GeForce GTX 1070 Founders Edition Overclocking Review @ [H]

FrgMstr

FrgMstr

Just Plain Mean
Staff member
Joined
May 18, 1997
Messages
55,634
GeForce GTX 1070 Founders Edition Overclocking Review - In our overclocking review of the NVIDIA GeForce GTX 1070 Founders Edition we will see how far we can overclock the GPU and memory and then compare performance with GeForce GTX TITAN X and GeForce GTX 980 Ti. How high will she go? Can the $449 GTX 1070 outperform a $1000 GTX TITAN X? The answer is exciting.
 
Biggest takeaway I got was "don't buy the founders edition". But otherwise, the overclocking performance is pretty good given the price so with a better cooler you might have a great card to overclock.
 
I see it more as, if you want to pay the early adopters tax and wear headphones, and can find one in stock, knock yourself out.

I personally do not see any other cards having a great advantage on clocks. Talking to NVIDIA about this, I got the feeling that clock limitations were more on the side of the architecture rather than cooling or power. I may however be proven wrong on this once we see custom AIB cards in the wild though. It will be interesting for sure to see what the likes of ASUS, MSI, and GBT come up with.
 
Very interesting your feeling that FE is being held back somehow. Another reason to avoid, I suppose.

Now if there was an AIB version available for under $500...
 
The few 1080 and 1070 AIB cards, even those with improved power delivery, all seem to be peaking at 2,100MHz ± 50. It must be a restirction placed by NVidia. At least with the Titan X, being reference only there was a perceived concern about overclocking frying the low-yield chips. That really shouldn't be an issue on these upper-mid range cards.
 
Well why do you bother to name the $379 price when there's no card available anywhere near this price ( not even mentioning reasonable quantities... ). Unless we can go out and buy this card it's not even $449, it's rather >500...

I have got two problems with this article:
a) You compare OC card to non-OC cards... why not OC both?
b) As we have two players ( AMD & NVidia ) I would take at least one card from the other company to compare ( looking at price Fury/FuryX would be reasonable ). With only two competitors doing a review with few cards from only one company makes little sense to me.
 
Is it just a coincidence that the overclocking headroom was limited at 12% and the power level increase was limited at 12%? Seems like Nvidia deliberately gimped these Pascal boards power-wise. Have there been any aftermarket boards confirmed that raise the TDP limit beyond 12%? There's probably a ton of headroom left for grabs in the architecture.
 
With the preview, full review, and now the OC review, I can say the 1070 is on my buy list. As noted above, while it would be interesting to see an OCed 980Ti to compare, at the current pricing it does not really compare.

I am seeing off/on availability of these on NewEgg. The Founders editions are available at $449. However, most interesting are the first "non" founder editions I am seeing from EVGA. They have a base card at $419 and their ACX Cooling version at $439 with a small built in OC. They actually had the ACX version available as a combo with a Razer Mouse for $519 last night.
 
Just IMO, but Pascal is the first architecture on a new process node, so it's not surprising that we will not see the same kind of overclocking advantage as we did with Maxwell, which was build upon a very matured and well understood process node.
 
Didn't NVIDIA say they spent a lot of time & money tuning Pascal to attain the clocks they arrived at? I think this is pretty reflective of that, they brought the floor up a lot and didn't leave a ton on the table. If the stock clock was 1300 mhz, and 2 Ghz was still attainable, then this would be seen as an overclocking god. As it is, this is better for most people (more performance and value out of the box), it just makes OC'ing not as worthwhile (at least for FE cards). I think Hornet is definitely right that we could see the ceiling climb as the process matures.
 
britjh22 said:
Didn't NVIDIA say they spent a lot of time & money tuning Pascal to attain the clocks they arrived at? I think this is pretty reflective of that, they brought the floor up a lot and didn't leave a ton on the table. If the stock clock was 1300 mhz, and 2 Ghz was still attainable, then this would be seen as an overclocking god. As it is, this is better for most people (more performance and value out of the box), it just makes OC'ing not as worthwhile (at least for FE cards). I think Hornet is definitely right that we could see the ceiling climb as the process matures.
Click to expand...

That begs the question, how close to the last generation architecture is this new architecture? If the out of box performance is better tuned than Maxwell was, are they really that far apart? So Pascal cards may leave 12% on the table while Maxwell cards might have left 15%+ on the table. I'm just not seeing a huge difference between the 1000 series and the 900 series. It's an improvement, but not a revolution.
 
Not really sure its fair to compare the 1070 Overclocked against stock 980TI and Titan X. If you're going to be overclocking one, you're going to be overclocking the other. The most fair comparison would be to 1070 OC vs the best 980TI OC you've gotten vs Titan X OC.
 
