Samsung wants to use AMD or Nvidia GPUs in its Exynos chipsets, plans to ditch Qualcomm for good in 2018

[Coaxialgamer]

Quote

According to a new "exclusive" report from behind the scenes, Samsung is in talks with both AMD and Nvidia for GPU tech. Namely, the Korean company wants to stop using the ARM-developed Mali GPUs in its Exynos chipset designs, and instead use ones created by AMD or Nvidia.

At the moment, the latter has an upper hand apparently, thanks to its "superior" Pascal architecture, but AMD can't be ruled out just yet since its Polaris architecture is powering the PS4 Pro.

 

Quote

Past rumors said that in 2017 we'll see an Exynos chipset with a Mali GPU still, but employing Heterogeneous System Architecture (HSA). This will let the CPU and GPU sit on the same bus and share memory and tasks, which should improve overall performance. Then, in 2018 we might see either a Samsung-developed GPU or one licensed from AMD or Nvidia if these talks pan out.

In parallel, the Korean company is slowly developing its CDMA modem tech, aiming to start testing support for that mobile network flavor in September 2017. If all goes well, then expect to see only one Galaxy S9 model in early 2018, powered by an Exynos chip - no more Snapdragon version for certain markets. With this move, Samsung will essentially ditch Qualcomm for good in the high-end mobile device space.

 

This is interesting,  but it raises quite few questions. If nvidia gets  the deal,  how will the put pascal on Samsung's 14nm ? 

 Will they  ( nvidia or amd)  use a completely different architecture? 

Always ,  however  does ploaris/pascal scale when restricted to a ultra low power consumption? 

 

 

Source :http://m.gsmarena.com/samsung_wants_to_use_amd_or_nvidia_gpus_in_its_exynos_chipsets_plans_to_ditch_qualcomm_for_good_in_2-news-20441.php

[Sakkura]

AMD may have an advantage if Samsung wants to bet on HSA, otherwise Nvidia looks like the better option.

[Coaxialgamer]

Just now, Sakkura said:

AMD may have an advantage if Samsung wants to bet on HSA, otherwise Nvidia looks like the better option.

Pascal does have the power advantage at 100w+.  We don't really know how either of those architecture will perform at ultra low power

[Sauron]

In response to your questions, first of all we must consider this will happen in late 2017/early 2018, so nvidia will most likely have moved on from Pascal by then. Besides, it is technically not mandatory to have the same production node on both the cpu and gpu as they are separate entities. Of course, it would make production harder, but maybe the r&d savings would be worth it - we can't know for sure without inside sources. Either way, the transistors used in gpus can be fundamentally different regardless of size. Remember, nVidia has manufactured their own SoCs in the past and have used kepler gpus in them, too.

[laminutederire]

5 minutes ago, Sauron said:

Remember, nVidia has manufactured their own SoCs in the past and have used kepler gpus in them, too.

Ahah considering this Samsung will go with AMD

[Misanthrope]

Hey Samsung. After having your recent flagship literally blow up you don't want one of them toasty AMD chips in there.

[Space Reptile]

i mean samsung already makes stuff for amd , so 2 bird one cannonshot really when samsung goes w/ amd  

[Bananasplit_00]

16 minutes ago, Misanthrope said:

Hey Samsung. After having your recent flagship literally blow up you don't want one of them toasty AMD chips in there.

nah im sure they will just throw a 480 at it  then your phone can be your stove too!

[Agosto]

Well, Polaris should scale down quite in terms of power

However, 2018 means Navi, not Polaris

[Sakkura]

2 hours ago, Coaxialgamer said:

Pascal does have the power advantage at 100w+.  We don't really know how either of those architecture will perform at ultra low power

The differences between Maxwell and Pascal are not that big, and we know how Maxwell does at low power. In any case, the 2016 architectures may not be directly relevant to the SOCs they want to launch in 2018.

[captain cactus]

2 hours ago, Bananasplit_00 said:

nah im sure they will just throw a 480 at it  then your phone can be your stove too!

Which one, the RX480 or GTX480?

[8BitBuilder]

2 hours ago, Sauron said:

In response to your questions, first of all we must consider this will happen in late 2017/early 2018, so nvidia will most likely have moved on from Pascal by then. Besides, it is technically not mandatory to have the same production node on both the cpu and gpu as they are separate entities. Of course, it would make production harder, but maybe the r&d savings would be worth it - we can't know for sure without inside sources. Either way, the transistors used in gpus can be fundamentally different regardless of size. Remember, nVidia has manufactured their own SoCs in the past and have used kepler gpus in them, too.

They will still be at pascal did you see how long we had to wait for the 1080 from the 980

[zMeul]

AMD ?! AMD doesn't have the know how to build ultra low power chips anymore, ATi sold their Imageon division to Qualcomm - the irony

on the other hand, nVidia has experience with Tegra - they even build Pascal based Tegra SoCs 

[Sauron]

12 minutes ago, 8BitBuilder said:

They will still be at pascal did you see how long we had to wait for the 1080 from the 980

Yeah, but pascal seems to be more like a maxwell die shrink than a completely new architecture. I think they were preparing their new stuff and saw it was taking too long, so they released pascal to keep people buying while they are finishing the rest. Either way, as I said an SoC doesn't necessarily have to be all on the same production node.

