Nvidia Pascal GPU to be built on TSMC's 16nm FinFET process

[GuidoLS]

Nvidia will only have enough HBM memory to send a few sample cards to journalists, so TSMC only have Apple to worry about

Seriously though, TSMC is the main player in this game, is it really surprising when they get big contracts?

I see a lot of people getting mouth-breathing enthusiastic about HBM this and HBM that... as far as I (and the rest of the world) were able to discern, Nvidia and AMD pretty much tied straight down the middle on this generation of cards - Nvidia did it on the old tech, and there's a LOT of questions about availability on any card that comes with HBM - to the point where AMD is using it as an excuse to not give out gpu samples to certain tech sites.

/waits with trepidation for the person with the username that resembles a cartoon baby Walt Disney elephant to pop in, preaching about how I don't worship tech...

[DanceswithUnix]

I see a lot of people getting mouth-breathing enthusiastic about HBM this and HBM that... as far as I (and the rest of the world) were able to discern, Nvidia and AMD pretty much tied straight down the middle on this generation of cards - Nvidia did it on the old tech, and there's a LOT of questions about availability on any card that comes with HBM - to the point where AMD is using it as an excuse to not give out gpu samples to certain tech sites.

/waits with trepidation for the person with the username that resembles a cartoon baby Walt Disney elephant to pop in, preaching about how I don't worship tech...

For years GPUs have been an exercise in getting around the memory bandwidth wall vs cost. Nvidia seem to have some really good texture compression which got them through this time, but if they make a GPU on 16nm with twice the shaders of the current 28nm parts then that runs out and HBM starts looking necessary.

For the mid range cards that I usually buy I don't see it as necessary, but if the volume gets high enough that it becomes normal then I won't say no.

[Corky34]

Wouldn't that be "fully accelerated"?

My shoes are fully compatible with a road in that there is no part that I am incapable of traversing, but I can't do 30mph unless I swap my shoes for a car

Yea I understand where you're coming from, but without looking are all the DX12 feature sets strictly to do with performance?
I was under the impression some higher levels features improve quality at the cost of performance.

I still don't understand why DX feature levels are getting so much attention for DX12...

Just guessing hear but with 11 people probably didn't care much because, among other things, DX11 was still single threaded, people didn't fully understand the different feature sets, and you had to be running an OS that received a lot of negative media attention. Again guessing but DX12 has probably received so much attention in part because of AMD's sterling job in showing people that there's a need for a multi-threaded API.

The whole feature level thing is a bit FUBAR if you ask me, but then again I didn't design the API.

[HalloweenJack]

with AMD in all the consoles - you can bet that if async speeds games up , then console ports to desktop will use it.

[DanceswithUnix]

The whole feature level thing is a bit FUBAR if you ask me, but then again I didn't design the API.

That is just how APIs work in general. The high level layers asks the low level layer what it can do, and works around the gaps.

The thing is, missing out a feature can be a completely valid thing to do. Let's say Nvidia had decided to have some really stonking async shaders in the 980. I am making figures up here as an example just to show that the world isn't a simple black and white place, just bare with me.

Every feature has a cost, you use a bit of silicon that could have been shaders and you make the circuitry more complex which means it doesn't clock so high and it uses power from your power budget that could have gone into shader clocks. So yes, when it wins you get a 30% gain in performance, but what if adding the feature has a background cost of 10%? Now in those cases when you don't use async you lose, and 10% is the difference between being faster and slower than Fury. Now it isn't looking so attractive. If the cost is 1%, then that sounds better. But still, sometimes the GPU guys would kill for a 1% performance lift. There is always a line, you have to guess where to draw it for each feature.

[Corky34]

with AMD in all the consoles - you can bet that if async speeds games up , then console ports to desktop will use it.

Async compute/shaders isn't a requirement for DX12.

@DanceswithUnix, My FUBAR comment was in relation to the unnecessary fragmentation (imo) that Microsoft decided to introduce with the whole feature level thing, not in relation to how certain features effect performance.

[HalloweenJack]

Async compute/shaders isn't a requirement for DX12.