Definitely you can see that the 1070 is much better bang for buck than the 1080, which is par for the course between top of the line and the rest of the cards, still kinda interesting how it also shows the non linearity of scaling with current games, and by that i mean that it was a +22% overclock yet the actual gain was between 10 and 16% depending of the game.

(2062/1683 = 1.22519 or 22.5% higher).
 
If you push a TITAN X (or 980ti) balls to the wall, is a 1070 still considerably faster? I bet it's not, or by a lot less than you'd expect. The rumblings on teh interweebs seem to indicate big Maxwell closes the gap considerably at high clock speeds. Any chance we'll see an article that compares OC to OC?

I'm not concerned with OC vs stock comparisons... they're not particularly useful. OC vs OC would be a very useful comparison, and actually give me some indication of whether I will consider jumping ship from my SLI TITAN X setup.
 
Dunnlang said:
That begs the question, how close to the last generation architecture is this new architecture? If the out of box performance is better tuned than Maxwell was, are they really that far apart? So Pascal cards may leave 12% on the table while Maxwell cards might have left 15%+ on the table. I'm just not seeing a huge difference between the 1000 series and the 900 series. It's an improvement, but not a revolution.
Click to expand...
Does it really matter though? The new process node has allowed substantially higher base clocks and big gains out of the box. How it arrived there is as much an academic exercise as anything else. Sure, if you OC Maxwell to within an inch of its life you might potentially shrink the gap because Pascal doesn't OC as well percentagewise, but will you ever overtake Pascal? No (keep in mind the shader difference between 1080/1070 and 980Ti/Titan). Its revolutionary in the sense that we got a new node and a lot more performance than if we hadn't.

Teenyman45 said:
The few 1080 and 1070 AIB cards, even those with improved power delivery, all seem to be peaking at 2,100MHz ± 50. It must be a restirction placed by NVidia.
Click to expand...
It might also just be where the architecture tops out rather than some kind of fabricated restriction. I do agree that the power envelope seems to be restricting, so you might be right, but that doesn't mean its entirely artificial.

Daroller said:
If you push a TITAN X (or 980ti) balls to the wall, is a 1070 still considerably faster? I bet it's not, or by a lot less than you'd expect. The rumblings on teh interweebs seem to indicate big Maxwell closes the gap considerably at high clock speeds.
Click to expand...
Hardly a surprise. Those cards have noticeably more shaders, so once you start clocking them at speeds approaching Pascal, of course they'll close the gap.
 
Last edited:
From the review here on [H], 980TIs selling for $630 three months ago, when overclocked, beat an OC 1070 in every game tested (where common tests exist). No, 1070 OC'd is not better than an OC'd 980TI, from data available here. Sometimes it's not even as good.

It is a valid point, though, that it gets pretty damn close, for considerably less money.
 
Activate: AMD said:
Hardly a surprise. Those cards have noticeably more shaders, so once you start clocking them at speeds approaching Pascal, of course they'll close the gap.
Click to expand...

Of course it's not a surprise. It would be nice to see the numbers, though. Preferably without buying the cards and performing the comparison myself.
 
Pretty much once the 1070's start showing up in decent quantities at close to MSRP. There is no reason to get a 980 ti unless you can score one for close to $300.

Good read. Always like the reviews here.
 
280-290W full system draw during OC'ing while skating by the Titan X. Holy shit that is awesome. :cool:

Thanks for the review, [H]
 
the gamer inside me (from 10 years ago) would really like to buy a 1070. But the availability (lack thereof) makes this whole launch feel like a paper launch. nVidia does great by having an initial launch of an extremely small quanitity of cards, then make it seem like there's too much demand, and they're probably sitting on a warehouse full of the cards. Artificial shortage creates price premium & more demand.

The price on 1070 is nowhere close to $450. it's $700 due to third-party profiteering.
 
jbc029 said:
It is a valid point, though, that it gets pretty damn close, for considerably less money.
Click to expand...
Well sure, that's the point. This is a x70-series card, so it's as fast as the highest-end from the previous generation.
 
As mentioned above it would have been nice to see a 980ti overclock vs the 1070 overclock.
Many people with 980tis want to know if selling their cards asap is the best option (while they can still get a slightly better price) or whether it has enough advantage to hang on for the 1080ti, or even keep.
 
J Macker said:
the gamer inside me (from 10 years ago) would really like to buy a 1070. But the availability (lack thereof) makes this whole launch feel like a paper launch. nVidia does great by having an initial launch of an extremely small quanitity of cards, then make it seem like there's too much demand, and they're probably sitting on a warehouse full of the cards. Artificial shortage creates price premium & more demand.

