[Various] NVIDIA GeForce GTX 1080 Review Thread

[Arachnotronic]

Well, this way some people might double dip. Doesn't work the other way around.

Eh, I think not putting your best foot forward leaves you vulnerable to surprises from the competition.

There will be some who double dip, but I doubt that they represent much more than a small fraction of the overall PC buying public.

I think my strategy this time around will be to buy one 1080 for each of my gaming systems. Once 1080 Ti or whatever comes out, I'll use two 1080s in SLI in one system and a single 1080 Ti in another.

Once I get a hint that the next gen is imminent, I will unload all of those GPUs and use an iGPU until the new stuff comes out. Rinse and repeat.

 

[Arachnotronic]

Mid means there is at least 1 better and at least 1 worse. GP104 is a mid range chip. Period.

The (gaming) stack is as follows, per 3DCenter:

GP102
GP104
GP106
GP107
GP108

"Mid-range" here looks like the GP106, performance tier looks like GP104, and extreme is GP102.

 

[.vodka]

You do realize that the reason NVIDIA puts out the GP104 parts first is that the "big" chips aren't ready until a good while later? Don't you think that if GP102 were ready to ship now, NVIDIA would love to have the 1080 Ti/Titan variant out now to grab even more money from customers?

The GP104 is the best that NVIDIA can release now, so calling it "mid-range" is just silly.

I do realise what you're proposing and the result of it all is that you're getting screwed if you upgrade each time they release a new card. Truth is, it's a strategy that just works for them and will probably work for the third time in a row.

I just don't agree with nV charging flagship prices for a card that will be utterly destroyed by their true flagship, the big chip in a few months' time, a card that you could buy just with a few more extra hundred dollars over a 1080. I mean, if one is to buy a 1080 they're going to part with $600-700, be it a scam edition or a proper aftermarket card. That's a lot of money, probably 80% at the very least of what a "1080Ti" will cost.


GK110 destroyed GK104 for a little more. GM200 made GM204 look like a joke only for $100 more ($550 vs $650 MSRP, right?). The same will happen to GP104. Each time the performance bar gets higher, that's true... whether that extra performance the mid range line brings to the table is worth it, that's up to one's wallet.


That's why I mentioned in my other post to stay into either of the product lines when upgrading to maximise bang for your buck, not jumping into each new release because then you're getting screwed hard.

 

[Flapdrol1337]

Eh, I think not putting your best foot forward leaves you vulnerable to surprises from the competition.

There will be some who double dip, but I doubt that they represent much more than a small fraction of the overall PC buying public.

I think my strategy this time around will be to buy one 1080 for each of my gaming systems. Once 1080 Ti or whatever comes out, I'll use two 1080s in SLI in one system and a single 1080 Ti in another.

Once I get a hint that the next gen is imminent, I will unload all of those GPUs and use an iGPU until the new stuff comes out. Rinse and repeat.

You're going to double double dip then?

I don't think amd and nvidia really surprise eachother. If one of them was producing a top end part in high volume the other would know.

I'm at least waiting for polaris before I decide when to get what.

 

[crisium]

That's right. Blowers aren't effective as the load approaches 200w and you're also trying to keep noise levels down.

We've all seen how GM204 did just fine on the blower while GM200 throttled like crazy. GP104 on the other hand behaves more like an overclocked GM204 (it's rated for "180w" vs GM204's "165w"), and the blower struggles.

If you want the blower to keep up, you have to drive fan speeds up too. Hello 290 blower of hell edition. It wouldn't surprise me if the 67°C load temperatures nV showed in the launch event had the blower running at 100%

The 1080 scam edition is actually a pretty poor value for the money. The same reference PCB with its crippled VRM section is better off with a decent open air cooler that can actually handle the GPU's thermal load. Proper PCBs with beefed up VRMs ready for >2.1GHz plus proper cooling are what one should aim to buy once they're available... if you're going to pay $600-700 for a mid range chip.

At what fan setting?

Both the default, and the default with maximum increased power target. The latter increases noise considerably by 5.5 dBs and still throttles 101Mhz below maximum boost.

http://www.computerbase.de/2016-05/geforce-gtx-1080-test/6/
http://www.computerbase.de/2016-05/geforce-gtx-1080-test/9/

Thus, it is a poor cooler for stock speeds, since stock speeds advertise 1886MHz and you will only boost there very briefly. Essentially, GPU Boost 3.0 should be called canned benchmark boost because this is the only scenario where it is appreciable.

Of course you can increase the fan speed manually to whatever loud setting you please, if you have top notch noise cancelling headphones this is certainly viable. And, while certainly not as loud as the 290 series references, the same applies there. Neither the 290 nor the 1080 reference coolers are admirable.

