Does the RTX series create an openning for AMD?

[DrMrLordX]

Nvidia are just giving their customers to AMD.

I'm gonna have to agree with @AtenRa here, that AMD is not getting any new customers unless they can launch 7nm cards to challenge Nvidia. They had a shot at GF 12nm and they didn't take it, why??? The market was ripe for product refresh and iterative improvements. 12nm Vega with working NGG etc. could be on the market RIGHT NOW. Instead consumers got nothing from AMD in 2018! Nothing. What is wrong with them?

We don't even know exactly what they are launching in 2019 or when.

 

[Despoiler]

Yeah, in candy land maybe .. the land of overzealous "analysts and their preposterous expectations.

Clearly you are the adult in the room that people should listen to. If you'd look at Nvidia's launch history it's always about ~1.5 years. Pascal is an outlier with 2+ years.

I'm gonna have to agree with @AtenRa here, that AMD is not getting any new customers unless they can launch 7nm cards to challenge Nvidia. They had a shot at GF 12nm and they didn't take it, why??? The market was ripe for product refresh and iterative improvements. 12nm Vega with working NGG etc. could be on the market RIGHT NOW. Instead consumers got nothing from AMD in 2018! Nothing. What is wrong with them?

We don't even know exactly what they are launching in 2019 or when.

12nm is like 20nm. The improvements aren't worth the investment. The cards would have only been +5-8% better perf and they still wouldn't have a 1080ti competitor. 7nm is shipping earlier than anticipated. Why go incremental when you can put your money into a real node? Yes AMD gives up some new product sales, but Vega is just now coming back to MSRP and people are buying them. AMD also knows that Nvidia is locked into a "12nm" RTX. The RTX product timeline is able to be anticipated. RTX is also huge and thus costly. AMD can swoop in with smaller chips that meet the price points customers want and they can make a bunch of money on. We've seen this same situation play out before. I think this is an easy call for AMD.

 

[ub4ty]

I'm gonna have to agree with @AtenRa here, that AMD is not getting any new customers unless they can launch 7nm cards to challenge Nvidia. They had a shot at GF 12nm and they didn't take it, why??? The market was ripe for product refresh and iterative improvements. 12nm Vega with working NGG etc. could be on the market RIGHT NOW. Instead consumers got nothing from AMD in 2018! Nothing. What is wrong with them?

We don't even know exactly what they are launching in 2019 or when.

Look at the Steam data on Monitor resolution and GPU usage.
90%+ are @ or below 1920x1080
Only 1% of people are on 1080tis. Another 1% or so are on 1080s. The max ceiling is at 1070. The majority are on 1060s and below.
AMD already serves this market.
12nm would be silly to launch on with 7nm right around the corner.
It would have made zero sense for AMD to make a move on 12nm when they are already shipping 7nm by the end of the year.
No one is begging for a GPU right now. What people are begging for are lower prices. AMD launched an HBM equipped GPU 2 years ago. 2 years ago...
Nvdia 2 years later at 12nm still doesn't have an HBM equipped GPU and their prices are insane.
Nvidia still doesn't have 2:1 FP16 even on a $1200 GPU.

AMD is going by the market and open standards. They are right where they want to be.
If they launch in 2019, I'll be interested. If they launch in 2020 I'll still be interested. I game a good deal on a Maxwell 2GB card. There's not a single game I play nor any popular game that I would have performance issues on.
AMD already has products for these markets. Vega 56/64 are a the higher end of performance for the broad based market as shown by the hardware surveys. They were priced great and inline with Nvidia.

I'm not paying more than $500 or so for a GPU ever.
RTX is a complete non-event for me.
AMD doesn't have the lead or money to make such stupid mistakes. Nvidia is.. and its historically why the company that remains in the lead for so long eventually begins falling and losing market share.
AMD's letting them gas themselves out and make all of the mistakes while learning/adjusting and carefully forming their line of attack.
Consoles are powered by AMD. At these stupid prices for modern GPUs people will go consoles even more. During this crypto mining idiocy, many people did.

You can't force progress when its unaffordable. Ray tracing is a dead feature because its unaffordable.
Tick -> tock ...
You get a major release and then you get a minor (repeat).

No educated consumer buys a minor release at a premium.
12nm is a minor release.
7nm is the real deal just like 14nm was.

The majority of my hardware is 14nm. I only upgrade at 7nm.
PC hardware is a 5 year thing.

If you're upgrading more frequently, you're blowing money away.
I have a range of hobbies but I don't waste money try to call such a thing a hobby that's simply called consumerism.

