• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

News [CNBC] AMD, Samsung partner on mobile graphics tech

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Very interesting point of view
ethernity said:
Samsung is pushing hard in the Automtive industry. One of the more common complaints however was the GPU on their SoCs is a bit behind when compared to the NVIDIA Xavier and Parker.
You have to consider that in Automotive the power envelope while generally not so high as desktop is much higher than smartphone and tablets.
One of the things that immediately comes to my mind is the integration of compute oriented architecture of AMD's Radeon in Samsung's Exynos Auto.
The requirement for compute power in Automotive is growing exponentially due to ADAS and Autonomous vehicles running a wide variety of inferencing and DNN algorithms and NVIDIA is quite strong in this area with their Tegra SoCs.
And it would not make sense for Samsung to license it from the competitor NVIDIA who is making Tegra SoCs for the GPU.
Couple this to the fact that AMD has a strong open source drivers and a budding compute stack it could be something to watch out.
You can guess who is the only closed source GPU vendor. Intel has massive open source components and DNN/compute stack as well.
Click to expand...
https://forum.beyond3d.com/posts/2071172/
 
RaV666 said:
You seem to be lost in the fact that samsung will use their architecture ,not just bought some patents to be free of litigation.
Also you seem to not get the fact that GPU in mobile sector is a pretty crowdy place, its not like they dont have a choice, the same goes for consoles.Samsung wouldnt buy it if it wasnt good for them.Neither would microsoft/sony and google.
And this is a big deal for amd no matter how much samsung pays them, they put their architecture in a new market, more developers will optimize for it.
Click to expand...
And if its Automotive market - its one of most growing markets out there. Having Samsung as partner wil open software ecosystem, which will attract even more partners for their GPU Semi-Custom business.
 
Hmm, hasn't Samsung been developing their own mobile GPU tech for something like 7 years? Did their homegrown solution just suck?
 
Glo. said:
Radeon is not profitable? You do realize that you are talking about IP that brings AMD to consoles?

There is much more than being profitable for AMD to keep RTG. Oh, and it is not RTG anymore. Its once again one company. Its AMD.

The reason why AMD wins, is because it is the most willing company to work with other companies. Its because they have Semi-Custom division. Its because they are the only one company who can bring high performance CPUs and GPUs to ANY market.

And yes. It means that AMD's IP is way better than perception there is, about them. Samsung did not bought IP. Samsung bought ARCHITECTURES! Which AMD will deliver.
Click to expand...
This is what I was thinking. How can you, at this point separate the two 'divisions', CPU and GPU, when they appear so intertwined. Real Siamese twins sharing most internal organs.
 
I'm pretty sure this is primarily a prelude to AMD moving from TSMC to Samsung. If Samsung's process can support the architecture... then everything will be at Samsung in the future.
 
Hitman928 said:
Couldn't AMD already use Samsung? I don't think there's anything stopping them if they wanted to.
Click to expand...

Samsung is trying to take some marketshare from TSMC so not only AMD, they will also try to get Nvidia for them to use their 7nm.
 
RIP ARM Mali.... With that, there are no longer a big exponent for their tech besides Mediatek.

Samsung will go AMD.
Huawei will go Power VR

and who will take Mali then?
 
guachi said:
"You get to use our IP, we get to use your factories" wouldn't be the worst trade AMD could make.
Click to expand...
Nah. Like @Hitman928 pointed out, Samsung would be happy to have AMD's money. No 'special' deal would need to be made.
Unless someone has read that AMD is somehow unhappy with TSMC (seems doubtful), I don't see why AMD would change foundries.
 
Ajay said:
Nah. Like @Hitman928 pointed out, Samsung would be happy to have AMD's money. No 'special' deal would need to be made.
Unless someone has read that AMD is somehow unhappy with TSMC (seems doubtful), I don't see why AMD would change foundries.
Click to expand...
AMD could use both foundries. That'll be beneficial to them from both deal and volume point of view, as long as Samsung's foundries are upto mark.
 
prtskg said:
AMD could use both foundries. That'll be beneficial to them from both deal and volume point of view, as long as Samsung's foundries are upto mark.
Click to expand...
IF AMD *needs* the volume. Otherwise, why spend the design $$s developing for a new PDK.
 
Ajay said:
IF AMD *needs* the volume. Otherwise, why spend the design $$s developing for a new PDK.
Click to expand...

Well obviously you would limit 1 "chip" to 1 foundry. But if you can make gpus at one and CPUs at the other for higher volume that would be a win.

