Nvidia GeForce GTX 1000 Series (Pascal) MegaThread: FAQ and Resources

[TehPenguin]

1080ti has like 2,5x the performance of a 1060! Even if you would OC the 1060 to 3GHz it wouldn't surpass a 1080ti. My only problem with a mini 1080ti would be the noise the little cooler makes.

 

[TechyInAZ]

1080ti has like 2,5x the performance of a 1060! Even if you would OC the 1060 to 3GHz it wouldn't surpass a 1080ti. My only problem with a mini 1080ti would be the noise the little cooler makes.

Yeah it would get quite warm. Though, I am curious. It seems like the 1080 Mini does a excellent job acoustics wise, so I wonder how much worse it will be on the 1080 ti.

 

[U6b36ef]

1080ti has like 2,5x the performance of a 1060! Even if you would OC the 1060 to 3GHz it wouldn't surpass a 1080ti. My only problem with a mini 1080ti would be the noise the little cooler makes.

Yeah it would get quite warm. Though, I am curious. It seems like the 1080 Mini does a excellent job acoustics wise, so I wonder how much worse it will be on the 1080 ti.

I was thinking the fan might be noisy. Zotac can whine/buzz a bit, anyway I think. If I am wrong I apologise, but that's what I think of them after watching Youtube.

The heatsink doesn't look extra deep like the Asus or Gigabyte either, meaning less cooling.

I think the only way that card would work quietly is in a well ventilated case.

 

[TehPenguin]

I have gotten myself a 1080 after the recent price cuts and done some overclocking, some interesting things I noticed:

By setting the voltage/frequency curve manually I managed to squish out extra ~100MHz(stable at 2063MHz)out of the card under ~3C lower load temepratures.
Which seemed weird to me. I always thought increasing the voltage would improve stability and increase your overclock together with the temperatures.

Is it something new and specific to boost 3.0? Before the manual curve I have found a stable +100MHz offset that levelled out at around 1960MHz and 1045mV at around 77C@load.
Now I tell the card to boost to 2101MHz at 1025mV and it levels at 2063, still at 1025mV at around 73/74C@load.

Both tests ran in Overwatch ~1hr of play at ~28C ambient.

Anyone had similar experience when overclocking pascal?

 

[Phaaze88]

Still bumping mine up bit by bit... but unfortunately(?) voltage is locked on my card. Even when I try to set it in Afterburner - nope, the bar stays grayed out. I guess because I bought the cheaper G1 gaming, and not the Aorus Extreme edition?

 

[TehPenguin]

Did you download the newest afterburner version? My voltage control was greyed out when I used the beta afterburner.

 

[Phaaze88]

Did you download the newest afterburner version? My voltage control was greyed out when I used the beta afterburner.

Yea, I'm on the latest version, but no dice. No big deal though, it's not like I'm missing much without it.

 

[James Mason]

My 1080 (EVGA ACX3.0) boosts itself to 2025 without me doing anything, so I don't bother to do anything to it.

 

[TechyInAZ]

TehPenguin, I've noticed the same with my GTX 1060. Not tempatures, but increased clock speed.

I can hit 2100mhz at stock voltage with a custom curve, then with max voltage, I can hit 2164mhz (can hit almost 2200mhz but isn't stable for long) without crashing.

But for now, I just use a regular 50+ offset overclock on the core for 2050mhz peak boost at peak temperatures. Going higher nets me 1 frame roughly so it's not worth it. Overclocking vram is more important.

 

[nikoli707]

kind of mind boggling how in 2 generations we went from a max air cooled overclock of about 1350mhz to 2200mhz. heck even one generation from the 900 series of a max overclock of about 1550mhz to 2200mhz for the 1000 series. the ipc is drastically lower to be fair though if we are comparing 900 to 1000 series.

 

[TehPenguin]

TehPenguin, I've noticed the same with my GTX 1060. Not tempatures, but increased clock speed.

I can hit 2100mhz at stock voltage with a custom curve, then with max voltage, I can hit 2164mhz (can hit almost 2200mhz but isn't stable for long) without crashing.

But for now, I just use a regular 50+ offset overclock on the core for 2050mhz peak boost at peak temperatures. Going higher nets me 1 frame roughly so it's not worth it. Overclocking vram is more important.

No, no. I am experiencing the exact opposite. Increasing the voltage did not give me more stable overclock, just increased temperatures and crashed during a stress test just the same. LOWER voltage improved pretty much everything. That's why I am confused.

EDIT: nikoli: he said he can HIT a 2200 overclock but crashes which should not count as an max overclock. You have to validate your boost by stress testing your GPU. I can, too, peak at 2150++ but it will not matter with boost 3.0, it will pretty much always level at around ~2050(assuming the temps are right).

VRAM was an easy +500MHz, no issues. All in all both overclocks net me an extra 15-20% (depending on the game) performance.

I will have to redo all the testing in a few months since now the ambients hit 30C what obviously reduces your thermal performance quite significantly.

 

[TechyInAZ]

Ah. Well umm.. I sorta have that issue?? maybe..

Basically my overclock gets unstable at 1.093v, tuning it down to 1.081v at same frequency yields slightly better stability. Even nvidia said Pascal was picky with voltage.

But huh, overvolting goes worse for you? Wow weird. Either way, I don't overvolt on Pascal, way to complicated for me. Stock voltage already does a very good job at overclocking.

