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

Breaking news: Volta Expected announcement in May and launch by Summer 2017 to compete with Vega's launch Q1/Q2 2017. ( Source)



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

Considering there are more than a half-dozen articles from WCCFTech, MotleyFool, PCGames, and so on, I felt it time to create a discussion thread for anything Volta! Feel free to discuss, speculate, share and debate. Continue reading for information regarding Volta and what we know. If you have more links to supporting articles regarding Volta, please share them here!



The first Volta-based GPU is expected to be shown at GTC next year, according to the rumor. Advanced Micro Devices is scheduled to launch its Vega architecture in the first half of 2017. " Considering the fact that the architecture will probably be shipping with HBM2 memory, it makes sense to differentiate it from Pascal. In many ways Pascal architecture was similar to Maxwell (albeit on the 16nm FF node) which raises the question whether Volta can be to Pascal as Maxwell was to Kepler." (Usman Pirzada, WCCFTech - Source)



It is rumored, in response to AMD's Vega 1H 2017, NVIDIA has pushed Volta's announcement up to GTC 2017, with a possible launch by the end of Q2 2017, but most likely by Q3 2017. Even though NVIDIA has yet to launch their entire Pascal GPU lineup, rumors have already emerged regarding their all new GPU architecture. The Volta GPU was to be NVIDIA’s next greatest chip architecture to replace Maxwell, but was delayed with Pascal taking its place. My opinion? Pascal is a way to milk GDDR5/GDDR5X once more before switching entirely to HBM. With Pascal research & development now complete, NVIDIA has set their focus and efforts in developing their latest and most advance GPU architecture, Volta. (WCCFTech)




​

Supporting articles:
http://wccftech.com/nvidia-roadmap-2017-volta-gpu/
http://wccftech.com/nvidia-volta-gv110-gv104-gv102-gpu/
http://www.pcgamesn.com/nvidia/nvidia-volta-release-early
http://www.fool.com/investing/2016/08/12/nvidia-corporations-volta-rumored-to-be-a-monster.aspx
http://www.fool.com/investing/2016/07/19/nvidia-corporation-may-launch-first-volta-processo.aspx

The poll references the announcement, not launch.

 

LOL!

CU at page >1000 for the release as usual!

 

For Volta, Nvidia has to deliver at least something quite soon (2017-ish).

They got a contract with US government for 2 supercomputers based on Volta - Summit and Sierra:



http://www.nvidia.com/object/exascale-supercomputing.html
https://blogs.nvidia.com/blog/2015/04/15/link-different/
http://www.anandtech.com/show/8727/nvidia-ibm-supercomputers

These two are $425M project, so there will be significant pressures to get this done.

 

Will the mobile models come with the 'm' suffix?


Sent from my iPhone using Tapatalk

 

Pascal is barely out, and we already starting a 1000 page speculation crusade on Volta, like Prema said...

 

This was a bit early but i dont see any reason not to open it.

 

I will do a tiny contribution and then log out in shame lol.

GV100 support was added to a Geforce driver all the way back in November 2015. So thats perhaps a sign on what will be the first card out since they havebeen testing a working product for a good while now.

Also the reason why they just reused the Titan X name with GP100. Because they are prepping a Volta Titan sometime in 2017 that is probably more important to give a new Titan name or something

Thats my 2 cents

 

You bring shame upon your house! With such dishonor, I believe you will do the right thing. I leave my tanto with you. The blade is sharp and true. I pray it finds your heart quickly. May the honor be restored upon your house!

Sent from my SM-G900P using Tapatalk

 

4Q 2017 with timed leaks around 3Q 2017 after the Pascal refresh in 1H 2017. Those are my predictions. High end cards will launch first, mainstream will hit 1H 2018

 

If they drop all of the Pascal Ti cards before the end of this year, I think they'll announce Volta 1H 2017 to compete* with Vega.

*By "compete," I really mean, destroy. But compete is nicer.

 

Seriously? Summit and Sierra won't even be in testing until 2H 2017 and you can bet that they are Nvidia's number one priority.

 

I did say announce, not launch. With all their fancy graphs and whatnot showing 500% increase in performance.

I think they spoke of Pascal twice, both at GTC 2015 and 2016, and are just now ready to launch it.

 

Ah, yeah that will happen but not til they drop the Pascal refresh. They're definitely going to milk a refresh out of it. I mean the Titan X isn't even a full GP102.

 

Volta will launch CES 2019 for sure. if it wants to use 10nm

 

Even Intel's CEO claims they'll be launching the first Cannonlake processors 2H 2017. I think things are moving faster than we expect.

