The AMD Execution Thread [2007 - 2017]

[3dilettante]

32nm is the last process that was well underway back when AMD spun off its fabs, so things can only diverge further from here.
Many key decisions in its feature set were made by AMD, which in turn can credit IBM for many of those decisions. The blame for that is more complicated.
Were it not for the consortium with IBM, AMD would not even have a problematic 32nm process.

The base problem is that both high-end processor design and process development take massive and ever-increasing investments, and the gulf between AMD and its competition is widening.

BD and 32nm is the best AMD could do. Much of what AMD disclosed indicates that a lot of what they did was to stretch their limited capabilities as far as they could.

 

[Deleted member 13524]

BD and 32nm is the best AMD could do.

How can this be true if BD's architecture seems to have about half the performance/transistor ratio of its predecessor?




And BD's performance brings another question the table:
- If Bulldozer sucks, how can Trinity not suck?

As far as I can tell, Llano failed to storm the notebook market due to supply shortages. Back in June I was pretty sure that by now, we'd have tens of mid-end models with quad-core Llanos + crossfired GPUs, but instead we only see Sandybridges with Optimus'ed GT5xx nVidia GPUs..
And then reports came up claiming that although Llano couldn't get its targeted marketshare (and mindshare), Trinity wouldn't suffer from the same problems so it'd be one Fusion chip to take the notebook market by storm.


By the looks of things, Trinity APUs may actually lower the CPU performance bar set by its predecessor... which wasn't all that great from the beginning.
That's a huge compromise to take, considering how GPGPU hasn't really set off to take advantage of the beefier iGPU.

 

[swaaye]

I really wanted to see some 13.3" Llano notes but it has just become the new ultra budget choice in loads of mediocre 15.6 and 17" machines.

Yeah I am not expecting much from Trinity. It doesn't make sense to assume it won't be middling against Ivy Bridge.

 

[3dilettante]

How can this be true if BD's architecture seems to have about half the performance/transistor ratio of its predecessor?

They have a lot of transistors invested in things that matter very little to the desktop, or make things worse, because it's a poor fit for the desktop space.
I said it's the best they can do, not that their best is necessarily competitive.
Merely shrinking a Stars core gets us Llano, or rather, Llano with pretty bad power management (and Llano already hits 100W).

And BD's performance brings another question the table:
- If Bulldozer sucks, how can Trinity not suck?

Trinity could dispense with a lot of the extra transistors the desktop doesn't need.
It also wouldn't take too much improvement to do better than Llano, which has peaked just below 3 GHz with a TDP in the same range as a massive BD die.

As far as I can tell, Llano failed to storm the notebook market due to supply shortages.

Due to yields, particularly yields of the top bins. Which rumors say is due to the CPU, the last gasp of Stars.

By the looks of things, Trinity APUs may actually lower the CPU performance bar set by its predecessor... which wasn't all that great from the beginning.

I don't know what Trinity's final configuration will be, so I can't rule it out. Llano doesn't set a very high bar to clear, however.
Llano does not clock to same levels as Phenom, so some regressions in single-threaded performance are already built into Llano.

 

[Raqia]

It seems Intel doesn't have to resort to rebates and intimidation to keep AMD down anymore; it's not that BD is a terrible chip but its overall execution was horrible which is in no small part due to AMD choking for cash after Intel's barrage of underhandedness. It's a shame that AMD took a measly 1 billion dollar settlement, but it probably needed that token amount just to ensure survival. I wouldn't be surprised if Dirk Meyer quit from just the stress of managing the fiascos of Barcelona, the Intel suit, 32nm yields, and the extended drama of BD during his tenure.

That's not to say he's faultless in his direction; he probably had the board is breathing down his neck because ARM scale CPUs totally left AMD (and Intel for that matter) behind in terms of relevance, and he must be kicking himself for selling the most mobile relevant Imageon branch of ATI to Qualcomm for a song years ago. I recall he made a presentation during an analyst day a few years back after going fab-lite that emphasized paring down and focusing the company to notebook and above CPU and GPU products; sadly it was exactly the wrong focus even though it was AMD's strong suit. If he kept that branch of ATI going and fabbed a die w/ k6 class cpus and a gpu for phones, AMD might have a few billion more in revenue, but very few would have had that kind of foresight and courage to do something like that if they were put in charge of a hemorrhaging company.

It'll be interesting to hear what Rory Read has to say on the next Analyst's Day.

 

[mczak]

I don't know what Trinity's final configuration will be, so I can't rule it out. Llano doesn't set a very high bar to clear, however.
Llano does not clock to same levels as Phenom, so some regressions in single-threaded performance are already built into Llano.

