see much fps increase in games?

 

[BobCharlie]

Nice work!

Do you see much fps increase in games?

Haven't tried gaming yet with OC'd settings. Was only using EVGA's "K-Boost" and 100% fans with GTA V, which is only game I've played since buying the card last week. Haven't played GTA V (only major PC I'm playing at the moment) since before I did the Passmark testing on May 1st as I've been busy. I originally wasn't even going to bother OCing this, but after seeing the huge gains reported in the forum, I thought I'd give it a try. Also, with having just bought this, wanted to see where it was at on it's own 1st compared to my 5 year old GTX 550ti was GTA V i.e. how many more setting could I enable, etc.

Anyways, did three back to back tests with Heaven just now. Gotta be the best looking bench test I've seen yet. Testing was done at my native resolution of 1920 x 1080. I left settings within Heaven whatever they are when you open it (i.e DX11, Quality high, Tessellation OFF, no stereo 3D, no multi-monitor, AA OFF, and full screen ON), selected the "RUN" button, then selected "Benchmark" so whatever it leaves off by default was left alone as I figure it'll match what everyone else benches at so the scores will reflect other's better. I'm not interested in maxing the settings at this point within Heaven, just seeing what highest possible OC is and providing the results. Once I find absolute max OC ability (will take awhile), I'll go ahead and max quality, AA, and Tessellation and do a stock vs. max OC for those interested in these things 😉 and to compare against their own cards.

Also, I used EVGA Precision X. for OC duty as I like it best as it has the wonderful K-boost, which locks the card to it's core speed w/o going into energy mode. It's the single best thing next to OCing as the card won't drop clock speed because it senses a 2D image in-game where it'll drop clocks to save energy as it would while opening folders on the desktop. With this OFF in a normal game, stuttering can occur as the core clock will fluctuate. If you still don't understand what K-Boost does, picture a gasoline engine car that makes differing levels of hp/tq depending on engine RPM, which changes with road speed and gearing, in heavy stop/go traffic on say a highway in Germany at rush hour, where your speeds go from 20 mph for a bit to 70 mph, to near stops with a couple spurts of 140 mph. Now picture a Tesla with an electric engine that makes peak hp/tq regardless of road speed, running flat-out on the Autobahn with no traffic doing 140 mph. That's what K-boost is like.

OK, now for the Heaven Test Results!

Test 1: GPU settings STOCK, other than 100% fans. Left K-Boost OFF

Results were less than stellar with a 2791 score. Min FPS dropped to a disgusting 9.1 and max were 233.7. Noticed some stuttering about halfway through as well, but forget to check which scene number it was. Think it was from atop one of the spires with the foreground in focus and background blurred briefly, and noticed it happened again near the end briefly. Couldn't tell if it was the card's eco settings or the program. Also noticed the test hits you with a FPS penalty drop at the very start of the test which should be patched or a delay added in as before the scene is even showing, you already have a min drop despite it soaring to like 150+ as the scene moves.


Test 2: GPU settings STOCK, K-Boost ON, Fans 100%

This is WHY I love K-Boost. Although I will note, that just like in the Passmark testing earlier, turning K-Boost on with this card locks the clock speed at 1328 mhz, which is technically an OC'd speed, I have ZERO understanding WHY it's happening, as the card's actual speeds are listed as 1140 MHz Core Clock, with a 1279 MHz Boost Clock. My GTX 550ti with K-Boost would lock in at core clock. This is my first GPU with a 'boost" clock ability, but I'd think it should be locking in at that 1279 MHz or the 1140 MHz. No idea where the 1328 MHz is ciming from as it's not listed anywhere in either the box or specs on Newegg. I noticed once you go pass +12mv, like +12mv to +18mv, THAT will cause the core clock frequency to increase by itself despite being the only thing adjusted (common knowledge), which suggests some sort of voltage to core clock ratio might be present? Maybe when K-Boost locks the clock rate of I'm guessing 1279 MHz, it's removing a soft voltage cap when the eco settings are on, voltage actually increases, and 1279 becomes 1328?? They are close enough it's possible as the jump from +12mv to +18mv in Passmark test showed similar increase.

