Overclocking a GTX 1070 Ti with Afterburner, pushing it higher... suggestions?

[Vayra86]

That's basically my method. If I goto +165 on the core, I start noticing the scores going down slightly.



It worked quite well to find the top Core / Mem clocks. Saw artifacts in Kombustor at +950 mem, so that's when I dialed it back. Also, testing with 3DMark can be a complete PITA when the system crashes and you have to reboot. With Kombustor, it crashes "gracefully" ... so it's a great tool for getting close to the "zone"

Do whatever you like but I can only add to this that Valley and Heaven NEVER fail when it comes to determining the maximum core clock. Your method is just wrong.

Another big advantage is that they will never force a reboot, the worst you'll get (unless you do something really stupid I suppose like set core to +400) is a locked application, close the process, done.

But above all: Valley and Heaven are very good at showing the minute performance differences that result from your OC. I mean stability is one thing, but if your highest clocks still result in lower performance (especially memory clocks too high can do this, they can even create stutter which is punished HARD in Unigine scores) then its pointless to push those clocks, all it serves then is an e-peen screenshot or something.

I have honestly never needed Furmark/Kombustor for any GPU ever to get it stable. The only purpose it has is frying your GPU back in the day, and today it really has no purpose at all. It doesn't test stability, it doesn't test actual clocks, and it doesn't test actual temps. All you realistically test is how fast can you yank a card into the highest power state and top-end temperature allowed in its BIOS. Well yay!

 

[Sasqui]

Your method is just wrong.

No, not "WRONG", there are better methods out there. Kombustor found a stability range, the next step was to run all 3DMark tests (DX11 and DX12) and see what ended up stable, it was only within 10% of what the Kombustor tests showed was stable.

So, Kumbustor is ok for "Quick" stability testing (yes it IS!). I don't disagree that Valley is better, I just didn't use it. Next time, I probably will.

 

[Vayra86]

The thing is that with Pascal you can get any card and put it within 10% of its max OC right from the get-go. There simply isn't more OC headroom to get.

Kombustor, once again, does not test a single thing, not even same ballpark.

 

[Sasqui]

Kombustor, once again, does not test a single thing, not even same ballpark.

Hey, it worked for me, showed memory artifacts at +900, which is why I dialed it back to +850. Core was off just a little more. What more can you ask for?

Yes, oranges and apples... Kombustor is OpenGL, where hardly any games are these days. That's why final testing was done with 3DMark.

 

[Vayra86]

No it didn't - you needed 3DMark to actually see what limited you.

 

[Sasqui]

No it didn't

You're saying I didn't see artifacts?

 

[Bamaofficial]

I have the 1070 ti Asus strix Advanced Binned I was able to hit 250mhz on the core and 500mhz on the memory. Never touched voltage and It is rock solid. I want to push it more because this thing never gets over 63c. This thing amazing me.

 

[Sasqui]

I have the 1070 ti Asus strix Advanced Binned I was able to hit 250mhz on the core and 500mhz on the memory. Never touched voltage and It is rock solid. I want to push it more because this thing never gets over 63c. This thing amazing me.

If it wasn't for the power limiter on those cards, you'd see a heck of a lot higher clocks, and probably still have nice temps too. But as it stands, you're in stock Vega 64 territory.

 

[Bamaofficial]

If it wasn't for the power limiter on those cards, you'd see a heck of a lot higher clocks, and probably still have nice temps too. But as it stands, you're in stock Vega 64 territory.

I have about $250 invested in this card. For that reason alone I am satisfied with the results. lol

 

[John Naylor]

According to TPU testing, the outta the box performance of the Strix is right on par with the stock 1070 Ti and 4% lowe than the stock Vega 64 ... ... difference being the Asus will go an extra 12.1% when overclocked ... Unfortunately at time of testing AMDs drivers were incapable of OCing. Other tests subsequently done on AIB Vega's showed a drop in performance and core clocks were actually lower. ... havent seen any subsequent testing so as to know whether anyone's been able to do better. Dosn't seem to be much interest.

