GTX 1080 Overclocking

[RussianSensation]

This isn't exactly the same argument. NVIDIA didn't charge an additional $100 for the reference/minimalist cooler before. Regardless, I do agree that this will all be forgotten in 1-2 months.

That's the point. This time they decided to add a $70/100 arbitrary premium for thermal throttling loud video card. When the card is upside down in my case, do you think I care if it's made from aluminium, plastic, gold, titanium, platinum? Unless I can sell that aluminium heatsink from GTX1080 for profit once the card is worthless in 6 years, all I care about are noise levels, minimums level of guaranteed performance (1733mhz boost -- FE 1080 cannot achieve that) and overclocking headroom at reasonable noise levels (FE 1080 fails miserably in this area). In this regard, FE 1080 fails all 3 of those parameters, just like the reference 290/290X did. The difference is those AMD reference cards never cost $70-100 extra over after-market cards. I paid less than $70 for the Thermalright Silver Arrow and Corsair H110i GTX -- both coolers that trash GTX1080's crappy blower into the ground (and both come with $30 fans alone!). What premium design? Does NV think we are stupid or something? :sneaky:

 

[airfathaaaaa]

It's 30% faster on more than a full node shrink. It offers same performance/$ as 980 or 980Ti after a full node shrink. That is what I call fail. If you as a consumer can't see this, I'm baffled. You do know that we get faster computers for same price because of new processes? What NV is doing is showing you the finger. With this practice, eg. same performance/$ for newer node we would be paying millions for CPUs today if we use same performance/$ as back then. So yes, this sucks pretty big. And if AMD repeats this, shame on them as well.

well considering that they used(well most of them) 4 games of 2011/2013 with huge average fps just to give a better overall score then no its not 30% faster
if you remove most of the games that arent relevant today and just bench it on games of 14/15 (without dx12 :hmm: )the card is from 18 to 22% faster...

 

[airfathaaaaa]

Im going to point this out again. This is exactly what was argued with the GTX680, GTX780, R9 290 etc. The same dead horse being beaten again and again. The throttling happens to all these cards under certain conditions mostly from bad airflow.

The same crowd hating on A,B,C from card from vendor X or Y. The same crowd hating on the reference coolers etc etc.

Just give it 1~2 months and this will be a non-issue (what throttling?) and the cycle is complete.

The amount of sensationalism on these boards is incredible! Im sure we can do better than this or is it just something to talk about til the actual cards (ref/custom) are in our grubby hands?

reaching the temp that the card starts to throttle on a pc case is easy VERY easy even if you case is the definition of the positive pressure
the problem is ok lets face it the card can be overclocked till a point that clearly is be chocked by the power delivery design and because of that even if you install a helium cryogenic system and you manage to keep the card under -100 the card will throttle up worse because you will eventually reach the power limit of that one mosfet...

now put that into perspective the people that live on sub tropical or tropical areas (90% of this rock population..)

and you decide to buy that card..having a house for 8 months on above 25c ambient inside is something natural.. can you imagine the outcry from those people?

 

[antihelten]

now put that into perspective the people that live on sub tropical or tropical areas (90% of this rock population..)

and you decide to buy that card..having a house for 8 months on above 25c ambient inside is something natural.. can you imagine the outcry from those people?

You might want to adjust that graph to only include people who have the purchasing power to even consider the 1080 in the first place. I'm pretty sure that if you did so you would find a very different pattern.

 

[Erenhardt]

I 100% disagree. The criticism is fully deserved because NV assigned a premium for basically nothing. If you have to explain why your FE card warrants a $70-100 premium, you automatically failed. If the product is worth it, the quality speaks for itself with real world results. My R9 295X2 runs at 1100mhz on both GPUs at 74C in the summer. Radeon Pro Duo maintains sub-60C at max load. That's premium design. FE cooler 1080 fails in ever way imaginable. For those who want to water cool, they'd rather take a bare PCB+chip design and save $100. The message that FE is a premium design pisses me off so much that even if FE GTX1070/1080 trash Polaris 10 into the ground, I'll buy slower AMD cards this generation just to prove a point that I am not a sheep that bends over for marketing gimmicks. Only AIBs having GTX1070 for $380-400 might sway my decision. Even if FE1080 was $499 and AIB 1080 was $599, I'd still pay $100 extra for the AIB card. That's how much NV got this wrong. Honestly I can easily afford to change every card in my house to a GTX1080 FE $699 but I just cannot do it. I cannot send a message to a corporation that charging me more $ for inferior design is OK. It's not OK.

