The AMD Execution Thread [2007 - 2017]

[Kaotik]

Was it that hard to compete with BayTrail, which is found on thousands upon thousands of tablets, including high-end ones?

Considering Intel sold them for less than it cost Intel to make them, yes, it probably was that hard. It's hard to compete when someone sells their goods on loss to gain market share

 

[swaaye]

AMD tried to do it to NV back in the RV770 days.

Maybe there's more to the lack of AMD tablets than meets the eye. Other companies (like NV) are obviously selling tablet SOCs with Intel's scheme in play.

 

[Kaotik]

AMD tried to do it to NV back in the RV770 days.

Maybe there's more to the lack of AMD tablets than meets the eye. Other companies (like NV) are obviously selling tablet SOCs with Intel's scheme in play.

All the rest are ARM, while Android works on AMD x86, too, to my understanding Intel has done quite big efforts to optimize it specifically for their chips with Google.
As for Windows, well, there really is only Intel and AMD after Windows RT got the bullet

edit:
Also, did they really? They sold them cheaper and smaller margins for sure, but for loss?

 

[Rikimaru]

AMD tried to do it to NV back in the RV770 days.

Maybe there's more to the lack of AMD tablets than meets the eye. Other companies (like NV) are obviously selling tablet SOCs with Intel's scheme in play.

RV770 was a good chip with right size. NV equivalent was harvested from large chip. That's why NV chip was much more expensive.

Read this interview http://www.anandtech.com/show/2679

 

[Alexko]

All the rest are ARM, while Android works on AMD x86, too, to my understanding Intel has done quite big efforts to optimize it specifically for their chips with Google.
As for Windows, well, there really is only Intel and AMD after Windows RT got the bullet

edit:
Also, did they really? They sold them cheaper and smaller margins for sure, but for loss?

At one point, they were basically giving them away for free.

 

[Deleted member 13524]

Considering Intel sold them for less than it cost Intel to make them, yes, it probably was that hard. It's hard to compete when someone sells their goods on loss to gain market share

I understand how that makes sense for $80-200 cheap chinese tablets. But for higher-end and premium tablets with larger margins like the Asus T100 Chi, the Lenovo Yoga 2 Pro and Thinkpad, HP Elitepad G2 and others (which came out ~6 months after Mullins' release), one would think the OEMs would have an interest in showing an edge in performance at the cost of a dozen bucks, at least for marketing's sake.

And now, even the chinese brands are starting to sell more expensive tablets on the $400-500 range and they're using Core M, so it's not like they're afraid of making pricier tablets.
Mullins would've been a great mid-term within the large price difference between BayTrail and Broadwell but still, no sign of it in any end product with.

Literally no one would touch Mullins.
It makes me think that either the chip was terribly flawed (over-heating, under-performing, etc.) or there seems to exist this weird industry-wide blockade against AMD chips among manufacturers.

And now Carrizo-L will be like the 3rd iteration of a seemingly promising tablet chip and yet again, zero signs of products using it. (EDIT: Carrizo-L is not a replacement for Mullins, but for the 10-15W Beema. My bad.)

Being a salesman for AMD within OEMs must be on the top 10 of most frustrating jobs in the world. Right next to warm soup vendors at the beach during a hot summer and selling orthopaedic boot soles in a convention for wheelchair-bound people.

 

[kalelovil]

At one point, they were basically giving them away for free.

And yet the HD 4xxx allowed AMD's graphics business unit to move from a loss to a profit. I don't see how they could have been selling them at a loss.
The RV770 was only slightly larger than the G92b the HD 4850 was priced against, and was significantly smaller than the GT200(b) the HD 4870/4890(RV790) was priced slightly under.

 

[silent_guy]

It's the first time that I hear RV770 being sold for cheap. Not saying it's not true, I just don't remember it being bad or uncompetitive.

 

[Alexko]

And yet the HD 4xxx allowed AMD's graphics business unit to move from a loss to a profit. I don't see how they could have been selling them at a loss.
The RV770 was only slightly larger than the G92b the HD 4850 was priced against, and was significantly smaller than the GT200(b) the HD 4870/4890(RV790) was priced slightly under.

