• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

AMD Reports 2016 First Quarter Results + Conference Call [3 semi-customs this year]

El_Chino

Member
Sorry MS fans, it's clearly PS4K, NX and NX Handheld.

;P

tumblr_mouhtc2qdm1s8qbmao1_400.gif
 
R

Rösti

Unconfirmed Member
Semi-custom and gaming the bigger driver for the coming quarters.
 
AMD wouldn't elaborate on whether the new design wins are iterative or would cannibalize existing sales. Exactly the question we were wondering and it got shot down. They want to wait for partners to make their announcements. So definitely sounds like consoles.

Follow up: past deals were negotiated at beginning and ASP were set. No renegotiation expected. As far as VR applies to consoles - they're excited for it in semi-custom space, but nothing to announce. Any new terms would have to be a new console proper. So reading between lines, if any design shrinks for planned, sounds like it was all negotiated up front.
 

Theorry

Member
AMD wouldn't elaborate on whether the new design wins are iterative or would cannibalize existing sales. Exactly the question we were wondering and it got shot down. They want to wait for partners to make their announcements. So definitely sounds like consoles.

Follow up: past deals were negotiated at beginning and ASP were set. No renegotiation expected. As far as VR applies to consoles - they're excited for it in semi-custom space, but nothing to announce. Any new terms would have to be a new console proper. So reading between lines, if any design shrinks for planned, sounds like it was all negotiated up front.

Jup.
 
Summary of forecast questions:

  • Semi-custom: 3Q is always peak. 2Q ramp starts up, but 3Q is peak this year as well. They think segment will grow year over year.
  • Total of 3 design wins lifetime revenue 1.5B, come across next 3-4 years. This is different than current console business. They expect to hit steady state in a few years.
  • Past semi-custom deals were negotiated at beginning and ASP were set. No renegotiation expected. As far as VR applies to consoles - they're excited for it in semi-custom space, but nothing to announce. Any new terms would have to be a new console proper.
  • The semi-custom business doesn't contain any VR content right now.
  • AMD wouldn't elaborate on whether the new design wins are iterative or would cannibalize existing sales. They want to wait for partners to make their announcements.
 

kami_sama

Member
Plot twist: it's everyone else but NX

With almost 100% probability the NX is counted in those 3. No other company can deliver a x86 SoC with enough GPU power for consoles.
Intel's integrated graphics are not enough, and Nvidia (I think) only has ARM SoCs.
 

Dovahking

Member
Summary of forecast questions:

  • Semi-custom: 3Q is always peak. 2Q ramp starts up, but 3Q is peak this year as well. They think segment will grow year over year.
  • Total of 3 design wins lifetime revenue 1.5B, come across next 3-4 years. This is different than current console business. They expect to hit steady state in a few years.
  • Past semi-custom deals were negotiated at beginning and ASP were set. No renegotiation expected. As far as VR applies to consoles - they're excited for it in semi-custom space, but nothing to announce. Any new terms would have to be a new console proper.
  • The semi-custom business doesn't contain any VR content right now.
  • AMD wouldn't elaborate on whether the new design wins are iterative or would cannibalize existing sales. They want to wait for partners to make their announcements.

Welp.
 
PS4K, XBox 1.5 2, NX

Would a straight shrinking of the PS4's APU to 14nm from 28 nm be considered a custom design?

No. It means at least some component of the APU changed. Sometimes, it's because of other constraints as in the XBox 360 slim. However, given there is nothing too exotic in XBox One or PS4 aside from XBox One's ESRAM, it'd be fair to say they are both getting a revised version with better specs.
 

AP90

Member
PS4K, XBox 1.5 2, NX



No. It means at least some component of the APU changed. Sometimes, it's because of other constraints as in the XBox 360 slim. However, given there is nothing too exotic in XBox One or PS4 aside from XBox One's ESRAM, it'd be fair to say the third could be an XBO Slim.they are both getting a revised version with better specs.

My thoughts for now.. So if the potential Xbox successor is ramping up production in 2017, that means release 2018?

So would that mean Polaris with GDDR5X and not hbm1 or hbm2?
 

The_Lump

Banned
My thoughts now.. So if the potential Xbox successor is ramping up production in 2017, that means release 2018?