Put the brand name behind. R9 290/290X reference were not able to sustain factory advertised clocks. 980 SLI thermal throttling. GTX1080 folding like a deck of cards after 10 min of gaming, failing to reach advertised boost clocks. To run 1080 FE overclocked at sustainable boost requires fan speed at levels only a gamer with closed headphones would find reasonable. The difference is 290/290X were the only cards hated full force by the online Internet community but GTX1080 gets to slide? Ya sure, no, not a chance. All of them are unified by a single factor - reference blower design on a $550+ card is garbage and a slap in the face after what we saw can be done with a Fuy X AIO CLC. The difference is 290/290X/980 (or put any other reference card in here) never charged $70/$100 premium for "premium" materials. GTX1080 is the worst reference card ever made because it comes with a premium for nothing. No one expects a reference blower card to actually be a good product in 2016 (lacking 0 dBA operation at idle, horrible overclocking headroom, loud at max loud) but no one expects a marketing $70-100 premium either. I don't care if AMD/NV/Intel/Matrox put out a reference board but when they release garbage and try to charge a premium for it, they better expect the biggest backlash ever. Had NV priced the reference 1080 for $599 and released the Founders Edition AIO CLC with binned chips (Gigabyte Gauntlet) for $699, I would have at least tried to understand.

The last reference card I had was HD6950 2GB and it was hot and loud. Never again, unless I am going water blocks or 3-4 cards in Quad-Fire/Quad-SLI. Time to bring AIO CLC as standard if they are going to charge us $70-100 premium. What NV did this round is nothing short of disgusting. The want us to finance their nonsensical 57.5% margins because the cooler is made out of aluminum but the card runs hot, loud, cannot overclock worth a damn? I can easily afford to buy 10 GTX1080 cards for mining but NV isn't getting $1 from me for the FE 1080 because of their arrogance of treating customers like Apple sheep.

Excellent post.

I would add one more thing that makes reference 290 >> 1080FE.
Both coolers are trash and can't handle the card, but at least r9 290 have a quality PCB that can supply immense amount of power:

290x:

1080FE missing one phase on VRM - quality premium design D: :

I guess you would have to pay $200 premium to get fully populated reference board! What a failure.

So here it is, for 290 you could get $40 -$70 aftermarket cooling:
http://www.tomshardware.com/reviews/r9-290-accelero-xtreme-290,3671-4.html
Or put it under water and OC it to the moon. Something you can't do with 1080Fe because the premium design is missing VRM phase!


They are asking $100 premium, yet they skipped a VRM components for a total of $1.5 savings. Give me a break!

 

[antihelten]

They are asking $100 premium, yet they skipped a VRM components for a total of $1.5 savings. Give me a break!

This to me is really one of the most egregious things about the founders edition. We can discuss all day about whether or not the cooler is decent/sufficient or not, but this PCB is a clear cut example of Nvidia cutting corners on a card where they have been bragging about craftsmanship and "crazy overclockability", which to me is just plain dishonest.

 

[Keysplayr]

What is the significance of 6 vrms over 5?
Are 6 needed? Is 5 plenty? What is it?
Why is 6 a magic number of VRMS/phases? Why not 20?
I'm not an electrical engineer. Just trying to understand.

 

[maddie]

What is the significance of 6 vrms over 5?
Are 6 needed? Is 5 plenty? What is it?
Why is 6 a magic number of VRMS/phases? Why not 20?
I'm not an electrical engineer. Just trying to understand.

For you, the simple answer is that you should know that 5 is less than 6. If not, I can recommend some books.


Personal attacks are not allowed.
Markfw900

 

[coercitiv]

I'm not an electrical engineer. Just trying to understand.

It means less crazy overclock ability and more responsible amounts of performance. It also means AIB premium versions will eat this premium reference card for breakfast.

