NVIDIA Volta with GDDR6 in early 2018?

[Rroff]

Of course they're going to have 7nm products, just after AMD. Due to their release cycle, and due to TSMC being later.

TSMC's first 7nm process is only for low-power chips, so their good yields and being ~3 months ahead of GloFo is only for mobile processors. Also it uses a very different track height, so is a different density and unlikely Nvidia will even use that process for a pipe-cleaner like the 750 Ti.

TSMC's high-performance 7nm will be available possibly as much as 6 months after GloFo's, since GloFo is going high-performance first.

Nope the site that is putting out that info is wrong - see the cadence link in the anandtech thread I linked above.

Both GF and TSMC are about on the same schedule for high performance variants - might be a month or two in it but both are running off test silicon for major clients and projecting volume production on about the same timeframe.

EDIT: Quoting some relevant sections to posts recently on this subject - this is straight from TSMC:

12FFC is an optical shrink from 16FFC, but some of the logic density and power reduction comes from the low-track standard-cell libraries, so it is best not to just shrink at the die level. Instead, logic should be re-implemented with the new libraries, but SRAM, analog, and I/O just require recharacterization. Comparing N12 to N16 there is a 20% area reduction with the 6-track library, or a 14% area reduction with the 6-track turbo library. There is also a higher performance 9-track library that obviously gives up more area. For HPC, there is a variant process with overdrive and larger contacted-poly-pitch (CPP). In the interconnect, there is wider metal, large vias and via pillars, where you can put vias on top of each other, and so get from low metal to high low-resistance metal without requiring much area and not requiring the risk of EM issues on intermediate layers. TSMC will be ready next month for N7 HPC. At N7+ HPC should get about another 4% improvement in performance or 15% in power versus N7.

• N10 is in volume production with all technical and customer product qualification complete and volume production in F12 and F15
• N7 is targeted at mobile, HPC and automotive and will be "TSMC's finest technology, serving all segments." For power sensitive applications it will be even better than 16FFC but there as the optimized performance version for applications that demand the most performance, and it also fulfills automotive needs. Risk production is April 2017. More than 20 tapeouts are planned this year.
• N7+ will be ready in 2Q 2018.

Someone is deliberately putting out FUD about TSMC/nVidia recently that I've seen a couple of other people repeating here and AFAIK its all BS.

 

[chrcoluk]

nvidia's biggest problem is getting existing nvidia users to upgrade, for that reason alone volta will be a decent jump.

 

[shamus21]

nvidia's biggest problem is getting existing nvidia users to upgrade, for that reason alone volta will be a decent jump.

Well lets hope so but the performance will not get much of a boost from the die shrink.

 

[Furbs]

nvidia's biggest problem is getting existing nvidia users to upgrade, for that reason alone volta will be a decent jump.

 

[Furbs]

nvidia's biggest problem is getting existing nvidia users to upgrade, for that reason alone volta will be a decent jump.

Surely they can achieve this with help from their driver team, I half expect the 9** series cards to start to go the same way the 7** did. Maybe I am just bitter as a previous 780 owner and how the performance stopped keeping up with the competition.

 

[greatash]

Surely they can achieve this with help from their driver team, I half expect the 9** series cards to start to go the same way the 7** did. Maybe I am just bitter as a previous 780 owner and how the performance stopped keeping up with the competition.

I can honestly see this happening with Maxwell/Pascal when Volta comes along with proper DX12 support and every game uses it. Once Nvidia supports full DX12 every game will properly support it and see big gains.

 

[AllBodies]

Nope the site that is putting out that info is wrong - see the cadence link in the anandtech thread I linked above.

Both GF and TSMC are about on the same schedule for high performance variants - might be a month or two in it but both are running off test silicon for major clients and projecting volume production on about the same timeframe.

EDIT: Quoting some relevant sections to posts recently on this subject - this is straight from TSMC:

--------------------------------------------------

12FFC is an optical shrink from 16FFC, but some of the logic density and power reduction comes from the low-track standard-cell libraries, so it is best not to just shrink at the die level. Instead, logic should be re-implemented with the new libraries, but SRAM, analog, and I/O just require recharacterization. Comparing N12 to N16 there is a 20% area reduction with the 6-track library, or a 14% area reduction with the 6-track turbo library. There is also a higher performance 9-track library that obviously gives up more area. For HPC, there is a variant process with overdrive and larger contacted-poly-pitch (CPP). In the interconnect, there is wider metal, large vias and via pillars, where you can put vias on top of each other, and so get from low metal to high low-resistance metal without requiring much area and not requiring the risk of EM issues on intermediate layers. TSMC will be ready next month for N7 HPC. At N7+ HPC should get about another 4% improvement in performance or 15% in power versus N7.

