When is Volta going to be released?

[TNA]

It feels like Volta will be with us before I get my custom cooled 1080Ti. Man I hate waiting

What are you looking forward to most about the 1080Ti. Coming from a 1080 it will be around 30% or there abouts boost to fps. That is the only change as far as I know. Don't take it the wrong way by the way, genuinely wondering

With Vega (if it is good), even though I may not necessarily need it on launch, all the new tech it brings may get the best of me and lead to a purchase.

 

[Gregster]

What are you looking forward to most about the 1080Ti. Coming from a 1080 it will be around 30% or there abouts boost to fps. That is the only change as far as I know. Don't take it the wrong way by the way, genuinely wondering

With Vega (if it is good), even though I may not necessarily need it on launch, all the new tech it brings may get the best of me and lead to a purchase.

Nothing really lol. Funds there for PC upgrades, so enough performance in the Ti for me to warrant it.

 

[TNA]

Nothing really lol. Funds there for PC upgrades, so enough performance in the Ti for me to warrant it.

Fair enough

 

[Seanspeed]

Although I love my 1080 Ti I still say maxwel was the dogs ******* as it was on the same 28nm process as Kepler they had to do some magic.
780 Ti - 980 Ti was a 65%+ bump in performance with a roughly 200mhz advantage to the 980 Ti. 1000mhz vs 1202 On the980. LESS cuda cores 2880 vs 2816 and same bandwidth and slightly more transistors.
However 980 Ti - 1080 Ti is another 65%+ though the 1080 has a 600mhz advantage 1202 vs 1800, 700+ cuda cores much more memory bandwidth and 50% more transistors. So 980 Ti in terms of specs was worse off vs the 780 Ti where as the 1080 Ti is better off than the 980 Ti though performance difference is the same

Kepler to Maxwell was all about architectural efficiency improvements.

Maxwell to Pascal was mostly about process efficiency improvements.

Maxwell was indeed an incredibly impressive jump and so I dont expect exactly that same improvement again with Volta's new architecture, but it will also be combined with a smaller process efficiency improvement and an ability to make bigger chips. 50% improvement again should be entirely do-able.

 

[shamus21]

"Seanspeed, post: 30630752, member: 151221"]Kepler to Maxwell was all about architectural efficiency improvements.

Maxwell to Pascal was mostly about process efficiency improvements.

Maxwell was indeed an incredibly impressive jump and so I dont expect exactly that same improvement again with Volta's new architecture, but it will also be combined with a smaller process efficiency improvement and an ability to make bigger chips. 50% improvement again should be entirely do-able.[/QUOTE]

I would be surprised if you see that sort gain, Maybe in 2019.

 

[Rroff]

Maxwell to Pascal was mostly about process efficiency improvements.

Quite a lot of architectural changes in Pascal - caches tweaked to better service the appropriate operators, improvements to how shared memory is handled in threading, improvements to command scheduling to better utilise the hardware (including stuff like pre-emption), tweaks to layout to increase max potential clockspeed and so on.

 

[Tommy_Here]

Kepler to Maxwell was all about architectural efficiency improvements.

Maxwell to Pascal was mostly about process efficiency improvements.

Maxwell was indeed an incredibly impressive jump and so I dont expect exactly that same improvement again with Volta's new architecture, but it will also be combined with a smaller process efficiency improvement and an ability to make bigger chips. 50% improvement again should be entirely do-able.

As the 1080 Ti isn't a full chip I expect a fully fledged Volta big gpu to be 70% faster than the 1080 Ti which would be a 5k card

 

[Seanspeed]

Quite a lot of architectural changes in Pascal - caches tweaked to better service the appropriate operators, improvements to how shared memory is handled in threading, improvements to command scheduling to better utilise the hardware (including stuff like pre-emption), tweaks to layout to increase max potential clockspeed and so on.

There were a number of improvements, but I'd classify most of them as fairly minor, relatively. The process improvement, a jump from 28nm to 16nmFF(skipping 20nm) really made Pascal what it is.

I'm not sure what layout differences there are, though. From what I've seen, it looks very much exactly like Maxwell.

Either way, my point is that Volta will likely be a significant jump in architecture over Pascal, which was only a minor blip on Maxwell. I'm sure a few things originally planned for Volta were brought in for Pascal, but the time difference should still give them plenty of time to do something very noteworthy.

 

[TheRealDeal]

Quite a lot of architectural changes in Pascal - caches tweaked to better service the appropriate operators, improvements to how shared memory is handled in threading, improvements to command scheduling to better utilise the hardware (including stuff like pre-emption), tweaks to layout to increase max potential clockspeed and so on.

Wasn't there a test done with the gtx980ti and the gtx1080 both around 1800mhz (gtx980ti was LN2 cooled off course). The gtx980ti was faster proving that architecture wise Pascal wasn't really much of a gain at least not like Kepler to Pascal.

 

[Rroff]

Architecture changes aren't just about straight up performance - some of the scheduler type changes for instance would only show up with advanced dx12 use.

 

[mahius]

As the 1080 Ti isn't a full chip I expect a fully fledged Volta big gpu to be 70% faster than the 1080 Ti which would be a 5k card

Lol, we wish. And knowing Nvidia, they'll price this 1180ti at £1.2k...

But I guess even a 1180 will manage to conquer most games in 4k60.

 

[Seanspeed]

I would be surprised if you see that sort gain, Maybe in 2019.

Why?

I dont get it.

 

[shamus21]

Architecture changes aren't just about straight up performance - some of the scheduler type changes for instance would only show up with advanced dx12 use.

And until DX12 becomes implemented across the board by all software developers you will not see much of the architecture performance improvements in every day use regardless of which make of GPU you use.

 

[shamus21]

Why?

I dont get it.

They will need to shrink the process more.

 

[TNA]

Why?

I dont get it.

I don't get it either.

 

[Seanspeed]

They will need to shrink the process more.

But it's also an architectural jump. And there's room to make bigger GPU chips since neither GP104 or 102 are that big for their respective slots in the lineup.

Add in a small process improvement and it seems pretty obvious big gains are inevitable.

 

[shamus21]

But it's also an architectural jump. And there's room to make bigger GPU chips since neither GP104 or 102 are that big for their respective slots in the lineup.

Add in a small process improvement and it seems pretty obvious big gains are inevitable.

Lets hope so seanspeed otherwise it will be a long wait for a bigger move down from 12nm

 

[Scooby-DoobyDoo]

Quarter 3 2017

https://www.fool.com/investing/2017/04/17/report-nvidia-prepping-new-volta-architecture-gami.aspx

 

[D.P.]

Quarter 3 2017

https://www.fool.com/investing/2017/04/17/report-nvidia-prepping-new-volta-architecture-gami.aspx

Volta chips for DriverPX are due to ship in volume then on TSMC 12nm FF++. Rumour might be confused with that, or there could be a 750ti type card with a similar chip as used in DriverPX. I doubt we will see big volta chips this year but could be possible.

 

[jjgreenwood]

Volta chips for DriverPX are due to ship in volume then on TSMC 12nm FF++. Rumour might be confused with that, or there could be a 750ti type card with a similar chip as used in DriverPX. I doubt we will see big volta chips this year but could be possible.

I heard only mobile companies had signed up for that so Samsung and Apple pretty much. Nvidia has yet to sign up for a smaller node hinting that Volta will be on current tech.