That sexy shroud though... makes you wanna reference everything

 

[Arachnotronic]

What is the significance of 6 vrms over 5?
Are 6 needed? Is 5 plenty? What is it?
Why is 6 a magic number of VRMS/phases? Why not 20?
I'm not an electrical engineer. Just trying to understand.

Here's a really good explanation on the topic:

To clarify power phases, each 'phase' is just another link supplying power.

Think of it like a fire truck filling a swimming pool. If the truck uses all it's pumping pressure to power one hose, that hose will squirt water out at incredible pressure, and the hose will fly about all over the shop. If the truck were instead to run 8 hoses, pumping the same amount of water as before only evenly distributed among all 8 hoses, each hose would sit there nicely and fill the pool. Both methods fill the pool just as quick, but using the extra hoses makes the job smooth and controlled.

It's the same with power phases. The power phases supply power to the cpu. The more phases there are, the less power needs to be drawn from each one. The less power being drawn, the cleaner that power is.

There is of course a limit where adding extra phases just becomes a gimmick. A current 8 phase board with digital vrms can easily do the same job as a 24 phase board from a few years ago.

Not only that, but for it to really be relevant to you you'd need to be overclocking the crap out of your cpu anyway. If you only plan to run your system at stock then it all becomes irrelevant, and you may as well buy the cheapest board that has the functionality (read: the right slots, right features etc) you need.

http://forums.overclockers.com.au/showpost.php?p=12858804&postcount=3

This was explained in the context of a motherboard for a CPU, but it applies here too.

 

[Mahigan]

What is the significance of 6 vrms over 5?
Are 6 needed? Is 5 plenty? What is it?
Why is 6 a magic number of VRMS/phases? Why not 20?
I'm not an electrical engineer. Just trying to understand.

VRM Phases are like tiny power filters which clean up variances in the power delivery to the GPU. Basically, the GPU board is fed 12v of power but the GPU needs far less than that (usually between 1-1.3v). What the VRMs do is phase the power down from 12v to the required 1.xv. Each phase step cleans the power to a larger degree and ensures less vDroop. VDroop happens when the supplied voltage drops unexpectedly. Such drops cause stability issues in the operation of the GPU which are more apparent when overclocking.

This means that NVIDIA have decided to omit one phase which results in a higher chance of vDroop on a card which has a high operating GPU frequency. This inhibits overclocking on a board being sold based on marketing fluff claiming that it is well engineered. Basically the marketing is false.

NVIDIA are cutting corners in order to improve their profit margins whilst also charging a premium tax on early adopters.

 

[ShintaiDK]

GTX680 is missing 1(got 5), GTX980 is missing 2(got 5).

GTX1080 is missing 1(got 5).

Nothing new as usual besides the usual FUD.

Mission critical servers dont have 8-32 phases either. They have 4-5.

 

[RussianSensation]

What is the significance of 6 vrms over 5?
Are 6 needed? Is 5 plenty? What is it?
Why is 6 a magic number of VRMS/phases? Why not 20?
I'm not an electrical engineer. Just trying to understand.

In this case, the power phases go together with the capped power delivery. At least it's not as bad as the GTX590.

When various sites tried using after-market cooling on the 1080 (Accelero/AIO CLC), the card reaches max power consumption limit NV set. Overclockers shows that with a base of 2000mhz, 1080 was using 220W:
https://www.overclockers.ru/lab/76273_4/obzor-i-testirovanie-videokarty-nvidia-geforce-gtx-1080.html

This means this reference/FE card is hopeless for extracting maximum potential out of the excellent GP104 chip. On one hand, this is a positive since it may foreshadow that GP104 has more headroom for overclocking with a proper power delivery and cooler. On the other hand, if AIB cards cost $650-700 and allow GP104 to overclock to 2.2-2.4Ghz, while running cooler and quieter, it means for most 1080 consumers the FE version is the card to skip.

In summary, the FE will run hotter, louder, thermal throttle, cannot turn off the fan at idle, and has a premium. NV says they made it so that people could "enjoy it" for the duration of a generation. All it is the Blower Myth kicked down another generation towards a group of GPU users that still believe a smaller fan blowing in a smaller heatsink is somehow superior to open air cooled cards -- they can afford a $700 card but cannot afford a $100 Phanteks Enthoo Pro or a similar modern 2015/2016 case?