@DanceswithUnix, My FUBAR comment was in relation to the unnecessary fragmentation (imo) that Microsoft decided to introduce with the whole feature level thing, not in relation to how certain features effect performance.

its one of the `optional` parts - like the other parts which NVidia support that AMD don't - hence the `fully compatible` comment from earlier

but , my comment still stands - with consoles being AMD powered and Async helps them , you can bet real money than ports will use Async.

[GuidoLS]

For years GPUs have been an exercise in getting around the memory bandwidth wall vs cost. Nvidia seem to have some really good texture compression which got them through this time, but if they make a GPU on 16nm with twice the shaders of the current 28nm parts then that runs out and HBM starts looking necessary.

For the mid range cards that I usually buy I don't see it as necessary, but if the volume gets high enough that it becomes normal then I won't say no.

Interesting - FF crashed on me, and Hexus went and saved what was basically an incomplete post that I didn't bother finishing....

At any rate, when Nvidia announced Pascal last year, it was also announced that they were going to use HBM, a la Hynix/AMD. Source. It's about 2/3rds down. I'm going to guess that if someone is willing to make that kind of announcement, they have the products involved already under contract. I can see it getting very messy, financially, if at any point there was some proof of an artificial shortage once a contract has been drawn.

Then again, until such time as prices come WAY down and production ramps WAY up, this is always going to be at the top end of the product lines, with a price that puts it out of reach of the normal person.

[GuidoLS]

Async compute/shaders isn't a requirement for DX12.

@DanceswithUnix, My FUBAR comment was in relation to the unnecessary fragmentation (imo) that Microsoft decided to introduce with the whole feature level thing, not in relation to how certain features effect performance.

/tongue in cheek -

Isn't DX12 basically Mantle? If so, doesn't that make this AMD's fault?

/tongue in cheek

[DanceswithUnix]

Async compute/shaders isn't a requirement for DX12.

@DanceswithUnix, My FUBAR comment was in relation to the unnecessary fragmentation (imo) that Microsoft decided to introduce with the whole feature level thing, not in relation to how certain features effect performance.

Fragmentation is just a reflection on it being a big complex world out there.

If you nail everything down hard then you stifle progress. Microsoft have done enough of that in the past with things like assuming all sound cards are a Soundblaster, I would have hoped they had learned their lesson.

[Corky34]

its one of the `optional` parts - like the other parts which NVidia support that AMD don't - hence the `fully compatible` comment from earlier

No it's not, the fact that people believe it's part of DX12, even just an optional part, is testament to AMDs marketing.

Async shaders and Asynchronous shader engine only concerns internal microarchitectural details of the GPU but has absolutely no relation to the optional features of the Direct3D runtime, that's unless you care to point out where in the Direct3D 12 feature options it talks about Async shaders or Asynchronous shader engines.

Fragmentation is just a reflection on it being a big complex world out there...

IMO it's not because it being a big complex world out there, that's unless things suddenly became more complex from DX10 to DX11, that being when feature levels were introduced by Microsoft, it was a choice to introduce tiered feature levels as in the past a card either did or did not support a feature.

Take Conservative rasterization for example, Microsoft could've said that for a GPU to support DX12 it must have the capability to do the highest level of conservative rasterization, just like they did with certain features of DX10, instead the have chosen to have low, middle, and high.

[watercooled]

A few posts behind here so bear with me:

I see a lot of people getting mouth-breathing enthusiastic about HBM this and HBM that... as far as I (and the rest of the world) were able to discern, Nvidia and AMD pretty much tied straight down the middle on this generation of cards - Nvidia did it on the old tech, and there's a LOT of questions about availability on any card that comes with HBM - to the point where AMD is using it as an excuse to not give out gpu samples to certain tech sites.

It's pretty much agreed upon that HBM or an equivalent memory capable of scaling bandwidth much higher than GDDR5 is or will be necessary very soon, that part really isn't in debate. Like how GDDR5 improved upon GDDR3 - AMD probably could have managed the 4870 with GDDR3 (and similar to what we're seeing now, GDDR5 was only used on their halo product to start with) but it was approaching the end of its scaling and a replacement would have been needed at some point. Same with HBM.