If you maintain a bleeding edge stance, this is more in line with a professional/business use case that will pay premiums for hardware because the resultant hardware results in maintaining the ability to make money.

No consumer on earth needs any of this stuff.
If you want to sell volume and get a technology adopted you reduce margin expectations and ship volume.

Nvidia is of the stance given their dominant market share that such a thing will only eat into sales their already making. It's why companies who maintain such leads eventually fall. At such a point, they have no where to go but down. So, they maintain even more insane margins.
Then an upstart comings along and cuts you at your knees. This is simply the nature and story of tech. No company has ever been able to defy this reality. Intel didn't and Nvidia wont.
Nothing can grow forever. Margins can't increase past a sensible point.
When you hit peak, there's no way but down.
I see nothing of grand excitement in Geforce20. I see everything that's indicative of a company hitting a peak and not knowing where to go from there.

As a quite avid gamer, I have no desire for even a 1080ti or a 1080 much less this new crop of cards.
Of course, if the prices fell, I would as the value would be clear. It's not clear for me or the grand majority of people.
So, this launch from Nvidia is literally nothing burger AMD honestly doesn't have to challenge. They already did 2 years ago. They're now refining drivers/support.
Show me an Nvidia card with 21 Teraflops of FP16 performance and post its price.
Tensor cores are meme.

 

[Head1985]

I'm gonna have to agree with @AtenRa here, that AMD is not getting any new customers unless they can launch 7nm cards to challenge Nvidia. They had a shot at GF 12nm and they didn't take it, why??? The market was ripe for product refresh and iterative improvements. 12nm Vega with working NGG etc. could be on the market RIGHT NOW. Instead consumers got nothing from AMD in 2018! Nothing. What is wrong with them?

We don't even know exactly what they are launching in 2019 or when.

Whats wrong with them?Money and lack of skilled ingeneers.They have ben stuck on 4096/256/64rops and 4x shader engines(that is very poor balance) since 2015 and they didnt do anything about it.
Vega 64 have same performance like vega 56 at same clock because there is tons of bottlenecks in currest GCN with 4096/256/64 4x SE setup.

 

[PeterScott]

Clearly you are the adult in the room that people should listen to. If you'd look at Nvidia's launch history it's always about ~1.5 years. Pascal is an outlier with 2+ years.

I agree that it will likely be 1.5+ years until NVidia 3000 series cards.

But it won't be 1.5+ years until NVidia has 7nm GPUs.

NVidia has in the past started a series of cards on one node, and shifted to smaller node later.

This is likely a perfect case for that and a sensible course of action.

Starting at 12nm gives NVidia a risk free earlier start on their 2000 series cards, now there is no pressure on getting 7nm designs working early. They can migrate some of the 2000 series to 7nm, when and where it makes sense, most likely after 7nm has good yields for lower costs.

The downside for this strategy is more money wasted on the high upfront costs of develop multiple dies at both 12nm and 7nm in the same series of cards. But NVidia has higher volumes and more money to make any waste on those extra die costs more palatable. So it is a reasonable choice for NVidia.

AMD OTOH is taking the delays and greater risks of skipping straight to 7nm. But it saves them those extra up front costs of creating dies that may replaced sooner rather than later. Since AMD has less volume and money, this strategy is a reasonable one for them.

Both companies made a reasonable, if not the most reasonable, choice for their position in the market.

NVidias choice has the most advantages, but that is what comes from having more money/resources to throw at a problem.

 

[AtenRa]

NVidias choice has the most advantages, but that is what comes from having more money/resources to throw at a problem.

I will say it here again, there is a possibility NVIDIA was forced to use 12nm because it didnt have capacity left at 7nm in 2018 and at least for the first half of 2019.
You dont use a 752mm2 die (TU102) for consumer products no matter who you are when 7nm is that close.

What we know is that Apple will use 7nm in 2018 for mobile SoCs and AMD will use 7nm for VEGA 20 but also ZEN 2 for EPYC 2 server SKUs and also consumer RYZEN 3 and TR 3.
Perhaps AMD will also have another die for consumer GPUs later on in second half of 2019, so that will make a lot of 7nm wafers in 2018 and 2019. There may not be enough 7nm capacity left for NVIDIA to use for those RTX dies at that time, so they had to go with 12nm that was cheaper, way more mature and already in production since 2017.