 

[sontin]

The stock speed is not advertise as max clock. It is advertised as base and boost clock.

And the reference cooler is quieter than custom 290X and 390X cards:
1080: http://www.hardware.fr/articles/948-13/nuisances-sonores-temperatures-photos-ir.html
290X: http://www.hardware.fr/news/13513/radeon-r9-290-290x-customs-nos-1ers-tests.htmlhttps://www.google.de/?client=firefox-b#q=hardware.fr+290x+custom&gfe_rd=cr
390X:http://www.hardware.fr/focus/110/msi-radeon-r9-390x-gaming-8-go-test.html

 

[tential]

Um, what? Maybe if you are making $750K-$1M+/year then $14K for a dGPU is fine, but for us peasants stuck in the realm of "normal" incomes, $14K is ludicrous.



Not sure if serious.

So you're now telling me it's unreasonable for Nvidia to price gpus based on what the market can pay? Why should Nvidia let scalpers benefit? After all Nvidia is a for profit company and by nature of demand we know people will buy across a large number of prices.

Auctioning the first 100-1000 gpus would give Nvidia the most profit.

Trust me. Nvidia will continue this style of premium pricing for early adopting?

What are you the gatekeeper of telling people what pricing model is exploitive but not too exploitive of Nvidia to use? Why shouldn't there be 2 founders models, the day 1 the day (release date unknown,) day +60 days after that, models? Allow Nvidia to get the max profit out as many people as possible.

I don't see how you can defend the founders model but then just decide to draw the line in the sand somewhere. If you're going to do something like this, you're going to continue down this line because it's profitable and works.

Expect more dynamic pricing from Nvidia because it's the intelligent thing to do.

 

[tviceman]

The (gaming) stack is as follows, per 3DCenter:

GP102
GP104
GP106
GP107
GP108

"Mid-range" here looks like the GP106, performance tier looks like GP104, and extreme is GP102.

I think the GP102 rumors finally need to die. If GP102 does exist, it's single precision would probably match GP100, making it redundant within the product stack. I doubt the cost to engineer a new 500mm2 chip would be cheaper than using an already existing 600mm2 chip.

 

[sontin]

I think the GP102 rumors finally need to die. If GP102 does exist, it's single precision would probably match GP100, making it redundant within the product stack. I doubt the cost to engineer a new 500mm2 chip would be cheaper than using an already existing 600mm2 chip.

GP100 is a HPC chip. GP102 will be used as a consumer and workstation chip.
nVidia had no problem to design and manufacture another Kepler chip while they were transforming to Maxwell.

 

[crisium]

The stock speed is not advertise as max clock. It is advertised as base and boost clock.

Right, and that advertised boost clock of 1886MHz cannot be achieved for more than a moment with default settings and even default setting + maximum Power Target.

Almost any blower can brute force cool temperatures to prevent throttling, including the infamous Hawaii reference, so it's hardly admirable that at default settings the 1080 FE throttles 200Mhz+, and even 100MHz with increased Power Target. So I don't see how this can be considered a good cooler for stock boost.

The stock speed is not advertise as max clock. It is advertised as base and boost clock.

And the reference cooler is quieter than custom 290X and 390X cards:
1080: http://www.hardware.fr/articles/948-13/nuisances-sonores-temperatures-photos-ir.html
290X: http://www.hardware.fr/news/13513/radeon-r9-290-290x-customs-nos-1ers-tests.htmlhttps://www.google.de/?client=firefox-b#q=hardware.fr+290x+custom&gfe_rd=cr
390X:http://www.hardware.fr/focus/110/msi-radeon-r9-390x-gaming-8-go-test.html

Even that "uber" setting is still throttling 100MHz+. Where's the noise measurement to get a mostly consistent 1886Mhz? I'm genuinely curious. A shame they didn't noise measure their OC 1080 which was hitting an impressive 1954MHz average, above the stock max boost which is very nice.

 

[Flapdrol1337]

Right, and that advertised boost clock of 1886MHz cannot be achieved for more than a moment with default settings and even default setting + maximum Power Target.

The advertised boost clock is 1733

 

[sontin]

Right, and that advertised boost clock of 1886MHz cannot be achieved for more than a moment with default settings and even default setting + maximum Power Target.

Almost any blower can brute force cool temperatures to prevent throttling, including the infamous Hawaii reference, so it's hardly admirable that at default settings the 1080 FE throttles 200Mhz+, and even 100MHz with increased Power Target. So I don't see how this can be considered a good cooler for stock boost.

There is only a base and a boost clock: http://www.geforce.com/hardware/10series/geforce-gtx-1080

 

[crisium]

Got it, so the 1886MHz is just to give the sites that still practice 30 second canned benchmarks an average increase. Thanks.

