Nvidia Pascal Speculation Thread

[lanek]

But for Pascal itself, while HBM might ideally make sense with its huge bandwidth and low power requirements, yes there's still supply constrain. SK Hynix is still the only supplier I know of within the next year+. But it is also a JEDEC standard, so I'd be surprised if someone else wasn't working on it.

Well, SK Hynix is the only supplier for the memory stack, but something tell me that, the memory stacks is not the bigger problem... many many other company are bringing the components, starting by the interposer ( who is much easy on paper to do, as you can take any 65nm lines for do it ).

Huum, i could imagine Samsung should be in the first line for make this production of stacked ram, as they do allready similar on other sectors of this industry, but i have heard nothing about them .. maybe they are not interested for the gpu side of it ..

 

[Dangerman]

Time to revive this thread. My info says big Pascal has taped out, and is on TSMC 16nm (Unknown if this is FF or FF+, though I suspect it is FF+). Target release date is Q1'16. This is a change from Kepler and Maxwell where the smallest chip (GK107 and GM107 respectively) taped out first. Maybe the experience with 20nm was enough for NV to go back to their usual big die first strategy. Given the huge gains in performance compared to 28nm, and the fact that the 16nm process is both immature and quite expensive, I suspect the die size may be a bit smaller than what we've seen with GK110/GM200.

Still wondering what they're doing on Samsung 14nm though, maybe just test chips for now?

What about HBM2? I'm guessing they'll be a GP2xxx with HBM2 later in 2016 and GP100 for the HPC market?

 

[Kaotik]

What about HBM2? I'm guessing they'll be a GP2xxx with HBM2 later in 2016 and GP100 for the HPC market?

They're supposed to go straight to HBM2

 

[Blazkowicz]

I could be wrong, but I think all recent GPUs support GDDR5 on a 128-bit bus, even if some SKUs only use 64 bits, or use the full bus with DDR3.

Slightly wrong : AMD went to 128bit on the low end (Oland, Iceland) - yet still sells older DX11 gens that have 64 bits ; nvidia has native 64bit width GPU, namely GK208 and GM108. (and GF119)

 

[Megadrive1988]

Time to revive this thread. My info says big Pascal has taped out, and is on TSMC 16nm (Unknown if this is FF or FF+, though I suspect it is FF+). Target release date is Q1'16. This is a change from Kepler and Maxwell where the smallest chip (GK107 and GM107 respectively) taped out first. Maybe the experience with 20nm was enough for NV to go back to their usual big die first strategy. Given the huge gains in performance compared to 28nm, and the fact that the 16nm process is both immature and quite expensive, I suspect the die size may be a bit smaller than what we've seen with GK110/GM200.

Still wondering what they're doing on Samsung 14nm though, maybe just test chips for now?

wccftech put up an article directly based on your post and they link right to it, as your report .

http://wccftech.com/nvidia-gp100-chip-taped-gp-104-arriving-q2-2016/

Nvidia’s Pascal ‘GP100 GPU’ Reportedly Taped Out on TSMC 16nm Node – ‘Big’ Pascal and GP104 Arriving Q2 2016

The age of the 28nm GPUs is finally drawing to a close and while the faithful node lasted for much longer than originally anticipated, 16nm GPUs are finally on the horizon. The first inkling comes from [URL='https://forum.beyond3d.com/threads/nvidia-pascal-speculation-thread.55552/page-8#post-1848329']a report by Beyond3D[/url] that the GP100 chip (First Generation Pascal Flagship Chip) has taped out on the 16nm FinFET+ node of TSMC. They also list that GPUs are expected by the second quarter of 2016.

Nvidia’s next generation pascal GP100 and GP104 chips with HBM arriving late second quarter of 2016

The GP100 chip is of course the ‘big’ Pascal chip. Nvidia’s nomenclature is as follows: the ‘G’ denotes the chip being a GPU while while the the second letter denotes the architecture, Pascal in this case. The first digit indicates the generation of the chip while as the remaining two indicate the relative performance of the lineup in an inverted order, with 00 being the highest, 04 being immediately below and so forth.Since the GX X04 chip is usually less powerful, it is also the first to be expected in the consumer side. However, if Beyond3D is right, then this might not be the case.