The price on 1070 is nowhere close to $450. it's $700 due to third-party profiteering.
Click to expand...

If you think they are sitting on a warehouse full of 1070s/1080s and don't sell them you can only hope AMD does very badly with their cards. Otherwise they would love to sell every single unit before AMD introduces new cards and forces them to lower price ( or simply lower market share ).

Guess what will be the price of 1070 if a $199 AMD card can come close like 20% to it? Will you pay ~$400 so 100% more for 20% additional perf. ? Even the difference between 1070 and 1080 is much better for perf/$ - ~55% higher price for ~20% perf.

If they sell very limited number of cards ( like few thousands ) and then they are forced to lower their price how many milions would they lose on that warehouse?

Another problem is 960/970/980/980ti. Any stock left will have to be sold with a lot lower price. If AMD was smart they knew what they will introduce and have tried to get rid off all old cards asap. For the moment NVidia will be left with possibly lots of hard to sell stock ( besides blind NV followers ) and nothing to sell ( besides few 1070/1080s, but can they make enought to really matter in market share? ).

If AMD new cards fail NVidia will have a nice price hike and make milions on blind followers. If they are good and NVidia supply of 1070s/1080s is low they might fall into their own trap - nothing great to sell and only empty promises for few more months... interesting times :)
 
Activate: AMD said:
Does it really matter though? The new process node has allowed substantially higher base clocks and big gains out of the box. How it arrived there is as much an academic exercise as anything else. Sure, if you OC Maxwell to within an inch of its life you might potentially shrink the gap because Pascal doesn't OC as well percentagewise, but will you ever overtake Pascal? No (keep in mind the shader difference between 1080/1070 and 980Ti/Titan). Its revolutionary in the sense that we got a new node and a lot more performance than if we hadn't.


It might also just be where the architecture tops out rather than some kind of fabricated restriction. I do agree that the power envelope seems to be restricting, so you might be right, but that doesn't mean its entirely artificial.


Hardly a surprise. Those cards have noticeably more shaders, so once you start clocking them at speeds approaching Pascal, of course they'll close the gap.
Click to expand...

It's not an academic exercise if the best performance from a 980Ti (OC and all) is very close to the best performance of a 1070 (OC and all). My point was that there is the appearance of a difference between the base clock 980Ti and 1070 OC. The base clock 1070 may appear close to the 980Ti, because it is benefiting from better out of the box optimization, not because the hardware is inherently that much better.
 
Ok, for a more apples to apples comparisson, this review appears to have the same Battlefield 4 settings for a pair of OC'd 980Ti's as the 1070 OC review has.
GeForce GTX 1070 Founders Edition Overclocking Review

The 980Ti's are ranging from 110 to 113 FPS. Per this review, the 1070 OC is running the same test at 109 FPS.

The 1070 offers excellent value and some potentially exciting features in a few years when developers integrate them. What it does not appear like is a viable upgrade path for any 980Ti, or better, owners. That's perfectly fine. It's just important to know.
 
That's how Nvidia told us it would perform-- if they thought there was any chance of getting people with 980ti and titan x's to upgrade to the 1070, they would have tried to make that play. But sure, it's nice to see the numbers and confirm that it isn't a viable upgrade path.

To be honest, the 1080 isn't really sexy for people with 980tis either.
 
Dunnlang said:
It's not an academic exercise if the best performance from a 980Ti (OC and all) is very close to the best performance of a 1070 (OC and all). My point was that there is the appearance of a difference between the base clock 980Ti and 1070 OC. The base clock 1070 may appear close to the 980Ti, because it is benefiting from better out of the box optimization, not because the hardware is inherently that much better.
Click to expand...

Fair enough, I see your point, and yes, the benefits appear to come more from the clock speeds than from architecture improvements. That said, the fact that a 1070 isn't an upgrade for owners of overclocked 980Ti's isn't exactly a shock. The fact that its even debateable is telling.
 
Hi guys, I understand what you are saying about overclocking the comparison cards as well. The plan is to do that with custom cards. I.E. I want to take a custom 980 Ti with custom cooling and overclock that to compare to 1070/1080 custom card from ASUS/MSI/GIGABYTE etc... Custom to Custom overclock comparisons to get the absolute best from both. That overclocking will come once we start testing custom cards, I will probably work up a nice big overclocking article for you guys that overclocks everything to the max, but I want ti do it with custom cards, not all reference cards which was used here today.

What you can get from this article is the potential for overclocking the Founders Edition GTX 1070 has, and how that would compare to a TITAN X or 980 Ti at stock. Just knowing how far the 1070 overclocks, is important. I want to move on from the Founders Edition 1070 now, I don't feel it is the best card for overclocking, and focus more on custom cards moving forward, comparing pricing and performance and overclocking with those. So yes, you will get your full "everything overclocked" comparison, I just want to do it with all custom cards versus reference cards so everything has the full advantage at its maximum potential. I think that will be more representative of the performance on all fronts.
 
