• Hey Guest. Check out your NeoGAF Wrapped 2025 results here!

Nvidia Volta is 16nm, expected in May 2017

TSM said:
Nvidia isn't going to let AMD have even one ray of sunshine. Hopefully AMD forces Nvidia's hand. Competition is good.
Click to expand...
What's your context here? The question was why you need HBM2 in a 10 TFLOP (or less) card. (The answer is that you don't -- you probably don't even need it below 15 TF if you put the fastest GDDR5X on a 384 bit bus)
 
Helacious said:
Don't think 2017 will be the year for us, those will be for workstations and super computers. Besides that makes the launch of 1080ti awkward, not enough space. Want to be proven wrong tho.



Yup
Click to expand...

1080ti I think will launch early, this side of the new year, especially with Pascal Titan being announced so early.


Volta end of next year, I think it will.

GDDR5x feels like a stop gap to me and not a long term memory solution, I will hold off on a 1080ti, will wait on Volta.
 
AmyS said:
Eventually, one day...

8PvpZc8.jpg
Click to expand...

i'm waiting for the nvidia faraday, which can't do any math
 
KurtSloane said:
1080ti I think will launch early, this side of the new year, especially with Pascal Titan being announced so early.


Volta end of next year, I think it will.

GDDR5x feels like a stop gap to me and not a long term memory solution, I will hold off on a 1080ti, will wait on Volta.
Click to expand...

Yep. I want HBM2 with my next gpu!
 
secretanchitman said:
Yep. I want HBM2 with my next gpu!
Click to expand...
To both you and the other posters, why? I mean, if you only want to buy a GPU again when it has 20+TFLOPs (or whatever the limit is where conventional memory can't keep up) then sure, that makes sense.

But if you want to buy a GPU below that threshold, why be so focused on what type of memory technology it uses? It's a bit like saying "I'll only buy a GPU built on Samsung's 14nm process!" -- it's not a fact that has a direct impact on performance, so why not actually make your purchasing decisions directly based on the outcome you are interested in rather than how it is achieved engineering-wise?

To put it differently, a Fury, thanks to its HBM, has a 60% (!) higher memory bandwidth than a 1080. Which one would you rather own?
 
Yeah the improved L2 cache in Maxwell and beyond plus the color compression really worked wonders. The GTX 960 gets along great with a mere 128-bit bus.
 
Durante said:
What's your context here? The question was why you need HBM2 in a 10 TFLOP (or less) card. (The answer is that you don't -- you probably don't even need it below 15 TF if you put the fastest GDDR5X on a 384 bit bus)
Click to expand...

afaik despite not getting any performance improvement, supposedly cards with HBM2 could be waaaay smaller than the ones we have now, isnt it?
I would really appreciate a GTX1180 the size of my 750Ti.
 
Izcarielo said:
afaik despite not getting any performance improvement, supposedly cards with HBM2 could be waaaay smaller than the ones we have now, isnt it?
I would really appreciate a GTX1180 the size of my 750Ti.
Click to expand...

They won't.
HBM2 requires way more power to drive than GDDR5.
Which in turn means more heat dissipation. Which means larger heatsinks and better fans.
 
Knurek said:
They won't.
HBM2 requires way more power to drive than GDDR5.
Which in turn means more heat dissipation. Which means larger heatsinks and better fans.
Click to expand...

Fury and the Fury X were both smaller than the High end GPUs. While the Fury X had watercooling from the get go, the Fury didn't
 
If you say something like "I want a GPU which fits in my ITX case!" then that's A-OK with me. That's actually exactly "making your purchasing decisions directly based on the outcome you are interested in".
 
Can't see this happening to be honest since there was nearly 2 years between the 970/980 launch and the 1070/1080 launch.
 
The 1070/1080 don't offer a good enough jump from my 290x at their current prices to justify it, especially since my card is still performing well.

Providing my 290x does well on BF1, Gears of War 4, Dead Rising 4 etc without too much of a compromise, I can wait for the next generation again.
 
sammyCYBORG said:
The 1070/1080 don't offer a good enough jump from my 290x at their current prices to justify it, especially since my card is still performing well.

