NVIDIA Pascal Mythbusting

[nycesquire]

180W TDP? Looks like Antimemetic is getting a GTX 1080... 

[vanished]

So far everything I've seen claimed is amazing!!    Now we just need 3rd party benchmarks of game performance to backup what we've been told (or disprove it, if it ends up that way )

[Krzych]

If someone told me year ago that price of my GTX 980 Ti is going to drop below $300, I wouldn't believe. But it looks like this right now, GTX 1070 will overperform it. 

[lukart]

Now lets see the performance, in which games and situations that plays out.

Usually they always play with the best case scenarios... But it's looking good for sure.

[HKZeroFive]

@Glenwing, any chance we're gonna get an update of the OP with the reveal of the GTX 1070 and GTX 1080? The marketing of NVIDIA has really done a number on this one.

 

'2s Faster than the Titan X!'. 'Faster than 980 SLI!'. 'Both the GTX 1070 and GTX 1080 are faster than the Titan X!'.

[Glenwing]

25 minutes ago, HKZeroFive said:

@Glenwing, any chance we're gonna get an update of the OP with the reveal of the GTX 1070 and GTX 1080? The marketing of NVIDIA has really done a number on this one.

 

'2s Faster than the Titan X!'. 'Faster than 980 SLI!'. 'Both the GTX 1070 and GTX 1080 are faster than the Titan X!'.

We'll see. Right now I don't really have any more insight than anyone else since information is limited, other than saying these claims aren't that unbelievable, personally.

 

If you compare the middle chip of each architecture to the middle chip of the previous architecture, Maxwell GM204 (GTX 980) is roughly twice as fast as Kepler GK104 (GTX 680), which was roughly twice as fast as Fermi GF114 (GTX 560 Ti). So it wouldn't be that strange if Pascal GP104 was twice as fast as Maxwell GM204.

 

Likewise, if you compare the middle chip to the previous architecture's flagship, Kepler GK104 (GTX 680) was roughly 20-40% faster than Fermi GF110 (GTX 580). Maxwell GM204 (GTX 980) was roughly 5-10% faster than Kepler GK110 (GTX 780 Ti / Titan Black) at launch. So if Pascal GP104 is 20-40% faster than GM200 (GTX 980 Ti / Titan X), again it won't be that strange.

 

Maxwell is the only one slacking a little in this pattern, the GTX 980 isn't quite twice as fast as the 680 though it's pretty close, and it's not quite 20-40% faster than the 780 Ti. But this makes sense when you consider that Maxwell didn't have a fabrication process advantage (it was on 28nm, same as Kepler). Fermi to Kepler also went from 40nm to 28nm, and Maxwell to Pascal is also going from 28nm to 16nm.

[HKZeroFive]

12 minutes ago, Glenwing said:

We'll see. Right now I don't really have any more insight than anyone else since information is limited, other than saying these claims aren't that unbelievable, personally.

 

If you compare the middle chip of each architecture to the middle chip of the previous architecture, Maxwell GM204 (GTX 980) is roughly twice as fast as Kepler GK104 (GTX 680), which was roughly twice as fast as Fermi GF114 (GTX 560 Ti). So it wouldn't be that strange if Pascal GP104 was twice as fast as Maxwell GM204.

 

Likewise, if you compare the middle chip to the previous architecture's flagship, Kepler GK104 (GTX 680) was roughly 20-40% faster than Fermi GF110 (GTX 580). Maxwell GM204 (GTX 980) was roughly 5-10% faster than Kepler GK110 (GTX 780 Ti / Titan Black) at launch. So if Pascal GP104 is 20-40% faster than GM200 (GTX 980 Ti / Titan X), again it won't be that strange.

 

Maxwell is the only one slacking a little in this pattern, the GTX 980 isn't quite twice as fast as the 680 though it's pretty close, and it's not quite 20-40% faster than the 780 Ti. But this makes sense when you consider that Maxwell didn't have a fabrication process advantage (it was on 28nm, same as Kepler). Fermi to Kepler also went from 40nm to 28nm, and Maxwell to Pascal is also going from 28nm to 16nm.

Ah, I was hoping if you could distinguish the misconceptions that the performance gains varies between VR and real-world gaming. Like you said, I'd be expecting the GTX 1080 to outclass the GTX 980Ti by 20% or so judging from NVIDIA's own graphs/slides.

[Undertow_]

I'm still waiting to see how they compare to the 900 series in real world tests. Hype is a bitch sometimes.

[Simon771]

I saw that Linus tech guys already have 1080 card.

Any idea when can we expect benchmarks of 1080 and 1070?

[Big-G]

8 hours ago, Simon771 said:

I saw that Linus tech guys already have 1080 card.

Any idea when can we expect benchmarks of 1080 and 1070?

Don't expect anything before 17 May

[Bhav]

9 hours ago, Simon771 said:

I saw that Linus tech guys already have 1080 card.

Any idea when can we expect benchmarks of 1080 and 1070?

According to google fu, 17th may is NDA lifted date.

[Simon771]

7 hours ago, Bhav said:

According to google fu, 17th may is NDA lifted date.

May I ask what is NDA?

[Big-G]

1 minute ago, Simon771 said:

May I ask what is NDA?