The report states that the GP100 ‘big’ pascal chip will hit the professional market first, giving time for green to roll out the GP104 chip to the mainstream consumer segment (followed by GP100 at a later date). The GM200 taped out in June and hit the shelves in March. That is a total time of approximately 9 months. The GM204 on the other hand took only five months to hit the shelves from its tape out date. At this schedule we can tentatively expect the Pascal architecture GPUs next year in late second quarter. The GP100 chip will also have HBM2 memory, which unlike the HBM1 specification can extend upto 32 GB of total vRAM size.

The report states that the gaming edition of the GP100 will be limited to 16GB of vRAM (which frankly is pretty effing huge considering the current standards. NVLink will also be supported for professional end users. While the node will shift downwards, due to the relative immaturity of the 16nm process I wouldn’t expect chips to exceed 550mm² initially. That means you are looking at around 5000-6000 CUDA cores. With the architecture improvement and and the process upgrade – an improvement of around 50 – 80% is on the cards depending on how well Nvidia handles it. There isn’t a single GPU card in existence that can handle 4K@60fps on its own. First or second generation Pascal however, should most definitely be able to hit the mark with the technological upgrades heading its way.

 

[mczak]

Slightly wrong : AMD went to 128bit on the low end (Oland, Iceland) - yet still sells older DX11 gens that have 64 bits

Actually AMD also has a "new" 64bit chip (Sun) though arguably it's not a "full" chip as it relies on display outputs, video decode etc. from the IGP.

 

[UniversalTruth]

According to the same article at http://wccftech.com/amd-working-entire-range-hbm-gpus-follow-fiji-fury-lineup/#ixzz3fpyrkUdp

Pascal will have troubles because:

AMD Has Priority Access To HBM2 Production Capacity And Is Aiming To Take Full Advantage Of It

Our sources have told us that AMD management has thrown significant weight behind this new range of graphics chips to accelerate its development and time to market. This is we’re told is to take advantage of a deal established with SK Hynix which gives AMD priority to HBM2 capacity which is going to be in limited supply initially. Capturing as much of the initial production capacity as possible would give AMD an edge against its main rival, Nvidia, going into the next generation of GPUs featuring second generation HBM technology. Which is exactly what the company is gunning for.

If successful this would push Nvidia’s Pascal launch schedule back than what the company originally planned. But the time waiting will not go to waste we’ve no doubt. The company has reportedly already taped out its first Pascal chip. However without access to HBM2 the graphics cards based on Pascal can’t produced. This may lead the company to consider the possibility of spending more time working on successive Pascal chips until it can secure enough HBM2 production capacity.

This fascinating dynamic is the result of AMD’s seven year involvement in the co-development of the High Bandwidth Memory JEDEC standard with its partner SK Hynix we’re told. Which is why AMD was the first company to introduce a product featuring HBM. And why the the first generation of the technology is exclusive to the company.

 

[Alexko]

This is Wccftech…

 

[Deleted member 13524]

This is Wccftech…

However, it'd make perfect sense to get contractual full priority on HBM2's production, since they co-developed the technology.

It'd also be quite the move from AMD to make a top-to-bottom GPU range with all of them using HBM2, (not only but also) to cock-block nVidia from getting enough quantities in the Pascal range for anything other than their top-end cards.

 

[silent_guy]

Conversely, it would be easy for Nvidia to pressure Hynix into providing them with a low volume part, by threatening to shift high volume parts to a second source.

 

[trinibwoy]

However, it'd make perfect sense to get contractual full priority on HBM2's production, since they co-developed the technology.

It'd also be quite the move from AMD to make a top-to-bottom GPU range with all of them using HBM2, (not only but also) to cock-block nVidia from getting enough quantities in the Pascal range for anything other than their top-end cards.

And piss off an even higher volume customer in the process? Yeah...not smart. AMD doesn't have the money or market share to make that happen.

