[ Various] NVIDIA could launch Pascal-based mobile GPUs by end-2016

[maddie]

Another article suggest NV's Pascal is due in the 2nd half of 2016, with validation of Pascal to be done in this first half. Samsung will be supplying them with HBM2 chips.

Looks like SK Hynix is supplying AMD, hence the exclusive rumor awhile back.

ST Micro is doing HMC (for Intel?).

http://www.kitguru.net/components/g...-gpus-rumoured-to-launch-in-2nd-half-of-2016/

The source is a Korean tech news, they have discussions with folks in Samsung to get the article with quoted sources from those in the industry.

This suggests NV is at least 1 quarter behind AMD, if AMD's plans to launch in Q2 isn't delayed.

Here's the interesting part, since the mid-range chip is unlikely to use HBM2, it means NV will be having a big Pascal w/ HBM2 sometime in 2H 2016. Likely Tesla, but maybe consumer Titan class!

IF AMD only has 2 Polaris in 2016, small and mid-range, they got nothing to compete with big Pascal. > This is potentially a major blow to AMD if NV release consumer Pascal Titan in late 2016, it's the perception, NV is supreme leader in graphics that will be very damaging. <

IMO, if AMD isn't working on a big Polaris for 2016 release, they will be in a world of hurt.

I'm curious to what you would class as a mid-range 14nm GPU, regarding the die size?

 

[tential]

Yeah. nVidia will own the $1000+ consumer GPU market.

If you own the high end the trickle down effect is real.

 

[3DVagabond]

If you own the high end the trickle down effect is real.

Depends on what they have to compete below it. Sure it has marketing value, but that's not going to make people buy noncompetitive sku's.

Besides, I'll believe it when I see it, as far as getting it to market in a timely manner.

 

[Silverforce11]

Depends on what they have to compete below it. Sure it has marketing value, but that's not going to make people buy noncompetitive sku's.

Besides, I'll believe it when I see it, as far as getting it to market in a timely manner.

Come on man, we already know the masses buy purely based on brand recognition. They do not go and read reviews in-depth to inform their purchase.

If AMD doesn't show up to the fight at the top end against big Pascal, the perception will be that NV is just vastly faster, superior = drive sales for the masses for the entire stack.

AMD wanted to change their image from being the cheap alternative, then the only way they can do so is take the leadership on performance.

 

[Silverforce11]

I'm curious to what you would class as a mid-range 14nm GPU, regarding the die size?

Next-gen on new node, mid-range chip should defeat current gen big-chip.

~300mm2 as it has been for many generations.

I am not expecting prices will be "mid-range" but towards the high-end.

 

[MrTeal]

Friend of mine was there, i'm not in this business.
You can hear it in this talk mentioned, that they have it in lab:
http://images.nvidia.com/events/sc15/SC5102-path-exascale-computing.html

Also discussing two new techniques which should be implemented in pascal.

Thanks, great link. Some things about that are crazy though and don't make any sense.
IE, 4:05, he hypothesises on a 10mmx10mm die, which would be 1/3rd the length and 1/9th the area of Pascal. That would be a 900mm² die for Pascal, which is outside the realm of possibility.
At 16:00 he says they have Pascal in the lab, but no indication of what one. It could be any of the production GPUs, or even a test chip for the architecture. Hopefully it is GP100 and we might see at least a Titan version late 2016/early 2017, but I'm not holding my breathe quite yet.

 

[tential]

Come on man, we already know the masses buy purely based on brand recognition. They do not go and read reviews in-depth to inform their purchase.

If AMD doesn't show up to the fight at the top end against big Pascal, the perception will be that NV is just vastly faster, superior = drive sales for the masses for the entire stack.

AMD wanted to change their image from being the cheap alternative, then the only way they can do so is take the leadership on performance.

This. Not that I personally care, I'll buy the best price/performance part.

But yes, I decided to read more into my purchases rather than just buy an nvidia gpu within my price range (nvidia the way it's mean to be played or whatever) , and that's how I got informed. Then it turned out amd was my best choice each time now. Otherwise I was about to get a gtx 680 just because who cares I have money and I used nvidia.

