How long is nvidia gonna keep milking pascal? the gtx1080 came out a year and a half ago

how long is nvidia gonna keep milking pascal? the gtx1080 came out a year and a half ago.

Other urls found in this thread:

youtube.com/watch?v=3aAEKRDhrj8
twitter.com/NSFWRedditVideo

As long as AMD doesn't get their shit together or games start getting too laggy on it

Volta is already out so not long

Since AMD can't manage to beat the 1080, Nvidia will keep selling them.
Their yields are stupid high right now.

At least another quarter, because Nvidia is the Intel of GPUs.

Until Navi.
They have no reason to reallt do anything until AMD makes them.

>how long is nvidia gonna keep milking pascal?
Not for long, they need to do something to not let the stock pop.

>need to do something
Pascal refresh

*sounds of $NVDA bubble popping*
Yeah right.

>ddr6
>more unlocked cuda cores
>unlocked overclocking
>20% gains
>sheeple are happu.

>ddr6
A memory?
>more unlocked cuda cores
Volta is also more SPs and nothing else (for gaymen that is).
>unlocked overclocking
Boost 3.0 says hello.
>20% gains
>sheeple are happu.
They are going to be angry though.

>Volta
Huang says NO VOLTA FOR GOYIM go google.

>pascal refresh
Gddr5x on new 2060-2070
Gddr6 on 2080
>boost 3.0
Pascal is hardware voltage locked.
Unlocking some 0.1-0.2v would enable a significant boost in performance
And you just dont know how many CUDA cores are still locked down in those 1080ti/70ti chips.
They may easily be sandbagging just like Navi 56 lasering off some cores to sell a full uncut product at a higher price every manufactorer has been doing it in the pasy.

youtube.com/watch?v=3aAEKRDhrj8

Nvidia already announced Volta, would not buy any graphics card unless you absolutely need 1 right now

Waiting for Volta GeForce will pay off

>waiting for bigger dies will pay off
?
It's bigger dies for bigger prices.

It’s not likely that Volta will ever come to the consumer market. The largest key advantages it holds over Pascal are either not applicable to typical use cases from an architectural standpoint, or hinge on the fact that it’s the largest processing die ever made both in density and size, so a cut-down variant would just be a slightly fancier Pascal.
Going forward it’s basically a dash to whoever can develop a valid multi-chip architecture first, because the prospect of going big has already been exhausted and is no longer economically feasible.
If Nvidia does actually end up releasing Volta in some non-enterprise capacity, you can bet that it’s just going to be a stall tactic while they cook something better up. They can afford this since the competition is at least a gen behind.

This.
Gamers dont have anything to gain from Volta.

Piling up on ALUs works to an extent.

Nvidia must have been scrambling to develop their MCM ever since AMD announced Navi.
Anything else is a losing strategy.
GPU workloads are super parallelized and scalable, an MCM GPU will literally scale indefinutely all the way up to ridiculous size and power and can be pretty cheap to manufacture.

If they arent ready to release an MCM gpu by the time Navi arrives they are boned.
Even if their Volta is better than Navi AMD can just put moar coars, as many as it takes to win and it's still gonna be cheap.
It's gonna be a Tripper vs i9 style massacre.

And they must be a few years behind in development too.

Well duh, MCM GPU is NUMA, and no one besides IBM can play hardball with AMD when it comes to NUMA.
But yeah, 815mm^2 die seems desperate.
Anyway, AMD needs to make Vega work first.

Considerin Navi roadmap says "leapfrogging dev teams" it would make sense to think Vega was actually developed by their B team while the real A guys were workng on Navi all along

No, Vega is their A team given the changes to geometry pipeline.

Consideting the size, HBM2, power consumption and technologies utilized it is suppose to shit around 1080ti by a great margin but it doesnt.

Its technologically superoor and just bigger, if it still cant win its either broken or stupid.

>Consideting the size
Yeah.
>HBM2
It's a memory. Memory is a memory.
>power consumption
All high-end GPUs are blast furnaces.
>technologies utilized
They kinda disabled right now.

