According to a report by BusinessKorea TSMC has been selected to produce the upcoming Pascal GPU after initially competing with Samsung for the contract.
Though some had considered the possibility of both Samsung and TSMC sharing production (albeit on two different process nodes, as Samsung is on 14 nm FinFET), in the end the duties fall on TSMC's 16 nm FinFET alone if this report is accurate. The move is not too surprising considering the longstanding position TSMC has maintained as a fab for GPU makers and Samsung's lack of experience in this area.
The report didn't make the release date for Pascal any more clear, naming it "next year" for the new HBM-powered GPU, which will also reportedly feature 16 GB of HBM 2 memory for the flagship version of the card. This would potentially be the first GPU released at 16 nm (unless AMD has something in the works before Pascal's release), as all current AMD and NVIDIA GPUs are manufactured at 28 nm.
Merge mobile and desktop GPU
Merge mobile and desktop GPU lines already…16nm is the RIGHT time to do it.
And its fine if mobile is missing the very top end one, and desktop is missing a few of the lowest end ones.
Desktop can be clocked a bit higher, but it should be the exact same gpu.
Ofcource you mean mobile as
Ofcource you mean mobile as in lap-top. At first I thought you were wanting to merge 250w and 1.2w parts, but that would be silly at this point.
Here’s the laptop gamer
Here’s the laptop gamer everyone! 16nm doesn’t mean the high-end chips won’t be consuming 250W and that is not really manageable in a laptop.
Any chance we can get a more
Any chance we can get a more detailed look into the differences in the actual processes and what this change means?
I know that 16nm/14nm etc doesn’t really mean much anymore as many of the features (initial metal layers) don’t scale down to that level but it would be really nice to have some idea of some of the trade offs involved. I think some of us might like the next step down look into how they compare.
Will this one have better
Will this one have better asynchronous compute ability, or will Nvidia green goblin gimp that ability in all but its HPC server line of SKUs. Arctic Islands is going to get even more asynchronous compute ability for the games makers to utilize, and the more the code can be run on the GPU independently of the CPU the less latency becomes an issue. The greatest latency issues are between the CPU down on the motherboard communicating with GPU on the PCI card, so the more asynchronous compute workloads that can be transferred to the GPU and run on the GPU’s asynchronous compute engines independently of the CPU the less that the CPU to GPU communication latency inducing steps become an issue. DX12 and Vulkan will make great use of any GPU asynchronous compute abilities for gaming and other GPGPU workloads.
Async was originally AMD
Async was originally AMD locked tech so can’t expect nvidia get it perfect when AMD has had a 4 year head start on their locked tech. It would be like physx becoming DX12 standard and expect AMD to support it perfectly from day 1.
AND Now the whining, and
AND Now the whining, and Nvidia did the asynchronous compute gimping as part of its plan to reduce the compute ability on its consumer SKUs, so anyone that wanted to use Nvidia’s GPUs for compute were forced to pay for Nvidia’s higher priced pro SKUs. This is a classic monopoly market segmentation tactic where Nvidia having the large market monopoly strove to fruther milk its customers for more profits. And Now that the gaming engine companies, and HSA enabled Graphics APIs(Vulkan, DX12) are able to and are making use of more GPU Hardware asynchronous compute abilities for gaming, and Nvidia’s apologists like you begin the same old FUD anew!
How many users that buy high
How many users that buy high end cards need high compute? Very SMALL %, they card is focused on GAMING not 5 diff things which makes card more complex then it needs to be. Most people that needed high compute speeds were ppl in pro end things that spending 2-3grand for a card is normal. Stop being a moron for 2 sec.
Stop being a shameless shill
Stop being a shameless shill for the hands that feed you, and most Nvidia gamers are clueless, but AMD’s customers do more with their GPUs than simply game. And now that the software is catching up to the hardware Nvidia sends you out for damage control! The entire mobile market will be using Vulkan and GPGPU for games on the mobile devices, and now even the big gaming Rigs will make use of more hardware asynchronous compute on AMD, like PowerVR, and Mali/other GPU/GPGPU graphics/compute for the mobile market. The games engines will use every bit of that hardware asynchronous compute from many GPU manufactures that do not gimp with the ACE types of hardware functionality on their GPUs. Even on Apples Metal will use those ACE units for AMD GPUs, and equivalent units on PowerVR GPUs will allow for better gaming, and other compute. Even the VR gaming folks are looking for the low latency run it on the GPU hardware asynchronous compute to make for better VR gaming.
