[Various] NVIDIA GeForce GTX 1080 Review Thread

[Head1985]

1080 is still memory bandwidth bottleneck.
OC memory from 10 to 11Ghz brings 6% AVG performance.
10 to 11Ghz is only 10% OC
http://www.computerbase.de/2016-05/...ramm-gddr5x-auf-der-gtx-1080-benchmark-part-2

 

[showb1z]

I had two MSI GTX 980 Ti (6GD5T) and they hit a brick wall at 1400MHz. Not the world's best aftermarket designs, but probably better than the ref blower. And these things ran hot & loud at those speeds.

Anyway, haters gonna hate. But I'm glad I sold the 980 Tis to fund the purchase of my 1080s. And I look forward to selling the 1080s when I get a whiff of the 1080 Ti release dates.

I like how the 980Ti is getting downplayed all of a sudden. :sneaky:
980Ti is still a beast, 1080 is very meh so far.

 

[mohit9206]

Gigabyte 980Ti extreme gaming-$690
Reference 1080-$700
Both have same performance yet you pay more for newer gen card.
Sure 1080 can OC too but here we are looking at out of the box performance of both these cards and the fact that 1080 loses to previous gen 980Ti is shameful.

 

[Erenhardt]

I like how the 980Ti is getting downplayed all of a sudden. :sneaky:
980Ti is still a beast, 1080 is very meh so far.

Well, I was lead to belive each gtx 980ti hits 1500mhz guaranteed.
Now we wait for those 2400mhz overclocks on 1080.
So far, waiting for a big boys seems to be the way to go.

 

[Freddy1765]

I like how the 980Ti is getting downplayed all of a sudden. :sneaky:
980Ti is still a beast, 1080 is very meh so far.

Yeah, totally. A chip half the size delivering 30% more performance is super meh.

:whiste:

 

[Noctifer616]

Yeah, totally. A chip half the size delivering 30% more performance is super meh.

:whiste:

Would that 30% performance difference still be there if 20nm actually happened and wasn't skipped? The jump from 28nm to 16nm FinFET is massive.

 

[antihelten]

3Dcenter just posted their launch analysis, collating results from over a dozen different sites.

Results:

So overall the 1080 is 60-70% faster than the 980 depending upon resolution, and 30-32% faster than the 980 Ti.

 

[selni]

Would that 30% performance difference still be there if 20nm actually happened and wasn't skipped? The jump from 28nm to 16nm FinFET is massive.

28/20/16nm are more marketing names than anything else these days, 28->20 was a node jump, but 16nm is very close to 20nm + finfets. Still a big jump, but not quite what the names imply.

 

[DDH]

Gigabyte 980Ti extreme gaming-$690
Reference 1080-$700
Both have same performance yet you pay more for newer gen card.
Sure 1080 can OC too but here we are looking at out of the box performance of both these cards and the fact that 1080 loses to previous gen 980Ti is shameful.

Come off it, your whole post is total fantasy. For the same price of course someone would purchase the 1080. Never mind that the 980ti is already over clocked balls to the wall. The 1080 can be pushed further still and the consumer gets more value for same cash. And how exactly does the 1080 lose to the previous generation?

 

[Qwertilot]

Would that 30% performance difference still be there if 20nm actually happened and wasn't skipped? The jump from 28nm to 16nm FinFET is massive.

Absolutely yes. NV have essentially done 30 per cent year on year for 5 years solid now! They'll probably achieve it for around another 5 or so too.

The slack they have to respond to competition is probably more about pricing/pulling things forwards/back etc than anything else.

They wouldn't have been able to make Maxwell (esp the 980ti) remotely as big on a new 20nm node of course. Getting their +30% gains even while still on 28nm was a serious achievement.

 

[know of fence]

So with the new cards people get even less transistors per dollar, less cores per dollar and less bandwidth.
It's completely weird to see performance gains achieved exclusively through higher clocks on a GPU. Isn't a new node supposed to increase / double the number of transistors running at similar clocks (around 1 GHz). I guess we'll have to look forward for incremental improvements for the next 5 years or however long it will take, which defeats the traditional rationale of buying a "powerful" card early in the cycle.

Hey I don't even care, nobody knows where video games are going and how long it will take to get there anyway.

 

[showb1z]

Yeah, totally. A chip half the size delivering 30% more performance is super meh.

:whiste:

It's a full node shrink, how is that impressive?
OC vs OC it's only 10-15%, and it's a regression in performance/$. We'll see what aftermarket cards bring, FE is a joke.

 

[cytg111]

Any of the reviews tackled SMTP yet?

 

[Aristotelian]

It's a full node shrink, how is that impressive?
OC vs OC it's only 10-15%, and it's a regression in performance/$. We'll see what aftermarket cards bring, FE is a joke.

The problem that I see from my end is that Nvidia is looking to charge (roughly) Ti prices for its mid range card. From a performance perspective, the 1080 does indeed stomp the 980, but is much more expensive.