People like to be on the winning team. Not necessarily making the best decision for themselves.
Ego is bigger than rational.

 

[maddie]

Thanks, great link. Some things about that are crazy though and don't make any sense.
IE, 4:05, he hypothesises on a 10mmx10mm die, which would be 1/3rd the length and 1/9th the area of Pascal. That would be a 900mm² die for Pascal, which is outside the realm of possibility.
At 16:00 he says they have Pascal in the lab, but no indication of what one. It could be any of the production GPUs, or even a test chip for the architecture. Hopefully it is GP100 and we might see at least a Titan version late 2016/early 2017, but I'm not holding my breathe quite yet.

I see you had the exact same reaction as me on viewing the presentation.
I bookmarked the link.

Great info on power saving though. I did not realize the immense savings potentially possible. Noticed when he claimed that node shrinks were becoming not as necessary as they used to be, as focus is increasingly being placed on architecture to save energy?

 

[3DVagabond]

Come on man, we already know the masses buy purely based on brand recognition. They do not go and read reviews in-depth to inform their purchase.

If AMD doesn't show up to the fight at the top end against big Pascal, the perception will be that NV is just vastly faster, superior = drive sales for the masses for the entire stack.

AMD wanted to change their image from being the cheap alternative, then the only way they can do so is take the leadership on performance.

This. Not that I personally care, I'll buy the best price/performance part.

But yes, I decided to read more into my purchases rather than just buy an nvidia gpu within my price range (nvidia the way it's mean to be played or whatever) , and that's how I got informed. Then it turned out amd was my best choice each time now. Otherwise I was about to get a gtx 680 just because who cares I have money and I used nvidia.

People like to be on the winning team. Not necessarily making the best decision for themselves.
Ego is bigger than rational.

It all depends on how the reviewers present them. Look at the 970. Except for perf/W it doesn't break any performance ground. Neither does the 980. Hawaii actually out performs them. When they were released though, the review sites made it look like they did (Let's stick performance numbers up for those old reference cards while they are throttling.) and we saw what happened.

While it might not make them happy, AMD's best marketing strategy is perf/$. It's when they got away from that (7970) the wheels came off. And like I said, I'll believe that nVidia can release a true flagship model in a timely manner when it happens. Nothing is pointing to that happening currently.

 

[maddie]

This. Not that I personally care, I'll buy the best price/performance part.

But yes, I decided to read more into my purchases rather than just buy an nvidia gpu within my price range (nvidia the way it's mean to be played or whatever) , and that's how I got informed. Then it turned out amd was my best choice each time now. Otherwise I was about to get a gtx 680 just because who cares I have money and I used nvidia.

People like to be on the winning team. Not necessarily making the best decision for themselves.
Ego is bigger than rational.

I guess you'll be waiting till the big Pascal is released before deciding.

 

[tential]

I guess you'll be waiting till the big Pascal is released before deciding.

No I'm locked into freesync.

 

[Silverforce11]

While it might not make them happy, AMD's best marketing strategy is perf/$. It's when they got away from that (7970) the wheels came off. And like I said, I'll believe that nVidia can release a true flagship model in a timely manner when it happens. Nothing is pointing to that happening currently.

The wheels never came off. Compare the marketshare of that era. Even after Hawaii's disastrous reference cooler launch, marketshare was climbing back to 37% for them, on a trajectory up to ~40%.

Then the 970/980 landed and boom, total collapse as everyone rushes out to upgrade to the 970, great bang for buck, great perf/w.

On Steam, there's more 970s than ALL of AMD's GPUs combined in recent times. That's the damage the 970 did and still does to this day.

If AMD wants to reverse their situation, they need to deliver on excellent low end and mid-range SKUs, with leadership in perf/w, while at the top-end, they need to be competitive with big Pascal. Else the avg consumer won't even take notice of their stack. For people to give them a second look, AMD has to come out and take the perf crown else it will be business as usual and they'll just get whatever NV releases.

Imagine headlines "World's fastest Graphics/GPU is AMD!", would that be an incentive for the avg consumer to check out AMD's products? You bet.