Anyways, with K-Boost being the ONLY setting change, I saw a huge improvement with a minimum FPS of 27.1, and max dropped slightly to a 232.8. Score jumped to a more realistic 2828. Temps stayed in the mid 50c range around 54-55 on average.

Test 3: +250 core, +460 mem, +18mv, K-Boost ON, Fans 100%,

Using the results from the Passmark testing earlier, I went straight for gold and tried the highest settings I used before stopping earlier. I fully expected it to crash to desktop, hang, or do something else equally saddening. But it just wasn't to be. Minimum FPS were near identical at 27.4 (I still question the validity of the minimum FPS as that penalty at the beginning while it's initializing the test seems like a developer bug), max FPS jumped to 254.6, which is about a 22 FPS jump. Score was now a more impressive 3174. Temps averaged around 63c throughout test.


I'm going to try a little more testing in Passmark (it's quicker for the 3D tests and will show if it'll at least run or not while seeing exactly what's improving, where, and by how much. Then I'll try those settings with Heaven and see what the absolute highest I can get with this is and report back. I thought for certain I'd need much higher than a +18mv to run on a more intensive bench test like Heaven, but it's apparently not the case. After skimming other's results, I suspect they are running their memory too high and are actually getting lower results than they would otherwise, along with a higher voltage requirement; remember, I stated with a +205 core going from +450 mem to +500 mem saw every single parameter in Passmark drop by 1-2%, so I backed down to +425 mem and ramped up the core instead, which paid off in dividends with each test, while I slowly bumped mem up at times. I've noticed many people were barely able to hit +225 core with many stuck around +200, which by all means is still a lot, but have a feeling their mem settings were killing it as I'm at +250 core. Then again it could be their power supply, motherboard's actual limitations, etc. I'm also not running a CPU OC at the moment for what it's worth., and I'm NOT touching the "Power Target" in EVGA either. Been leaving it at 100 which is default. With K-Boost on, I don't think it's needed. The only time I tried changing that 110%, saw the only crash in Passmark.

I suggest downloading Passmark's bench test suite (it tests everything from RAM, to CPU, GPU, Hard drive, etc.) and running the 30-day trial period. It's not as pretty as Heaven, but it's fast and gives results against other cards from low-end to mid-end to a 980 being top end (at least this is what mine does) and it takes 2 minutes to run just the 3D GPU tests which covers DX9 through DX11 (I listed all the results at first post to give an idea, but left weaker cards being compared against for obvious reasons. For giggles, do the 2D GPU test and watch a lowly GT 630 rape your 900 series nvidia ;(

I'll get into my photobucket account today (been awhile) and upload all the screen shots from Passmark (this is a more recent Passmark btw, as I seem to remember an older version that was different. Once 30 days is up, you can still do the GPU tests, etc.) tests earlier along with all the Heaven screen shots. I saved the .html links to the Heaven tests as well and will try and see if there's an easy way to upload them so you can see the the pics are genuine.

Anyways, off for some more testing!! This card is the ultimate OC's far as I'm concerned. Oh yeah, wanted to ask, I noticed people mentioned Heaven can supposedly check for artifacts? Where is that in the settings?

 

[BobCharlie]

Alright, this was frustrating. Used the +250 core +460 mem +18mv as the watermark, I tried increasing the core by 10, got a crash. Tried lowering mem, lost performance on the DX10 and DX11. Tried even lower memory, then bumped core (+257 core +455 mem +25mv and everything improved except DX11 which degraded enough I wouldn't want to run it. For the next hour or so, everything I tried either crashed on the DX11 test, causing a driver crash, or would improve DX9 stuff (by as much as 5% at one point). Finally got the idea to put the settings back at the watermark i.e. +250 +460 +18mv, and saw that DX11 was now at nearly 18% difference from a GTX980 whereas yesterday it was at 11.9%. Tried a 2nd time with +249 and +459, and it crashed on that. So something was clearly off. Rebooted PC just in case the last driver crash did something, and stuck one of my room fans up against PC (I leave cover off normally). Tried again with +249 and +459, and DX11 was back at being within 12% of a GTX 980 and DX10 and Direct Compute were exactly 1% off from where they were with the watermark test.