--------------------------------------------------

• N10 is in volume production with all technical and customer product qualification complete and volume production in F12 and F15
• N7 is targeted at mobile, HPC and automotive and will be "TSMC's finest technology, serving all segments." For power sensitive applications it will be even better than 16FFC but there as the optimized performance version for applications that demand the most performance, and it also fulfills automotive needs. Risk production is April 2017. More than 20 tapeouts are planned this year.
• N7+ will be ready in 2Q 2018.

--------------------------------------------------

Someone is deliberately putting out FUD about TSMC/nVidia recently that I've seen a couple of other people repeating here and AFAIK its all BS.

This is a bit confusing. And I can't find anything up to date myself which says TSMC are now planning to enter volume production of mobile and high-power at the same time.

Do you know what date that first quote is from? Since it says they'll be ready (implied risk production) with 7nm HPC a month after the quote.

Also the "N7+ will be ready in 2Q 2018" is referring to risk production of their 7nm+EUV enhanced node, not volume production of their two 7nm nodes.

 

[Rroff]

The implied month after the quote was May 2017 - there is a little difference between mobile and HPC readiness but its in the context of days not 6+ months.

 

[AllBodies]

The implied month after the quote was May 2017 - there is a little difference between mobile and HPC readiness but its in the context of days not 6+ months.

Well I'll give you the benefit of the doubt until I can find something more up to date.

At the same time though, what would this mean for consumer Volta if you think TSMC's readyness date is around Q2-Q3 2018?

It means the Volta coming Q1-Q2 2018 definitely can't be 7nm, so it must either be 16nm or 12nm. But then if they launch in early 2018, they're not going to launch something new till end-2018 at the very earliest.

So do we expect only a Volta Titan to launch until end-2018, where consumer Volta would launch on 7nm? And therefore the GTX 2080-Volta would smash the Volta Titan because it had a huge process advantage?

It doesn't make sense from a launch-cadence point of view.

Either they launch nothing till Q4 2018, would would mean an unprecedented ~20 months in-between launches. Or Volta launches on 16/12nm in Q1 2018 and then has a 7nm refresh, which would be a massive performance boost, in Q1 2019 at the very latest.

I just don't see either of those happening.

I think Volta is coming in Q1-Q2 2018 on 12nm, and then due to release cadence AMD will have a process advantage for ~6+ months unless they get delayed.

This does also mean Nvidia have a process and massive design advantage for 6-9 months in 2018.

Volta on 12nm will make everything AMD has on 14nm look pathetic.

 

[Rroff]

Any GeForce type Volta will almost certainly be 12FF out the gate - just don't assume that means Volta is "stuck" on 12FF for its entire life cycle - for a long time they targetted 10nm for it and compared to a new design will be relatively easy to rework for 7nm.

 

[AllBodies]

Any GeForce type Volta will almost certainly be 12FF out the gate - just don't assume that means Volta is "stuck" on 12FF for its entire life cycle - for a long time they targetted 10nm for it and compared to a new design will be relatively easy to rework for 7nm.

I don't think it's stuck logistically, I just think it's 'stuck' as a business decision due to their release cadence.

The only way I can see it makes sense is something like this:

• The 2080, 2070 and Titan Xv come out Q1-Q2 2018 on 12nm
• 2060 and below come out Q2-Q3 2018
• 2080 Ti comes out Q1 2019 on 7nm, is faster than the Titan Xv but is actually smaller than even the 2080. This is done as it works out cheaper than building lots of Titan Xv-sized cards on 12nm, but obviously adds to R&D, but then this is justified as the 7nm 2080 Ti also acts as a pipe-cleaner for Einstein
• The 2080 Ti is the only 7nm Volta card we ever get
• The 3080, 3070, and Titan Xe come out in Q4 2019-Q1 2020 on 7nm+EUV.

