AMD and Samsung Announce Strategic Partnership in Mobile IP

Nobody cares about marketing names. If your phone/GPU is hot and slow that takes precedence and you'll get a reputation from that. There's no indication as of yet that Samsung has any branding obligations, if Qualcomm can buy themselves a branding with minor contracted changes then I sure hope Samsung is able to do the same with an architecture license.

That's just such a stupid comparison on so many levels. One is a minor revenue boost (btw, the stock soared a whopping +0.6% by end of day, whohooo...), while the other one was an existential threat to a company.

 

Are you sure? I only saw Qualcomm mentioned at least as far as SoC's go. Samsung has shown Windows on ARM laptop(s?) but that is (was?) using some Qualcomm chip.

 

Windows on arm only supports Qualcomm as far as I'm away, at the moment anyway.

MS worked with them, bit like they did when they was doing phones.

 

It would be wiser to learn a thing or two about marketing before stating something like this.

No need to be dickish about it.

And here I thought nobody cares about marketing names...

 

Found it:
https://www.sammobile.com/2016/09/0...lks-with-nvidia-and-amd-for-gpu-technologies/

That's dated September 2016. Earlier that year in March, Navi had already appeared in a roadmap for late 2017 / early 2018 stating "scalability" as you mentioned.



At the time (and until very recently) we thought that scalability would mean some kind of multi-chip GPU to scale up. Probably because in 2015-2016 when doing rounds about Zen, AMD also said they would stay away from ULP SoCs in the foreseeable future. It would seem that AMD already had the idea of making Navi more downwards scalable instead, to be able to market it as IP to SoC makers.
At the moment, I can't find any news about AMD stating they were trying to sell GPU IP, but IIRC those appeared around 3 years ago.

 

Believe me, companies when they license IP they don't give a crap about marketing names of that IP. In fact most companies don't even know what the marketing name will be until about a year later after they sign a contract.

If you think that whole story was about marketing names then lol. Apple stopped marketing names 3-4 years prior.

 

It's not about which marketing names are given.
It's about company A being contractually obligated to say their new product X is carrying technology Y from company B, which works as promotion of B's brand.
This is obviously an integral part of the licensing contracts.


According to your previous post, apple is still paying IMG for GPU IP but they've taken away every mention of IMG or PowerVR when promoting the later iphones and ipads.
Also per your previous post, you think Samsung is doing a GPU IP licensing from AMD that in technical terms is similar to apple's current deal with IMG.
If that's the case, the greatest discernible difference is the fact that samsung.com has an official announcement stating that Samsung will be using AMD Radeon graphics technologies, specifically based on the "recently announced, highly-scalable RDNA graphics architecture".

Samsung is cross-promoting with AMD. Apple is not cross-promoting with IMG.
It's a very important difference, despite your personal opinion on the matter.

 

The only difference is Apple is a secretive company while Samsung does Hotchips talks about their IPs. I don't see how this company culture difference has any material effect on either product. You don't sell IPs by brand popularity.

 

Apple has the most valuable brand in the world, so the more they talk publicly about adopting other companies' tech into their products the more they lose in brand value. It's essentially the same reason why you don't see those pesky Intel or Radeon stickers in macbooks, and apple is probably paying a premium for not putting those stickers.

Brand value is always one of the major points when selling whatever to whomever.

 

No thread would be complete without including content from areas everyone loves to hate, analysts.

Here is Jon Peddie Research Publication article on the AMD and Samsung agreement: https://gfxspeak.com/2019/06/03/announce-strategic-partnership/

What do we think?

Lots still to learn. For instance, will this deal kill Samsung Semiconductor’s GPU project?

Samsung is the number one mobile phone supplier using ARM processors and a Mali GPU. ARM, therefore, is one of the losers from this deal. If Samsung Electronics pushes the new SoC into other segments such as automotive, other IP suppliers in those segments could get squeezed out.

It’s a huge win for AMD. You don’t make a lot of money on individual IP sales — you make volume (not a joke), and longevity. The overhead is very low so the margins can be very good. AMD could easily make a couple hundred million a year off this deal once it goes into production. When is that? Hard to say, but assume the design qualifications have been done, and AMD has already given simulators to Samsung, they could start pouring silicon within a year, an oh my oh, does Samsung know how to pour silicon — you may recall they built a few chips for AMD.