 

[PeterScott]

I will say it here again, there is a possibility NVIDIA was forced to use 12nm because it didnt have capacity left at 7nm in 2018 and at least for the first half of 2019.
You dont use a 752mm2 die (TU102) for consumer products no matter who you are when 7nm is that close.

Forced? Come on. NVidia has been a very important TSMC customer for many years, being almost exclusively TSMC, and now suddenly they are shut out?

7nm mobile, is not necessarily the same process as 7nm for GPUs that suck down 200+ watts, and the state of readiness for high volume, high power GPU cores on 7nm is still an unknown.

Jumping to 7nm is a delay and a risk. NVidia chose early and certainty.

But they will still likely have their first gaming 7nm GPU within months of AMD, unless AMD quickly re-purposes Vega-20 as gaming card.

 

[AtenRa]

Forced? Come on. NVidia has been a very important TSMC customer for many years, being almost exclusively TSMC, and now suddenly they are shut out?

7nm mobile, is not necessarily the same process as 7nm for GPUs that suck down 200+ watts, and the state of readiness for high volume, high power GPU cores on 7nm is still an unknown.

Jumping to 7nm is a delay and a risk. NVidia chose early and certainty.

But they will still likely have their first gaming 7nm GPU within months of AMD, unless AMD quickly re-purposes Vega-20 as gaming card.

Companies book time on TSMC and any other Fab a long time ahead. It seems that AMD knew about GF 7nm was a dead end and re-arranged their strategy accordingly booking 7nm capacity in TSMC ahead of NVIDIA. Apple is also an early new process adopter and they would have booked 7nm capacity ahead of NVIDIA.
Now, since all 7nm at TSMC in 2018/2019 is build on the same FAB 12/15, it doesnt matter if you are going to use 7nm SoC or 7nm HP for your dies, you are still booking time and use the FAB 12/15 instead of your competitor. And because TSMC doesnt allocate the same capacity for 7nm today as 16nm, if you dont book early you are simple shut out until next availability.

 

[PeterScott]

Companies book time on TSMC and any other Fab a long time ahead.

Who says NVidia didn't also book time in the slots they wanted? Just because NVidia recently came out with 12nm parts, doesn't mean they don't have 7nm slots also booked for 6-9 months later for their 7nm parts.

NVidia is profitably enough to work on more than one process node at a time.

 

[Hitman928]

I can't tell you for TSMC 7nm, but for another advanced node that I have access to that is certainly in less demand than TSMC 7nm, there is a minimum 4 month lead time to schedule fab space and it fills up fast (usually if you wait until the 4 month minimum, it's already full).

I'd imagine TSMC 7nm is on "back order" for 7nm by 2+ runs right now.

 

[AtenRa]

Who says NVidia didn't also book time in the slots they wanted? Just because NVidia recently came out with 12nm parts, doesn't mean they don't have 7nm slots also booked for 6-9 months later for their 7nm parts.

NVidia is profitably enough to work on more than one process node at a time.

You said it yourself. That is the thing, they were shut out of early 7nm and they had to use 12nm until they would have 7nm capacity 6-9-12 months later than the others.
Nobody said they were shut out indefinitely from 7nm, they were forced to use 12nm TODAY because they didnt have 7nm capacity to use now.

 

[PeterScott]

You said it yourself. That is the thing, they were shut out of early 7nm and they had to use 12nm until they would have 7nm capacity 6-9-12 months later than the others.
Nobody said they were shut out indefinitely from 7nm, they were forced to use 12nm TODAY because they didnt have 7nm capacity to use now.

I didn't say that. It was a rhetorical question indicating the opposite. "Come on" was the hint, if you didn't get it.

They are not shut out. The went with 12nm for their first 2000 series cards, because it is a sure thing delivering now. I don't see any GPUs on 7nm right now. Do you?

Also I didn't say 6-9-12 months later than the others. I said 6-9 months later than their own 12nm parts.

Who knows when you see gaming cards from AMD at 7nm, that could still be 6 months from now.

So AMD and NVidia 7nm gaming cards could be arriving within a couple of months from each other.

Heck, NVidia could release GTX 2060 as their pipe cleaner part at 7nm in Q1-2019, beating AMD to 7nm gaming cards.

The bottom line is that NVidia's 12nm cards, don't block them from building 7nm GPUs just as early as anyone else. Though they obviously have less pressure to do so.

 

[Muhammed]

LOL people are already forgetting NVIDIA has Ampere in the pipeline .. This one will show on 7nm for professionals first as well. And Unlike AMD. NVIDIA currently don't announce their future plans so as not to distract people from current offerings. Turing came out of nowhere without pre-announcement. And Ampere will too.