At this point, it seems they are on track (or not far off). Kaby Lake is expected to launch at the end of September.

 

Volta is supposed to stay on 16nm:

It would make sense: it was a long and painful wait for 16nm, everybody involved would like to amortize costs, get yields up, increase margins.

Let's hope HBM2 costs would go down by then too (AMD folks have been saying right now it's super expensive, hence only very high-margin GPUs can justify it).

 

Yes, I also heard that, with a planned refresh in 2018 on 10nm.

These rumors are screwing with me.

 

TSMC has already said 10nm would be short lived in favor of 7nm so it probably doesn't make sense for Nvidia to move to 10nm when there's still more that can be squeezed out of 16nm.

 

Then mid 2018 to Q3 2018 for volta not earlier because in laptop gpus at least nvidia has no competition

 

Oh my!

http://www.digitimes.com/news/a20160418PD206.html

Whatever will come in 2018 (Volta shrink or Einstein on 7nm) will be pretty nuts.

 

Because TSMC is always on schedule and never screws anything up...

 

They seem pretty confident this time :

http://www.eetimes.com/document.asp?doc_id=1330117

 

Well TSMC and ARM successfully pushed a test 10nm chip through in May and if 7nm uses most of the same equipment, it makes sense that 7nm wouldn't be far behind.

 

Hey guys give me your honest opinion :

Based on what I red on this forum, you all think that if invest now in a laptop with Pascal architecture, I won't have the possibility to upgrade components like CPU and GPU, because in a year (max 1.5 year), new Cannonlake CPU and Volta GPU won't be supported by the current laptop models configured with Skylake and Pascal.

Is that what you say ? because I waited until now to buy a new gaming laptop, and knowing that its death has already been decided before it was born ... it's sick

I don't know what to do

 

They're rumors, that's all. You can buy and resell, worst case scenario. You may lose some money but that's the price you pay for always needing the latest and greatest hardware as soon as it is available.

My advice would be to purchase something you can easily afford, so that you minimize your loss and liability later on. Don't stretch your budget or open new credit cards for a possible "single-generation" product.

 

NVIDIA may be going directly to 7-nanometer manufacturing technology for its post-Volta graphics processor family.

http://www.fool.com/investing/2016/08/14/nvidia-corporation-likely-to-skip-10-nanometer-in.aspx

This is speculation based on changes in Nvidia's job post on LinkedIn (they replaced 10nm with 7nm in more recent update):

Spoiler: Nvidia job post

 

That seems more like they're searching for a development & quality assurance team. That doesn't suggest they're skipping ahead to 7nm, but rather they're just entering development for that process. Companies develop products years in advance of launch. If they're still hiring, well... It'll be a while before we see 7nm.

 

I believe it was meant more something like just a supporting evidence pointing to 10nm being skipped (together with TSMC's claims about 10nm being focused on mobile).

Which indeed means the next node after 16nm would come a bit later (7nm compared to 10nm). So we can add maybe ~1 year to the estimates of when 2nd gen Volta / post-Volta architecture could come.

 

Honestly, I think since Pascal was never meant to exist and primarily does be cause HBM wasn't ready, Volta will be around for at least two or three generations. So, in essence, I can totally see the third generation of Volta being 7nm. However, I find it hard to believe they'll skip 10nm when Intel is working with 10nm as well.

2017: 16nm or 14nm (with the introduction of HBM2 and new socket type).
2018: 10nm (true Volta w/same socket & substantial gain in performance).
2019: 7nm.​

If they follow this, the performance of first-generation Volta will likely be around 50% greater than Pascal. But the primary point and most important factor, in my opinion, is future upgradeability (or compatibility) with upcoming product releases. I do not believe any Pascal-based product released during 2016 will be upgradeable to Volta.

For the record, none of this is confirmed. Readers, let it be known that this is speculative and based on rumor. We'll know more after Pascal's launch.

 

Can someone please explain the benefits of stacked VRAM? And is stacked system DRAM in our future as well for the same reasons?

 

Able to communicate much, much faster to handle requests. Bandwidth exceeds 1TB/s. GDDR5X runs at about 320GB/s on the 1080.

There are many benefits, though. You can find them here.

 

I believe that there will be indeed another iteration of Pascal for 2017. And by the summer of 2018, the Volta GPUs should have flooded the market.

 

I'm actually hoping for an early announcement. Not launch, but announcement. There's a difference.

 

Wasn't Volta announced alongside Pascal? In which case, maybe we should expect the next architecture after Volta to be announced alongside a second, more detailed Volta announcement?