Other posters on this forum are sourcing prices of EUR 700 and more per card, which is insane for a mid range (in Nvidia's lineup) card. So if I'm going to buy a 1080 hydrocopper from EVGA I should pay nearly EUR 1000 per card? From my posts here months ago I was hoping to get the Pascal Titan for around EUR 1000 per card, but I guess I was hopelessly optimistic.

I would have much preferred seeing something like this:

Pascal Titan (HBM2 16gb) = EUR 1099
Pascal Ti (HBM2 16gb) = 799
Pascal 1080 = 599
Pascal 1070 = 399

Roughly something like that. I do not mind (at all) paying for performance but I'll repeat my point here that asking even someone like me to pay EUR 700 plus for a mid range card is just insane, and I really mean 'mid range in the pascal lineup'.

 

[Timmah!]

3Dcenter just posted their launch analysis, collating results from over a dozen different sites.

Results:

So overall the 1080 is 60-70% faster than the 980 depending upon resolution, and 30-32% faster than the 980 Ti.

Its funny that when you take that Nvidia roadmap slide with major Pascal improvements over previous gens into consideration (FP16,HBM2,NVlink), GP104 boasts none of them.

 

[beginner99]

3Dcenter just posted their launch analysis, collating results from over a dozen different sites.

Results:

So overall the 1080 is 60-70% faster than the 980 depending upon resolution, and 30-32% faster than the 980 Ti.

And if we look at that table more closely we also see that the 1080 pretty much the same amounts more expensive than 980 and 980 TI ("Mehrpeis der GTX 1080").

So after more than a full node shrink Nvidia offers you the same performance/$ as before. I anyone still thinks this is great, I don't really know what to say to them...

 

[Grooveriding]

Has anyone seen a review of the 1080 where they used the built in benchmark of a game ?

 

[imaheadcase]

i cant believe the tech press isnt up in arms about the outrageous price. truly outrageous

Yah $50 more than a card released a long time ago, such pitchforks are needed..lol

 

[DamZe]

Yah $50 more than a card released a long time ago, such pitchforks are needed..lol

It's not that pitchforks are needed, but what seems to get to people is the way NVidia is pretending to have launched something truly revolutionary and pricing their mediocre reference cards at a unreasonable price Premium, we can all see from the specs alone that the 1080 is not a monumental leap from the 980Ti/Titan X. Sure it brings much better performance/watt metrics, and remarkable clock speeds but features less of everything else, less shaders and a 256-bit bus. It absolutely looks like a Maxwell Card on steroids (clock wise, and the IPC performance over Maxwell is questionable at best, it might even be identical).

 

[Qwertilot]

Roughly something like that. I do not mind (at all) paying for performance but I'll repeat my point here that asking even someone like me to pay EUR 700 plus for a mid range card is just insane, and I really mean 'mid range in the pascal lineup'.

People really have to get away from thinking of these medium cards and the really big cards as part of the same line up. Maybe they were 5+ years ago.

Now, no. NVidia have a rolling, annually refreshing (at an absurdly regular +30% increase/year) line up of GFX cards. The fastest card at a given time alternates in size between big and medium but has a roughly fixed price.

The 1080 is at the top of the 2016 line up.

The 2017 line up will near certainly be Big Pascal of some sort, maybe cut big P, 1080, 1070 etc. Possibly a rebadge, and likely price cuts for the 1080 at least.
(This could vary a little depending on how AMD's competition goes. Didn't happen with the 970/80 of course but then AMD didn't really turn up.).

If you think about things the way NV are pricing things is forced given the way they're releasing cards.

Now if people want to complain that could/should be releasing the big cards sooner after the medium ones then go ahead. That might not be entirely fair - big Pascal will really gain from having HBM2 say, ~early 2017 earliest for that - but it would at least be fully coherent

Marketing is always daft.

 

[Abwx]

. Isn't a new node supposed to increase / double the number of transistors running at similar clocks (around 1 GHz).

The transition to Finfet is not an usual shrink, density wise it s about a full node shrink, so you can pack twice the transistor count on a same area..

Electrical characteristics wise this is equivalent to almost two nodes shrinks, that is, you could pack 3.33x the transistor count at same TDP if frequency is left unchanged.

Hence density is lacking to make a full use of the electrical parameters improvements, so if it s 2x the transistors count there will be a margin left (sqrt(1.665) = 1.29) to increase frequency, if the transistor count is left unchanged this allow to increase frequency by 65% at same design, at least with TSMC s 16FF+.

 

[Det0x]

The problem that I see from my end is that Nvidia is looking to charge (roughly) Ti prices for its mid range card. From a performance perspective, the 1080 does indeed stomp the 980, but is much more expensive.

Other posters on this forum are sourcing prices of EUR 700 and more per card, which is insane for a mid range (in Nvidia's lineup) card. So if I'm going to buy a 1080 hydrocopper from EVGA I should pay nearly EUR 1000 per card? From my posts here months ago I was hoping to get the Pascal Titan for around EUR 1000 per card, but I guess I was hopelessly optimistic.

