Volta: NVIDIA's Next Generation GPU Architecture (2017-2018)

Better make that thread now then!

 

@J.Dre @Ethrem @D2 Ultima

Couple of questions:

1. How much hotter than Pascal is Volta going to be?
2. How much more performance will Volta give over Pascal?

I know it's early days yet, but even rough/speculative/vague answers are ok.

Sent from my SM-G935F using Tapatalk

 

Use this chart to help you conceptualize the Volta Phase Change - from Solid to Plasma



Nobody knows... there's not even any way to guess... we don't even know if it's gonna be 14nm, or 10nm, or 7nm.

It's too early to do anything except joke about it

 

That is asking like, how much hair am I going to lose within the next 2 years, or will I win the mega millions. WHO KNOWS!!!!

 

Fair enough. However, in an earlier post, @Ethrem said Volta is hotter than Pascal - just wondering if he had any more info. Also, way before the Pascal launch, we had people guessing reasonably accurately about performance gains. Just wondering if any knowledgeable people could do so here.

Sent from my SM-G935F using Tapatalk

 

We guessed based off of existing card samples, and what we knew about the architecture. Right now, there is little to know about Volta. Ethrem says it's hotter because he likely has information about early tested samples, but it's too far away to judge. I will say however if volta is hotter than pascal, air coolers are going to start dying, especially for the top end chips (GV100, GV102). Handling not-even-full-die GP102 on air is pretty difficult in the first place, especially in a warm (far less hot) room. GV104 would need vapor chamber contacts and very effective cooling systems. It'll also mean GV104 will not SLI in notebooks, and single GPU won't work in most machines either.

But we'll see what happens later down the line. However since nVidia can shove flaming dog feces in a black shroud and charge $700 for it and the world compliments them on their "aggressive pricing" and people scramble to buy them quick quick quick till they're sold out for months straight, they have ZERO need or desire to actually fix Volta. It's not like AMD is going to realistically surpass them like nobody has ever seen, and they refuse to sell reference watercooled cards, so... that's that.

 

Thanks for the useful opinion!

Sent from my SM-G935F using Tapatalk

 

My guess on my hair is that I will lose more hair and feel colder in the winter and cooler in the summer.

My Volta guess is that it will be faster than pascal.

 

Safe guess, but is it going to be 40-50%+ or 30% ? That is the kind of speculation I'm looking for.

Sent from my SM-G935F using Tapatalk

 

I have a cousin that knows a guy who lives next to an AMD engeneer who thinks he heard something like 27,82% difference.

 

Wow! That is sure to be accurate!

/s

P.S.- hahahahaha

Sent from my SM-G935F using Tapatalk

 

1. Depends on the process but it should be manageable.

2. Again, depends on the process. If it is 16nm, then maybe 30%~50% is fair. We haven't seen the full potential of 16nm, yet. If they skip to 10nm, then I would expect another 50%-70% on top of that, over Pascal.

I think 2017 will be 16nm Volta with HBM2.

 

Thanks a lot. Useful info. Probably 16nm, I hope they go down to 10nm but I suspect they'll next go down a process node at the same time as 6-core Coffee Lake.

Sent from my SM-G935F using Tapatalk

 

http://www.tweaktown.com/news/54147/nvidia-reveal-next-gen-volta-architecture-gtc-2017/index.html

http://www.tweaktown.com/news/54148/volta-based-geforce-cards-coming-2018-48gb-gddr6/index.html

http://www.tweaktown.com/news/54144/nvidia-refresh-pascal-gpu-architecture-2017/index.html

Looks interesting to me!

 

We saw very similar reports with Pascal and hardly any became true. That said, if they do plan to launch HBM3 and GDDR6 in 2018, then I'd be happy to wait for it!

 

If they use the same nm for the 1st gen Volta (which they most likely will), I would estimate a 20 - 30% boost over Pascal. If they decrease the nm, then I would estimate a 50% - 70%

 

The most important link IMHO was the one saying a Pascal refresh is coming. Maybe Volta will be to refreshed Pascal what the R9 Fury/Fury X/Fury Nano were to the R9 390X and the other non-HBM cards.

Sent from my SM-G935F using Tapatalk

 

Sorry, I missed that link, what Pascal Refresh??

The rumored 1080ti slotted between the current 1080 and Titan Pascal that's supposed to come out in Jan/Feb 2017?

Or, the low end Pascal release filling in below the current 1060?

 

Read the third link in that post of mine.

Sent from my SM-G935F using Tapatalk

 

ha-ha after following the pascal thread for SOOOO long just to be let down with the non-mxm modules I can't even...

 

Never doubt the milk....

http://wccftech.com/nvidia-pascal-volta-gpu-leaked-2017-2018/

 

That article supports my prediction that Volta will be announced at GTC in May.

It's basically a summary of rumors.

 

I was refuting the fact that the 1080ti would be ''the final consumer (gaming) Pascal-based GPU EVER''

Kind of like taking a page outta Intel's playbook and just release modest refreshes. Instead of pushing full steam ahead when they don't have to. Unless amd lights a fire under there ass then they don't have to.

 

Refreshes aren't new GPU's, so I don't count those. But even if they did ruin the 11-series name with straight refreshes, they'd have what, 3-5 months of sales at the most? Who in their right mind would buy it? Volta will be announced in May and launched by the end of Summer.

The whole idea of a Pascal refresh doesn't make sense. I think it's just rumor and it's wrong.

 

IDK people bought TITAN X pascals and the 1080ti is supposed to launch in January that like 5 months of not waiting....That's what I'm saying about the MILK!!! Never doubt the milk with NVidia!!