Since NV doesn't officially support/encourage 3-4 way SLI anymore, the argument for a blower is worse than it's ever been. Once you install the card, you cannot even see the face of the cooler either.

If AMD/NV cannot make a good $550-700 reference card, don't make one AT ALL. Sapphire's Fury Air cost $569 a year ago and had an air cooler that makes 1080's look like a $15 aluminum gimmick. At max load, max overclocked, this card is more or less as quiet as the 980Ti is at idle! That's premium air cooling:
http://www.anandtech.com/show/9421/the-amd-radeon-r9-fury-review-feat-sapphire-asus/2

Remember when higher end coolers started to have copper and heatpipes over the far cheaper aluminum? The reason copper fell out of popularity was due to excessive weight for larger sizes coolers such as Noctua NH-D14. The sheer mass of such coolers ensured that copper was no longer the optimal material since then not all common motherboards could handle such loads.

If you took the Phanteks PH-TC14 and sliced it in the middle/half, you could make 2 separate coolers, each with a 140mm fan, that have better cooling capacity than the Titan blower. This cooler with 2x140mm fans sells for $60 in retail!
http://m.newegg.com/Product/index?itemnumber=N82E16835709002

NV needs to stop lying and just say we charged $70-100 more for a $20 air cooler + $80 worth of expensive marketing material because they knew a small fraction of their customer base will pay. The $100 premium could have been $125-150, but NV decided to try it out first to see how the market reacts.

By NV setting the MSRP of $699 for the FE card it has a ripple effect of some AIBs pricing their cards at $699 and higher. It's a double hit to the consumer between an overpriced and mediocre blower reference card that cannot even maintain advertised factory boost and a risk of waiting for $599 AIB cards that may cost more than that because NV's $699 MSRP made it so easy to justify > $599 MSI Gaming, Asus Strix, Gigabyte G1.

Because the FE 1080 cannot guarantee reference boost clocks are reasonable noise levels after hours of gaming, this point alone means the card isn't even worthy of NV's $599 MSRP as the customer isn't getting 100% of the promised spec 1080 performance.

 

[showb1z]

GTX680 is missing 1(got 5), GTX980 is missing 2(got 5).

GTX1080 is missing 1(got 5).

Nothing new as usual besides the usual FUD.

Mission critical servers dont have 8-32 phases either. They have 4-5.

Just keep ignoring all JHH's claims. Here's a reminder.

http://forums.anandtech.com/showpost.php?p=38242302&postcount=87

 

[Arachnotronic]

The message that FE is a premium design pisses me off so much that even if FE GTX1070/1080 trash Polaris 10 into the ground, I'll buy slower AMD cards this generation just to prove a point that I am not a sheep that bends over for marketing gimmicks. Only AIBs having GTX1070 for $380-400 might sway my decision. Even if FE1080 was $499 and AIB 1080 was $599, I'd still pay $100 extra for the AIB card. That's how much NV got this wrong. Honestly I can easily afford to change every card in my house to a GTX1080 FE $699 but I just cannot do it. I cannot send a message to a corporation that charging me more $ for inferior design is OK. It's not OK.

If the FE1070/FE1080 ain't your jam, why not buy a card that better suits your needs from one of the AIB vendors?

I can't honestly believe that somebody who goes to such great lengths to try to educate people on how to get the best value for their $ on these boards would purposely buy an inferior product at a worse price just to send a message to a corporation that will not be received .

 

[Mahigan]

That's an Asus GTX 980 Ti. Not only does it contain 12 VRM phases but they're higher quality phases.

Another example is the MSI Lightning boards..

Reference designs have historically contained 6 high quality phases. NVIDIA have just dropped this down to 5.

 

[Arachnotronic]

NVIDIA are cutting corners in order to improve their profit margins whilst also charging a premium tax on early adopters.

The same reference board will likely be used in the $599 variants, so keep that in mind. NV is only claiming a premium cooler shroud, not anything out of the ordinary on the reference PCB.

 

[Erenhardt]

What is the significance of 6 vrms over 5?
Are 6 needed? Is 5 plenty? What is it?
Why is 6 a magic number of VRMS/phases? Why not 20?
I'm not an electrical engineer. Just trying to understand.

