Pascal: What do we know? Discussion, Latest News & Updates: 1000M Series GPU's

Based on 1080/1070 price performance - a 980ti is worth what?

1/3 of a 1080, which is $200?, or less?

 

Boy do I feel bad for Titan X owners, it's been a while since I've seen new gen tech murder the previous gen

 

30% increase from a 980 ti non VR performance. 980 TI can overclocked up to 30% on air for a decent card. I would buy a 980 ti for what? 250$?

Good time to pick some cheap 980 TI since everyone and their dog is dumping them. And this is why I dumped laptop setup before...

 

$300 or less Titan X's... Worth less than 1/2 1080.

 

thats for VR only performance....

 

The 980ti and Titan X aren't just markedly slower, they draw more power and block slots that could hold 1070/1080's.

And most everyone with a Titan X or 980ti's are going to want to dump them for 1080/1070's, a buyers market

 

Mobile pascal is going to be insane. The power effiency and low heat is going to make for some extremely thin and portable monsters. I'm imagining there being an aorus with 1070m sli and a 6820HK. Or a sager with 1080m sli and a 6700k!

 

Keep in mind that nVidia always overblows the actual performance and take it with a grain of salt until it's out in the wild. Besides, they can build a video card 100 times faster than the Titan X and it still won't mean crap if they can't make drivers for it.

 

I might be interested in a aorus with 6820hk and 1070m SLI if the damn thing holds its clock and dont sound like a leafblower on overdrive as soon as I put some load on it.

 

So a 980 is worth what? 1/4 or less of a 1080? $150 or less?

And 970's are those GPUs that aren't worth selling, so you keep it in the drawer for testing old MB's?

 

I don't see why they can't put a desktop 1080 in a laptop and just take it down a little from 2ghz, power and heat don't look like a concern at all in a big laptop

 

Ofcourse they cant. That means they cant milk the 1080m for as much as possible.

And 2ghz was the overclock.

 

Well AMD...it's do or die. Pascal brought the heat

 

It doesn't look like Nvidia wants to give up any market share.

Nvidia didn't deliver watered down Milk, not even Whole Milk, they brought nothing but the Cream off the top.

Full strength.

I'm feeling a full 1080 desktop for mobile coming.

No mobile 'm' low fat, it's Cream off the Top of Pascal for mobile.

Full strength

Maybe 1080m/1070m's too? For slim laptops.

A 1080 desktop at 180w is less than the 980 desktop mobile 200w, but more than the 165w 980 desktop spec.

 

180watt tdp?
Single 8pin power.
Been sittin' here w/a smile on me face. uh huh. Bodes well for lappys too. Maybe not the best comparison, but I wonder what one 'll do to a $1500 Radeon ProDuo. Two for that matter --> for $300 less ...

or three 1080's ... for $300 more, lol ...

 

This, seriously. From the graphic bars shown on their own page, 1080 is barely faster than a 980ti. It's what, 60% to 80% faster than a 980 for the shown FX12 titles? We don't know much info yet. They definitely created hype though.

It seems VR will get a marked improvement though.

And the GPU seems big. Up to 94 celsius and two slot size.

 

It's fun to put Jen-Hsun's words to the measure against the old generation; 1080 = 2x Titan X is big news.

The truth in gaming performance will come out soon enough.

The 2 1080 vs 980 gaming scores given @ 1733mhz would be 2x+ @ 2144mhz 1080 OC vs 980 stock.

It's going to be fun. Like Jen-Hsun said, smiles on all GTX 1080 owners faces

 

That 1080 founders edition is mine if they don't hint at mobile by June

 

They needed room for the massive vapor chambers cooling the thing lol.

Seriously though, scaling clock for clock, Pascal really isn't that impressive. It's the high clock speed of FinFET that brought the most gain. The VR optimization is obviously excellent but outside of that they didn't really bring a whole lot to the table that Maxwell didn't already have. GP100 is where the real gains will come and we won't be seeing the Ti come out until the Tesla orders are taken care of.

 

Well i've seen 980's at 1.6ghz already, but the point was about the actual performance. The slides definitely don't show 2x over a titan x, if they can't get over 60% on a single 980 on stock. I did watch the stream and was very intrigued but in the end felt disappoinment because it was just a hype machine. No actual numbers until the very end. Relative, Efficiency, Performance/watt. You can easily manipulate data.

I so very much want Pascal to be a powerhouse because I want super awesome GPUs. I am actually excited for the VR improvement since I am a big fan of Vive. But until I see actual numbers... I will believe. I am sure Pascal has much more to offer with the proper configuration and HBM2 memory, but for the time being, I will remain skeptical until full review

Yeah, VR optimization was a big hit for me so I am happy in that regard. Everything else, I am very skeptical for the time being. The released slides are not very encouraging...

Though in a way, there is no good excuse not to have a decent mobile GPU.

They could have easily kept the 384bit bus but I supposed the tradeoff in power efficiency was not worth it.