So would that mean Polaris with GDDR5X and not hbm1 or hbm2?

Not necessarily. These things won't be sat around in boxes for long. Plus it seems to be talking about early 2017 on the slide.
 
PS4K, XBox 1.5 2, NX



No. It means at least some component of the APU changed. Sometimes, it's because of other constraints as in the XBox 360 slim. However, given there is nothing too exotic in XBox One or PS4 aside from XBox One's ESRAM, it'd be fair to say they are both getting a revised version with better specs.

I would expect a shrink to be par of the original negotiations. They seemed to indicate that by stating how the negotiations are for the life of the deal. Naturally, any console maker would want to guarantee shrinks so they could lower their own console ASP down the road.
 

AP90

Member
Not necessarily. These things won't be sat around in boxes for long. Plus it seems to be talking about early 2017 on the slide.

So building up stockpiles for release day and next few shipments after release does not take long anymore?
 

TSM

Member
The high performance PC pillar descriptions seem a bit anemic. I expect the performance per watt metric to have diminishing sales returns outside of laptops. We are reaching the point where people will be able to put a decently powerful video card in a computer with a 300 watt power supply.

The other three bullet points are:
-People will buy our products for the obvious reasons
-Our products are comparatively less shitty than they have been in 10 years
-We'll sell more stuff this year. We promise.

I was hoping for them to be much more competitive to keep Intel and Nvidia honest.
 

No_Style

Member
So would that mean Polaris with GDDR5X and not hbm1 or hbm2?

My bet is on GDDR5X because I'm assuming they'll continue down the Jaguar architecture path as well. HBM1 has a 4GB limit and HBM2 will require memory controller changes to interface with Jaguar. There's also manufacturing cost considerations between HBM2 and GDDR5X.

The high performance PC pillar descriptions seem a bit anemic. I expect the performance per watt metric to have diminishing sales returns outside of laptops. We are reaching the point where people will be able to put a decently powerful video card in a computer with a 300 watt power supply.

The race for performance per watt will pay off if they can get themselves into embedded systems (think autonomous cars) and high-performance computing sector. If you can squeeze more computing power out of the same electrical and thermal conditions as your competitor, you're going to find yourselves customers.
 

Yoday

Member
So building up stockpiles for release day and next few shipments after release does not take long anymore?
Consider also that an upgraded system launch isn't going to necessarily need the level of stock that a console at the start of a generation would need. A good amount of consoles sold are still going to be the lower priced models that already exist. They could easily launch with a more limited supply while ramping up next year.
 

tuxfool

Banned
The high performance PC pillar descriptions seem a bit anemic. I expect the performance per watt metric to have diminishing sales returns outside of laptops. We are reaching the point where people will be able to put a decently powerful video card in a computer with a 300 watt power supply.

You need to increase performance/watt in order to keep your TDP under control in larger dies with a lot more transistors. Otherwise you have to end up resorting to expensive cooling solutions.

If your 28nm GPU with 6000M transistors has a TDP of 300W then using, say, double the transistors at 14nm without caring about TDP leads to a device with a TDP of 600W (running at full load).
 

nib95

Banned
I'm going with NX, Neo, and One 1.5.

Exciting times ahead :)

This on top of PSVR, Vive and Occulus? The fuck. I need to come up with some good excuses for the wife, and make more money lol. Adding to the gaming expenses, I just recently bought a new racing wheel and an Xbox Elite controller. Oh dear….
 

blu

Wants the largest console games publisher to avoid Nintendo's platforms.
With almost 100% probability the NX is counted in those 3. No other company can deliver a x86 SoC with enough GPU power for consoles.
Intel's integrated graphics are not enough, and Nvidia (I think) only has ARM SoCs.
That surely is their main problem stopping NV from console wins /rolleyes

I would expect a shrink to be par of the original negotiations. They seemed to indicate that by stating how the negotiations are for the life of the deal. Naturally, any console maker would want to guarantee shrinks so they could lower their own console ASP down the road.
Exactly. Shrinks are never announced as design wins, as the former are negotiated with their original design wins. And essentially, a design win is your part going into a new product, and a shrink is not such an instance, as you already have your part in the product, to boot.
 