 

They just updated the roadmap. Never officially announced it. Pascal is a placeholder.

Basically, "Hey, look, we're introducing Pascal now and delaying Volta because HBM isn't ready, yet."

HBM is finally ready!

 

So how powerful is the Volta? 10x more powerful than Pascal?

 

You can bet that their graphs will be skewed to show it that way but there's no way that will actually happen.

 

Not that much, haha. But it will depend greatly on what process they decide to use: 16nm, 10nm or 7nm.

You can expect at least 50% over Pascal, but I'd assume it will be more.

 

Pascal 1080 is 100% more powerful than 980, right? or is it 80%?

 

It's actually just ~60% - 70% on average over multiple games, depending on which ones you check and at which resolution (less difference at 1080p, more at 4K):

http://www.pcgamer.com/geforce-gtx-1070-performance-preview/
http://www.hardwareunboxed.com/update-nvidia-geforce-gtx-1080-gtx-1070-benchmarks/

Volta is supposed to launch at 16nm, the same as Pascal (at least the first batch), but with a different architecture.

So maybe we could get something like Kepler => Maxwell change?

Volta being 50% faster than Pascal then sounds about right (maybe more, or even dramatically more at higher resolutions thanks to HBM2; this is where "CEO math" type of charts may happen, with exaggerated multipliers for specific use cases ).

 

I'm expecting to see Volta around summer of 2018.

 

Agreed, I expect Pascal to be milked to 2018 at least!!

Sent from my LG-H850 using Tapatalk

 

Considering Pascal didn't exist until after Maxwell was announced, when they updated the road map because HBM wasn't ready, it's my belief that there's not much of a reason to stay with Pascal 16nm when you can launch Volta 16nm, and milk Volta for three generations . Out with the old, in with the new.

AMD is about to release Zen and their HBM2 GPU lineup Q1 2017: Vega. NVIDIA will likely follow and compete with Volta.

 

Ah, and I'm expecting to buy a laptop from you a that time. Need that time to save haha

 

We'll be here when you're ready!

 

Firstly...

In essence, the graphics processor now becomes an SoC instead of a PCB, where everything is laid out wide on a PCB, with the VRAM chips separate from the GPU, and connections between the two. This has several benefits:

1. Smaller physical footprint. Video cards have rivalled their motherboards in size in recent years. I remember having an 8400 GS that was slightly smaller than two credit cards lengthwise. Now, GPUs are behemoths with massive coolers. With HBM/HBM2, something the size of the card above, or even smaller (possibly with the area of a UDIMM or even SO-DIMM) can house processing power well in excess of the current Titan XP.
2. Waaaay faster memory bandwidth, on the order of a few TB/s.

That would mean RAM upgrades would be tied to CPU upgrades, as the system is now an SoC. Then we'd be locked down even more. As far as desktops are concerned - this won't happen for a long, long time, possibly a couple of decades at least.

 

Actually, current plans for stacked ram/3d ram/hbm2 on CPUs are to use it for the iGPU with HMA so that when not used by the igpu, it becomes a fourth cache level while maintaining memory on the board for when it is exhausted. So, buying a processor with 16-32GB HBM2 would still allow for however much ddr4 to also be used...

Sent from my SM-G900P using Tapatalk

 

Second point is false. First point is kinda true...

1. Reduce size is true but thermal will still be an issue. Expect similar size as R9 Fury X for HBM2 cards. You still need spaces to lay out your VRMs/cooler/voltage controllers/etc.

2. Hardly way faster.... 3072bit 12gb HBM2 on PCIE P100 12GB gets you around 540gb/s performance, within reach of good overclocked 10ghz gddr5x on a 384bit. 4096bit 16GB HBM2 on 16GB P100 would give you 720GB/s where a 14ghz GDDR5x on a 384bit bus would give you 672gb/s. Ofcourse HBM2 could be clocked faster in future giving it more edge.

The power consumption reduction/efficiency and bandwidth are likely to be key points. With bandwidth probably not being a keypoint for consumers right now.

The con of HBM2 right now is likely to be horrendous yield. NVIDIA have 2 different PCIE P100 configuration with one of them having only 3/4 stacks of HBM2 enabled.

Do you mean HSA?

 

Yep, sorry. Was typing fast and have been off my game lately. HSA. I got heterogeneous memory architecture in my mind somehow.

Sent from my SM-G900P using Tapatalk

 

The thing is, I think the earliest we will see 3d stacked memory on CPU is Zen+ and thats even a very optimistic guess.

One thing would be interesting... watching Nvidia and Intel fight out the HPC market.