 

Well looking at the numbers, basically 30% of the performance increase can simply be attributed to the boost increase (1216 vs 1766). You get another 20% from the increase in CUDA cores from 2048 to 2560. Realistically they just refined Maxwell, shrunk it, and added a few optimizations then spun the numbers.

 

it's maxwell with higher clocks and an msrp of 699 for a midrange card... get your stuff together AMD

 

Not too sure we will see decent gaming perf increase on GP10p either tbh. Maybe 30-40% over 1080 at best...

 

http://arstechnica.com/gadgets/2016/05/nvidia-gtx-1080-1070-pascal-specs-pricing-revealed/

The Founders Edition:

Also, Founders Edition can only purchased from nvidia.com.

 

Basically a 90-100% gain over the 980 which isn't really anything to sneeze at. The problem I see though is that I doubt the GP100 can handle the clocks the GP104 can. It wouldn't surprise me to see GP104 at 2.3GHz with a volt mod and liquid cooling.

I guess we have to wait for the numbers to really start coming out. AMD better come out swinging though. Even with the misleading product placement with Pascal, the 1070 is tempting at its price point. More than enough for me anyway. I've gotten used to turning down settings on my games anyway since my 980Ms run hot as the core of the sun and I won't be gaming over 2K. 4K scaling is just disgusting.

 

I am getting 4k for screen real estate and I will probably pick up a pair of 1080 if only to test it out. Mostly waiting for an official launch of 6950x.

I dont want to wait another 6 month for gp100.

 

Guys from this release any guesses about how the 1080M would be ?

 

Cut down 1070 as usual. Full die in 2017, possibly even a 1080 notebook edition. There's no way they're going to give us the full die right off the bat, there would be no way to milk it and nothing to do with the dies that can't be used in desktop cards which they're sure to have a lot of with the massive clocks on the chips. GP106 should take care of the lower end of the spectrum up through the 1060M.

Nvidia won't want to close the gap between the desktop and laptop after seeing that they can charge such a premium for a full desktop card.

 

I've worked out that Pascal is only about 65% more performance per watt than Maxwell. Firstly, apologies if anything I'm about to mention has already been discussed or concluded, but haven't read the large number of pages created since yesterday yet. I read the article on Guru3D about the Announcment, and from the Performance vs Power graph showing Maxwell vs Pascal in that article (Image 8 shows the graph at the following link: http://www.guru3d.com/news-story/nv...and-geforce-gtx-series-1000-announcement.html) it can be extrapolated the Performance increase Per Watt, this is what I wrote over on Guru3D (cut & pasted):

That graph where they show Power vs Performance for Maxwell cards vs the GTX 1080, I've extrapolated that this configuration of Pascal has 'only' 65% more performance per Watt than Maxwell:

Titan X = 250W
GTX 1080 = 180W
Relative performance of 1080 to Titan X read from graph = 4.3/3.6 = 1.19 (or 19% more performance)
GTX 1080 uses 72% ( 0.72) of the power of Titan X: (180W/250W = 72%)

Using the above determined variables (in bold font) Extrapolated performance of GTX 1080 at same wattage as TitanX = 1/0.72*1.19 = 1.65 or 65% more performance per Watt.

I'd expected the performance per Watt to be at least double that of Maxwell due to the 2 node die shrink from 28nm to 16nm (they skipped the 20nm node), and also given that it is a new architecture I thought they might have had IPC improvements too. A 65% performance increase per Watt seems a little disappointing.

 

Darn it, I was wrong. No mobile Pascal yet. I hate being wrong. Oh well. Soon. It's gotta be soon now...

GTX 1080M at 980 Ti level of performance with ~100W? Is that what we're likely looking at?

How would that compare to the leaked PS4 Neo?

 

16nm isn't really true 16nm anyway, it's in the middle between 20nm and 16nm but yes, the performance per watt is quite disappointing. The vast majority of the performance increase can simply be attributed to the increase in clocks and the CUDA core count. There's only roughly 10-15% due to optimizations and besides that, they cherry picked games. The presentation mentioned nothing about increased async performance either which basically makes Pascal Maxwell 2 with tweaks, nothing special except for GP100. I already knew GP106 was essentially Maxwell but I had no idea nVidia pulled it for GP104 too. AMD could potentially decimate nVidia this time around until Volta if they play it right, not to mention that 14nm FF is much more mature thanks to Samsung's Exynos chips which should translate into lower costs and higher yields. Unfortunately, I believe they will both just milk the living hell out of their respective architectures. We really need a third major player in the arena.

 

I was going to post something like this anyway but you did a good job! So yeah I did those same calculations and came to the exact same conclusion and I just looked at the graph and estimated 4.3 for 1080 and 3.6 for Titan X, which Anandtech also posted later, and yeah efficiency came up at 66.67% over Titan X and less over 980.

 

If AMD doesn't release a Full Polaris 10 mobile GPU and decides to milk then it's gg, they won't have a chance at gaining mobile marketshare back. They say they have an extremely efficient architecture, top desktop chips are only going to have about 135W TDP, if after that they don't release the full chip their product placement team needs to be fired.