As for performance versus clocks I have found it hard to make any hard and fast rules other than Highest clocks don't = highest fps. And each test I use has usually produced different results, non of them (Heaven, Valley, Firestricke, etc) showing any consistency as the "best tool". Other things I have found...

a) Some Ocs may be fine for a game in general .. but I sometimes have to drop to another profile to get thru certain parts ... example when you coine out from underground in metro 2033 and when you have that charging bull thing on ice.

b) Any OC that is stable in every other game has a solid likelihood of crashing in BF anything

c) Game betas are particularly problematic

 

[Assimilator]

Sadly NVIDIA cards are voltage/power/temperature limited very hard nowadays, and there's no way to overcome that unless you're able to figure out how to flash modified Pascal BIOSes.

 

[Deleted member 67555]

My 1070ti will OC 250mhz on the core and a measly 33mhz on the memory.....
I can get the memory up and I can run Time spy etc and it's fine until I game then it's intsa-crash...
Every single crash is a driver failure

 

[Vayra86]

My 1070ti will OC 250mhz on the core and a measly 33mhz on the memory.....
I can get the memory up and I can run Time spy etc and it's fine until I game then it's intsa-crash...
Every single crash is a driver failure

Points to a weak IMC I guess. Most memory does get further, at least +100 should be easy.

As for @John Naylor 's findings about OC reliability, I can't really say I can draw a similar conclusion. My 660's in SLI back in the day ran on the OC I had set right from the start until the moment I sold them. The 770 after that crashed once in Crysis 3, I dropped one bin (13mhz) and it was fine ever since; the 780ti after that didn't even OC proper so it remained at stock (Gigabyte Windforce + meh chip probably did the trick), and the 1080 I have now has been running +120 core / +500 mem ever since I bought it. I had one TDR crash with the 1080, I did not dial down and hasn't crashed since.

And I play an extremely wide variety of games too, from alpha's even up to open/closed beta, lots of indies, old and new, to day one patched triple A junk.

I think the trick is not to push it too far, and when you find your top end clocks, take one small step back for stability.

 

[Bamaofficial]

Points to a weak IMC I guess. Most memory does get further, at least +100 should be easy.

I settled on 600mhz. I just need to push it any further when I reached the core overclock. I would say it would go a little further but not sure how much. It was a advanced Binned card also tho.

 

[Vayra86]

I settled on 600mhz. I just need to push it any further when I reached the core overclock. I would say it would go a little further but not sure how much. It was a advanced Binned card also tho.

Don't let the marketing fool you. Binning on Pascal is heavily overrated. All cards clock to 2 Ghz at least and all memory goes to 11Gbps / 9Gbps (GDDR5X/5) regardless of whether you have a newer version of it or not. The best ones *may* hit 2148 or 2176 but that's all and they will also need more volts to do it, raising temp = dropping the clock faster. The headroom beyond that is 2-4% at best and core clocks depend more on temps than they do on voltage; Pascal will force voltage down and drop a bin every 5 C you gain. All this combined makes any sort of binning rather pointless.

And as others have pointed out as well, clocks do not always translate to performance, especially when it comes to memory. You can use Valley scores to get a good handle on performance gain/loss with little memory tweaks. Heaven as well, to a somewhat lesser degree.

Another thing of note is that 1070's and possibly all GDDR5 based GPUs are likely to clock higher, because less power is reserved to feed VRAM and shader count is lower, so clock bumps are easier.

 

[Bamaofficial]

2165mhz is stable at 250mhz. That's about where It seems to range.

 

[Sasqui]

Sadly NVIDIA cards are voltage/power/temperature limited very hard nowadays, and there's no way to overcome that unless you're able to figure out how to flash modified Pascal BIOSes.

Actually, it's not the BIOS, but hard wired power limiters on the card itself, at least most of the 10xx series. There are some guides out there that are very much like pencil mods to increase the power limit.

Here's a tutorial, there are more out there too: http://www.overclock.net/forum/69-n...shunt-mod-titan-x-many-other-nvidia-gpus.html

2165mhz is stable at 250mhz. That's about where It seems to range.