I think roadmaps always showed Trinity would be one or two modules. With anandtech hints of no L3 cache even a two module Trinity would be quite small area-wise on the cpu side (assuming same L2 cache) (comparable to 4-core Llano).
I would think performance would be faster or at least as fast for single-threaded case, not only should it have (hopefully!) some little improvements, but imho the clock really should be quite a bit higher than Llano, both on the desktop and mobile side, not least because Llano clock is low (and with turbo which doesn't do much if anything which is fixed with BD). Multithreaded though I'm a bit less optimistic (less help from turbo and after all 4 cores still do scale better than 4 half-cores) but overall it could be an improvement over Llano. But I'd think it will look (on the cpu side) just as silly compared to Ivy Bridge as does Llano to Sandy Bridge now...

 

[CarstenS]

Q3 results have been posted:
http://www.amd.com/us/aboutamd/newsroom/Pages/newsroom.aspx

Just a quick c'n'p:
Quarterly Summary

Gross margin was 45 percent.
Cash, cash equivalents and marketable securities balance, including long-term marketable securities, was $1.86 billion at the end of the quarter.
Computing Solutions segment revenue increased 6 percent sequentially and 5 percent year-over-year. Sequentially, higher mobile and server microprocessor revenues were partially offset by lower desktop revenue. The year-over-year increase was primarily driven by higher mobile processor and chipset revenue.
Operating income was $149 million, compared with $142 million in Q2 11 and $164 million in Q3 10.
Microprocessor ASP increased sequentially and decreased year-over-year

Seekingalpha-Transcript:
http://seekingalpha.com/article/303...sses-q3-2011-results-earnings-call-transcript

 

[NathansFortune]

Just as everyone figured, AMD are doing well in the mobile sector because of Zacate but poorly in the desktop sector where they aren't even competing.

I really hope Piledriver delivers the goods next year because AMD can't go on like this for much longer.

 

[Alexko]

Just as everyone figured, AMD are doing well in the mobile sector because of Zacate but poorly in the desktop sector where they aren't even competing.

I really hope Piledriver delivers the goods next year because AMD can't go on like this for much longer.

Well, they made a profit, even though they suffered from capacity constraints pretty much across the board, and Bulldozer sucks. So maybe they can.

 

[NathansFortune]

Well, they made a profit, even though they suffered from capacity constraints pretty much across the board, and Bulldozer sucks. So maybe they can.

Bulldozer is out this quarter so revenue will be booked now, and we will see how it affects their gross margin. They have moved from a medium sized chip with Phenom to a large 2bn transistor chip and haven't been able to raise their prices very much.

 

[no-X]

In fact Bulldozer (8C, 315 mm²) is smaller than Thuban (Phenom X6, 346 mm²).

 

[3dilettante]

Until the process matures, AMD may not see a cost savings since yields are lower and other costs higher than they would be for 45nm.
BD could raise the ASP, which might make it an improvement regardless.

 

[NathansFortune]

In fact Bulldozer (8C, 315 mm²) is smaller than Thuban (Phenom X6, 346 mm²).

Oh wow. They really packed those transistors in!

 

[no-X]

Yes, density is really good, that's their strong point. But it's completely wasted, because Intel has ~ twice as good performance-per-transistor ratio.

 

[Alexko]

Until the process matures, AMD may not see a cost savings since yields are lower and other costs higher than they would be for 45nm.
BD could raise the ASP, which might make it an improvement regardless.

Under the current agreement, they only pay for good dies. They do pay for the poor yields in terms of capacity constraints, though.

 

[3dilettante]

Rory Read seems to have tripped the flux capacitor, because I thought it was three years ago, and then two years ago.
Time for another decimation.
http://www.amd.com/us/press-releases/Pages/pr-3-11-11.aspx

 

[Miksu]

Any ideas if AMD will completely kill one or more of their current product lines? Laying off 10% of workforce is a big step and I'm wondering if they are going to completely get rid of one or more of their current business units.

 

[Alexko]

I really can't see anything to get rid off. They've already sold Imageon, they only have their core businesses left.

 

[Miksu]

I really can't see anything to get rid off. They've already sold Imageon, they only have their core businesses left.

I wouldn't be suprised if they decide to alter their core business and get rid of something large. For example the discrete GPU business in general seems to be spiraling down.

 

[karlotta]

Other than OEM fusion products, Discrete GPUs ARE there core/income/Image. Getting rid of Discreat GPUs before ARM saves there ass, is end of story. Server Biz is the big appendage cut/ loss.
Should of, would of, could of, got 10billion out of Intel, as it stands now that One Billion is just a Intel Marketing cost for Dell and company to keep AMD out of the Public mind share back when it counted... Tick tock