AMD CPU speculation... and expert conjecture

[mayankleoboy1]

http://semiaccurate.com/2012/12/12/nvidias-audi-win-goes-to-tis-jacinto-5-soc/

Nothing significant, except that Nvidia is more sh!tty than other companies.
Also, Tegra3 based phones are not in the top category anymore. So yes, Nvidia has made some money this quarter, but its going to lose money if Tegra4 is not present in top tier phones and tablets.

 

[Cazalan]

http://semiaccurate.com/2012/12/12/nvidias-audi-win-goes-to-tis-jacinto-5-soc/

Nothing significant, except that Nvidia is more sh!tty than other companies.
Also, Tegra3 based phones are not in the top category anymore. So yes, Nvidia has made some money this quarter, but its going to lose money if Tegra4 is not present in top tier phones and tablets.

It's a false title once you read the update. The TI SoC didn't displace the Nvidia one. It's being used for a more advanced digital radio alongside Tegra.

 

[fazers_on_stun]

Higher demand from earlier adopters left nVidia somewhat behind the 8 ball early on.

As for costs, everyone had to pay those rates, as newer nodes are simply more expensive to make.

Perf wise, its been a very good node

And as for yield ramps, fairly dismal for both NV and AMD from what I remember, which is the main reason why the new GPUs were in short supply for a good while.

 

[Cazalan]

IIRC the justification for SOI was lower power consumption, due to lower leakage. Of course you could also oc the heck outta it, on LN2 anyway

It's all tied together though. Less leakage means you don't have to push the voltages as high as you ramp the clock speeds, which keeps you lower on the TDP curve.

That's one reason why we don't see 4Ghz ARM chips yet.

Even ARM concluded that SOI gives a sizable benefit over bulk.

http://www.arm.com/about/newsroom/26070.php

"The silicon results show that 45nm high-performance SOI technology can provide up to 40 percent power savings and a 7 percent circuit area reduction compared to bulk CMOS low-power technology, operating at the same speed. This same implementation also demonstrated 20 percent higher operating frequency capability over bulk while saving 30 percent in total power in specific test applications."

 

[JAYDEEJOHN]

Like I said, 2-3 months, easy to fudge when youre an Intel, not so when youre Apple,AMD,nVidia etc etc
Of course, you could wait til ramp is full...but then again, how happy would Apple,AMD etc etc be?
http://www.semiwiki.com/forum/content/1054-tsmc-28nm-yield.html

 

[mayankleoboy1]

If you are Intel, any fudge you make, still far ahead of the industry. Instead of leading them by 70%, you lead them by 50%.

 

[de5_Roy]

AMD: We Would Like to Use Process Technologies for Longer Periods of Time.
AMD Unsure About Transition Timeframe to 20nm Node
http://www.xbitlabs.com/news/other/display/20121213170238_AMD_We_Would_Like_to_Use_Process_Technologies_for_Longer_Period_of_Time.html

....Earlier this year we have brought our GPU products to 28nm. You will see [Kabini and Temash] APUs in the first half of next year on the 28nm. […] We have not said when we are going to 20nm, which would be the next node,” said Ruth Cotter, vice president of investor relations at the Raymond James conference.

they also have not said when when kaveri will come out

 

[fazers_on_stun]

^ My take is that (1) AMD doesn't have the money to move the next-gen CPUs to 20nm or lower, as presumably they would have to pay extra, and (2) AMD doesn't have a lot of confidence that GF and/or TSMC is gonna get to the next node when they say they are.

Also, didn't AMD recently state something about being able to realize almost a node's worth of power savings merely by using a high-density macro library? If that is true, then they could achieve good results entirely in-house and avoid paying for the next node. Let the rich cellphone customers pay for TSMC's or GF's R&D at 20nm or 14nm..