EDIT: I ONLY overvolted on my 1060 Pascal because i was testing, i don't overvolt for 24/7 use.

 

[TechyInAZ]

To give you an example, I am able to hit this without much issue (however I HAVE NOT tested for a while, so i'ts most likely unstable):

 

[renz496]

kind of mind boggling how in 2 generations we went from a max air cooled overclock of about 1350mhz to 2200mhz. heck even one generation from the 900 series of a max overclock of about 1550mhz to 2200mhz for the 1000 series. the ipc is drastically lower to be fair though if we are comparing 900 to 1000 series.

IPC between maxwell and pascal most likely about the same. most often we only look at core clock and ignore everything else. true GTX1060 need to be clocked much higher than GTX980 to reach 980 like performance but 1060 are doing it with much less CUDA cores than 980 (1280 vs 2048).

 

[nikoli707]

kind of mind boggling how in 2 generations we went from a max air cooled overclock of about 1350mhz to 2200mhz. heck even one generation from the 900 series of a max overclock of about 1550mhz to 2200mhz for the 1000 series. the ipc is drastically lower to be fair though if we are comparing 900 to 1000 series.

IPC between maxwell and pascal most likely about the same. most often we only look at core clock and ignore everything else. true GTX1060 need to be clocked much higher than GTX980 to reach 980 like performance but 1060 are doing it with much less CUDA cores than 980 (1280 vs 2048).

for sure its a trading around of cores to mhz etc. im curious where we will be in a few years. will graphics cards start shipping with 3.0ghz? really doesn't matter if they do or dont, the performance will be there... its just a curiosity of mine.

 

[TehPenguin]

@TechyInAz: That's pretty neat in itself. The temperatures look also really nice. What AIB did you get?

 

[TechyInAZ]

@TechyInAz: That's pretty neat in itself. The temperatures look also really nice. What AIB did you get?

Gigabyte G1 Gaming.

So...after that, I tried Unigine Heaven, and I immediately got crashing, so I guess Superposition isn't as demanding? Idk.

Anyways, the max I can really get is 2164mhz. Any higher and I get temperature limits and reliability voltage limits then the card starts to freak out (really odd, my custom fan curve never allows my card to go above 70C.

Update: Did some more testing. So it's sora interesting, so even if you manually force the GPU to clock at specific voltage and specific clock speed until you hit max temp (92C), the GPU will still downclock, for me that is 70C or lower for vcores of 1.075v, it sorta goes into this "emergency mode" where it downclocks to a specific clock speed and voltage that you did NOT even define. Even MSI Afterburner's graph shows you this second curve which you cannot control. I guess FinFET is very picky.

 

[TehPenguin]

Yeah I noticed that, too. I think this is a feature of boost 3.0

 

[renz496]

kind of mind boggling how in 2 generations we went from a max air cooled overclock of about 1350mhz to 2200mhz. heck even one generation from the 900 series of a max overclock of about 1550mhz to 2200mhz for the 1000 series. the ipc is drastically lower to be fair though if we are comparing 900 to 1000 series.

IPC between maxwell and pascal most likely about the same. most often we only look at core clock and ignore everything else. true GTX1060 need to be clocked much higher than GTX980 to reach 980 like performance but 1060 are doing it with much less CUDA cores than 980 (1280 vs 2048).

for sure its a trading around of cores to mhz etc. im curious where we will be in a few years. will graphics cards start shipping with 3.0ghz? really doesn't matter if they do or dont, the performance will be there... its just a curiosity of mine.

while gaming pascal is very similar to maxwell architecture wise pascal have the refinement that allow for very high clock. nvidia use this high clock to reduce the amount of the CUDA cores to reach similar performance to maxwell. this with 16nm node allows nvidia to bring 980 performance using only half the die size of 980 (398mm2 vs 200mm2). so is 3ghz will be finally possible with volta? honestly i don't know. gaming pascal probably the best form of maxwell. just increasing the clock probably will not going to yield more performance because performance increase are not linear with clock increase (we can already see this with maxwell and pascal). i think nvidia need to use new approach with volta instead further refining pascal existing design.

 

[axlrose]

I haven't been able to find the 1080 ti seahawk ek x anywhere. I can only find it one place where it is out of stock and not expected to be back in stock.

Can anyone give me other any other 1080 ti cards that are ready for a custom loop with full water block? I don't want a hybrid and I'm not really interested in trying to take a 1080 ti apart to install a water block on my own. It's my first custom loop and I already soaked my new build once when I screwed down the water block too tight and cracked it.

Thanks.

 

[TehPenguin]

If you're from Europe there are these two that seem to be in stock in a few places, and if not maybe some of them can ship worldwide:

MSI
Zotac

 

[axlrose]

I was wondering about the MSI in europe. It seems to be there. It's about $200 more before figuring out shipping though. The Zotac looks even more expensive, but I haven't seen that before. I'll look into both of them. Thanks for the help. I'm buying from the U.S.

 

[axlrose]

The Zotac has my attention, but I can't find it over here in the U.S. either. Odd...

 

[TehPenguin]

Yeah sadly electronics are more expensive in Europe.

 

[axlrose]

Does anyone have any good links to 1080ti's that are water cooled? When I filter on newegg I only seem to find one, and I think my understanding is that pcpartpicker doesn't include water cooled cards for some reason. I want a 1080ti to put into my loop to replace an old air cooled card. Thanks for any help finding a way to compare some and find prices and availability.