Presumably many non FE versions should hit 1886 consistently. So, even with it not advertised it still seems to point to the fact that the FE cooler is not up to snuff, since by default stock settings 1886 is technically achievable... but this cooler cannot really demonstrate it.

 

[sontin]

Got it, so the 1886MHz is just to give the sites that still practice 30 second canned benchmarks an average increase. Thanks.

Yes, it is basically a 3DMark setting. :thumbsup:

This is how Boost 2.0 is working since Titan: http://www.hardware.fr/articles/887-4/gpu-boost-overclocking.html

 

[thesmokingman]

Right, and that advertised boost clock of 1886MHz cannot be achieved for more than a moment with default settings and even default setting + maximum Power Target.

Almost any blower can brute force cool temperatures to prevent throttling, including the infamous Hawaii reference, so it's hardly admirable that at default settings the 1080 FE throttles 200Mhz+, and even 100MHz with increased Power Target. So I don't see how this can be considered a good cooler for stock boost.



Even that "uber" setting is still throttling 100MHz+. Where's the noise measurement to get a mostly consistent 1886Mhz? I'm genuinely curious. A shame they didn't noise measure their OC 1080 which was hitting an impressive 1954MHz average, above the stock max boost which is very nice.

It's sort of ironic in a way, Nvidia pulled a hawaii reference cooler hat trick. Who would have imagined that one?

 

[tential]

You do realize that the reason NVIDIA puts out the GP104 parts first is that the "big" chips aren't ready until a good while later? Don't you think that if GP102 were ready to ship now, NVIDIA would love to have the 1080 Ti/Titan variant out now to grab even more money from customers?

The GP104 is the best that NVIDIA can release now, so calling it "mid-range" is just silly.

Actually no they wouldn't. There is no competition for those big chips. Why put them out when Nvidia showed us in multiple generations now how segmenting their releases can get high end customers to buy multiple of their high end skus.

So no you're absolutely wrong and Nvidia's own track record shows it. Nvidia is maximizing the amount of gpu revenue they can get, hence why they don't follow the ridiculous plan you proposed of selling a ti/Titan now instead of waiting too after they've sold their 1070s/1080s to launch higher performing skus.

Posters here already intend to buy the 1070/1080 then upgrade to the Ti later so even more proof of how ridiculous the plan proposed is. Nvidia makes smart decisions, not ridiculous ones. Hence why they record amazing revenue figures.

 

[Arachnotronic]

Actually no they wouldn't. There is no competition for those big chips. Why put them out when Nvidia showed us in multiple generations now how segmenting their releases can get high end customers to buy multiple of their high end skus.

There absolutely will be competition for those big chips. Vega. Imagine if NVIDIA could get Pascal GP102 out now? AMD would literally have no answer for about six months.

The fact that NVIDIA hasn't struck in such a fashion tells me that GP102 simply isn't ready.

So no you're absolutely wrong and Nvidia's own track record shows it. Nvidia is maximizing the amount of gpu revenue they can get, hence why they don't follow the ridiculous plan you proposed of selling a ti/Titan now instead of waiting too after they've sold their 1070s/1080s to launch higher performing skus.

They are maximizing the GPU revenue that they get, but do you really think that people who "double dip" are so numerous so as to materially impact NVIDIA's bottom line? I just don't think that's very likely.

Posters here already intend to buy the 1070/1080 then upgrade to the Ti later so even more proof of how ridiculous the plan proposed is. Nvidia makes smart decisions, not ridiculous ones. Hence why they record amazing revenue figures.

OK, but those people will sell their 1070/1080 cards most likely and in effect "cancel out" the demand for a new card by selling a used 1070/1080 to a potential customer. I don't think it's as lucrative for NVIDIA as you think it is.

 

[tential]

There absolutely will be competition for those big chips. Vega. Imagine if NVIDIA could get Pascal GP102 out now? AMD would literally have no answer for about six months.

The fact that NVIDIA hasn't struck in such a fashion tells me that GP102 simply isn't ready.



They are maximizing the GPU revenue that they get, but do you really think that people who "double dip" are so numerous so as to materially impact NVIDIA's bottom line? I just don't think that's very likely.



OK, but those people will sell their 1070/1080 cards most likely and in effect "cancel out" the demand for a new card by selling a used 1070/1080 to a potential customer. I don't think it's as lucrative for NVIDIA as you think it is.

Amd has no answer for the 1070/1080...there is no need to launch a ti/Titan Nvidia is sitting on it until before the vega launch. Then they'll sell it too.