Providing my 290x does well on BF1, Gears of War 4, Dead Rising 4 etc without too much of a compromise, I can wait for the next generation again.
Click to expand...
The 1070/1080 are just way to expensive right now. No competition and all...
 
There is a possibility for me to grab a Gtx1060 if a friend of mine does buy my gtx970. He needs a second one so he offered 210€ for mine. Seeing i can get a Gtx1060 for 270€, the 2 extra GBs of vram plus the minor extra performance do seem nice for 60€.

If not for that, i would probably use my gtx 970 until volta.
 
Sweet. I might just get a RX 480/ GTX 1060 to tide me over the next ten months on ultra 1080/ high 1440 in games. I'll make a 4k build next year when shit gets serious with HBM2 GPU's and 120hz 4k monitors. Extra time to save for it now.
 
Renekton said:
HPC - mid 2017
Desktop - early 2018?
Click to expand...

Why? There was like two months between GP100 and GP104 introduction. If Volta will be out mid 2017 for HPC then you can be pretty certain that it will be out for gaming markets till the end of 2017 as well.
 
Nvidia are going GPU crazy.

They'll probably launch the full stack of Volta GPUs in 2017 like they will have done with Pascaly.

The 1080 Ti is coming to fleece those keeners by the end of this year and that only leaves the 1050, which will still probably release before AMD gets their 460/470s out as AMD is terrible.
 
RE4PRR said:
People's obsession with HBM2 is fascinating to read.
Click to expand...

AMD marketing spent a lot of time distributing HBM kool-aid. And it makes such a nice big numbers in bandwidth column..

Hmm but this is interesting question - should I go 980ti ->1080ti or wait for Volta GV104 which should match big Pascal.
 
So what was the point in releasing GTX 10 series this year if presumably new Volta lineup coming next year? We could've just easily waited for Volta one more year with ours 970/980/980Ti-s.

So it's like what... one year gap between new GPU lineups now? I don't like this at all.
 
michaelius said:
AMD marketing spent a lot of time distributing HBM kool-aid. And it makes such a nice big numbers in bandwidth column..

Hmm but this is interesting question - should I go 980ti ->1080ti or wait for Volta GV104 which should match big Pascal.
Click to expand...
AMD's marketing is weaksauce (or nearly non-existant) lol, we can't give them too much credit when techsites and enthusiasts pine for the latest buzzword.

Maybe you can go by numbers instead of naming. Upgrade based on a fixed % benchmark improvement, e.g. 50%.
 
HBM2 wouldn't even be that much of an improvement in many cases. The 12GB Tesla P100 has 538GB/s memory speed. That's not much more than GP102 Titan X at 480GB/s with 12GB GDDR5X. Probably with a hefty premium too.
 
Agent_4Seven said:
So what was the point in releasing GTX 10 series this year if presumably new Volta lineup coming next year? We could've just easily waited for Volta one more year with ours 970/980/980Ti-s.

So it's like what... one year gap between new GPU lineups now? I don't like this at all.
Click to expand...

The point is not for the consumer. Its for nvidia to extract as much money out of you as possible.
 
Agent_4Seven said:
So what was the point in releasing GTX 10 series this year if presumably new Volta lineup coming next year? We could've just easily waited for Volta one more year with ours 970/980/980Ti-s.
Click to expand...
Speak for yourself, I am pretty happy I got to upgrade to a 1080 for Raw Data rather than playing it on a 970 :P

Agent_4Seven said:
So it's like what... one year gap between new GPU lineups now? I don't like this at all.
Click to expand...
Were you in the GPU market in the 90s/most of the 00s?
 
nOoblet16 said:
I don't get what this is suppose to mean.
Click to expand...

Einstein - Nvidia's GPU architecture that was meant to succeed Maxwell. It's been assumed that Pascal and Volta were inserted into the roadmap, and it was rumored Nvidia said Einstein is still coming.

Green tint to the Einstein picture meant to signify Nvidia.