 

[xpea]

If you own the high end the trickle down effect is real.

^^ This
Marketing 101

especially when you are already considered as the best brand/ecosystem

 

[Pariah]

If AMD wants to reverse their situation, they need to deliver on excellent low end and mid-range SKUs, with leadership in perf/w, while at the top-end, they need to be competitive with big Pascal. Else the avg consumer won't even take notice of their stack. For people to give them a second look, AMD has to come out and take the perf crown else it will be business as usual and they'll just get whatever NV releases.

Imagine headlines "World's fastest Graphics/GPU is AMD!", would that be an incentive for the avg consumer to check out AMD's products? You bet.

No one cares about perf/w. Even fewer care in the low to mid-range. The only reason anyone cared about power usage for a while is because of how tragically bad the 290x reference cooler was which falsely gave people the impression that the card was using such absurd amounts of power that AMD couldn't come up with a HSF that was capable of keeping the heat and noise to an acceptable level.

What card is the current perf/w leader among high end cards? I don't know offhand (I don't really care), but I'll guess it is the Nano without looking it up. Who bought a Nano? No one. So long as the ergonomics of the card are acceptable, performance is the only metric that matters.

The only way AMD gains back any significant market share is if NVidia screws up. AMD set the precedent with their last round of cards that they are no longer going to try and beat NVidia by being the value leader. You can't beat the incumbent by simply matching or slightly beating them in performance if you're charging the same amount. It's unlikely that either side is ever going to have a significant advantage in performance comparing competing cards from the same generation, so AMD's only hope of getting market share back is either NVidia getting Pascal months late to market, or NVidia releasing their cards slightly after AMD and pricing them way higher at similar performance levels. The 2nd isn't very likely, so AMD has to hope they have no delays in their release schedule while not AMD'ing any other aspect of the release (bad ergonomics, not enough product, blatantly false product claims, etc..) while NVidia runs into some non-trivial delays in their release.

 

[3DVagabond]

^^ This
Marketing 101

especially when you are already considered as the best brand/ecosystem

The bold is much more important than the "trickle down" is.

 

[AtenRa]

IMO, if AMD isn't working on a big Polaris for 2016 release, they will be in a world of hurt.

So if they will have a better Laptop product and sell more volume at higher margins than NVIDIA in that segment and loose the extremely low volume of High-End Desktop segment how they will be hurt ???

I believe AMD will leave NVIDIA play with huge dies for now and strongly attack high volume high margins Laptop segment and low-mid Desktop.
I also believe that NVIDIA will do the same, im not expecting a huge NV die at 16nm before H2 2017 for retail use.

 

[Silverforce11]

No one cares about perf/w. Even fewer care in the low to mid-range.

Fully disagree.

Especially the low to mid-range. Perf/w is a very important metric as it determines where they folk out more for quality PSU and better cases & airflow.

 

[Silverforce11]

So if they will have a better Laptop product and sell more volume at higher margins than NVIDIA in that segment and loose the extremely low volume of High-End Desktop segment how they will be hurt ???

I believe AMD will leave NVIDIA play with huge dies for now and strongly attack high volume high margins Laptop segment and low-mid Desktop.
I also believe that NVIDIA will do the same, im not expecting a huge NV die at 16nm before H2 2017 for retail use.

You're assuming they will have better Laptop products?

NV will be a quarter late with Pascal, but when they show, I fully expect them to be very competitive. Just take Maxwell and shrink it will result in excellent performance and perf/w. They already have a good foundation to build on.

Then the HBM2 big Pascal ships in late 2016, cements performance crown for the generation.

They are the incumbent for performance, perf/w and brand. If AMD wants to change the status quo like they've been talking so hard, they have to not only match NV but exceed them. Missing the top-end next-gen chip is a recipe for failure.

 

[AtenRa]

You're assuming they will have better Laptop products?

NV will be a quarter late with Pascal, but when they show, I fully expect them to be very competitive. Just take Maxwell and shrink it will result in excellent performance and perf/w. They already have a good foundation to build on.