If I had to guess, I'm at the limit of my cheap power supply which might have been experiencing some sort of heat soak, as that's the weak link at this point. I had tried rebooting twice before as well during the testing, and DX11 remained high (i.e. worse). Once the fan was blowing in, it was within 1% of watermark, so it's making sense. I have a ton of fans in the case, but PS only has it's one fan. I did a complete dust blow out about 2 weeks ago. Unfortunately, wasted something like 4 hours on this w/o realizing.

Obviously can't do anymore testing as it won't be reliable until a new power supply is ordered. Will come back when that's done. Also, in case you were wondering which Passmark bench test I was using, it's their "Performance Test 8.0".

 

[BobCharlie]

Ended up ordering an EVGA Supernova G2 1000w power supply 😉 Supposed to arrive later today (Monday). Assuming it can get shoehorned into the case (one way or another it'll fit), I'll update later on the OC. I realize the card is essentially "old news" to most, but it's still Christmas for here and figured it'll be helpful to someone at some point, and given how well it was doing before, I think it's fun to talk about.

 

[BobCharlie]

Did my 1st test since power supply upgrade with the "watermark" settings (+250, +460, +18mv) and something odd and quite simply amazing happened.

DX9 simple 615 (29.9% less than 980)
DX9 (complex) 110.5 (58.1% less)
DX10 162.5 )0.9% less)
DX11 184.3 (20.5%)
Direct Compute 4854 vs a 980's 4848!! Finally went neg as in -0.1%.

Pretty amazing how a weak power supply alone can alter numbers. Not sure what's going on with DCX11 & DX9 complex. I'm wondering if my "Global" settings within nvidia panel are messing with the numbers? I'll try defaulting global and try again, then bump voltage just in case. Still floored by these numbers though. I mean with DX10 games as-is, it's identical to a 980 and can compute just as fast too. Simply astonished.

 

[BobCharlie]

WOW!! Impressive turnaround results! I went into global nvidia settings, and either turned everything OFF, or set to "performance", NO DSR, etc. Then, went back and upped voltage to +25mv. Dunno if the voltage was main reason for such a huge turnaround or if it was indeed the global settings. I just didn't feel like running back to back tests, but it can possibly help the next guy regardless. I'm in utter disbelief though after testing.

Passmark's Performance Test 8.0 is great as it's actually telling you something if you know how to read the results and pay attention to what's changing and what's not. If you score low on DX9 complex + DX11, but everything else is OK, up voltage a notch and re-test and might as well set global settings in nvidia to performance or OFF as I think it might have been influencing the scores. Odds are, the card's getting starved but not enough to crash with a set too low voltage. DX11 FPS skyrocketed in-test, and never dipped below 173 FPS when the giant jellyfish shows up. You'll also know if the OC is bad if it freezes and stops DX11 test on the giant Jellyfish. Here's a pic as I finally got something into photobucket:


+250 core, +460 memory, +25mv, k=boost ON, v-sync OFF (very important to have it off for benching scores) and no frame limit in EVGA PX 16, etc.

Want to point out, how drastic of change there was changing power supplies and might as well ignore all testing prior to this one as the results weren't accurate with weak PSU. Your 700w PSU might actually be producing less cont. amperage than you think, causing for either erratic tests, early crashes, really LOW offsets, etc. My Thermaltake, with it's 32 amp cont. and 50 amp "max", ran this card, but it suffered when pushing the OC offsets and going over 12mv. It was having to dip into it's "max" reserves constantly to try and keep up. I did the math and 50 amps is 64% over 32 amps. Applying the same 64% to this PSU (EVGA Supernova 1000w G2) I ended up with 53.3 amps continuous. That's 11 amps more than recommended. If I'd gone with a 750w PSU like this http://www.newegg.com/Product/Product.aspx?Item=N82E16817139006
it's rated at 62 amp MAX. Using the 64% difference, that's roughly 39.68 amps continuous!! Using say a 10% +/- difference, 39.68 +10% = 3.968 amp increase on continuous, or 43.66 amps.That would be 1 amp over what MSI suggests at STOCK speeds! No head room there.