I was was just being silly, put 4 pictures together, connected, representing 4-way SLI w/ next-next gen Nvidia cards :)

This article is 3-4 years old,

http://www.3dcenter.org/news/nvidias-maxwell-nachfolger-hoert-auf-den-namen-einstein

(Google translation from German).

NVIDIA Maxwell successor goes by the name "Einstein"
Even if at present nothing concrete for the current Kepler architecture subsequent Maxwell architecture (2014, 20 nm) is a long known, nVidia still busy of course, already with the long-term future and is researching the planned for chips. For a long-term project now the code name became known: The Einstein architecture should follow the Maxwell architecture. This code name nVidia remains rest among the younger tradition: The architecture Fermi, Kepler, Maxwell and Einstein were now devoted to all known scientists
Click to expand...

By means of the Einstein architecture should nVidia improve the overall usability of graphics chips for GPGPU tasks likely to continue, but all been said so far this has probably only Study character, should not long be gearbeit on a real circuit diagram of Einstein graphics chips. To produce larger and hence have Erscheingstermin Einstein architecture there are however contradictory statements: While some Einstein considered 14nm- or 16nm chip in 2016, others speak of a 10nm chip in 2018. If, however, the Einstein architecture really follows the Maxwell architecture right, is probably first resolution to be correct - with terrible 14nm manufacturing since the contract manufacturer TSMC and Globalfoundries will in future follow in this matter the Intel manufacturing steps.
Click to expand...

Guys (IIRC) Einstein is a GPU architecture / chip that would power a super computer project Nvidia was working on, called Echelon

http://insidehpc.com/2010/11/nvidia-reveals-details-of-echelon-gpu-designs-for-exascale/


2aec3Kc.jpg


http://www.nvidia.com/content/pdf/sc_2010/theater/dally_sc10.pdf

Nvidia was describing a 20 TFLOP single GPU in 2010.

vSTB3SM.jpg


BN22Hin.jpg


That's the Einstein GPU on paper, back then.

Pascal Titan X for gamers is 11 TFLOPs in 2016.
 
Agent_4Seven said:
So what was the point in releasing GTX 10 series this year if presumably new Volta lineup coming next year? We could've just easily waited for Volta one more year with ours 970/980/980Ti-s.

So it's like what... one year gap between new GPU lineups now? I don't like this at all.
Click to expand...

They want to make more money out of you, so they will push product out at a higher frequency to achieve those aims. Maxwell to Pascal was a longer gap than they would have liked no doubt (28nm to 20nm was skipped) and yearly product stack releases is much more pleasing to the big bosses at NV HQ. This would go for AMD too but they struggle to release a single card without running into problems because of poor management and being broke.

Seeing as Volta is 16nm too, I don't see it being hugely faster than Pascal unless there is miraculous overhaul of the architecture. That will come in 2018 with the Steve arch.
 
Durante said:
To both you and the other posters, why? I mean, if you only want to buy a GPU again when it has 20+TFLOPs (or whatever the limit is where conventional memory can't keep up) then sure, that makes sense.

But if you want to buy a GPU below that threshold, why be so focused on what type of memory technology it uses? It's a bit like saying "I'll only buy a GPU built on Samsung's 14nm process!" -- it's not a fact that has a direct impact on performance, so why not actually make your purchasing decisions directly based on the outcome you are interested in rather than how it is achieved engineering-wise?

To put it differently, a Fury, thanks to its HBM, has a 60% (!) higher memory bandwidth than a 1080. Which one would you rather own?
Click to expand...

The one that has a chip from the 1080 maker alongside the higher memory bandwidth.
 
You must log in or register to reply here.

Similar threads

Nvidia RTX 6000 Series Expected to Launch in Early 2027
Hardware  2 3
107
12K
CuNi replied
After 20 years, Nvidia retires the NVIDIA Control Panel
2
77
273
Celcius replied
Super Metroid is expected to get a remake sometime in 2026
Rumor 
47
851
schaft0620 replied
NVIDIA DLSS 5 Settings Appear in the Latest Drivers
Platform 
11
99
Wolzard replied
  • Locked
Nvidia is building its second biggest Campus, in Israel
2
83
1K
DonkeyPunchJr replied
Top Bottom