AMD CPU speculation... and expert conjecture

[8350rocks]

Jaguar servers are replaced by ARM servers and Temash tablets are replaced by ARM tablets. Links and roadmaps were given.

However, your claim was "Why does that mean AMD is scrapping x86 products?", which is not true, because AMD continues selling and developing many x86 products. Did you hear about Kaveri and Steamroller? Warsaw? Berlin?...

Where did I say AMD was scrapping ALL x86?

Temash, jaguar, kabini, ect are x86 products are they not?

If they are scrapped according to your wishes, then AMD is scrapping x86 products are they not?

i am just trying to jog your memory, i could remember it wrong myself, so...
anywho, iirc, the original claims(made by juanrga) were: (edit:compilation of, not direct quotes)
amd is replacing jaguar with arm cores in microservers.
warsaw is for legacy support.
temash is being replaced by arm soc.
amd will not make 4+ core cpus, no need for them anymore.
kaveri, berlin, warsaw do not have (much of) a future. <- this may have been very, very, very, very strongly implied instead of being explicitly stated, i can't remember clearly.

so... the ones that are left.... i guess... are the discreet gfx cards for x86 platform, which, amd are not abandoning, apparently...

....
wait. somehow, the 4 core cpus (apus sans igpus) seem to not have been abandoned by amd. edtit: so did single/dual/tri core cpus. yay. on a less happy note, amd seems to have entirely abandoned L3 cache.<sadface>

The thing that someone wants to ignore is the fact that SR-FX or whatever they decide to call it was never slated to be released before 2015. AMD has since released an APU roadmap and has been taken to mean that is the only products that will ever exist. No products other than those on the APU roadmap are in development. They also have gone as far as saying the existence of high end ARM will cancel the existence of low end x86.

AMD supposedly abandoning high end cpus leaves only middle performance APU for x86 support or Intel for ALL high end computing. ARM is not high end. Maybe 2, 5, 8 years from now, who knows how long that will take. Some people are convinced that is already here today based on personal opinion.

If this truthfully is the case, I may have purchased my last AMD cpu. I am not interested in a downgrade powered cpu to play games at medium to low settings. If ARM is the only choice to make in the future, ill go with Samsung or Qualcomm since AMD is forcing the choice (or rather lack of choice).

AMD needs to play both markets. They can't survive with a middle performance only x86 cpu, (think VIA) and ARM is full of competition already.

They will have something for HEDT, I have assurances of that, I can't get any details as to what the plan is yet...however, I am confident the source is being as upfront as they can at this point.

EDIT: My understanding is that a migration to FD-SOI (UTBB most notably) is not out of the question for these parts. I couldn't get outright confirmation in the conversation; however, he said they were considering alternatives to bulk process and wouldn't state specifically which one. Based on the history they have with SOI, I am inclined to think it would be that direction...they could go to FinFETs, but I doubt that would make any sense given their production partners.

 

[-Fran-]

This is an interesting JINX for Intel.

Back when Intel was fighting RISC from MIPS (I think), IBM was the one who let Intel be the monster it is today and let AMD (forced Intel in other words) have the X86 ISA as you all might already know. VIA was something in between as well, but I don't remember that part, lol. Currently, IBM doesn't give a damn about Intel in the server space (that I know of, please correct me if I'm wrong) and Intel doesn't have anyone big enough to play ball with them.

Anyway, the current landscape is getting muddied by Microsoft's incompetence to keep the Windows momentum into new versions. Meaning, it's letting rivals get more appealing, ironically, from the ground up.

Back when MS won (again, lol) the big IBM contract that made it jump to stardom, they were inside every "IBM-Compatible" machine out there. That was something important in the consumer segment, if not the most important "checkbox" of them all. With the introduction of Mobile computing (smartphones in other words) the landscape is changing at a blazing fast speed and, I'll make a bet, that Mr. Otellini knows this and he dropped the ball on it, since Intel did not react fast enough to that swift change.

Let me explain that: people is getting accustomed to a new UI experience (Android and iOS mainly; BB, WinMo and Symbian are small players now) which is basically a new OS underneath (light or not, is another topic) that gives support to a wider variety of hardware combinations (except iOS). People is shifting from the Wintel paradigm. Even more, they're shifting from the WinX86 paradigm from the small OS to eventually land in a big OS with whatever it has inside.