 

Jen sounded like trash and the awful presentation of slides with no proper relative reasoning and explanation, clarity, depth tbh..
Except for the fact that we all were waiting for the Pascal. All those claims over 980 SLI, too much. Just hoping that they don't gimp maxwell and deliver some true performance from the 1080 GTX, GP104. Though I really want AMD Polaris to kick ngreedia this time, the 14nm Samsung FinFet node is really better over this 16nm TSMC FinFet.

 

Two things about the livestream really annoyed me, the terrible music (oh god, I have Audio-technica ATH-AD700X and you can hear everything so clearly it sounds amazing, yet that music was so bad my ears were almost in pain) and that overexcited audience, especially that one girl back there, made me think they payed them to say(scream) certain things.

 

I have a strong feeling this guy would love to be able to disable reserve feature on his lot

 

I didn't say that they wouldn't release full Polaris but there are ways to hold back performance artificially. AMD doesn't nerf compute, they can easily get a leg up of 10-15% over 1080M, and since 14nm will be cheaper than 16nm (more mature process so cheaper cost and higher yields than 16nm), they can be competitive on pricing. AMD securing the Neo, the NX, and presumably whatever Microsoft comes up with has once again infused them with the cash to fight. Neither nVidia or AMD are likely to drop their cards on the table with their mobile offerings if only for playing on the ignorance of the masses who believe that a new name on a GPU means a new product. AMD's failure in the mobile market was probably much more because of their failure to address their GPUs that kept failing than it was the performance. OEMs had to get tired of replacing all of the failed parts. When nVidia took the performance AND reliability crowns, there wasn't much incentive for OEMs was there?

 

You're probably right, though I doubt they'll delay full 1080 for too long. The reason the 980 came so late and still could command it's absurd price was that it came with whole lineup of enthusiast class notebooks designed to support it, creating essentially new notebook category to catch the fancy of the rich enthusiasts. This time around most of these enthusiasts already have MXM laptops capable of running a full 1080 card, also the 1070m & 1080m would probably render the 980 obsolete. So if there is ever a time to demand a premium for a full 1080 MXM card, it is now, along with its desktop counterpart, when the hype for Pascal is at its peak. The standard MXM cards (1080m/1070m) would probably be priced reasonably to increase sales of VR Ready laptops.

Sent from my Micromax AQ4501 using Tapatalk

 

I read this and the first thing that came to mind at that level of power was the killer instinct announcer screaming "ULTRAAAAAAA!!!"

 

I am so worried that laptop version will not be dtrong enough to satsfy our enthusiast needs. I mean, even 20% faster than actual 980 in laptops is enough, but not slower than 980 in laptops.

 

I'll just leave it here guys...

http://deliddedtech.com/2016/05/07/hidden-information-found-in-source-code-of-nvidias-gtx-1080-page/


Sent from my iPhone using Tapatalk

 

Commenting something out doesn't mean much of anything except that it may need verified, isn't ready to be added (legally), or may not be added at all. It may also just be a placeholder and unrelated altogether.

Love how journalists and writers blow things way out of proportion to gain attention... The desperation lol.

 

No news on mobile pascal yet?

 

Not until the end of the month.

 

Well 980m is quite the burner, but I don't see separate topics about it. Just occasional mention here and there, inside unrelated topics. Same goes for older GPUs (680m and 780m) as well. It's like there's unwritten rule (or it's written?) to not post new topic if one's nGreedia fails (this could've been posted in the nGreedia conspiracy as well). As for Pascal - really high clocks and worse performance per watt than initially stated, sounds like someone is pushing it really hard, since they don't have a real new architecture. Time will tell, but at least in my mind, this is all nGreedia has on their hands for foreseeable future. I guess Polaris' rumors scared them quite a bit.

 

I am not sure what to believe anymore.

I don't want to trust myself with their own graphics and benchmarks, which are 100% sure to be influenced by them. I want real life benchmarks.

This one image best expresses my feelings.

 

I'm putting money aside for this starting now. I don't feel desperate to upgrade like I did with Kepler (my 780M ran way too hot) as the 980M is nice and cool running and still plenty powerful. But I shall upgrade nonetheless because I want to aim for that 1080p/60fps sweet spot with the latest games. And because I like new and shiny toys!

We can expect at least 50% improvement in performance over 980M assuming same 100W TDP, right?

 

I really want that best smooth 60fps 4K experience.

 

If you don't believe what NVIDIA is saying about Pascal, dig deeper into what they said and the reality of Maxwell to get an idea of how "blown" up it is:

Spoiler: NVIDIA Claims - 980M vs. 680M

I think people are underestimating Pascal. It may not be what they claim TODAY (first release), but it will be. That's what they're saying - what it's capable of. Though I am excited to see an actual review of the cards by someone other than NVIDIA!

 

Let's assume the 1080M will be the GTX 1070, clocked differently. Anyone want to hold onto the belief that it won't surpass the GTX 980?