Not everybody drinks Nvidia’s Green kool aid like you do! Some people do non gaming graphics with their GPUs, and why pay extra for compute that used to come standard. AMD spent the time and effort, and now AMD reaps the rewards, as will the other members of the HSA foundation with their GPU/GPGPU products.
I just picked up the
I just picked up the GTX980-Ti, please come out next year around this time. I don’t want a situation were I picked up my 4790k 3 weeks before Skylake launched.
That sounds like a personal
That sounds like a personal problem.
#Humblebrag
#Humblebrag
It will be a while before
It will be a while before come out yet. On the brand you bought, evga has a setup up program so that is limited to 90 days if i remember.
First world problems.
First world problems.
PUMPED!!! And ready for
PUMPED!!! And ready for Pascal! bring it NV!
If this is true and they are
If this is true and they are about to ramp up production for HBM2, then this also means that production of AMD’s HBM2 chips is about to go underway too(if it hasn’t already) since AMD is allegedly supposed to get the “first dip” of HBM2.
SK hynix Inc. and AMD
SK hynix Inc. and AMD partnered to bring HBM to the market, they even did the work to make HBM a JEDEC standard. Now SK hynix’s and AMD’s exact process for the making of the now JEDEC standard HBM is patentable, but the JEDEC standard is open to all memory manufacturers to develop their own patentable processes for HBM memory, the standard is open but not a company’s individual process, and AMD and SK hynix have the advantage of having a working proprietary HBM process that implements the JEDEC HMB standard.
Care to guess if AMD is going to share any of that patentable process that it worked so long and hard to get up and running with SK hynix Inc. Nvidia and its HBM partner are going to have to reinvent the wheel so to speak in developing a working and patentable HBM process of their own, and AMD and its memory partner have a sizable HBM process lead! HBM 2 will be a drop in replacement for AMD’s current Fury line of products should AMD want to make a revised series of SKUs that could utilize HBM 2, before the arrival of Arctic Islands.
Actually, amd confirmed this
Actually, amd confirmed this week that there will be no licenses or fees to pay for HBM memory chips….all though they should’ve. Those money would be well spent indeed
That’s because it is a JEDEC
That’s because it is a JEDEC standard, but did you read the post that you replied to! The JEDEC HBM standard requires no licensing fees to implement, but SK hynix Inc. and AMD partnered to create their own patented process to implement the Free and Open and FEE-LESS JEDEC HBM standard. The JEDEC standard requires no fees to implement, but it’s not the same thing as a company’s or companies’ proprietary process implementation like that proprietary Hardware/firmware HBM process that AMD and SK hynix Inc. partnered to create.
JEDEC only has the standard, a set of rules and data protocols that are the HBM open standard. It’s like the PCI-SIG standards any company can implement the PCI rules, and data protocol chips to implement the PCI standard, but the chips/firmware that a company creates to implement the PCI standards from PCI-SIG are the company’s proprietary property I.E. the created chip/firmware(that implements the standard).
Do you understand the concepts behind the industry standards that goes into the many computing devices that are available on the market today, and the standards that are open and free to use, and the actual parts that a company spends millions/billions to create to implement a device that utilizes the open/free standards. The company/ies have proprietary rights to the hardware/processes that they created, but the open standards the proprietary hardware implements are still open and free.
Irony in all that, if they
Irony in all that, if they charged for it it would be against their claims of wanting stuff to be standard and free.
Oh boo hoo, its not AMD’s
Oh boo hoo, its not AMD’s standard to charge for It’s a JEDEC HBM standard now, so the charging for use of the standard needs to be directed at JEDEC and its board. Nvidia and its memory partner has to come up with their own proprietary hardware to implement the JEDEC HBM memory standard! AMD’s and SK hynix Inc. spent the time/money to develop their proprietary hardware/firmware, some packaging and fab processes, and licensed some micro-bump processes from other third parties. So Nvidia, and possibly Samsung will have to get their engineers to make their own proprietary equivalent Chips/firmware from scratch to do the same thing to implement the JEDEC standard. A lot of that packaging/micro-bump IP is licensed through industry third parties, but the AMD/SK hynix Inc. patented processes developed for their HBM product are theirs to use, and not to give/share with the competition. Ask JEDEC what they will charge for usage rights on the JEDEC part of the standard.