So, the jinx is basically that Intel is pursuing ARMs territory, so desperately that they're neglecting consumer space and will neglect Microsoft eventually in favor of the new paradigm OSes. We all know Haswell is a landmark for this point and Intel themselves have said it. They don't want the penny-change consumer market, but AMD wants it. They have console design wins and have APUs coming as well. Intel's extreme platforms are stupid expensive, so they cannot compete with AMD in a price/performance war without dumping cash, and Intel doesn't do that, not even to save face (Pentium 4 EE, remember that? it's the equivalent to the FX 9xxx). AMD won't disappear from either market, since they still make money out of them.

Thinking about the company world, that's dominated by Wintel-boredom-boxes alright. That won't change soon, if MS knows how to keep them, but at the same time, the pace in what companies shift to new OSes massively is slow. So slow that MS cannot depend on that, I'd say. If they would like to adapt to keep the companies happy and neglect (just like Intel, for different reasons) the consumer market, they'll have to change how they develop Windows. Meaning an impact in consumer OS relevance, possibly making other OSes more attractive due to shorter release cycles and new bling bling.

Hell, at the end of this brain fart, I'd say its also a JINX for Microsoft, lol.

I need to re-think all this, but I'd say I raised a few good points.

Cheers!

 

[8350rocks]

Well, and the funniest part is, Intel has admitted that the Ultrabook was a failure after 3 years of sinking money into a product line that was bearing no fruit. So, now they're going to 2-in-1's or detachables.

So they're taking everything people hate from the Ultrabook, and removing the things that made it PC-esque, and promoting that instead. By trying to aggressively cram product down consumer's throats, they're emulating the M$ paradigm of trying to tell people what they want. That is, instead of listening to the consumer.

M$ seriously underestimated mobile, by even their own admission. Now they're scrambling to play catch up, and are *FAR* behind because of it.

Other OS's are rising up like crazy now because the little guys have realized that the behemoth that is M$ has a cracking façade, and no one seems competent to actually go in and repair the damage that has already been done.

I think the issue here lies in 2 things specifically:

1.) PC market is not actually shrinking or failing, the driving force behind them (Wintel) is not innovative enough to keep people happy. Add in the fact that new hardware and software are not compelling upgrades over systems from 3-4 years ago for most people, and you can quickly see why no one is buying aggressively right now. The evidence of this is HEDT (i.e. gaming PCs). HEDT is growing at a pace that exceeds the overall shrinkage of the DT market in general. This means that people who are actually buying DT PCs, are buying HEDTs looking for all out raw performance for gaming or heavy workloads.

2.) With the stagnation in software and hardware improvement, a great deal of the populace is becoming more discontent with Wintel. This is evidenced by the resurgence of Linux distros, and increasing popularity of alternatives. A new API like MANTLE would not be given a glimmer of a snowball's chance in hell if M$ still had an iron fist style grip on the DT PC market. However, software that is compelling, like Ubuntu, Linux Mint, Fedora and Arch Linux provide compelling alternatives with powerful systems and zero compatibility with the M$ gaming infrastructure. Instead of someone trying to reassert dominance, M$ will have to capitulate because it is now in only 24% of devices around the world. Instead of the prime of the DT PC surge when that number was hedging 90%.

 

[Cazalan]

A phone chip is one with a phone-class TDP rating. If you can introduce more performance than competence in that thermal package that is welcomed.

And what do most Tegra 3 and Tegra 4 chips go into? Tablets and their hand held gaming platform Shield. These are clearly more than just phone-class TDP parts. They're a family of parts. There is a Tegra 4 and a 4i. The 4i is the one going into phones. The Tegra 4 is going into Surface 2 and Shield.