 

[de5_Roy]

^^ iirc it was during steamroller's unveiling few months ago. i think amd claimed 30% improvement... in something. people who covered that speculated that high density library won't debut until excavator. now that sr is delayed....
edit: found it here:
http://www.anandtech.com/show/6201/amd-details-its-3rd-gen-steamroller-architecture/2

We won’t see these new libraries and automated designs in Steamroller, but rather its successor in 2014: Excavator.

lol
anyway, i'd love to see amd applying design tweaks like these in low power apus and mobile sr first, if possible.

 

[gamerk316]

^^ iirc it was during steamroller's unveiling few months ago. i think amd claimed 30% improvement... in something. people who covered that speculated that high density library won't debut until excavator. now that sr is delayed....
edit: found it here:
http://www.anandtech.com/show/6201/amd-details-its-3rd-gen-steamroller-architecture/2

We won’t see these new libraries and automated designs in Steamroller, but rather its successor in 2014: Excavator.

lol
anyway, i'd love to see amd applying design tweaks like these in low power apus and mobile sr first, if possible.

AMD claims a lot of things. I very much doubt 30%, except maybe in one benchmark that gets near perfect CPU scaling.

 

[mayankleoboy1]

AMD is going to release SR sometime in 2014, right ?
And on 28nm, even with 30% IPC improvement over PD, it cant be competitive with Intel's 14nm.

Only if AMD sells 8core unlocked CPU's at the price of Intel Pentiums, can they make some money.

 

[griptwister]

AMD is going to release SR sometime in 2014, right ?
And on 28nm, even with 30% IPC improvement over PD, it cant be competitive with Intel's 14nm.

Only if AMD sells 8core unlocked CPU's at the price of Intel Pentiums, can they make some money.

Or, if intel screws themselves over with the integrated RAM and CPUs. LOL! Smaller doesn't mean faster, it just means smaller... But it will be interesting to see the performance increase Intel Gets out of their CPUs due to the smaller die size, I'm hoping that they don't rip us off "If or once AMD goes under." And, we don't even know if that's official. In case you didn't notice, this isn't the "AMD Won't Top Intel" thread, this is the "AMD Steamroller Speculation and Expert Conjecture" thread.

 

[fazers_on_stun]

Hmm, considering the rumour du jour in the Haswell thread, that Intel may buy out nVidia and then make Jen-Hsun "Can o' Whoopas" Huang CEO of the combined firm as Otellini is retiring next year, that would probably ensure AMD's demise..

Imagine a GT 780 or 880 on Intel's 14nm process in 2014, since out of the 3 of them - Intel, TSMC and GF - I would bet Intel most likely to succeed getting to 14nm when they say they will (esp. seeing as how Intel already has working Broadwell CPUs on 14nm, booting into Windows).

 

[viridiancrystal]

Hmm, considering the rumour du jour in the Haswell thread, that Intel may buy out nVidia and then make Jen-Hsun "Can o' Whoopas" Huang CEO of the combined firm as Otellini is retiring next year, that would probably ensure AMD's demise..

Imagine a GT 780 or 880 on Intel's 14nm process in 2014, since out of the 3 of them - Intel, TSMC and GF - I would bet Intel most likely to succeed getting to 14nm when they say they will (esp. seeing as how Intel already has working Broadwell CPUs on 14nm, booting into Windows).

XD I like that.

Intel is no doubt in the led as far as process goes. They had the best 32nm, and they are the only ones using tri-gates AFAIK, and they are on a secure 22nm, while GF and TMSC are on 32nm and 28nm.

Steamy is planned for 28nm, so that shouldn't be the reason for it's push back to 2014.

 

[noob2222]

XD I like that.

Intel is no doubt in the led as far as process goes. They had the best 32nm, and they are the only ones using tri-gates AFAIK, and they are on a secure 22nm, while GF and TMSC are on 32nm and 28nm.

Steamy is planned for 28nm, so that shouldn't be the reason for it's push back to 2014.