TSM

Member
You need to increase performance/watt in order to keep your TDP under control in larger dies with a lot more transistors. Otherwise you have to end up resorting to expensive cooling solutions.

I meant specifically as a sales tool. Performance per watt is obviously important for progress, but once a certain threshold is reached most consumers will have far less of an interest. AMD's video card's reputation for generating heat like a small space heater was a big part of what put them in the position they are in. Once cards run cool and quiet people will just take it for granted.
 
If your 28nm GPU with 6000M transistors has a TDP of 300W then using, say, double the transistors at 14nm without caring about TDP leads to a device with a TDP of 600W (running at full load).

No. 14nm FinFETs are much more power efficient. You can assume that power density per unit area will roughly stay the same.
 
I would expect a shrink to be par of the original negotiations. They seemed to indicate that by stating how the negotiations are for the life of the deal. Naturally, any console maker would want to guarantee shrinks so they could lower their own console ASP down the road.

Die shrinks are implicitly part of any original negotiations as AMD just sells a design. The manufacturers deal with the die shrinks of that architecture themselves.

New custom design wins means AMD had to create a new design. Now, it doesn't necessarily mean higher specced revisions. However, if I was a console manufacturer, I'd like to get the most value for my money especially if I expect a platform to have a long life-span.
 

teh_pwn

"Saturated fat causes heart disease as much as Brawndo is what plants crave."
I certainly hope the +40% IPC holds true and they can keep clocks 4.5+GHz. They'd actually be competitive with Intel's core microarchitecture. AMD hasn't matched/beaten Intel since Netburst. Though I feel like that advantage would be short lived as I've heard rumors a new microarchitecture that's not entirely silicon is coming from Intel in a few years.

I doubt they'd get server marketshare because power consumption. But they could get some desktop marketshare back.
 

tuxfool

Banned
I meant specifically as a sales tool. Performance per watt is obviously important for progress, but once a certain threshold is reached most consumers will have far less of an interest.
This is a shareholder document. All other slides are also more technical in nature.


No. 14nm FinFETs are much more power efficient. You can assume that power density per unit area will roughly stay the same.

Of course. I'm just highlighting the fact that Perf/Watt is an important consideration. You get some/most of that for free in the die shrink, but with smaller transistors you also get more leakage, so you do also need to look at architectural solutions to manage power.
 
Die shrinks are implicitly part of any original negotiations as AMD just sells a design. The manufacturers deal with the die shrinks of that architecture themselves.

Who are you calling the manufacturer here? That same semi-custom team would be responsible for shrinking the design, too. They'd have to synthesize the design with new transistor libraries and re-run all their worst case analyses to make sure the design closes. It's not just pushing a button.

Of course. I'm just highlighting the fact that Perf/Watt is an important consideration. You get some/most of that for free in the die shrink, but with smaller transistors you also get more leakage, so you do also need to look at architectural solutions to manage power.

There are a lot of ways to mitigate leakage. FinFET by design helps. Lower gate voltages (which FinFET enables) help. Advances in doping help.
 

TSM

Member
I certainly hope the +40% IPC holds true and they can keep clocks 4.5+GHz. They'd actually be competitive with Intel's core microarchitecture. AMD hasn't matched/beaten Intel since Netburst. Though I feel like that advantage would be short lived as I've heard rumors a new microarchitecture that's not entirely silicon is coming from Intel in a few years.

This is where the conspiracy theorists will see if they were right. Intel will either suddenly come up with a CPU design that gains much more than the usual 10% per iteration for once, or we'll get the usual 10% and people will have to face the fact that we have finally reached diminish returns on silicon.
 
Who are you calling the manufacturer here? That same semi-custom team would be responsible for shrinking the design, too. They'd have to synthesize the design with new transistor libraries and re-run all their worst case analyses to make sure the design closes. It's not just pushing a button.

The manufacturer would typically be TSMC, FoxConn, etc depending on who Sony,. Microsoft, and Nintendo partner with to actually build the APU.
 
Is the 'taking semi cutom business beyond gaming' still a thing for this year or was that a misquote?

That surely is their main problem stopping NV from console wins /rolleyes


Exactly. Shrinks are never announced as design wins, as the former are negotiated with their original design wins. And essentially, a design win is your part going into a new product, and a shrink is not such an instance, as you already have your part in the product, to boot.