What exactly is phone-class anyway when battery sizes keep going up? The silvermont chip is a 2W part and you can see iPhone4 reviews drawing 2.8W.

http://www.anandtech.com/show/4971/apple-iphone-4s-review-att-verizon/15

 

[-Fran-]

What MS has in favor of them to keep the fight with a small effort, is the ridiculous amount of patents for OS-related stuff. The Patent system would need a really big re-haul for any other player to be as big as MS to play a one-on-one with them. I mean, IBM also has a bazillion patents (I think they're still number 1 YoY for that), but MS packs OS-centric ones. The second big argument (that gamerk will use, I'm sure) is that companies are cheap enough to not port stuff. Most windows centric software really needs a handy work to be ported out from it (think dotNET stuff, THE HELL I TELL YOU), even with a lot of good open libraries that work in most OSes (GTK and QT comes to mind). Also, in that same topic, remember Linux can not be a serious alternative due to patents and how MS has been taking advantage of the Patent System to keep competition away from it (at least, not as bad as Apple has).

I don't want to make this thread into a political propaganda, but the US really needs a Patent System rehaul IMO. With big Corps """"""sponsoring"""""" politicians, it's going to be quite hard, haha.

In all fairness, Windows is not a bad product. I really think is a good OS for whats intended: cover all bases. It's its great plus and also it's biggest problem, haha. We love those dichotomies, right?

Cheers!

 

[juanrga]

At the end of the day, in regards to CPU performance, X86 wins, easily. By how much is up for debate, but even the highest tier ARM looses to the lowest tier X86 (Silvermont). How is ARM going to compete against HW/SR if it can't even beat Atom?

ARM isn't displacing X86 anytime soon. The only reason ARM wants to get into the server market is because of the insane margins available. And lets not forget MIPS and PPC are always around the corner in the low-power market...

The Apple phone chip humiliated the tablet Silvermont chip, but why do you insist on comparing a phone chip to desktops chips like Haswell and Steamroller? Why don't use AMD Seattle?

Some basic numbers:

Phone (ARM A8): 4 GFLOP
Phone (ARM A9): 24 GFLOP
FX-8350 (Piledriver): 256 GFLOP
Opteron (Interlagos): 384 GFLOP
Seattle (ARM A57): 256 GFLOP

Phone (ARM A8): ~2500 DMIPS
Phone (ARM A9): ~15000 DMIPS
FX-8350 (Piledriver): ~120960 DMIPS
i7-3960X (Sandy-E): ~185200 DMIPS
Xeon E5-2687W (Sandy-EP): ~233500 DMIPS
Seattle (ARM A57): ~131200 DMIPS

The pretension that ARM can't even beat Atom was _very_ funny.

<...>

Good to know that you could find the word Apple. I suppose that the bold format did help. You calling them "Crapple" explains your hostility here.

Yes, the Apple A7 is using custom cores. We know this since first day.

Contrary to your beliefs, there is no cheating in the AT review of the new Apple SoC.

The TR article is rather clear about why Temash tablets were not competitive:

AMD has historically had a hard time getting its low-power chips into tablets. The old Brazos-based Hondo APU barely barely made it into any slates, and the newer Temash and Kabini chips don't have many design wins. The lowest-power Temash variant still has a relatively high 3.9W TDP, though; AMD can surely do better with ARM-based chips. Radeon graphics technology should make those offerings unique, and it could help AMD gain a foothold in the tablet market.

If you believe that translates to "AMD will continue pushing Temash tablets", that is fine for me

Android runs on x86.

Proof?

Are you kidding?

http://www.technobuffalo.com/2013/04/23/lenovo-k900-android-launch-date-may-6/

http://www.intel.com/content/www/us/en/smartphones/power-of-intel-inside.html

LOL...are you seriously suggesting that the tablet OS is a full blown desktop OS???

You might be crazier than I expected...LOL!!!!

That's a good one, Android tablet OS is the "ARM OS" solution....

Have you used a chromebook? Talk to me when you have...Chrome OS is roughly 2-3x more features than Android, and it's not even a rough fill in for a real desktop OS.

This is the fifth time that you ask me if Android is a full blown desktop OS and this is the fifth time that I say you: NO

And why do you insist on confounding Android with some of the ARM OS that are ready or are being developed?

Ask this question rocks (if you can): do you really believe that the ARM supercomputer that is being developed will run Android?