If you can't get your GTX 970or 980 to OC, look further into you power supply. It's better to have power to spare than not enough. It'll only use what it needs 😉

Now, off to get a Heaven test out of the way to see where that score is now.

 

[BobCharlie]

And here's the Heaven test :

Again, pretty significant jump over previous test. Minimum FPS didn't crash to low numbers at onset of bench test this time either. It wasn't until about half way through they dipped to 30 FPS.

Tell me people, are these good scores or what? I have no idea what most of you are getting out there. As can be seen with Heaven pic, I'm running 1920 x 1080 with default settings. Are these good scores? Looking for input as there hasn't been much in the way of responses with this yet.

I'm going to go back to Passmark to see how high I can get it. I won't list all, only highest. Will come back with Passmark screen shot and another Heaven screen shot. Before, hit an absolute wall trying to go higher, but I think it can do it with this power supply as it seemed mainly to crash on DX11 Passmark's giant Jellyfish before (seemed like it *could* go higher as it was passing most everything else) which clearly needs more juice to render (I tried earlier tonight with +250 core, +0 mem and stock voltage, and it crashed on giant jellyfish once more, like it did being pushed too hard with old PSU, so it infers a voltage bump is needed if goes fine through all tests until giant jellyfish shows up. Wish me luck!

 

[BobCharlie]

Alright, highest actual stable OC was +260, +460, +43mv, k-boost on. I was able to eventually take core up to 470, but at 50mv, and it crashed on the jellyfish in Passmark's DX11 tester. I wasn't willing to go any higher with voltage. While overall Passmark score increased, DX10, 11, and DC either stayed the same or decreased slightly at this OC. DX9 simple and complex improved, though the card's stock speed can slaughter any DX9 game, so why bother.

Passmark's "Performance Tester 8.0" is an excellent, quick tester for GPUs. Do the 30 day trial, after it expires, you can still do the test benching for video cards but can't save anything. Screen shot is only option. I say it's excellent because it tests DX 9, 10, and 11 very quickly if you select all of them which is more important as you can see immediately what's changing and what isn't, for better or worse, and can use that to get a feel of where to go, whether voltage, less memory, more core, etc. DX 11 was the only one that would sometimes crash due to the OC. Rather than sit through ALL tests, I'd just select the DX11 test, and go at it until either stable or improvement occurred, then would run ALL 3D tests. The results with this simple and fast DX11 bench, translate directly to whether it'll score well with Heaven or not. That highest OC was actually -1% from the +250, +460, +25mv watermark. Heaven was -1 point with virtually identical numbers so both benches came to same conclusion, just one did so faster. Here's that Heaven shot:

+252, +460, +25mv was best I could get w/o DX 10 and 11 w/o getting worse. Sometime the DX11 test would be +/- 2 points. It wasn't enough to warrant a Heaven bench though as it was hardly enough of a change. Here's that :

And just to show you how lousy these cards are at "2D" benching, take a look at this comical score. I actually had a high OC +265, +465, +50mv that was stable enough for the 2D, so keep that in mind when comparing against the GT 630 beast and the GTX 550ti:

My take away after spending numerous hours OCing it are:

1. Your power supply MUST be up for it. You need to know it's continuous amperage rating (max amp x 0.64 = rough estimate of continuous to get you in ballpark) is high enough. If card company says 42 amp, give yourself some headroom. If your highest OC is +200 core, something isn't up to snuff. The fact EVGA for example wouldn't even admit to or disclose "continuous ratings", says something right there. When I forced their rep into a corner about it, he outright lied about the capability of my PSU with regards to SLI and how much amperage was needed. Play it safe and go larger if you plan to upgrade.