Sorry, I was talking about Intel's tablet chips, not AMD's GPUs, as I thought that was Kaotik's question. I've just re-read his post and he was probably talking about AMD's GPUs indeed. RV770 was probably sold with very reasonable margins.

 

[Grall]

Can we make up a new rule that says it's forbidden to use GPU codenames from products half a decade or more into the past ty kkthxbai.

 

[Lightman]

Can we make up a new rule that says it's forbidden to use GPU codenames from products half a decade or more into the past ty kkthxbai.

So how we are going to discuss chips like R200 or R100 without confusing them with more modern Radeons 7000 or 8000? People are going to ask what do you mean Radeon 8500 supported pixel shader 1.3 if everyone knows it's 5.0?

 

[swaaye]

All the rest are ARM, while Android works on AMD x86, too, to my understanding Intel has done quite big efforts to optimize it specifically for their chips with Google.
As for Windows, well, there really is only Intel and AMD after Windows RT got the bullet

AMD could have Android tablets too. I suppose they do sell those Mullins/whatever chips in various ultra low end machines though. Tegra can't do that without x86 compatibility.

 

[silent_guy]

Can we make up a new rule that says it's forbidden to use GPU codenames from products half a decade or more into the past ty kkthxbai.

Objection! I know RV770, but I have no idea what product it ended up in...

 

[Grall]

I know RV770, but I have no idea what product it ended up in...

Radeon 4870 and similar cards, IIRC. Anyway, old codenames are hard to keep track of. Hell, I find NEW codenames hard to keep track of. In essence, codenames suck.

 

[UniversalTruth]

RV770 and RV790 were sold for cheap. Actually, they were perhaps the best value cards for quite long time. Even now and in recent times nothing can beat this.

They ended in cards like 4830, 4850, 4870, 4890, 4870X2.

...old codenames are hard to keep track of. Hell, I find NEW codenames hard to keep track of. In essence, codenames suck.

I agree not here. Codenames are fine, perhaps you do not find those products interesting, for you.

 

[Rikimaru]

Numerical code names were much better.

 

[Deleted member 13524]

I used to have 2*RV790 HD4890 in Crossfire, back when they cost like 190€ each.
Then I switched for a single Cypress HD5870 for about 50€ more than the ~300€ I got for the HD4890s, but I remember actually losing performance in some titles with this transition, which I found really weird..

 

[Razor1]

I think AMD lost marketshare in both CPU's and GPU's this quarter,

according to the q2 financial call

inventory is way up.

q3 and q4 they don't expect be making a profit, even though gross margins will be up 3-9%

They didn't really answer Fury supply question, they said HBM is ramping well

Nano coming in August.

Low gross margins on semi custom wins (console GPU's)

Taped out a couple of 14/16nm designs... which one of those, Zen and next gen GPU?

HBM targeted at Enthusiast segment. Didn't answer on HBM vs older ram margins.

R&D budget will be protected as much as they can even with the down turn.

Q4 pc market should stabilize for AMD. (Analyst responded AMD's expectations are not good in the past, how is that going to change, AMD responds they need more information to give better guidance, and they are being cautious)

They will be able to get cash ABL,- Asset based lending, pretty much getting money for assets (collateral) or possible capital markets (shares for cash or taking on more debt -- ABL)

ARM CPU ramp up will be slow, they need to build the ecosystem first. ARM server market will take 3-5 years to develop. x86 will hold most of the market for the near feature. (PS i think this is stupid for them to go into the ARM server market, if Zen brings competitiveness back for AMD with Intel btw, because it just creates internal competition and that will reduce gross margins, increases R&D, marketing expenses etc.)

Zen and next gen graphics will use 14/16nm sometime in 2016

That's it.

 

[Alexko]

Did they say anything about their 2016 APU and FinFETs? It is looking increasingly likely that they'll just vaguely re-heat Carrizo on 28nm, which I think would be disastrous.

 

[silent_guy]

Terrible results. What I don't get is how they were expecting better results in H2 during last investors call (don't remember if it was previous quarter or their investors day.) At that point, they should have gotten a pretty clear insight into their competitive positioning both for CPU and GPU. And now all that is suddenly gone.

HBM targeted at Enthusiast segment.

To the surprise of no-one.