Non Disclosure Agreement - basically a gagging order till they feel the market is ready or enough hype is created 

[Simon771]

4 hours ago, GidonsClaw said:

Non Disclosure Agreement - basically a gagging order till they feel the market is ready or enough hype is created 

So if I understand that right ... Linus and like 100 other people have that card and can test it, but need to wait till 17th May to relsease benchmarks and further informations about it?

I'm sure LinusTech won't leek informations before that, but if realy like 100 card are already out, I belive someone will leak it before  

[vanished]

1 hour ago, Simon771 said:

So if I understand that right ... Linus and like 100 other people have that card and can test it, but need to wait till 17th May to relsease benchmarks and further informations about it?

I'm sure LinusTech won't leek informations before that, but if realy like 100 card are already out, I belive someone will leak it before  

Most people try to not break NDAs because it makes the chance that they will ever get another product ahead of time almost nothing, since no company would trust them from then after.  And for a channel like LTT or anyone else who does this for a living, that's a death sentence since it means they would have to line up and buy the product like everyone else when mass availability hits, and that would mean they can't get their review out in time with everyone else's, and often times being really late with these kinds of things means no one will care.

[Simon771]

4 minutes ago, Ryan_Vickers said:

Most people try to not break NDAs because it makes the chance that they will ever get another product ahead of time almost nothing, since no company would trust them from then after.  And for a channel like LTT or anyone else who does this for a living, that's a death sentence since it means they would have to line up and buy the product like everyone else when mass availability hits, and that would mean they can't get their review out in time with everyone else's, and often times being really late with these kinds of things means no one will care.

Yeah I get that, but someone could leak infos annonimusly.

[vanished]

58 minutes ago, Simon771 said:

Yeah I get that, but someone could leak infos annonimusly.

But then what would be the point?  Usually, the point of giving up the info (writing an article, making a video, etc.) is to get views - attention - recognition.  There would still be some risk with getting found out, and for what personal gain?

[lukart]

Not sure if you guys seen this video, but sure it's good food for thought

 

[vanished]

25 minutes ago, SettlingAbyss96 said:

http://www.overclock3d.net/articles/gpu_displays/gtx_1080_ashes_of_the_singularity_benchmarks/1

 

just putting this out there...

So clearly still not handling DX12 well...

[superfly06]

would pascal cards require pci express gen 3

[Mira Yurizaki]

On 5/12/2016 at 3:34 PM, SettlingAbyss96 said:

I'm honestly not surprised, AMD is years ahead of Nvidia when it comes to DX12 (which is heavily based off mantle). Also another factor to think about is that AMD is producing both major consoles. Facts are Facts and significantly more games are sold on consoles than on PC so logically I would expect to see the majority of the future titles to be more optimized towards AMD hardware than Nvidia. More than ever actually especially with the Xbox 1's performance compared to the PS4.

The reason why AMD is killing NVIDIA on DX12 is that NVIDIA switched to a software scheduler whereas AMD uses a hardware scheduler. DX12 seems to want a hardware scheduler more than a software scheduler. Though all this really means is NVIDIA needs to step up that aspect of their drivers.

 

Also console optimization doesn't mean anything. Game developers on consoles optimize for the entire system, not just the GPU. Very few, if any, console optimizations are going to work on a PC unless it's a generic algorithm. Not to mention Sony doesn't use AMD's APIs, they used a custom in-house one. So none of those optimizations are going to work for any PC system (not to mention PS4 uses BSD as the OS).

 

On 5/12/2016 at 6:10 PM, superfly06 said:

would pascal cards require pci express gen 3

No. No graphics card has actually needed anything beyond PCIe 1 in terms of bandwidth. Well, maybe except the dual-GPU cards.

[Bhav]

I'll add this information to this thread too ...

 

1080 founders edition throttles within 5 minutes of gaming when used inside a PC case, and disables the boost clock.

The throttled performance at 1607 Mhz is roughly 10% SLOWER than a factory overclocked 980 Ti at 1440p.

If the 1080 doesnt throttle (open bench results), its about 2% faster than a factory overclocked 980 Ti, more or less depending on how overclocked.

When overclocking the 1080 without throttling, performance increases minimally / negligably - 1-2 FPS. It does not scale in performance anywhere close to previous generations.

 

Most of that information is explained with sources in this video:

 

 

Only buy the founders edition if you plan to use a custom water block, or wait for the better custom designs with coolers that can prevent the throttling on air.

 

Also dont 'upgrade' to one from a 980 Ti unless you only have a single brain cell.

[vanished]

1 hour ago, Bhav said:

*snip*

very interesting.  One would hope nvidia would have bothered ti make the reference version actually functional in and of itself, but I guess they're taking a pass on that this time around.  Hopefully OEMs will be able to make air cooled versions that actually perform as advertised.

[Bhav]

1 minute ago, Ryan_Vickers said:

very interesting.  One would hope nvidia would have bothered ti make the reference version actually functional in and of itself, but I guess they're taking a pass on that this time around.  Hopefully OEMs will be able to make air cooled versions that actually perform as advertised.

Well it is functional at the boost clock if you turn the fan up. But those blower fans are noisy.

[lukart]

It's funny to see Nvidia rushing now to come up with the 1060, which was meant to be later on around sept / oct now to be announced later in July as the latest update.

They won't use their 106 chip but instead they will use cut down 1080/1070 GPUs to go to 299$ price point. 

 

Interesting times...