I don't recall seeing HBM supply used as an excuse by AMD themselves, regardless of what the net's rumour mills imply, and it doesn't make sense as one either.

Yea I understand where you're coming from, but without looking are all the DX12 feature sets strictly to do with performance?
I was under the impression some higher levels features improve quality at the cost of performance.

Well a lot of it is hardware acceleration of things which are already possible, albeit more slowly, in software. The actual impact of performance either way will likely depend heavily on exactly how it's used in a given game, and whether a software implementation would be possible is a question for someone more knowledgeable of the subject. I know it's been done, but I don't know how feasible it is in software for games.

Async compute/shaders isn't a requirement for DX12.

But it's potentially really useful.

/tongue in cheek -

Isn't DX12 basically Mantle? If so, doesn't that make this AMD's fault?

/tongue in cheek

Not sure if the tongue in cheek part meant this, but DX12 is not Mantle. It seeks to achieve many of the same goals, but they're not the same thing.

No it's not, the fact that people believe it's part of DX12, even just an optional part, is testament to AMDs marketing.

It not being a part of DX12 is largely academic; games can and do make use of the feature regardless. It doesn't matter that MS don't mention it.

[GuidoLS]

Not sure if the tongue in cheek part meant this, but DX12 is not Mantle. It seeks to achieve many of the same goals, but they're not the same thing.

That's why the tongue in cheek was even necessary. There are more than a few persons here that will insist that not only is DX12 the exact same thing as Mantle, they'll insist, with RIGHTEOUS INDIGNATION, that Microsoft stole the technology, or at the very least, were dead in the water until AMD, in their most gracious of benefactorial selflessness, gave Mantle to the world, the very act of which deserves its own mid-winter holiday... no names needed, as there's not a soul here on Hexus that has read more than 2 threads on this topic who don't know the players.

Am I still being sarcastic? In part, yes, and in part, I'm being dead serious.

As for the HBM thing, and the net rumor mill, that's where 90% or so of every quote here on the forums come from. Sadly, it seems it's where a good bit of the news on the main page comes from at times as well. I don't know why some sites got GPU's and some didn't, and I don't really care. Just about every site on the web has some sort of bias, or they wouldn't exist. Every site that has more than 1 reporter (or repeater, in many cases) has multiple biases. And what they report will always lean in the direction of the bias. And once again, I don't care. All I want is for something to work. If product A plays my favorite game at 60+FPS at 1080p, and product B does the same, but for $20 less, and is as reliable, I'll buy product B. That's all I want or need to know (well, I am somewhat socially responsible, and won't buy Apple products, or anything else stamped Foxconn, because a lot of their innards are made by what is, essentially, slave labor, which is why they have a $200billion cash surplus - easy to make $$ when you aren't paying labor)

[Corky34]

...But it's potentially really useful.
...
It not being a part of DX12 is largely academic; games can and do make use of the feature regardless. It doesn't matter that MS don't mention it.

Totally agree, that it's not part of DX12 is academic but we are nerds are we not
TBH I'm surprised its taken so long to move away from using a single thread to dispatch jobs to the GPU (excluding the unknown secret sauce possibly being done in drivers), isn't it almost a decade since we've had multi-core CPUs?

[DanceswithUnix]

Totally agree, that it's not part of DX12 is academic but we are nerds are we not
TBH I'm surprised its taken so long to move away from using a single thread to dispatch jobs to the GPU (excluding the unknown secret sauce possibly being done in drivers), isn't it almost a decade since we've had multi-core CPUs?

Yes, but Iris-GL that became OpenGL that started all this is *much* older than that

[shaithis]

but , my comment still stands - with consoles being AMD powered and Async helps them , you can bet real money than ports will use Async.

I wouldn't bet on it.

When we saw XB1 and PS4 were using x86 and DirectX parts, we all thought that this was going to be a new era in amazing console ports. So far I have found the ports to generally be worse than ever.