On the conclusion page you say this
. We don't think it is the Founders Edition cooler holding it back at 100% fan speed
Click to expand...
Then further down you say this
We feel that GeForce GTX 1070 is being held back in its overclocking potential on the Founders Edition and its cooler.
Click to expand...
Sounds like a contradiction.
 
Having a good chuckle about the folks with Titan's in SLI wondering if an OCed 1070 is faster than a Titan X. If you are running dual Titan's are you really concerned about the 2nd tier 1070? More relevant would be a OC comparison to the 1080. Flagship vs. Flagship.
 
rinaldo00 said:
On the conclusion page you say this

Then further down you say this

Sounds like a contradiction.
Click to expand...

It is being held back by the cooler, at default fan speed, but not at 100% fan speed. In the first sentence you quote I note its not held back at 100% fan speed. But in the second sentence I note it is being held back by the cooler, cause at default fan speed, it is. You have to manually increase the fan speed else it will hold you back, on the Founders Edition. Both statements are true in my testing and experience.
 
Nice article and cant wait to see the one for aib overclocks. This seems like an obvious question but do you get a reference card for review purposes or did you hit refresh until you could get one added to a cart to buy? Finally can I have the card now you are done with it :)
 
schizrade said:
All the 980Ti's just became obsolete as shit.

lmao
Click to expand...

wareyore said:
Pretty much once the 1070's start showing up in decent quantities at close to MSRP. There is no reason to get a 980 ti unless you can score one for close to $300.

Good read. Always like the reviews here.
Click to expand...

How so? This thing is still slower than custom 980 Ti out of the box and it will take custom 1070s to maybe match or beat them. Considering FE is $450 and even the EVGA SC is hitting $440, good luck finding anything under $450 that beats a custom 980 Ti. So why would anyone be dumb enough to sell a 980 Ti that has more OC headroom for $300? :rolleyes: Even my reference Titan X's (1450 MHz on air and 1520 MHz water) whip this FE's ass when comparing OC vs OC. The 1070 is a nice card but let's not get carried away here.
 
Last edited:
Yet again I would have loved to see the 1080 numbers side-by-side with the OCed 1070, as I imagine a lot of us are wondering if it's worth the price difference to go for the 1080.

Still, seems like a decent performing card.
 
Brent_Justice said:
It is being held back by the cooler, at default fan speed, but not at 100% fan speed. In the first sentence you quote I note its not held back at 100% fan speed. But in the second sentence I note it is being held back by the cooler, cause at default fan speed, it is. You have to manually increase the fan speed else it will hold you back, on the Founders Edition. Both statements are true in my testing and experience.
Click to expand...
So a better cooler will just help with noise levels not with getting a better overclock.
 
rinaldo00 said:
So a better cooler will just help with noise levels not with getting a better overclock.
Click to expand...

Most custom 1070s are peaking at 2050 MHz and aren't doing any better than FE at 100% fan. This and the 1080 need a custom unlocked vbios and who knows if/when that will come. I have a guy (Prema) on my forum that specializes in vbios/bios unlocks and he tells me Pascal is a lot tougher to crack than Maxwell because of NVIDIA messing with it.
 
From all reports I've seen, Pascal clocks aren't constrained by power or voltage. They are constrained by temperature with the reference cooler, but increase its fanspeed and temperature doesn't constrain clocks either. Pascal is constrained by its architecture. Most chips top out around 2.05Ghz, and if you're lucky you can get to 2.1Ghz.

Fancy AIB cards with more power phases, more power plugs, giant fans, and custom BIOS won't change that one bit. The giant fans will be quieter, that's it. You're still gonna be stuck at 2.05Ghz, or 2.1Ghz if you win the silicon lottery.
 
schizo said:
From all reports I've seen, Pascal clocks aren't constrained by power or voltage. They are constrained by temperature with the reference cooler, but increase its fanspeed and temperature doesn't constrain clocks either. Pascal is constrained by its architecture. Most chips top out around 2.05Ghz, and if you're lucky you can get to 2.1Ghz.

Fancy AIB cards with more power phases, more power plugs, giant fans, and custom BIOS won't change that one bit. The giant fans will be quieter, that's it. You're still gonna be stuck at 2.05Ghz, or 2.1Ghz if you win the silicon lottery.
Click to expand...

Custom vbios is still up in the air based on conflicting reports (e.g. Galax). We won't know until one is released that addresses this.
 
You must log in or register to reply here.
Back
Top