That's exactly where I ended up too (exactly where NVidia put up the wall). I got more out of the memory tho, I was up close to +500mhz. Beat the Vega 64 in all 3D mark tests. However, when I overclocked the Vega, it left the 1070 ti in the dust.

 

[Junkie XL]

That's exactly where I ended up too (exactly where NVidia put up the wall). I got more out of the memory tho, I was up close to +500mhz. Beat the Vega 64 in all 3D mark tests. However, when I overclocked the Vega, it left the 1070 ti in the dust.

Until you put the two in actual games. I could care less about a card(s) performance in a software benchmarking suite. I care about what actually matters, gaming. Which is ironically also the best place to test stability. Idk how many times I've tuned my cards stable in 3DMark or Heaven (Valley is trash), only for it to crash in games, sometimes immediately.

In games, no way is the Vega 64 leaving the 1070ti in "the dust". Granted I don't own a Vega (I'm done with anything AMD GPU related), but I've seen enough gaming benchmarks to know both Vegas are underwhelming, esp for the price. I can only assume it's due to the majority of games being optimized for the green team. Which would make sense as last time I looked nvidia holds a solid 60% of the marketshare, vs AMD's 19% (Intel has 17% lol).

 

[Sasqui]

Idk how many times I've tuned my cards stable in 3DMark or Heaven (Valley is trash), only for it to crash in games, sometimes immediately.

Agreed, the benchmarks are a sort of "fleshing out" of what clocks and voltages look stable. Ultimately, if it can't play a game without crashing or artifacting, then what's the point?

Keep in mind, part of the process is the thrill of the chase in itself

 

[Deleted member 67555]

I couldn't agree more with using benchmarks to test stability as that's not going to happen..
I had mine "Benchmark Stable" and I was so impressed then I went to play a few games and instacrash... Literally as soon as I would start playing it would crash...
I'm really starting to distance myself and my expectations away from benchmarks... They are becoming quite pointless.

 

[Junkie XL]

Agreed, the benchmarks are a sort of "fleshing out" of what clocks and voltages look stable. Ultimately, if it can't play a game without crashing or artifacting, then what's the point?

Keep in mind, part of the process is the thrill of the chase in itself

Oh yea. Overclocking is like gambling. You never know what the end result will be, chip from chip. Even though GPU overclocking is hit and miss for 5-6 FPS increase on average it's still fun to do.

I used to use benchmarking software to find a chip stables OC, but to me, they are a waste of time. I just boot up a few of my most sensitive games and go from there.

 

[Sasqui]

I just boot up a few of my most sensitive games and go from there.

What are your most sensitive games? I'm assuming it's not solitaire lol

 

[Junkie XL]

What are your most sensitive games? I'm assuming it's not solitaire lol

MMO's tend to be pretty sensitive for whatever reason. Though Final Fantasy 15 with everything maxed out 3880x1440, worked wonders for tuning my current 1070Ti

I have the 1070 ti Asus strix Advanced Binned I was able to hit 250mhz on the core and 500mhz on the memory. Never touched voltage and It is rock solid. I want to push it more because this thing never gets over 63c. This thing amazing me.

Wow an Asus GPU worth a crap? I just had to replace an Asus Strix R9 390 that kicked the bucket. No surprise, that thing gamed between 89-92c even with me replacing the thermal paste to Kryonaut. My roommate picked up the Sapphire Nitro R9 390 and it stayed cool and quiet. My other previous Asus GPU was a GTX 660Ti TOP DirectCU II.was nothing but a PITA. Constantly crashed everything. Defective memory controller I think it was, and it was a 'binned' GPU. RMA'd it only for a worse replacement. Love their everything else though.

I wouldn't bother adding voltage to the 1070ti. I found worse OC headroom when I added voltage to mine. Thought I had defective power delivery. Did some research and seems it's an issue with Pascal in general. I cannot complain really. Mine was able to boost up to 2,106 mhz stable in games 14.4k Passmark 3D score. Though in some games the boost will droop a bit under 2.1, but only 3 of my games.