Given the leaks of Navi pricing if they come true it's a clear hint AMD is supply constraint.
 
beginner99 said:
Well obviously you would limit 1 "chip" to 1 foundry. But if you can make gpus at one and CPUs at the other for higher volume that would be a win.
Click to expand...
I had thought about that, but Zen2 processors are more likely to have "break out" volume than Navi, IMHO.
 
Pretty sure AMD is already working together with Samsung Semiconductor if only as a fail safe whenever TSMC doesn't deliver. This is (should be) just basic prudence, something AMD already showcased with GloFo's 7nm.

May well be the case that that contact led to Samsung licensing the RDNA IP.
 
Yeah. AMD actually fabbed some chips at Samsung when GlobalFoundry couldn't keep up on 14nm (think that was earlier on, and the leadup to Zen 1 release). So they already had some relationship. I don't think there was anything preventing them from working again other than timing.

The main reason why AMD didn't use Samsung for 7nm is that Samsung chose to go EUV with 7nm so AMD couldn't just migrate what they had intended for GF there, it basically had to be TSMC or would be delayed for probably a year while Samsung got their 7nm up and running (imagine if they were just now sampling Vega 20, and then Navi was pushed to probably mid 2020). AMD definitely was not going to start Zen 2 on Samsung 7nm for both the delay, and because EUV production is likely still limited by the amount of EUV equipment, meaning that even if AMD wanted to sell more Zen 2 chips, they couldn't because the production would be limited. And even though it might have been possible to produce Zen 2 both at Samsung and TSMC since they both have 7nm EUV process, it still would have been late (plus the extra learning curve of EUV in general) would mean Zen 2 wouldn't be til 2020.

Now, it will be interesting to see where things go from here. But the thing is, TSMC is still ahead so I kinda don't see AMD doing much at Samsung. TSMC has 7nm, 7+ with EUV, 6 (which is just 7nm improved, making easy transition), and then already starting on 5+ (7+ EUV improved so offering another easy transition).

Not sure how long AMD is gonna stick out with GF (seems like they're planning on fulfilling their wafer deal, so I could see them do their next I/O at 12nm with GF), so there's potential options after that. Samsung has 10nm, 8nm, so there's definitely options, but its not likely going to be for AMD's cutting edge chips.

Which, AMD has plenty of ways they could go about things. They could split consumer and EPYC Zen die (where consumer stays on cheaper processes, like say Zen 3 EPYC goes to 7nm EUV, while consumer just goes to 6nm). They could do the I/O chip (or make different I/O chips for different market segments) different places. They could do multiple APUs (i.e. the small monolithic ones would lag behind on cheaper processes; then they could have higher end ones that use slightly behind the curve CPU die - so say Zen 3 moves to 6nm or 7 EUV next year, but their APUs mixing CPU and GPU die next year would use the current Zen 2 7nm die; likewise for the GPUs where they'd pair it with 7nm GPU die, but the dGPUs would be on 6nm or 7EUV).
 
Last edited:
Can anyone speculate why Samsung didn't buy Imagination Technologies rather than have this deal with AMD? Was PowerVr not good enough for gpu heavy stuff like VR?
 
Yeah, its really weird. Why go through and setup a launch less than 2 weeks ago for yesterday, if they knew it was not going to be unveiled? Its pretty normal for sales to disregard engineering, but this seems like it takes this to a new level.
 
They canned it coz it would have gotten more bad press than good. Looks like RDNA2 doesn't work for mobile. How could it? Their silicon is not power optimized for handhelds. Samsung should have gone with Tegra something.

But if I were at Samsung, I would just release it as a budget option, instead of discarding all the money and effort spent on the thing.
 
igor_kavinski said:
They canned it coz it would have gotten more bad press than good. Looks like RDNA2 doesn't work for mobile. How could it? Their silicon is not power optimized for handhelds. Samsung should have gone with Tegra something.

But if I were at Samsung, I would just release it as a budget option, instead of discarding all the money and effort spent on the thing.
Click to expand...

What are you basing this off? There have been zero indicators that RDNA2 won't scale down.
 
igor_kavinski said:
Their lowest desktop option, RX 6400, consumes 53W. Compare that to the best mobile GPUs sipping max ~6W.
Click to expand...

By that logic, then AMD's own RDNA2 based APUs should be an absolute failure, but that doesn't seem to be the case at all. . .

Edit: BTW, that 53W is typical board power, so it includes the RAM, VRMs, etc. needed for the whole card. It also has more CUs and higher frequency than I would expect a mobile GPU to have, so I don't see any reason it couldn't scale down, just like they can fit an RDNA2 GPU into a 15W APU (which includes the full SOC) and have class leading performance.
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top