2. Don't go overboard with voltage. I've seen people say "I upped voltage to +81mv with a +200 core. Do NOT do that. Voltage should be the LAST thing you touch, and ONLY after core isn't stable. With old power supply, I didn't touch voltage until +240 core with +425 memory. +205 core & +450 mem still had stock voltage!! Also, voltage adjustments with this card at high OC was like fine-tuning a guitar string. Too tight or too loose hurt bench scores. Only increment ONE pre-set at a time, and only after crashes with core increments. Also, keep in mind if you are running an OLD motherboard or one that's not very great, with lousy RAM sticks, etc. you might not be able to OC very far, especially if you ignore no.1 above.

3. Don't go overboard with memory. I saw plenty of people saying "I have memory OC'd to +500". Well, guess what? You are probably benching a lower score because of that and aren't even realizing it. I went from +425, to +450 and improvements. Then tried +500 memory. All offsets were stable, but bench scores actually DROPPED going from +450 to +500!! I backed off to +450, got core high enough, then bumped mem in increments of +10. +460 was sweet spot for this card. Going to +462 did not help it, nor did anything higher but it CAN go higher.

4. Core you can go pretty crazy with. If you can't go over +200 on core, upgrade the rest of your system to be on-par with the quality of the card. If you have the cheapest reference card, chances are it was a lower quality piece off the assembly line, hence it's being sold for less as it couldn't reliably take a factory OC. If OCing the card is your goal, spend the extra $30+ and get one that's either known to OC well, or already has an OC applied to it.. So far, I have to say the "Gaming 4G" GTX 970 from MSI is a top shelf GPU within the 970 range based on how well it OC'd. Even running the highest OC with the +50mv, it never went pass 58c with fans at 100%, but ambient in-room was chilly.

5. K-Boost is your BEST friend. It'll stop the card from bouncing clock speeds to desktop speeds, it'll greatly improve in-game response, reducing stuttering, and at least with the case of the MSI 970, gives you a tiny overclock over your boost core but don't expect that with older cards as it won't OC them. My bench scores improved significantly between having 100% stock card settings to turning it on. Minimum FPS went from 9 to 27 in Heaven with only k-boost. That says something right there.

6. Don't rely on one bench test that only measures DX11 like Heaven. Use something like Passmark's "Performance Test 8.0" and run the gambit of 3D tests while paying attention to each adjustment you make, get the DX11 and DX10 as high as they'l go, then use Heaven or another that tests only DX11. I noticed that at some point, the card didn't want to improve DX10 and 11 and either stayed the same, or decreased, while DX9 increased. It isn't worth it. I was able to actually hit 1.6ghz, but it ran better slightly under around 1589 mhz.

7. Run the fans manually on high OC. Fans are fairly quiet, and better to keep card cool with a touch of noise, vs. it dying 2 years prematurely due to heat. I noticed it ran better under 50c. Once it hit 50c, it typically dropped a point. And no, power management wasn't to blame as it was set to 86c un-linked.

8. "Power Management" to 110% did absolutely NOTHING with this card when OC'd. Score did not change when I tried setting it to 110%. It might help or do something when k-boost is off, I dunno. All I do know is I left it at "100" the entire time. I did in-link it to set the temps a little higher just in case as it'll throttle the gpu if temps hit the number you set.

9. Try and keep in-game memory usage to 3.5ghz, or you'll end up with a performance hit. Also, DSR in nvidia settings will have an impact on performance. Other settings also seem to impact bench tests, as evident by DX11 and 10 going from barely out of reach of the 980 to being on top of it. Consider going into the global settings, defaulting everything, then applying and retest. I seriously doubt going from +18mv to +25mv was enough to pick up nearly a 10% drop in in 10 and 11. DSR increased the difference between the DX 11 score in Passmark. 1st test I did earlier DX11 was extremely far apart from where it was old PSU, which was got me thinking as I had added DSR and upped a bunch of stuff in global. If benching, disable everything.If not benching, turn DSR back on, etc. I suggest leaving the rest of the settings to the actual game tab instead. It'll help reduce load on card. DSR with the DSR filtering smoothness OFF improves even VLC videos, but can' be adjusted per program for non games unfortunately.