I have 2 years of precedent on my side. There is nothing you have presented showing Nvidia will do the opposite of what theyre doing for the 3rd year in a row. Sorry. Nvidia is right on this one. You're not arguing against me you're arguing against Nvidia. Good luck arguing against a money making machine. When you run a business more successful than Nvidia maybe you can chime in and say Nvidia isn't doin what it can to maximize profits. Right now, I trust Nvidia to know how to maximize profits more than you.

If Nvidia believes in the double dip, then it's real.

 

[Pneumothorax]

I've been suckered twice already by Nvidia on paying 'big-die' prices for their overpriced mid range cards in the 680/980. Not going to do it again...especially that they've tacked on an extra $200 due to some clever 'marketing'

Nvidia has done the 'double-dip' for 2 generations of GPU's and have gotten away with it as their competitor is hardly keeping up with them and realistically won't be for some time if not ever. I'm sure Nvidia has a pretty good info on what AMD is going to release considering how they timed the release of the 980ti so perfectly to steal Fury's thunder.

I really think these FE's are really for Nvidia/AIB's to cash on Newegg/Amazon price-gouging that is usually prevalent on a big new release as the first new node in almost 5 years. Typically, the retailers are the one's overpricing the cards and raking in the cash. Not saying that this isn't going to happen this time, but it's allowing the gfx card manufacturers to take a piece of the pie.

When I retire my 290x's it's going to be either for Vega or big Pascal.

 

[maddie]

I've been suckered twice already by Nvidia on paying 'big-die' prices for their overpriced mid range cards in the 680/980. Not going to do it again...especially that they've tacked on an extra $200 due to some clever 'marketing'

Nvidia has done the 'double-dip' for 2 generations of GPU's and have gotten away with it as their competitor is hardly keeping up with them and realistically won't be for some time if not ever. I'm sure Nvidia has a pretty good info on what AMD is going to release considering how they timed the release of the 980ti so perfectly to steal Fury's thunder.

When I retire my 290x's it's going to be either for Vega or big Pascal.

What if small Vega comes in around 5-6 months with no big Pascal in sight.

I can see this scenario playing as there only appears to have 1 pascal die above the GP104, but there are 2 Vega models. The small one should improve quite a bit on the 1080. Around 400mm^2 with (2) 4GB HBM2 stacks. Should have fair yields by Christmas, at least much better than a 600mm^2 die.

Would you still wait to see big pascal, even if several (5-6) months away?

 

[tential]

What if small Vega comes in around 5-6 months with no big Pascal in sight.

I can see this scenario playing as there only appears to have 1 pascal die above the GP104, but there are 2 Vega models. The small one should improve quite a bit on the 1080. Around 400mm^2 with (2) 4GB HBM2 stacks. Should have fair yields by Christmas, at least much better than a 600mm^2 die.

Would you still wait to see big pascal, even if several (5-6) months away?

Remember when people thought they caught Nvidia with their pants down with fury x? Launch 980ti! Nice try.

Remember when people thought they caught Nvidia pants down with Polaris? Pascal with the quick attack!

Nvidia isn't someone you're going to surprise with 0 counter.

 

[Pneumothorax]

Wow, Nvidia could've priced these at $999! Amazon sold out within minutes on a surprise pre-order. Man Nvidia, you just weren't greedy enough lol.

Linky

 

[tential]

Wow, Nvidia could've priced these at $999! Amazon sold out within minutes on a surprise pre-order. Man Nvidia, you just weren't greedy enough lol.

Linky

Like I said, 700 was way too low. It would have sold out at 999 I agree.
Why did Nvidia even include the cooler? They would have sold out at 999 with a founders edition to found a decent cooler.

 

[Timmah!]

There absolutely will be competition for those big chips. Vega. Imagine if NVIDIA could get Pascal GP102 out now? AMD would literally have no answer for about six months.

The fact that NVIDIA hasn't struck in such a fashion tells me that GP102 simply isn't ready.

I am pretty sure they are not releasing it now, cause they dont have to. 1080 is already faster than previous gen and there is no competition either to release anything faster than GP104. If there was, i am pretty sure we would see GP102 already by now, or simply GP104 would be bigger than 314mm2 and have more cores.

In other words, all that is happening is pre-planned and does not happen ad hoc. Watch GP102 to be released sometimes later, most likely next year and be pretty much GP104 + 1/3. Pretty much same scenario as GTX980 -> 980Ti.

 

[tviceman]

GP100 is a HPC chip. GP102 will be used as a consumer and workstation chip.
nVidia had no problem to design and manufacture another Kepler chip while they were transforming to Maxwell.

They didn't design another Kepler chip. They slightly modified an existing chip, which is a $100,000,000 difference.