AMD Reportedly Scraps 28 nm APUs at GlobalFoundries

Page 3 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Status
Not open for further replies.
[citation][nom]ashven23[/nom]AMD should be thinking more along the line of 22nm if they are thinking starting from scratch, though i do not know if TMSC could provide them with a solution.[/citation]
I was just thinkiing the same thing, if they are moving from one company to another to manufacture then why not go directly to 22nm? As long as TSMC can give assurances that they can do it the jump in with both feet and make up some lost ground.

Who the hell is running AMD, circus clowns?
 
People put way too much emphasis on holding the "performance crown." Never mind the fact that this accounts for less than 1% of the income for Intel/AMD. (and way, way less than that in terms of number of CPUs sold) And also never mind the fact that for the most part they're pointless CPUs to have anyway; most of us have agreed that while sure, the i7 3960X may be king of the hill, it's just not worth the $1,049US price it asks. Instead, we look at a much cheaper end: most of us are saying that there's no point for even an Intel fan to get something beyond the 2500K, which is a COMPLETELY different market at $225US. And of course, even at that level, it's been seen that it's still a waste for Battlefield 3.

At this range, AMD does have competitors; it doesn't matter that it's their very BEST CPUs are what it takes to compete, but as long as they provide comparable performance for the price, they're competitors. All this emphasis on Bulldozer is forgetting that in terms of price-point competition, AMD's still had competitors in the form of both Deneb and Thuban, ALL of which can be had for well under $200US.

And more importantly yet are the sub-enthusiast markets, which are bigger still: in that market, Intel's not been able to compete in YEARS: Celerons only sell because they have the Intel brand name on them, and I think everyone here can agree that they are by and far a joke. I'll take a cheap Athlon/Phenom II or Llano over a Celeron ANY day of the week.

And as Kyuuketsuki pointed out, in the low-end market, having a good APU is more important than good CPU. And again, Intel's GPUs have always been an utter joke; it's just been whether it's a big crack-up (like Intel "Extreme") or a running gag. (like Larrabee) Simply put, they have no hope of competing with the integrated Evergreen-based GPUs on Bobcat APUs. For the mainstream-level stuff, the 320+ stream processors to be found in a Llano APU actually make a strong argument for a real GAMING computer without a discrete GPU.

And likewise, at the low-power end so critical for netbooks and the like, Atom, which was once hailed as a masterpiece, is woefully inadequate to compete with AMD's netbook-level fusion APUs; even on the CPU level, an Atom rings vastly worse per-clock performance without being able to up the clock to make up for it... And often while consuming MORE power. Once you throw in the GPU suitable for low-end gaming on the AMD chip's side, it's no contest.

And yeah, Intel's had some embarassing flops with their technology in the past, even with their esteemed flagships. Some might know, as mavikt does, that it was actually Tom's itself, for instance, that discovered the flaws that kept Pentium III from passing 1 GHz, while AMD's Athlon blew past it and beat it in every way, and then a few years later, Pentium 4 hit a wall before 4 GHz, and got thoroughly trashed by the Athlon64. It's only a matter of time before it happens again; it just appears that Bulldozer wasn't to be that.
 
Finally! If it's true, GF will finally start to get their game up.

AMD lost a lot cause of poor yields at GF. Not that TSMC does any better when moving to a new node and process, but at least AMD has more options now.

Good move, Mr. Read, good move.

Cheers!
 
[citation][nom]wiyosaya[/nom]Intel has developed their 3d gate technology, and it might do AMD well to start a similar research effort, either in-house, or a collaboration with a university.[/citation]

FinFET is nothing new. AMD was playing with 10nm double gate lengths back in 2002 (Intel responded saying they were looking at tri-gate) for use on a 22nm process.

http://www.amd.com/us/press-releases/Pages/Press_Release_42454.aspx

As Wikipedia puts it...

In current usage the term FinFET has a less precise definition. Among microprocessor manufacturers, AMD, IBM, and Motorola describe their double-gate development efforts as FinFET development whereas Intel avoids using the term to describe their closely related tri-gate [1] architecture. In the technical literature, FinFET is used somewhat generically to describe any fin-based, multigate transistor architecture regardless of number of gates.

http://en.wikipedia.org/wiki/Multigate_device#FinFETs
 
[citation][nom]DavidC1[/nom]What are you talking about? Pentium 4 was very successful back in the day. They also took the lead back from AMD when they brought 0.13 micron in and then Hyperthreading and 800MHz FSB.Oh, and Itanium's server sales are better than AMD's Opteron sales. It might not be a success for Intel but not too bad either.[/citation]

What you're saying isn't true, you've got a distorted perspective.

The Pentium 4 sold well because it was an Intel chip. Yes, it was more successful than the Athlon or Athlon 64 by that measure, but anything Intel made would be, because it was from Intel.

But, if you look closer, and use a more intelligent metric, it was very unsuccessful. Extremely. Measure it against what Intel sales would have been if it were an enhanced Pentium III (like Conroe), for example. Intel slowly lost market share, and even Dell deserted them. The reputation for the Pentium 4 was so poor, Intel had to use a new name for the newest generation Pentium III, calling it Core. Intel still had good market share, but declining slowly but surely as long as the Pentium 4 was reigning.

Technically, it was a horror. It's not as bad as the Bulldozer, which fails at virtually everything, while being huge, but it sucked too. There were times where it was faster than the Athlon, like when the Northwood came out, but most of the time it was behind much smaller chips, that used much less power, and gave off much less heat. Making a bigger chip that uses more power, and is faster is not a victory, it's expected. Nonetheless, Intel failed to do this the vast majority of the Pentium 4's life.

Big, power hungry, hot, and slow are a failure. Losing market share is a failure. Hurting one's hard fought for reputation is a failure. Compromising the name of the processor so badly it needs to changed, is a failure.

The Pentium 4 was a failure.
 
an APU with power comsumption below 2W to smartphone makers they'll have won the game. Until then I'll be browsing the internet on a bulky Intel notebook.
 
[citation][nom]ta152h[/nom]It's not as bad as the Bulldozer, which fails at virtually everything, while being huge, but it sucked too.[/citation]

I'm not sure what you're getting at.

When P4 came out, it offered SSE2 which was exclusive, however it performed slower or on par with CPUs clocked 75% of its speed for most workloads. Bulldozer is in the same boat, generally speaking; it offers FMA4 and XOP which are both exclusive for the time being, and clock-for-clock per-core performance is notably inferior to Sandy Bridge. Both have their strong points; the P4's ability at encoding was its only saving grace until Northwood came out, and Bulldozer (admittedly, a server CPU) is strong at heavily threaded workloads along with newer instruction sets.

P4 got better with time thanks to larger caches, the change to DDR2 and, at times, HyperThreading, as well as the introduction of 64-bit processing capability, more instruction support and improved manufacturing, though it still didn't shake its reputation for being power hungry.

Bulldozer is in its infancy. There's a lot to be worked out, and it will be. AMD couldn't very well go with an improved version of Stars again as it would only take them so far. The comparisons with Willamette are obvious - late, hot, and underwhelming in all but a few tasks.

In any case, saying Bulldozer was worse than P4 is very far off the mark. At least it didn't require a special kind of RAM, nor is it priced astronomically.
 
if TSMC turns out to be a successful alternative to GF then maybe AMD can have more confidence in designing a more innovative architecture and not worry too much whether what they come out with can be efficiently manufactured
 
Status
Not open for further replies.