 

[griptwister]

My question for you guys is, Why are you debating about ARM when you should be playing the BF4 BETA?

@Paladin, O.O You're like Batman.

@The Q6660 Inside, I got some good news coming Late January

@juanrga, Look, It's obvious that the future is ARM as Obama Voters love their smart phones and tablets. But in HIGH END PCs Like, I don't give a crap about power consumption, I just want power, PCs.

 

[juanrga]

amd is replacing jaguar with arm cores in microservers.

This is not exact. The exact claim made was that AMD will replace Opteron-X servers (jaguar) by Seattle servers (ARM). But it is not a mere replacement of cores. There are more innovations in Seattle than just ARM cores...

warsaw is for legacy support.

Yes, Warsaw is _only_ for legacy support in a declining market. This is what AMD said:

He also told us that “Warsaw”, AMD’s follow up in the 2P/4P server market, will cater for a declining but still large market with institutional customers that will take longer to migrate to an ARM ecosystem.

Andrew Feldman also said:

I think the largest customers want ARM to win

It is evident why customers want the better product. Only 3 persons who still believe on nonsense such as "ARM cannot beat an Atom" or "x86 is a 180% more efficient" disagree.

temash is being replaced by arm soc.

Again this is not exact. The claim made was that AMD will replace Temash tablets by ARM tablets.

amd will not make 4+ core cpus, no need for them anymore.

This is incorrect. Warsaw is 12/16 cores CPU. Seattle is 8/16 cores CPU, HieroFalcon is 4/8 cores CPU...

kaveri, berlin, warsaw do not have (much of) a future.

As explained above, Warsaw is a temporary product before the customers migrate to ARM products.

Kaveri will be replaced by Carrizo in 2015.

 

[juanrga]

@juanrga, Look, It's obvious that the future is ARM as Obama Voters love their smart phones and tablets. But in HIGH END PCs Like, I don't give a crap about power consumption, I just want power, PCs.

Learn this. The HPC community want much much more power than you. That is the reason why they are developing an ARM supercomputer that will be about 1000x faster than the fastest x86 supercomputer today.

Also the AMD Seattle chip is not a phone or tablet chip. Some basic numbers:

FX-8350 (Piledriver): 256 GFLOP
Seattle (ARM A57): 256 GFLOP

FX-8350 (Piledriver): ~120960 DMIPS
Seattle (ARM A57): ~131200 DMIPS

 

[juanrga]

A phone chip is one with a phone-class TDP rating. If you can introduce more performance than competence in that thermal package that is welcomed.

And what do most Tegra 3 and Tegra 4 chips go into? Tablets and their hand held gaming platform Shield. These are clearly more than just phone-class TDP parts. They're a family of parts. There is a Tegra 4 and a 4i. The 4i is the one going into phones. The Tegra 4 is going into Surface 2 and Shield.

http://www.techradar.com/news/phone-and-communications/mobile-phones/nvidia-s-tegra-3-to-run-30-smartphones-in-2012-1082199

http://www.digitaltrends.com/mobile/zte-geek-announced-again-becomes-the-worlds-first-tegra-4-smartphone/

 

[8350rocks]

The Apple phone chip humiliated the tablet Silvermont chip, but why do you insist on comparing a phone chip to desktops chips like Haswell and Steamroller? Use a server chip such as Seattle

Some basic numbers:

Phone (ARM A8): 4 GFLOP
Phone (ARM A9): 24 GFLOP
Desktop FX-8350 (Piledriver): 256 GFLOP
Server Opteron (Interlagos): 384 GFLOP
Server Seattle (ARM A57): 256 GFLOP

Phone (ARM A8): ~2500 DMIPS
Phone (ARM A9): ~15000 DMIPS
Desktop FX-8350 (Piledriver): ~120960 DMIPS
Desktop i7-3960X (Sandy-E): ~185200 DMIPS
Server Xeon E5-2687W (Sandy-EP): ~233500 DMIPS
Server Seattle (ARM A57): ~131200 DMIPS

The pretension that ARM can't even beat Atom was very funny.

Source? Benchmarks? You can't provide any real world tests of "Seattle" and it's performance or power consumption.

Point made yet?