>disabled
Hopefully.
Im started to think they just shat the bed on silicon level can fix it and chose to release as is because fuck it deadline has passed anyway

>Im started to think they just shat the bed on silicon level can fix it and chose to release as is because fuck it deadline has passed anyway
It works in CAD.
Just look at wireframe shit performance of Vega.

>it
Its not the primitive shader and tiled rasterizer.

It totally is their new geometry pipe.
Wireframes are pure poly throughput, a workload that absolutely chokes anything Radeon before to silly death.

vega was a ballpark 500mm part, its not that they laser off shit for the sake of a 56, even with excellent yields there are still flaws, and 500nm~ is the area where even great yields yield few flawless.

as for what nvidia has left... not much

this is the titan xp
3840:240:96
its the same core configuration as the p6000

this is the 1080ti
3584:224:88

there is very little wiggle room for another sku in there

and with the die size being 471, its not realistic they will just make bigger dies, to make new skus they would be bringing bigger dies down to lower segments. if they do this, they may bring out a titan that is an unfucked double precision, and the refresh will likely be double half precision, as games are going to start using that for their lighting methods as even single precision is overkill and gpu tech isn't getting to single card path tracing fast enough.

GV102 will have 5376 CUDA cores, just like GV100

Stupid people that don't learn from GP100(3840 CUDA cores, 3584 enabled) and GP102(3840 CUDA cores) just won't learn and will be stubborn as hell to accept facts

nah, if they want people to upgrade they will have to either bring out better performance per mm, or bring higher mm, nvidia needs to sell gpus, and with the 1080 being midrange to low midrange at best based on die size, they can take a hit for a generation just to keep sales up

>... not long

another half a year at least

AMD is not gonna rest for long and you know that.

Slightly off-topic. How big are the stockpiles of Tegra X1? Could Switch sales hypothetically force Nintendo to build them with more processing power down the line?

Who the fuck cares about shitty tablets?

>Unlocking some 0.1-0.2v would enable a significant boost in performance

and start fires?

>how long is nvidia gonna keep milking pascal?
Until TSMC is ready with 10nm.

No reason fab anything new on 16nm, AMD dropped the ball and spending 3 gens on 28nm was costly enough.

They have no plans to make a consumer (aka "gamer") discrete volta GPU, though

Yes.
But CLN10 was ready ages ago.
Besides it's nVidia, they need a miracle to not fuck up the node jump.

AMD has dropped the ball going back 2 real generations (Vega, Polaris) for gamers - who are the real bellwether of the market, as juvenile as they may be.

Vega is squarely aimed at all things 3D, they simply need to get it working.
Besides, what's wrong with Polaris?

source or gtfo

>Have 500 spare euros
>Want to buy a new GPU
>Upgrading from GTX770
>Don't know when Volta is out

Fuck guys, what would you do?

It didn't provide anything that wasn't already served by AMD themselves.
No performance improvements for the price, even at SRP and a massive amount of the older product.

It's great for AMD, since they can sell the same performance point at the same price, but cut their production costs - for the short term.

Sadly nothing published, but a group of engineers from NVIDIA gave a talk at my university mostly to to try to sell their $150k volta server to researchers. Somebody asked them specifically if there would be GeForce card with Volta, and they said they didn't think so.

what's so bad about waiting for what you would feel is a good moment to upgrade?

Volta is 1H 2018, no?

I think that delaying the update was a mistake, now I feel the urge to change my GPU asap.
I don't know, but upgrading right now seems a really bad idea.

It offers ~Hawaii performance for $200, which is exactly what it was intended to do.
You want super ebin flagship novideo killer? I doubt current AMD will greenlight it.

Why isn't right now a good moment to upgrade? I have a laptop with Broadwell and Intel 5300 integrated graphics. I'm thinking of upgrading to a desktop with 7700 and GTX 1080. Is that a bad idea or could it be ok?

honestly right now it seems best to just wait a few months

Nvidia could sell the gtx 1070ti for $200 and still make bank, probably even lower now. Its such a small chip. They're making so much money its disgusting

No, but ~$300 would be a fair (read ATi) price.