I would have much preferred seeing something like this:

Pascal Titan (HBM2 16gb) = EUR 1099
Pascal Ti (HBM2 16gb) = 799
Pascal 1080 = 599
Pascal 1070 = 399

Roughly something like that. I do not mind (at all) paying for performance but I'll repeat my point here that asking even someone like me to pay EUR 700 plus for a mid range card is just insane, and I really mean 'mid range in the pascal lineup'.

World prices then (suggested MSRP for Founders edition):

Serbian Dinar RSD 96,900
Czech Koruna CZK 21,400
Danish Krone DKK 6,150
European EUR 789
British Pound GBP 619
Hungarian Forint HUF 259,850
Norwegian Krone NOK 7,599
Polish Zloty PLN 3,599
Romanian New Lei RON 3,499
Russian Rouble RUB 54,990
Indian Rupee INR 63,250
Swedish Krona SEK 7,699
Turkish Lira TRY 2,850
South African Rand ZAR 13,599
Switzerland CHF 790
UAE AED 2,850
USA USD 699

Better make that ~1200 EUR for the Pascal Ti, and ~1600+ EUR for Pascal Titan in mid/late 2017 :whiste:

Its simply not worth the money.. Especially when you consider that a max oced 1080 is only ~13% faster then a max oced 980ti

 

[Aristotelian]

People really have to get away from thinking of these medium cards and the really big cards as part of the same line up. Maybe they were 5+ years ago.

Now, no. NVidia have a rolling, annually refreshing (at an absurdly regular +30% increase/year) line up of GFX cards. The fastest card at a given time alternates in size between big and medium but has a roughly fixed price.

The 1080 is at the top of the 2016 line up.

I get the point that you're trying to make, but please try to understand mine.

From the German link above they argue:

"Damit bleiben noch einige Fragen offen, welche sich dann aber erst mit dem Auftauchen der Herstellerdesigns zufriedenstellend beantworten lassen. Dazu gehört auch die allerwichtigste Frage – diejenige nach dem Preis/Leistungs-Verhältnis. Eine Antwort hierauf kann derzeit nur sehr eingeschränkt gegeben werden, da die Herstellerdesigns fehlen sowie zum Referenzdesign der Founders Edition nur Listenpreise vorliegen, aber noch keine echten Straßenpreise. Davon ausgehend, das sich die nVidia-Vorgabe von 789 Euro tatsächlich so im deutschen Einzelhandel zeigt (der US-Listenpreis würde direkt umgerechnet eigentlich eher ~735 Euro ergeben), kostet die Karte gegenüber Radeon R9 Fury X und GeForce GTX 980 Ti ziemlich exakt so viel mehr, wie viel sie auch mehr an Performance erbringt. Dies ist oberflächlich in Ordnung, normalerweise bezahlt man für Spitzenperformance gern auch einmal klar mehr als was es an Leistungsschub gibt."

My courtesy translation as a native German speaker will be: it's normal to pay for more performance, and the 1080 brings quite a bit more performance than a 980 Ti and a Fury X. Fantastic - generation X+1 produces graphics cards faster than those in generation X. If your argument (and the one from the website) were extended, it would follow that graphics cards continue to increase in price forever (with each generation). The 'buying now' element that you cite in your post about timing (that this will be the fastest single gpu card in 2016) is simply timing manipulation by a company to try to get as much money as it can for products that it wants to time right - milk a bunch of money from (legitimate upgraders and those who want more but don't want to wait much longer) and prolong the 'need' element so that someone like me would upgrade repeatedly.

They are free to do that - businesses exist to maximise profits and to survive as long as possible, but I am also free to call them out on it. Buying a 1080 for something like EUR 800 is a joke, considering where it sits in Nvidia's product lineup - which they should not release if they themselves don't want us to see the card as anything more than stopgap milking. Was there anyone happy they bought a 980 over a 980Ti? I doubt it.

Did anyone pay as much for a 980 as a 980Ti? I'm not sure - potentially the day the 980 was released?

 

[TestKing123]

Seems like your 1300 MHz is the outlier, over 50% of the cards in these TPU reviews clocked over 1500 and more than 80% clocked over 1450 MHz. Small sample size of course but 1500 MHz or close to it is certainly not a golden sample. I'd call 1450 Mhz "routine clocks".

And exactly how stable are those clocks? Sorry, but overclocking a card for a quick benchmark run doesn't cut it. I can clock to 1450 for maybe a quick Firestrike run, but playing Witcher 3 for 2 hours straight? Not without some throttling and eventual crashes/reboots that miraculously disappear after backing down to normal clocks.

Oh, and how many of those are SLI friendly? What's the point of slapping on a cooler that allows 1400 stable clocks but requires 3 fans to blow into your case rather than out the case? That won't work with a tight SLI setup, that requires a rear exhaust like in my case otherwise heat is going to be an issue real quick (at which point the card will throttle).

 

[sirmo]

@Aristotelian

Yup, that's Nvidia for you, they will squeeze every cent out of you they can. Other companies aren't much different per se. But Nvidia takes the cake when it comes to deceptive pricing or marketing.

There is nothing major in 1080 manufacturing costs which warrants a higher cost than say 980, and even 980 had large profit margins.

It's their prerogative however. And it's our prerogative to vote with our money.