LOL. This is the fifth time that you ask me if Android is a full blown desktop OS and this is the fifth time that I say you: NO

And why do you insist on confounding Android with some of the ARM OS that are ready or are being developed?

Ask this question rocks: do you really believe that the ARM supercomputer that is being developed will run Android?

Who is developing an ARM OS? M$? Red Hat? Apple?

No, none of those are even considering an ARM DT OS. Canonical has a mobile OS called Ubuntu Touch that will run on ARM. Their DT version isn't nearly as capable as the x86 version.

Additionally...you are counting chickens *way* before they hatch. NVidia's so called ARM supercomputer has to have the chips delivered before it ever gets off the ground. Plus, that will likely be the lion's share of the ARM chips that NVidia sells for 3 quarters after that. Notice they haven't achieved any design wins in mobile formats? There's very much a reason for that. NVidia's ARM adventure cruise has run into a Hurricane called Qualcomm. That's not likely to change any time soon either.

Additionally, I would wager it still won't top the #1 Super Computer in the world at this point. 17 petaflops is a *LOT* of compute power. Little ARM cores would need thousands upon thousands to reach a number like that, even with Tesla GPGPUs.

 

[juanrga]

The Apple phone chip humiliated the tablet Silvermont chip, but why do you insist on comparing a phone chip to desktops chips like Haswell and Steamroller? Use a server chip such as Seattle

Some basic numbers:

Phone (ARM A8): 4 GFLOP
Phone (ARM A9): 24 GFLOP
Desktop FX-8350 (Piledriver): 256 GFLOP
Server Opteron (Interlagos): 384 GFLOP
Server Seattle (ARM A57): 256 GFLOP

Phone (ARM A8): ~2500 DMIPS
Phone (ARM A9): ~15000 DMIPS
Desktop FX-8350 (Piledriver): ~120960 DMIPS
Desktop i7-3960X (Sandy-E): ~185200 DMIPS
Server Xeon E5-2687W (Sandy-EP): ~233500 DMIPS
Server Seattle (ARM A57): ~131200 DMIPS

The pretension that ARM can't even beat Atom was very funny.

Source? Benchmarks? You can't provide any real world tests of "Seattle" and it's performance or power consumption.

Point made yet?

Sources were given before.

It is evident that you have double standards. It has been acceptable for you to discuss performance of Kaveri, Steamroller, new radeons, MANTLE... but it is not acceptable to discuss ARM, Seattle...

The same regarding power. When people claims that top Kaveri will be a 100W TDP rated APU, you say nothing, when people claims TPD for Seattle you solicit real world tests.

Double standard noted, rocks.

LOL. This is the fifth time that you ask me if Android is a full blown desktop OS and this is the fifth time that I say you: NO

And why do you insist on confounding Android with some of the ARM OS that are ready or are being developed?

Ask this question rocks: do you really believe that the ARM supercomputer that is being developed will run Android?

Who is developing an ARM OS? M$? Red Hat? Apple?

No, none of those are even considering an ARM DT OS. Canonical has a mobile OS called Ubuntu Touch that will run on ARM. Their DT version isn't nearly as capable as the x86 version.

Additionally...you are counting chickens *way* before they hatch. NVidia's so called ARM supercomputer has to have the chips delivered before it ever gets off the ground. Plus, that will likely be the lion's share of the ARM chips that NVidia sells for 3 quarters after that. Notice they haven't achieved any design wins in mobile formats? There's very much a reason for that. NVidia's ARM adventure cruise has run into a Hurricane called Qualcomm. That's not likely to change any time soon either.

Additionally, I would wager it still won't top the #1 Super Computer in the world at this point. 17 petaflops is a *LOT* of compute power. Little ARM cores would need thousands upon thousands to reach a number like that, even with Tesla GPGPUs.

You miss lots of people who are developing ARM desktop OS, including that link that was given to you in a previous reply.

Again: Mobile OS != ARM DT OS

Nvidia is already shipping dev kits for prototyping the ARM supercomputer and for developing software that will be run in the supercomputer. Nvidia already ported CUDA to ARM, for instance.