This is the only way it could make sense release cadence wise. But I'm not sure it really makes sense to use the 2080 Ti as a pipecleaner, and this also represents a delay of Einstein.

The other way it could make sense is if Einstein is pushed back on purpose now, and they have a Volta-2.0 whole generation on 7nm. Kind of like Maxwell -> Pascal.

But then in order to go through 2080 12nm -> 2080 Ti 12nm -> 3080 7nm -> 3080 Ti 7nm, you'd be talking about the 3080 Ti launching around Q2 2020. And again this doesn't make sense because that would push back Einstein to Q1 2021 at the earliest, and would also mean skipping 7nm+EUV as 5nm would be ready by then.

The only other time-point to keep in mind is that Nvidia will want to launch some kind of top-end HPC card for compute around the time HBM3 is available (i.e. P100 and V100 tier). So this could mean Einstein is scheduled for Q4-2019 to Q1-2020 on 7nm+EUV + HBM3.


TL;DR Unless Nvidia push back Einstein, have a weird half-generation 7nm Volta refresh, or have a VERY accelerated release schedule of Volta 12nm and Volta2.0 7nm (like 6-9 months maximum between xx80 -> xx80 Ti -> xx80), then it doesn't make sense for Nvidia to NOT have a process disadvantage to AMD for 6-9 months

(but they will then have a forward-loaded huge advantage over AMD for 6-9 months from early-2018)

 

[Rroff]

Volta is a bit of a wildcard to me - not sure if they were talking about the tensor cores or something but those I talked to awhile back a bit more in the know were saying people had the totally wrong understanding of the architecture/end of the stick when it came to Volta - its possible it might not see much use on the GeForce side at all or something.

 

[Shin1234]

here fill ye boots. https://devblogs.nvidia.com/parallelforall/inside-volta/

 

[beany_bot]

Any news on Volta? Been quiet on here for a while.
Been saving every month for what seems like an eternity. I'm looking to upgrade to 4K gaming and Vega being a disappointment (in 4K gaming terms). My eyes are now on Volta.

 

[TNA]

Any news on Volta? Been quiet on here for a while.
Been saving every month for what seems like an eternity. I'm looking to upgrade to 4K gaming and Vega being a disappointment (in 4K gaming terms). My eyes are now on Volta.

I am in the same position, I game at 4K and Vega disappointed me so I am waiting for Volta now. I cannot see Volta coming out this year. Most people estimate Q1 2018.

 

[Rroff]

Any news on Volta? Been quiet on here for a while.
Been saving every month for what seems like an eternity. I'm looking to upgrade to 4K gaming and Vega being a disappointment (in 4K gaming terms). My eyes are now on Volta.

TSMC boosting production of 12FF Q4 this year which probably means there is a new nVidia product coming early next year but doesn't mean that is necessarily GeForce - nVidia have stated they are going to see out the year on Pascal for GeForce products but didn't rule out a potential Titan type card on something else.

 

[r7slayer]

TSMC boosting production of 12FF Q4 this year which probably means there is a new nVidia product coming early next year but doesn't mean that is necessarily GeForce - nVidia have stated they are going to see out the year on Pascal for GeForce products but didn't rule out a potential Titan type card on something else.

Just dominate on professional cards? AI is making them big money too.
Pascal is just fine at the moment. They can just keeping making money off it for a while longer sit out till mid 2k18 before releasing anything. Still come in before AMD and perhaps give them a good time scale to refresh volta on a node shrink.

 

[Telecaster]

The higher end Pascal cards perform such that I can only see the 1050ti type replacement coming first, like the 750ti did with Maxwell. They're in no rush now that Vega has flopped.

 

[HRL]

Can't really see them stalling their release schedule just because of AMD.

Business as usual. Just means there are less likely to be any price drops before Volta.

 

[nashathedog]

Just dominate on professional cards? AI is making them big money too.
Pascal is just fine at the moment. They can just keeping making money off it for a while longer sit out till mid 2k18 before releasing anything. Still come in before AMD and perhaps give them a good time scale to refresh volta on a node shrink.

The way things have been for RTG I bet Navi's close to two years away still. I don't know what to make of Vega, It's not worth what they want so the smart money is on waiting it out, Pascal's over a year old now so it shouldn't be too long till we see Volta in some form.