If you try to deliver the power you transfer with 6 phases through 5 phase power delivery circuit, it will go pop.

Plenty of reviews overclocking 1080fe with aftermarket cooler hit power limit. Additional phase would definitely help lift that overclocking bottleneck.

FuryX was an overclockers dream - when looking at the pcb and cooling solution compared to 1080fe.

1080fe is an overclockers dream too! But that dream is a Nightmare, because your GP104 is chained, closed in cage, which is again cuffed in chains and dropped deep in the ocean.

 

[Mahigan]

GTX680 is missing 1(got 5), GTX980 is missing 2(got 5).

GTX1080 is missing 1(got 5).

Nothing new as usual besides the usual FUD.

Mission critical servers dont have 8-32 phases either. They have 4-5.

The GTX 1080 FE is selling for more than the GTX 980 Ti did. So compare it with a board of similar pricing..

http://www.techspot.com/photos/article/1011-nvidia-geforce-gtx-980-ti/

See, 6 VRM phases.

So it's not FUD.

 

[Arachnotronic]

1080fe is an overclockers dream too! But that dream is a Nightmare, because your GP104 is chained, closed in cage, which is again cuffed in chains and dropped deep in the ocean.

This is a great piece of imagery :thumbsup:

 

[Mahigan]

The same reference board will likely be used in the $599 variants, so keep that in mind. NV is only claiming a premium cooler shroud, not anything out of the ordinary on the reference PCB.

That's not true. NVIDIA claimed top engineering from the FE.
http://forums.anandtech.com/showpost.php?p=38242302&postcount=87

Watch it yourself.

Clearly, the AIB vendors will be the ones to offer that once the FE reference run is over.

 

[RussianSensation]

Just keep ignoring all JHH's claims. Here's a reminder.

http://forums.anandtech.com/showpost.php?p=38242302&postcount=87

What makes it worse is that Computerbase used a low air flow case - Fractal Design R5 - exactly the "optimal" case which calls for a blower. Under these conditions, the card failed to reach 1733mhz boost after 20 minutes of gaming in every game other than the AofS. This proves the card fails exactly under the conditions it was designed to excel in over an open air cooled card. That's an engineering failure then because in a case with good to great airflow, the open air cooled card will destroy it. If the blower cannot even deliver promised spec performance in a case with poor airflow, then it has no purpose other than for those who water cool and/or intend to use miniATX boards in SLI. I bet NV loves that it will now shove $449/699 FE cards down BestBuy and OEM markets where most consumers aren't tech savvy to know how much of a blunder the FE cooler is.

Notice Computerbase has Max results in their chart that prevents throttling? To get there took 48 dBA in the same case/test bed where Sapphire Fury Nitro is 36 dBA and Fury X is 37 dBA:
http://www.computerbase.de/2016-05/geforce-gtx-1080-test/9/

I guess now noise levels and temperatures and GPU Boost throttling don't matter either, just performance. Goal posts moved again.

 

[Mahigan]

In this case, the power phases go together with the capped power delivery. At least it's not as bad as the GTX590.

When various sites tried using after-market cooling on the 1080 (Accelero/AIO CLC), the card reaches max power consumption limit NV set. Overclockers shows that with a base of 2000mhz, 1080 was using 220W:
https://www.overclockers.ru/lab/76273_4/obzor-i-testirovanie-videokarty-nvidia-geforce-gtx-1080.html

This means this reference/FE card is hopeless for extracting maximum potential out of the excellent GP104 chip. On one hand, this is a positive since it may foreshadow that GP104 has more headroom for overclocking with a proper power delivery and cooler. On the other hand, if AIB cards cost $650-700 and allow GP104 to overclock to 2.2-2.4Ghz, while running cooler and quieter, it means for most 1080 consumers the FE version is the card to skip.

In summary, the FE will run hotter, louder, thermal throttle, cannot turn off the fan at idle, and has a premium. NV says they made it so that people could "enjoy it" for the duration of a generation. All it is the Blower Myth kicked down another generation towards a group of GPU users that still believe a smaller fan blowing in a smaller heatsink is somehow superior to open air cooled cards -- they can afford a $700 card but cannot afford a $100 Phanteks Enthoo Pro or a similar modern 2015/2016 case?