 

Brand value is utterly meaningless in business-to-business and is nothing more than an afterthought or footnote. The actual product matters and that's where the money flows. The customer isn't going off buying Exynos chipsets, they'll buy whatever is in the actual end-product and if it performs well. In the case of Apple and IMG if anything the association likely damaged IMG because customers didn't want to be second fiddle to the big Apple (I remember MediaTek talking about this few years ago). In the case of Samsung and AMD if Samsung's Mobile division ever makes mention of Radeon i'll be the first to admit I'm wrong but given the history of the consumer division sabotaging the silicon division for this many years I'd unfortunately be very surprised if this ever happens. Samsung Mobile hasn't made mention of Exynos in years.

 

Personally I don't see the GCN/RDNA brand being relevant for this deal beyond some PR announcements and don't really think it's a very interesting discussion here either.

 

You are correct, thanks for reminding me. So this could potentially be Samsung trying to get a piece of that if they feel Windows on ARM has a chance of taking off.

Thinking long term, this also gives them a possible in for the round of consoles after the next one if MS or Sony decide to ditch x86. MS will have more experience with x86 to ARM recompilation or real time translation, so they'd be the more likely one to potentially pursue that while potentially keeping backwards compatibility. Sony OTOH likely doesn't feel as much of a need to keep BC.

This also give them an opportunity to try to get into the next Nintendo console, which I think someone mentioned previously.

Automotive, however, is likely the next big target as some have mentioned. This potentially allows both Samsung and AMD to position themselves more advantageously against NV, although that's still an uphill battle.

Regards,
SB

 

Even if the first sentence was remotely true, both Samsung and AMD use direct B2C channels for marketing. Neither company is strictly B2B.

History is full of B2B and B2C examples of better products that didn't make it due to unsuccessful marketing, which makes your position even less grounded IMO.

Regardless, I agree that this topic has run its course. Arguments were presented, each one can form their own opinion.

The problem I see with nvidia is that Tegra for mobile applications has apparently hit a dead end in 2016 which was the last time nvidia tried to make a sub-30W SoC.

If Nintendo ever wants a Switch successor using nvidia they either pay up big money to order a semi-custom part (Nintendo paying big money haha..) or they go with 2016's Parker which is hardly a generational upgrade from TX1.

OTOH, nvidia is supposedly offering a "vertical stack" of hardware and dev tools, which is something Samsung can't do AFAIK. Backwards compatibility from nvidia could also be a huge pain in the ass too.

 

Qualcomm WOA soc's has some customizations to help with X86 emulation, what it amounts to can't remember.

Reckon MS won't support any other Arm soc's for a while.
Hopefully won't take as long as I suspect though.

 

If Samsung's recent attempt to design its own GPU flopped too, then the licensing deal with AMD hardly comes as a surprise. What one needs to keep in mind is that both Apple and Samsung want to compete with Qualcomm's GPU IP on every PPA level and so far neither ARM nor IMG with their last designs can hold a candle against the former for high performance ULP SoC GPU cores.

 

I don't think it flopped, AMD's deal was probably just that much better.

 

All-around, GPU performance itself seems to be pretty close between S855 and Exynos 9820, though I heard the 9820 is fairly larger and uses up a larger proportion of its die area for the G76-MP12, whereas the Adreno 620 is much smaller block.
The difference in performance between S855's Adreno 640 and S845's Adreno 630 is less than 20% despite going from TSMC's 10nm to 7nm.
It's possible that Qualcomm did a little bit of sandbagging on Adreno 640 to save die area and drive up the profits.

And they could be doing it for quite a while to be honest. It's not like there's a bunch of games for Android that are demanding faster hardware.

 

IMHO it's less about a sterile performance or sterile die area perspective. Look at the sustained performance ratio of the latest Adreno variants and their power consumption. In the grander scheme of things with such constants it's times easier for Adrenos to scale up higher in performance to something like low end notebooks f.e. and that's a territory where a company like AMD could eventually be a far better horse to bet on.

By the way a place to also have a look at is sustained performance in combination with frequencies on each side (and yes you should also compare to Apple's high end ULP GPU blocks, things aren't much better there either). I don't even recall what the latest Adreno clocks a,t but someone correct me if I'm wrong but I'd be very surprised if it's much above the 700MHz region if at all. Now see where other frequencies are at and think what the performance of the Adreno would look like if they'd clock it by 200MHz higher f.e. which makes it a far better candidate for anything mobile over the tablet format.

If your competing solutions throttle on average by say 30-40% and your own solution at just 10% or less (all freely invented numbers), don't you have quite a bit of frequency headroom in theory to scale both power and performance higher if needed?

As we all know PPA stands for Power, Performance, Area and I severely doubt that it's just a coincidence that power comes in first

 

Sounds like an oxymoron to me. If something is "that" much better how can it not be a flop?