[Trixanity]

1 hour ago, zMeul said:

AMD ?! AMD doesn't have the know how to build ultra low power chips anymore, ATi sold their Imageon division to Qualcomm - the irony

on the other hand, nVidia has experience with Tegra - they even build Pascal based Tegra SoCs 

Nvidia who had to exit the market because their chips used too much power to comfortably fit in a phone? 

 

They both suck if that's what you're getting at. 

 

Regardless, we're talking years from now and it is possible to scale either architecture down. It's just a matter of if performance is intact, which remains to be seen. Also, it sounds more like Samsung is looking for a license to make their own rather than a semi custom collaboration. So pipe down, little buddy. 

[Doobeedoo]

Seeing this being way later in date, I remembered Navi architecture with 'scalability' and 'next gen memory' traits. So Samaung may be targeting that.

 

Oh yeah ATI Imageon thing that happened too though.

 

P.s.

Exynos is a great chip, though I always wonder how come they're not going with Apple way, 2 really strong big cores, and now 2 extra power saving cores in big.LITTLE A10 SoC bit odd. 

I can't find much info about their core design.

[Prysin]

5 hours ago, Sakkura said:

AMD may have an advantage if Samsung wants to bet on HSA, otherwise Nvidia looks like the better option.

however AMD sold their mobile GPU IP to Qualcomm years ago... Adreno is just a anagram for Radeon after all.

So im not sure AMD has any GPUs for the mobile sector that they can sell to Samsung atm due to their previous IP being sold.

[Sakkura]

9 minutes ago, Prysin said:

however AMD sold their mobile GPU IP to Qualcomm years ago... Adreno is just a anagram for Radeon after all.

So im not sure AMD has any GPUs for the mobile sector that they can sell to Samsung atm due to their previous IP being sold.

Qualcomm only got their old Imageon architecture, AMD can still get back in that business with their regular GPU architecture. They just don't have the money or any reason to do that on their own, but licensing to Samsung could definitely be interesting for AMD.

[mrchow19910319]

3 hours ago, lots of unexplainable lag said:

Which one, the RX480 or GTX480?

Both.

Kappa

[Trixanity]

There is absolutely no reason for AMD to try to make a mobile SoC on their own or for Samsung. They lack the resources and IP to do it. It's also a cutthroat business that even Intel couldn't pull off. 

 

However it's very likely that they'll license directly to Samsung or get a contract to design a GPU for them. I find it a bit odd with the timing though. ARM's Mali chips seems to be blossoming with the upcoming G71. But perhaps they think the two big players can do better than what's on the Mali roadmap. 

[Bananasplit_00]

9 hours ago, lots of unexplainable lag said:

Which one, the RX480 or GTX480?

lol GTX480 ofc, how else are you going to be cooking bacon with your phone?

[Tomsen]

Lets all come down for a second.

They are interested in HSA? Well, noone would be better to deliver fully functional HSA system than AMD.

Why is people fixated with radeon or paxwell? Doubt those design scale down to the TDP/power-consumption area that samsung targets.

[Tomsen]

5 hours ago, Trixanity said:

There is absolutely no reason for AMD to try to make a mobile SoC on their own or for Samsung. They lack the resources and IP to do it. It's also a cutthroat business that even Intel couldn't pull off. 

 

However it's very likely that they'll license directly to Samsung or get a contract to design a GPU for them. I find it a bit odd with the timing though. ARM's Mali chips seems to be blossoming with the upcoming G71. But perhaps they think the two big players can do better than what's on the Mali roadmap. 

It would be stupid, since samsung is well on making their own. If AMD were to design the SoC, samsung would help with resources and IP (just like sony and microsoft). Also AMD got plenty of IP that are of relevance. Intel isn't really in semi-custom business at all, so they wouldn't pull that off anyway.

 

Or they just want their own GPU team. Instead of having to reinvent everything, they buy out a design from AMD/Nvidia/whoever, and build upon that.

[maulemall]

AMD is really going to go after this I bet..

Their new facilities they are building in India are going to need to be put to work.

The question you need to ask is how much higher will thier stock go. 

[Curufinwe_wins]

Just letting everyone know, it isn't a signficant design challenge to migrate Pascal from TSMC's 16nm to Samsungs 14nm process... I mean crap... Both AMD and Apple have in recent memory dual sourced from both without significant R&D overhead.

 

Anyways, this is a good idea for Samsung right now since mobile GPU's are starting to move into the more desktop featureset space (with things like tesselation showing up in the most modern mobile gpu's.)

 

It would probably be a large design win for whoever gets the contract, so I will look forward to hearing more about this in the future. 

 

I wouldn't count AMD out on undercutting Nvidia guys, they did that with the last two generations of consoles and with Apple's product lineup.