EDIT 1080TI**

 

Don't get me wrong AMD milked the crap outta Hawaii. But in a different way. Pushing out new hardware to fast like NVIDIA is doing is obsoleting cards even faster as gimpworx will require the latest TITAN to run (an exaggeration). Pushing for people to buy there overpriced latest and greatest faster and faster and alienating people who spent large sums of cash o have the fastest only to be obsolete in 3-5month(another exaggeration) just like sony and micro$haft pushing new consoles out faster forcing upgrades to use the latest graphical advancements. There is method to these large corporations madness and that madness is greed.

 

If they launch the 1080Ti with other Ti series cards next month, then I can see a refresh happening around January or February through late Summer '17. In business, we don't usually launch products whose shelf-life is less than six months due to the lower ROI potential.

 

I will probably still get the titan xp personally even if I know a 1080 ti out in january back in august. The 1080 ti is probably gonna be cut down titan.

 

I was expecting the Titan to have more cores. Bit surprised.

 

There's a new rumor
http://www.tweaktown.com/news/54285/nvidia-geforce-gtx-2080-ti-lead-20-series/index.html

 

So what happens to Volta then? 2018?

Sent from my SM-G935F using Tapatalk

 

Volta was always expected to arrive in 2018. I always assumed it would be announced in May 2017.

I'll be surprised if they refresh Pascal with 10%~ improvement. Why would anyone buy it? They're better off leaving the 10-series on the market. And we still have yet to see the Ti series cards, so I would expect those to arrive sometime in Q4 2016 or Q1 2017.

 

its not that bad actually, ud get a 10-30 % perf bump in parallel with significantly lower prices compared to the 10 series.

Sent from my Huawei Mate 8 NXT-AL10

 

Yeah, 1080 Ti is supposed to be out in January. Saving my pennies to upgrade to that from my 980 Ti. Otherwise it's just another cash grab by nVidia to milk the cow.

 

I think a Pascal refresh is only worth doing if AMD becomes competitive again. As long as Nvidia is still on top and the 10-series cards are selling, why waste the money?

 

I am looking to upgrade my desktop from a GTX 690. From reading these past 3-4 pages I think I can wait a bit. The desktop is made out of eBay parts that I am sniping.

 

http://wccftech.com/nvidia-volta-gv110-gv104-gv102-gpu/

http://marketrealist.com/2016/10/nvidia-may-accelerate-gpu-roadmap-compete-amd/

Worth reading.

 

Interesting. I wonder if it will come with mxm3b or which ever is the standard we are using now.

Sent from my SM-G930W8 using Tapatalk

 

I'm sticking with my original prediction of a 16nm (architecture) refresh but with Volta cores and stack memory. It's not full-pledged Volta, but still a nice step up over Pascal. I don't believe we will see 10nm or 7nm until late 2018, early 2019. However, if production starts mid-2017, we may see them launch by Fall 2017. All depends on GTC in May!

 

Useful information. Thanks.

Sent from my SM-G935F using Tapatalk

 

Yup.

Also sheesh these people have no respect for my upgrade cycles.



Sent from my SM-G930W8 using Tapatalk

 

Only thing companies care about is $$$$$

Sent from my LG-H850 using Tapatalk

 

The first iteration of Volta may have GDDR6. HBM2 may not come to the GeForce series until 2018 (or possibly on the Volta Titan). This memory type is much cheaper to manufacture, which is why I'm assuming they want to keep it with 16nm until 10nm or 7nm in 2018.

 

"THE AMD - NVIDIA TECHNOLOGY WAR!"

http://seekingalpha.com/article/4022484-amd-coming-technology-war-nvidia

Well, at least we know Volta is coming soon. Unfortunately, it won't be "true" Volta. It'll be on the Pascal 16nm FinFET process. But should be a nice improvement, nonetheless.

 

Interesting. Do you think its possible that Volta will launch next year? And if it does, can we expect a 50%+ performance improvement over Pascal?

Sent from my SM-G935F using Tapatalk

 

It'll most likely be on the 16nm process. So, yeah, I think it'll launch sooner than we all expect. Instead of refreshing Pascal, they'll just launch a Pascal - Volta Hybrid with marginal gains in performance (up to 50%). But at least we'll see stacked memory, and maybe even HBM.

It's worth waiting for. We're what... Seven months away? Most systems are no longer upgradeable and become obsolete when new technology arises. Volta will be around for much longer than Pascal.

 

Question for you - what will 10nm Volta's gains in performance be over 16nm Volta (an estimation is fine) ? If it is 35%+, I might wait for 10nm Volta at the end of 2018 paired with 6 core Coffee Lake processors.

Sent from my SM-G935F using Tapatalk

 

They may skip 10nm and go straight to 7nm. Cannonlake is 10nm and so is Vega. If Volta is going to be 16nm for 2017, then NVIDIA may jump ahead and move straight to 7nm for 2018/2019. I'd expect nearly double the performance of 16nm with 7nm. This seems to be the most likely scenario.

You should not wait that long unless you are very happy with your current setup. Beyond GTC in May 2017, it's entirely speculative.

 

I don't mind waiting for 7nm Volta, I'll just upgrade my desktop for now and keep my old Sager with an 860M as long as it lasts, and reap the benefits of 7nm/10nm in end-2018. If it really ends up being 7nm in 2018, I think my decision would be vindicated.

Sent from my SM-G935F using Tapatalk

 

By end-2018, we may even get things like 17" OLED displays, more mature Optane and other cool things.

Sent from my SM-G935F using Tapatalk