Im wondering if AMD got wind of this early. http://news.cnet.com/8301-1001_3-57559079-92/the-pcs-past-and-intels-future/

Would make some sense as AMD would be the only company to release a NEW DT CPU (marketing hype) in 2014.

 

[viridiancrystal]

That seems like an okay idea, but why they wouldn't stick to yearly releases and release Exxy when Intel isn't making a desktop update doesn't make much sense. Unless they feel the need to hype that they've been working for years on this processor arch. We all know how well that works out.

 

[de5_Roy]

That seems like an okay idea, but why they wouldn't stick to yearly releases and release Exxy when Intel isn't making a desktop update doesn't make much sense. Unless they feel the need to hype that they've been working for years on this processor arch. We all know how well that works out.

rumor is that steamroller - as it is now, has not passed some kind of internal evaluation and is deemed unfit to take on haswell. delaying sr should push excavator (if it even comes out) back even further. s/a has published a few articles on why sr was delayed. the rumored 'new' apus in 2013 are called richland - pd cores with gcn igpus. s/a speculated (long before they wrote about steamroller's delay) that trinity might receive an update before sr comes out/gets delayed.
i assume there could be one more reason - sr seems to be tied with glofo and glofo doesn't seem to have a working (afaik, i haven't checked lately) 28nm node.
imo if the current sr arch won't come out, amd might design something else and call that steamroller, for example - jaguar cores tweaked for performance use.

 

[jdwii]

Oh yeah since we all GF knows what their doing, Amd stays with them because of their great track record i'm sure.

I'm probably going to say SR will be delayed over performance issues just like BD was. Amd doesn't like history i guess since if they even looked back 2 years they will notice they keep doing the same mistakes over and over again.

 

[noob2222]

That seems like an okay idea, but why they wouldn't stick to yearly releases and release Exxy when Intel isn't making a desktop update doesn't make much sense. Unless they feel the need to hype that they've been working for years on this processor arch. We all know how well that works out.

the thing is all we know is 2014. truthfully, that could be as little as 3 months from their normal schedule.

 

[bawchicawawa]

Does anyone know the wattage of the 7660d in trinity?

 

[mayankleoboy1]

A generation of AMD GPU's built on intels 22nm process would kill Nvidia for a generation atleast.

 

[guskline]

Looks like PileDriver may be the last "big" desktop chip.

 

[Cazalan]

Steamy is planned for 28nm, so that shouldn't be the reason for it's push back to 2014.

The analysts expected an 18 month cycle to switch from GFlo to TSMC.

http://www.eetimes.com/electronics-news/4230857/Opinion--AMD-roadmap-change-benefits-TSMC-

I don't think they'll ever recoup the money they gave away to switch process. With the penalties, broken contracts, and a 18 month delay on top.

Whomever at TSMC convinced AMD to switch must be getting some huge kickbacks. You'll probably see some high level AMD exec(s) go to TSMC next year.

 

[fazers_on_stun]

The analysts expected an 18 month cycle to switch from GFlo to TSMC.

http://www.eetimes.com/electronics-news/4230857/Opinion--AMD-roadmap-change-benefits-TSMC-

I don't think they'll ever recoup the money they gave away to switch process. With the penalties, broken contracts, and a 18 month delay on top.

Whomever at TSMC convinced AMD to switch must be getting some huge kickbacks. You'll probably see some high level AMD exec(s) go to TSMC next year.

Hmm, that article is from 13 months ago, so I dunno if still relevant or not. But it is true AMD is paying an arm + leg to GF to get out from under.

 

[Cazalan]

Hmm, that article is from 13 months ago, so I dunno if still relevant or not. But it is true AMD is paying an arm + leg to GF to get out from under.

The divorce looks to have started sometime in Q1 2012 when the wafer contract was renegotiated and ended with a $700 mil charge.

http://blogs.barrons.com/techtraderdaily/2012/03/05/amd-street-says-deal-with-globalfoundries-encouraging/

The 18 month was speculation and probably didn't account for the serious layoffs.