I will say it here again, there is a possibility NVIDIA was forced to use 12nm because it didnt have capacity left at 7nm in 2018 and at least for the first half of 2019.

NVIDIA created custom TSMC 12nm long before any GPU maker did. Specifically for Volta. They weren't forced to do anything. They planned for their releases accordingly.

You are telling me that after creating 12nm exclusively for them, they were abandoned by TSMC on 7nm? that's non sense.

 

[ub4ty]

So the consensus is : As a consumer, you'd have to have a screw loose in your head to buy 12nm GPUs at a premium, a 1st gen launch of ray tracing, + all of the other issues when you have a much more longer lasting and sound release coming right around the corner : QED : Geforce20 is an absolute joke as I felt it was.

Catch you guys in 2019 or 2020.
I'm stocked on GPUs/CPUs/Storage for now.
I was holding out for a miracle and Nvidia delivered a turd.

 

[Kenmitch]

The usual bickering back and forth, displeasure, no leaks, and 1 guy pimping nvidia to all no end....Did I miss anything the last couple days?

 

[GodisanAtheist]

A question for those that might be more familiar with the GCN arch than I am: what is the roadblock that has prevented AMD from scaling up, restructuring the layout of GCN from it's cap of 4096:256:64 SP:TMU:ROPS?

To the layman it seems odd that AMD seems to have settled on this incredibly shader heavy, obviously under utilized design for Fury, then Vega 10, and likely Vega 20 (7nm) while NV apparently can scale their arch up and down and left and right and mix up their Streaming Multiprocessor (SM) counts every generation with 6 Graphics Processing Clusters.

I understand that AMD has been strapped for cash, but a simple remix of the parts that are already there to extract better utilization, then adding more of those parts, seems like it the sort of things you'd be paying your engineers to do first as low hanging fruit prior to trying to strap a basket full of untested efficiency goodies to an unbalanced design.

 

[Head1985]

A question for those that might be more familiar with the GCN arch than I am: what is the roadblock that has prevented AMD from scaling up, restructuring the layout of GCN from it's cap of 4096:256:64 SP:TMU:ROPS?

To the layman it seems odd that AMD seems to have settled on this incredibly shader heavy, obviously under utilized design for Fury, then Vega 10, and likely Vega 20 (7nm) while NV apparently can scale their arch up and down and left and right and mix up their Streaming Multiprocessor (SM) counts every generation with 6 Graphics Processing Clusters.

I understand that AMD has been strapped for cash, but a simple remix of the parts that are already there to extract better utilization, then adding more of those parts, seems like it the sort of things you'd be paying your engineers to do first as low hanging fruit prior to trying to strap a basket full of untested efficiency goodies to an unbalanced design.

That is very good question.
1-lack of skilled ingeneers.I think all ingeneers who worked back in ATI on GCN left AMD during Rory reed was CEO-after hawaii launch and they dont know how to do it.
2-they know how to do it but dont have money to do it

With 6x shader engines they could use up to 96Rops, but they will need change SP numbers because 4096SP cant be divided by 6.Most close number divided by 6 is 4224SP and it would by 704SP by 1x shader engine.Same ratio like hawaii.(furyx and vega64 have 1024SP per shader engine so 45% more...)

4224SP
264TMU
96Rops
That GPu will have 50% more geometry performance because each Shader engine have one geometry engine.Also 50% more pixel fillrate.

btw pixel fillrate and geometry performance are two bottlenecks what current GCN have.Next is shader underutilization because there is too much shaders in single shader engine.All those bottlenecks will fix 6x shader engines.
4th is memory bandwidth.Delta color compression is way worse than nvidias.

 

[beginner99]

they were abandoned by TSMC on 7nm? that's non sense.

No one said abandoned. it's more like too late to the party. Booking production windows isn't a charity and first come first served. 12nm might actually be the reason NV was late to the party. it's at least a reasonable possibility compared to the "NV is perfect and planned this all along". 1000-series for sure lasted longer than usual.

 

[krumme]

A question for those that might be more familiar with the GCN arch than I am: what is the roadblock that has prevented AMD from scaling up, restructuring the layout of GCN from it's cap of 4096:256:64 SP:TMU:ROPS?

To the layman it seems odd that AMD seems to have settled on this incredibly shader heavy, obviously under utilized design for Fury, then Vega 10, and likely Vega 20 (7nm) while NV apparently can scale their arch up and down and left and right and mix up their Streaming Multiprocessor (SM) counts every generation with 6 Graphics Processing Clusters.