Then the HBM2 big Pascal ships in late 2016, cements performance crown for the generation.

They are the incumbent for performance, perf/w and brand. If AMD wants to change the status quo like they've been talking so hard, they have to not only match NV but exceed them. Missing the top-end next-gen chip is a recipe for failure.

I didnt say NV will not be competitive in laptops, what im saying is what if AMD manage to sell lots of laptop SKUs now they will have a ways better Laptop product than what they had the last years and regain Laptop GPU share.

By big pascal you mean a 200-300mm2 die or 400-500mm2 die ??? because I cant see NV or AMD bring 400-500mm2 dies before H2 2017 or early 2018. By then we may see a big AMD GPU as well.

 

[Silverforce11]

By big pascal you mean a 200-300mm2 die or 400-500mm2 die ??? because I cant see NV or AMD bring 400-500mm2 dies before H2 2017 or early 2018. By then we may see a big AMD GPU as well.

Big >500mm2.

Samsung is mass producing HBM2 NOW, with only one customer: nVidia.

Do you think GP204 will be HBM2 or GDDR5X? Right. So there's only one chip that has HBM2, GP100, the big one.

You don't mass produce HBM2 unless there's a strong demand. This suggests NV will be ready with its big Pascal in late 2016, in volume. There's only one scenario where that could happen, a consumer Titan SKU alongside all the Teslas.

 

[Pariah]

Especially the low to mid-range. Perf/w is a very important metric as it determines where they folk out more for quality PSU and better cases & airflow.

No, it doesn't. The low to mid range cards use so little power that PSU and airflow is largely irrelevant.

If perf/w is such an important metric, why does Anandtech not bother to report it in any of their video card reviews?.. Or TomsHardware?... Or Techreport?... or [H]HardOCP?... Or Bjorn3D? Surely if anyone cared, at least one of the major sites would bother to report this metric, either because of reader demand or because they themselves deemed it important. Yet, none do. Why?

 

[AtenRa]

Big >500mm2.

Samsung is mass producing HBM2 NOW, with only one customer: nVidia.

Do you think GP204 will be HBM2 or GDDR5X? Right. So there's only one chip that has HBM2, GP100, the big one.

You don't mass produce HBM2 unless there's a strong demand. This suggests NV will be ready with its big Pascal in late 2016, in volume. There's only one scenario where that could happen, a consumer Titan SKU alongside all the Teslas.

Such a product will be severely volume constrained for a desktop retail release at the end of 2016. Also HBM2 mass production doesnt mean they will produce the same volume as GDDR-5 in 2016. It only means they have started production and HBM2 will be available in 2016.

But let me say here that i believe the 200-300mm2 dies will use HBM2 in 2016, at least the AMD ones. If Polaris 11 is a 250-300mm2 die the performance with HBM2 will be higher than Fury X at way lower power consumption. That GPU will also be perfect for high-end Gaming LapTops that are low volume high margin products.

 

[3DVagabond]

No, it doesn't. The low to mid range cards use so little power that PSU and airflow is largely irrelevant.

If perf/w is such an important metric, why does Anandtech not bother to report it in any of their video card reviews?.. Or TomsHardware?... Or Techreport?... or [H]HardOCP?... Or Bjorn3D? Surely if anyone cared, at least one of the major sites would bother to report this metric, either because of reader demand or because they themselves deemed it important. Yet, none do. Why?

Of course perf/W is important. Are you serious? In the end it is what defines the limit of overall performance.

 

[Pariah]

Of course perf/W is important. Are you serious? In the end it is what defines the limit of overall performance.

It doesn't matter to the enduser. No one sits around calculating perf/w at home and buys based on what card finishes on top. I already posed the question of why the major review sites don't track the metric, and as usual you ignored it and tried to change the angle. You can twist what I said however you feel to try and make an argument out of it.

 

[Silverforce11]

They don't need to track the metric because its self explanatory. Similar performance, similar price, one card uses much more power, it's worse.

Many reviews devote paragraphs or even conclusions based on perf/w with mentions of performance and power consumption.

You would be one of the few who suggests perf/w isn't important.