I've been wondering about product binning, do you think it would make sense or be possible to create regular ps4 14nm chips from substandard ps4k chips if low yields are going to be a problem? I feel they could be killing two birds with one stone here but I am clueless on the technology.
 
The manufacturer would typically be TSMC, FoxConn, etc depending on who Sony,. Microsoft, and Nintendo partner with to actually build the APU.

Ok, in that case, it's not true. TSMC is not responsible for synthesizing the design in a smaller process. That onus is on AMD. TSMC doesn't have the necessary expertise, nor would AMD/Sony/Microsoft want to share more design details than necessary to a foundry partner.
 
Ok, in that case, it's not true. TSMC is not responsible for synthesizing the design in a smaller process. That onus is on AMD. TSMC doesn't have the necessary expertise, nor would AMD/Sony/Microsoft want to share more design details than necessary to a foundry partner.

Thinking about you are correct. However, the fact remains AMD would not announce something that was agreed upon in a previous contract as a design win to investors. As it would be fraud to report it as such since it gives the impression of another revenue stream.
 
So it sounds like one of these semi-custom chips isn't gaming related, and at least one won't see the light of day in 2016. I'd say NX is guaranteed 2016, and from what we've heard so far, PS4K would be the other one releasing maybe spring 2017 latest. This could leave MS with no new console until late 2017 earliest.

There are rumors AMD is designing semi-custom x86 SoCs for Apple's iMac line for 2017 and 2018: http://wccftech.com/amd-making-custo...acs-2017-2018/

I could totally see Apple just handing out a big pile of money for AMD to manufacture a semi-custom Zen/HBM/Navi SoC for the next "revolutionary" OS X device.
 

MuchoMalo

Banned
Sorry MS fans, it's clearly PS4K, NX and NX Handheld.

;P

I'm going with NX, Neo, and One 1.5.

Exciting times ahead :)

Yeah, about that... There have been multiple rumors of Apple having a design win. If that's the case, it means one of two things:

A.) PS NEO and/or XBox 10 don't/doesn't actually count as (a) new design win(s) because they are just new and updated versions of existing products and don't have any architectural changes (or MS really doesn't have an upgraded XBox planned), or
B.) Nintendo does not have a design win with AMD.

The former is more likely because one of the wins is ARM-based and Apple is going to use an x86 APU based on the rumors.

Here's the funny thing, though. If "A" applies to both Sony and Microsoft's new hardware, it would mean that either there's another company with a deign win that we don't know about, or that Nintendo has two design wins: one x86 and one ARM. That would be rather interesting, wouldn't it?


Here's what doesn't make sense though: AMD seemingly has very low revenue expectations out of two of those chips: $1 billion over the course of 3 years. (This was brought up in one of the NX threads.) There's something funny about that, and Kitguru had an explanation for why:

KitGuru Says: In fact, even combined total lifetime revenue of approximately $1 billion over approximately three years for two chips indicate that both products are rather basic and inexpensive. It is unlikely that they will power devices that are sold in very high volumes (i.e., tens of millions of units).

So, that makes it even less likely that any of these are referring to the upgraded twins unless this is a case of extreme lowballing. It would also be a huge sign that NX is vastly underpowered, again unless AMD is playing it very safe with these numbers.


I wish AMD would say if the chips use Polaris or not. Then things would really get interesting!

The two 2014 semicustom contract wins were rumored for Nintendo and Apple.The mid 2015 win was rumored for Sony.

Oh, and if this is true it confirms ARM for NX and possibly that the Ori dev was right that we're only getting a handheld, though it's possible that Nintendo's win is for two different but similar chips.
 
Yeah, about that... There have been multiple rumors of Apple having a design win. If that's the case, it means one of two things:

A.) PS NEO and/or XBox 10 don't/doesn't actually count as (a) new design win(s) because they are just new and updated versions of existing products and don't have any architectural changes, or

The CPU in Neo would not be a design win. However, as the new GPU contains a higher CU count that does not exist in the generation of GPUs from which the PS4 GPU that would definitely be a new design and not counted in the original contract for PS4 APU.
 
Top Bottom