I understand that AMD has been strapped for cash, but a simple remix of the parts that are already there to extract better utilization, then adding more of those parts, seems like it the sort of things you'd be paying your engineers to do first as low hanging fruit prior to trying to strap a basket full of untested efficiency goodies to an unbalanced design.

As i understand it from the original vega whitepaper the ngg path would have removed that problem of underutilized shaders.

Ngg was borked. No bonus for raja. Rajs goes Intel.

 

[DrMrLordX]

12nm is like 20nm. The improvements aren't worth the investment.

What? GF 20nm didn't even work. 12nm works, they've already made the investment. GF has the wafers available, and has had them available since . . . whenever full production of retail Pinnacle Ridge began.

The cards would have only been +5-8% better perf and they still wouldn't have a 1080ti competitor.

I mean, okay? Not that you know that for certain, since Vega's promised performance is never going to emerge until they bugfix the damn thing. 5-8% plus bugfixing could have produced interesting results. Besides, producing a 1080Ti competitor was not exactly the point. Providing additional volume to a market stripped bare by heavy demand as recently as a few months ago would have made them a lot of money, and they could have avoided rehashing Polaris again by instead replacing it with a shrunk, stripped-down Vega product (think of it as a precursor to Navi10).

AMD should have updated Vega 64/56, and run a stripped down Vega to replace their Polaris lineup, all on 12nm. Who cares if they couldn't top the 1080Ti? The main thing is to outperform their own lineup and to keep the product lineup fresh in the face of possible/inevitable Navi delays.

The only reasons NOT to do so are to commit all dice to pro graphics cards (including custom/semi-custom jobs we may never hear about) or to simply abandon the consumer market.

7nm is shipping earlier than anticipated.

What are you talking about? Navi is where exactly? What consumer 7nm GPUs will AMD launch in 2018? Or ever? We don't even know! 7nm consumer Vega may never happen, and Navi is God-only-knows when. If we are lucky it shows up in Q2 2019. Maybe. It might not show up until Q4 2019.

Why go incremental when you can put your money into a real node?

AMD went incremental with Ryzen and it worked for them. That is why you do it, to continue innovating and releasing products. Instead of rehashing Polaris.

AMD also knows that Nvidia is locked into a "12nm" RTX. The RTX product timeline is able to be anticipated. RTX is also huge and thus costly. AMD can swoop in with smaller chips that meet the price points customers want and they can make a bunch of money on. We've seen this same situation play out before. I think this is an easy call for AMD.

AMD is not going to swoop in with any smaller 7nm product. Their roadmaps are terribly muddy when it comes to the consumer market.

If they wanted to take advantage of the RTX situation as you frame it, they would need product ready to go right now. They do not have it. They will not have it. They may not launch another consumer card until late 2019.

Look at the Steam data on Monitor resolution and GPU usage.
90%+ are @ or below 1920x1080
Only 1% of people are on 1080tis. Another 1% or so are on 1080s. The max ceiling is at 1070. The majority are on 1060s and below.
AMD already serves this market.
12nm would be silly to launch on with 7nm right around the corner.

AMD is trying to serve this market with Polaris, which is already out-of-date. The market is about to be/is being saturated with used Polaris products from miners dumping cards. If AMD wants to continue selling cards in this price category, they need an actual product refresh . . . not just more Polaris. 12nm Vega in various forms fits that bill.

7nm is not "right around the corner" for AMD's consumer GPU lineup. Why do people assume Navi10 is knocking on the door?

It would have made zero sense for AMD to make a move on 12nm when they are already shipping 7nm by the end of the year.

They are not shipping consumer graphics cards on 7nm by the end of the year. Only 7nm Vega ships in 2018, to the pro/AI market. That's it.

No one is begging for a GPU right now. What people are begging for are lower prices. AMD launched an HBM equipped GPU 2 years ago. 2 years ago...

AMD isn't lowering MSRP on anything. Bringing 12nm GPUs to the market would at least lower the market value of their older Vega and Polaris cards as well.

If they launch in 2019, I'll be interested. If they launch in 2020 I'll still be interested. I game a good deal on a Maxwell 2GB card. There's not a single game I play nor any popular game that I would have performance issues on.
AMD already has products for these markets. Vega 56/64 are a the higher end of performance for the broad based market as shown by the hardware surveys. They were priced great and inline with Nvidia.