Since NV doesn't officially support/encourage 3-4 way SLI anymore, the argument for a blower is worse than it's ever been. Once you install the card, you cannot even see the face of the cooler either.

If AMD/NV cannot make a good $550-700 reference card, don't make one AT ALL. Sapphire's Fury Air cost $569 a year ago and had an air cooler that makes 1080's look like a $15 aluminum gimmick. At max load, max overclocked, this card is more or less as quiet as the 980Ti is at idle! That's premium air cooling:
http://www.anandtech.com/show/9421/the-amd-radeon-r9-fury-review-feat-sapphire-asus/2

Remember when higher end coolers started to have copper and heatpipes over the far cheaper aluminum? The reason copper fell out of popularity was due to excessive weight for larger sizes coolers such as Noctua NH-D14. The sheer mass of such coolers ensured that copper was no longer the optimal material since then not all common motherboards could handle such loads.

If you took the Phanteks PH-TC14 and sliced it in the middle/half, you could make 2 separate coolers, each with a 140mm fan, that have better cooling capacity than the Titan blower. This cooler with 2x140mm fans sells for $60 in retail!
http://m.newegg.com/Product/index?itemnumber=N82E16835709002

NV needs to stop lying and just say we charged $70-100 more for a $20 air cooler + $80 worth of expensive marketing material because they knew a small fraction of their customer base will pay. The $100 premium could have been $125-150, but NV decided to try it out first to see how the market reacts.

By NV setting the MSRP of $699 for the FE card it has a ripple effect of some AIBs pricing their cards at $699 and higher. It's a double hit to the consumer between an overpriced and mediocre blower reference card that cannot even maintain advertised factory boost and a risk of waiting for $599 AIB cards that may cost more than that because NV's $699 MSRP made it so easy to justify > $599 MSI Gaming, Asus Strix, Gigabyte G1.

Because the FE 1080 cannot guarantee reference boost clocks are reasonable noise levels after hours of gaming, this point alone means the card isn't even worthy of NV's $599 MSRP as the customer isn't getting 100% of the promised spec 1080 performance.

Bingo.

The $599 cards will actually turn out to be superior with the AIB vendors absorbing the extra costs associated with high quality electrical components.

 

[Piroko]

The 195W GTX 680:

Four VRM phases for the GPU plus two smaller ones for Ram, also 3 (?) MOSFETs per Inductor

The 165W GTX 980:

Four VRM phases for the GPU, 3 (?) MOSFETs per Inductor.

The 185W GTX 1080:

Five VRM phases for the GPU, only one MOSFET per Inductor.

I'm actually curious why they've used three MOSFETs on the 680/980 and only one on the 1080. I guess using more than two improves temperature and stability and using less than two might marginally improve efficiency. Two is something I would have expected, as to why - here's a decent ling to read up on: http://www.overclock.net/t/943109/about-vrms-mosfets-motherboard-safety-with-125w-tdp-processors

 

[Mahigan]

What makes it worse is that Computerbase used a low air flow case - Fractal Design R5 - exactly the "optimal" case which calls for a blower. Under these conditions, the card failed to reach 1733mhz boost after 20 minutes of gaming in every games other than the AofS. This proves the card fails exactly under the conditions it was designed to excel in over an open air cooled card. That's an engineering failure then because in a case with good to great airflow, the open air cooled card will destroy it. If the blower cannot even deliver promised performance in a case with poor airflow, then it has no purpose other than those who water cool and miniATX boards in SLI.

Notice Computerbase has Max results in their chart that prevents throttling? To get there took 48 dBA in the same case/test bed where Sapphire Fury Nitro is 36 dBA and Fury X is 37 dBA:
http://www.computerbase.de/2016-05/geforce-gtx-1080-test/9/

I guess now noise levels and temperatures and GPU Boost throttling don't matter either, just performance. Goal posts moved again.

Overclocking doesn't matter anymore either. Oh and heat doesn't matter too.

I swear, you can't make this up. I'd like to create a collage of folks who make changing claims as generations of GPUs release in an effort to back their favorite side. No doubt I'd get banned though.