Anyways, it was fun to OC. Never seen anything like it before. Spent several hours OCing it and got dead on or BEAT a GTX 980 with DX 10 and Direct Compute, and kissing DX 11, all for nearly $250 less? Just can't beat that type of performance considering it's already fast as-is assuming you use k-boost to defeat the clock throttling vs. gpu demand, or go the risky route and tamper with it's BIOS, which I wouldn't risk. Hope this helps someone shave a few hours off their OC and hopefully it's been useful. The earlier results with old PSU may or may not be accurate, so disregard those.

 

[SteffzStarkz]

This was the best I could do without toucing the volts:

My overclock:

• 
  - GPU Clock +200 Mhz
  - Power Limiter 110 %
  - Mem Clock +700 Mhz
  - Voltage +0 mv

Pretty good or what?

 

[Sybnios]

Ok I was hoping you could help me out. I read your wonderful results and thank you for sharing. Its my first time in OC. So after reading your post I got my hopes really high about my MSI 970 4G Gaming. I have a 4790K with Corsair Vengeance 16GB 1866 MHz, Seasonic 860XP2 Platinum and ASUS Maximus Hero VII.

So my best results are +160 core clock +200 memory clock with 110% power limit and +12mv core voltage at 60% fan speed. On my GPU-Z results I get a 109.8% power consumption.

Anything after that i.e. +170 core +200 memory with 110% power +18mv core voltage and I get 110.9% power consumption which I assume is not something safe to keep right? This result is from 3DMARK Fire Strike benchmark. Nevertheless when I play GTA V and I stress out the settings in Ultra/V.High and play around 2-3 hours the highest power consumption I get is 94%.

What should I do? This is it for me? It feels I got so low results and I built this PC with components that as far as I know are quality components and should give me better results. Any help would be deeply appreciated since you've spent many hours on this and seems your knowledge on the matter is way past mine.

 

[Sybnios]

Also I forgot to mention that my CPU temps were reaching 93 Celsius! That happened after I upped the volts to 18mv and the core clock to 170. My best OC without the volts was +140 core +200 memory. I have Coolermaster Hyper EVO 212 for CPU cooler.

 

[SteffzStarkz]

Every card overclocks different and your overclock isn't bad, around average I guess. But you could try to flash your VBios with for example 125% Power Limit. I've never done it before so I don't know how to do it.

Actually I was kinda thinking of do some bios flashing myself today. I've never reached higher than 108%, but I want a bit more space. I have only tested my overclock with Unigine Heaven Benchmark so I'm gonna do some more testing with 3DMark FireStrike, which is more demanding in my experience.

I bet that my overclock gonna crash with FireStrike or some really heavy gaming. Maybe my sweet spot would be +200 Mhz and +500/600 Mhz.

 

[Arronleeds]

you guys ever hit 1600mhz? mine craps out at about 1560, +200 core and +500 memory. i wanna flash my bios but i dont know how to, unless someone can send me a video on how to do it.

ive tried to get 1600mhz by putting the mv to the full 87 but i just get "driver has crashed".

ive had my gpu hit 13600 score in firestrike.

 

[Sybnios]

Every card overclocks different and your overclock isn't bad, around average I guess. But you could try to flash your VBios with for example 125% Power Limit. I've never done it before so I don't know how to do it.

Actually I was kinda thinking of do some bios flashing myself today. I've never reached higher than 108%, but I want a bit more space. I have only tested my overclock with Unigine Heaven Benchmark so I'm gonna do some more testing with 3DMark FireStrike, which is more demanding in my experience.

I bet that my overclock gonna crash with FireStrike or some really heavy gaming. Maybe my sweet spot would be +200 Mhz and +500/600 Mhz.

I see.. I was hoping something more than average.. I have no idea how to do that through the bios and sounds like I could damage my GPU. I just did a little research and it didn't seem that simple and as its my first time overclocking. Also I couldn't find a detailed tutorial for my mobo. Guess I have to accept these results.