I like my Vega and all but if AMD wants to take market share from Nvidia, sitting around and not launching anything is not how you do it. And they are not taking much market share with Vega, sorry. Maybe they should be, but they aren't. They could probably do it with Navi if it could launch right now. It is not ready.

AMD's letting them gas themselves out and make all of the mistakes while learning/adjusting and carefully forming their line of attack.

AMD isn't doing anything on the consumer front. At. All.

Consoles are powered by AMD. At these stupid prices for modern GPUs people will go consoles even more. During this crypto mining idiocy, many people did.

And? I like that they got console design wins, but AMD's margins on dGPUs are much better than anything they get from console hardware sales.

You get a major release and then you get a minor (repeat).

AMD hasn't released a GPU since 2017. 2017! Where is the minor release? There isn't one!

Whats wrong with them?Money and lack of skilled ingeneers.They have ben stuck on 4096/256/64rops and 4x shader engines(that is very poor balance) since 2015 and they didnt do anything about it.
Vega 64 have same performance like vega 56 at same clock because there is tons of bottlenecks in currest GCN with 4096/256/64 4x SE setup.

Eh the whole "hate GCN" thing is a bad trope at this point. I will agree that they may not have enough money to do entirely new designs on 12nm. But they sold enough Polaris and Vega cards to bankroll some Vega rehashes + bugfixes on 12nm, easily. And they have (and had) the talent to manage all that.

 

[Hitman928]

snip

Based on the latest comments from Lisa Su and another executive (Devinder Kumar), I'm not expecting Navi until Q4 2019.

I think the lack of improvement on the Vega architecture stems from a few things, namely, lack of R&D funds, a poor plan by Raja, and a shift of priority to data center needs.

 

[Kenmitch]

Seems like the majority just wants AMD to make Nvidias prices lower. If you wouldn't buy the products then why bitch they haven't made them? I've read too many comments around the net to think otherwise.

 

[Flayed]

I would buy an AMD card if they release one with 1080 ti level of performance. I'd consider one with less performance if the price was right.

 

[tajoh111]

You said it yourself. That is the thing, they were shut out of early 7nm and they had to use 12nm until they would have 7nm capacity 6-9-12 months later than the others.
Nobody said they were shut out indefinitely from 7nm, they were forced to use 12nm TODAY because they didnt have 7nm capacity to use now.

Everyone is shut out for 7nm for the most part today. Aside from apple and telecoms, there is little capacity for anyone else. What it means is there is no volume for anyone including AMD and Nvidia. You can launch an ultra low volume part and pay an arm and a leg, but nothing mainstream because your supply and cost will limit your potential revenue and sales while potentially cannibalizing your existing supply unlimited cards already on the market. A 7nm card from AMD would obsolete it's current line and the volume of such a product at this time, would not offset the revenue loss from this obsolescence/price drops. It's why vega 20 is launching purely in the professional market where it does not touch the rest of amds consumer lineup.

What is likely going to happen is AMD will focus everything they have on getting Zen 2 out the door. This is because the market opportunity is better because of Intel's vulnerability(difficulty with 10nm and being stuck on their own node), the market being 5x to 10 larger and the margins being a whole lot better(213mm2 ryzen(initial price 500 to 200 dollars) vs 231mm2(Polaris 240 to 200 dollars with board parts and ram eating into that margin.

Navi will likely come in 2019 2nd half because strategically, they want to use all wafers on ryzen 2. Its the more profitable move and will generate a whole lot more money because it is less risky(because an intel counter is more difficult), per wafer generates more money and it is easier to grab marketshare from intel.

Navi is likely to beat nvidia 7nm products to market but the lead will be short lived because of the amount of resources Nvidia is pouring into its r and d along with the sheer man powers. AMD is not locking nvidia out of anything with leverage. Them using hbm2 before AMD is an example of this and their profits afford them much more leverage compared to amd.

Nvidia has the most aggressive release strategy and they will not allow AMD to have exclusitivity on 7nm for very long. They are incredibly aggressive in preventing market share gains in competitors. Turing will have a short life and it exists to take advantage of the gap in AMD lifecycle and to do a product refresh ala tick tock to stimulate demand when product sales slow which is definitely occurring.

 

[AtenRa]

A 7nm card from AMD would obsolete it's current line and the volume of such a product at this time, would not offset the revenue loss from this obsolescence/price drops. It's why vega 20 is launching purely in the professional market where it does not touch the rest of amds consumer lineup.

A 7nm card from AMD now will not obsolete anything, it will only be a high-end product selling at 600+ USD leaving the rest of the line intact, much like NVDIA did with RTX cards.