” Nvidia and its HBM partner
” Nvidia and its HBM partner are going to have to reinvent the wheel so to speak in developing a working and patentable HBM process of their own, ”
Don’t forget that Nvidia is ALSO a member of the HBM working group. They have just as much assess to the development and standards documentation as AMD does.
For the standard/HBM working
For the standard/HBM working group, but not for AMD’s and SK Hynix’s proprietary implementation of the Open To All HBM standard, and the somewhat more loosely defined HBM standard is a different thing from that which one company, or group of companies that partner to create a proprietary implementation that uses any industry standard. And Everybody in the computing business is in with the JEDEC standards organization, the PCI-SIG standards organization, VESA, etc. etc. The standards are usually just incomplete outlines with some less exact electrical/engineering guidelines and some and more exact electrical/engineering guidelines depending on the standard. But they do not define a companies exact product that implements the/any standard and that exact product is protected by patent. Nvidia, and any partner, as well as any Industry third party is welcome to use the JEDEC industry standard, but not any other companie’s/group’s exact proprietary implementation Hardware/firmware that implements the standard, that is proprietary property. Nvidia/partner(if any) will have to do their own engineering and expend their own monitory/other resources or license the design form an third party industry IP provider. Hell Intel and the other Fabs license a lot of that third party IP for their fab businesses, and there are about six main chip fab third party IP providers that provide the bulk of the chip fab IP to the entire world.
edti: monitory/other
to:
edti: monitory/other
to: monetary/other
If Samsung had won the
If Samsung had won the contract, would that have meant that the devices could also be produced at Global Foundries since they are using Samsung’s process?
Wouldn’t that have been funny?
Not really, it’s just
Not really, it’s just business and Apple will do business with Samsung on one end, and be in court with Samsung the very same day on the other end. Hell Samsung and GlobalFoundries will both be getting some Power8 fabrication business, GlobalFoundries is contracted for IBM’s internal/systems Power8s and Power9s, but Samsung will be open to get some of the OpenPower power8 licensees business, as that is up to the licensee to choose who they have as a fab partner. The foundries will cross license processes if it means more revenues, and GlobalFoundries will not have enough fab capacity for Licensed power8s if Google decides to Go with power8s for their server farms.
Samsung, GlobalFoundries, and IBM have been in a technology sharing association for a good while now.
Yes, business is business.
My
Yes, business is business.
My point is that it would be funny if GF were producing both AMD GPUs and nVidia ones.
About as funny as TSMC
About as funny as TSMC producing both AMD GPUs and nVidia ones.
What is so unusual about
What is so unusual about that! GF is a Chip foundry business. What about TSMC for the same thing!
Why do you think that IBM
Why do you think that IBM started a technology sharing partnership/organization with GlobalFoundries, and Samsung well in advance of IBM’s getting out of the chip fab business(Except for IBM’s research fabs). So that IBM would have access to a future competitive chip fab capacity that used the same process. Samsung’s 14nm process is the result of that technology sharing partnership with IBM/others. Why do you think that IBM made Intel cross-license the x86(16/32 bit) ISA to AMD’s and others at the very beginning of the PC market, that lead to what you see today. It sure was not out of any great concern on the part of IBM for Intel, AMD or any of the other x86 cross-licensees, it was Just IBM not wanting any one parts supplier to get IBM behind the 8 ball, or the devil and the deep blue sea. Just look at how two of those lowly(CPU, OS) part suppliers have the PC/laptop OEM industry by the jewels today. Not so for IBM back in the day, at least not until the clones took over.
This is a month old news.
This is a month old news. Google: nvidia pascal tsmc
Rumors from Taiwan are Pascal
Rumors from Taiwan are Pascal will be utterly unbeatable.
Rumors from Taiwan will be
Rumors from Taiwan will be that Pascal still won’t have any fully hardware based asynchronous compute abilities, and more compute will be stripped from the GPUs. That Green Goblin gimping will still continue, and you’ll have to pay top dollar to get it back. Rumor also has it that all the dead scientist’s ghosts are demanding that Nvidia stop using their names on a product so devoid of hardware asynchronous compute ability, and so popular with the brainless gits.