 

[Deleted member 13524]

And piss off an even higher volume customer in the process? Yeah...not smart. AMD doesn't have the money or market share to make that happen.

What do customers have to do with AMD getting priority over the HBM orders in exchange for investing and working on the actual development of the technology?

If nVidia is scratching their balls while AMD and Hynix are hard at work developing HBM (or just pouring money on Gameworks-infected developer contracts instead), then they can use GDDR5 like they've been using until now. Besides, isn't GDDR5 so much better, according to silent_guy?

Not everything that AMD develops must be for nVidia do use right away. The opposite certainly doesn't happen, does it?

 

[Deleted member 2197]

Not everything that AMD develops must be for nVidia do use right away. The opposite certainly doesn't happen, does it?

Oh, is AMD living up to their Open Source ideals with HBM as well? NV likely has had working prototypes for quite a while and just waiting for volume to ramp up.

 

[Razor1]

its highly unlikely nV will be supply constrained at least for the high end, HBM2 comes into mass production when q4 2015 and when are the next gen cards supposed to come out end of first half 2016? Common sense dictates it doesn't' take 6+ months to get production up to the point they can supply AMD and nV at only the high end......
If AMD is planning on releasing all their cards top to bottom at the same time, no company would shoot themselves in the foot by cutting off its own market to 20% of sales.

 

[UniversalTruth]

Oh, is AMD living up to their Open Source ideals with HBM as well? NV likely has had working prototypes for quite a while and just waiting for volume to ramp up.

I would cheer AMD if they start with dirty tricks to the competition to spoil their progress.
As history teaches us, it has always been the opposite (Intel and nvidia both doing against AMD).

I would think that in the next two-three years it would be fine with AMD providing the only solutions.

 

[Deleted member 13524]

If AMD is planning on releasing all their cards top to bottom at the same time, no company would shoot themselves in the foot by cutting off its own market to 20% of sales.

As I mentioned, that wouldn't happen on cheer choice. It'd happen over a contract where AMD pays for and helps with the development in exchange for priority (or even time exclusivity) on the orders for HBM2.
It wouldn't be the first time for something like this to happen in the tech world, and it wouldn't be the last either.

Court documents show that Nokia and ST Microelectronics signed a contract which gives Nokia 12 months exclusivity on the microphone chips

And you can bet the HTC One sold a lot more than the Lumias during 2013.

 

[Razor1]

So you are saying exclusivity now wouldn't that have been taken care of with HBM currently?

And I think AMD does have exclusivity for the first generation of HBM for one year.

 

[lanek]

Oh, is AMD living up to their Open Source ideals with HBM as well? NV likely has had working prototypes for quite a while and just waiting for volume to ramp up.

What the open source have to do with hardware components supply ?

This said, reading the article i got the feeling that in reality AMD want, try to release early thoses gpu's for keep the benefit of the actual Hynix exclusivity contract who will run to current 2016. Not extend it further.

 

[3dilettante]

What do customers have to do with AMD getting priority over the HBM orders in exchange for investing and working on the actual development of the technology?

If nVidia is scratching their balls while AMD and Hynix are hard at work developing HBM (or just pouring money on Gameworks-infected developer contracts instead), then they can use GDDR5 like they've been using until now. Besides, isn't GDDR5 so much better, according to silent_guy?

Not everything that AMD develops must be for nVidia do use right away. The opposite certainly doesn't happen, does it?

Hynix has indicated that there are non-graphics targets for HBM2, such as networking. Perhaps those wins are further off, but unlike graphics there is competition from competing high-bandwidth memory types, so Hynix would be interested in expanding HBM beyond graphics before something like HMC becomes completely entrenched. Whether it is plausible that AMD could demand that Hynix service all its other customers before servicing Nvidia, I am not sure.

 

[RecessionCone]

If nVidia is scratching their balls while AMD and Hynix are hard at work developing HBM (or just pouring money on Gameworks-infected developer contracts instead), then they can use GDDR5 like they've been using until now.

What evidence do you have that AMD and Hynix are the only ones driving HBM? Other than marketing droids?