I notice that you missed how a pair of posts above I gave links to mobile formats (smartphones) that Nvidia won. In any case the phone market is irrelevant to the ARM supercomputer that is being developed in collaboration with Nvidia.

Since Nvidia custom ARM will be competing with Xeons and Opterons in raw performance (their claim) the supercomputer will have less cores than what you believe.

 

[jdwii]

Man i am sick of reading about Arm, Does anyone here even think Arm is going to become mainstream just curious.

 

[juanrga]

LOL. This is the fifth time that you ask me if Android is a full blown desktop OS and this is the fifth time that I say you: NO

And why do you insist on confounding Android with some of the ARM OS that are ready or are being developed?

Ask this question rocks: do you really believe that the ARM supercomputer that is being developed will run Android?

Who is developing an ARM OS? M$? Red Hat? Apple?

No, none of those are even considering an ARM DT OS. Canonical has a mobile OS called Ubuntu Touch that will run on ARM. Their DT version isn't nearly as capable as the x86 version.

Apparently, you are caught in an endless loop:

while 2 != 3 {
ARM is RISC
RISC is "incomplete" // (first mistake)
ARM is "incomplete" // (second mistake)
ARM cannot run a complete DT OS // (third mistake)
ARM only can run Android // (misguided conclusion)
}

I take note that you didn't offer any answer about if you believe that the ARM supercomputer will run Android or not.

 

[Cazalan]

EDIT: My understanding is that a migration to FD-SOI (UTBB most notably) is not out of the question for these parts. I couldn't get outright confirmation in the conversation; however, he said they were considering alternatives to bulk process and wouldn't state specifically which one. Based on the history they have with SOI, I am inclined to think it would be that direction...they could go to FinFETs, but I doubt that would make any sense given their production partners.

Everything at 14nm will be FinFET including Samsung, TSMC, GF, IBM, Intel.

Looks like only IBM is doing SOI at 20nm. AMD is cutting too much money from GF for them to ramp 20nm SOI. Bulk is the choice for 20/28nm.

 

[de5_Roy]

Spoiler

The thing that someone wants to ignore is the fact that SR-FX or whatever they decide to call it was never slated to be released before 2015. AMD has since released an APU roadmap and has been taken to mean that is the only products that will ever exist. No products other than those on the APU roadmap are in development. They also have gone as far as saying the existence of high end ARM will cancel the existence of low end x86.

AMD supposedly abandoning high end cpus leaves only middle performance APU for x86 support or Intel for ALL high end computing. ARM is not high end. Maybe 2, 5, 8 years from now, who knows how long that will take. Some people are convinced that is already here today based on personal opinion.

If this truthfully is the case, I may have purchased my last AMD cpu. I am not interested in a downgrade powered cpu to play games at medium to low settings. If ARM is the only choice to make in the future, ill go with Samsung or Qualcomm since AMD is forcing the choice (or rather lack of choice).

AMD needs to play both markets. They can't survive with a middle performance only x86 cpu, (think VIA) and ARM is full of competition already.

to make sure, i am discussing marketing p.r. and promo slides here, not exact engineering.
i've noticed that amd has seperate roadmaps (that they update regularly) for seperate products and platforms. the foremost might be the cpu uarch roadmap, then the cpu roadmap, followed by gpu roadmap, server (high end opterons), mainstream apu, professional/workstation gfx, mobile (laptop and smaller) roadmap. the newest addition is the microserver roadmap and the revamped embedded roadmap. having this many promo slides give amd (and the likes) enough leeway through their disclaimers and confusing marketing for pluasible deniability.
about low end - we speculated that apus like llano, trinity will cannibalize low end discreet gfx card sales within 2-3 years. now, 2-3 years later, low end dgfx is still alive and kicking, with new gfx cards taking their place. per trend, entry level price point offers more performance than before, but it's not dead.

....

ah. thanks for the corrections. ofc you're the one who'd remember the claims better than i. thanks.

EDIT: My understanding is that a migration to FD-SOI (UTBB most notably) is not out of the question for these parts. I couldn't get outright confirmation in the conversation; however, he said they were considering alternatives to bulk process and wouldn't state specifically which one. Based on the history they have with SOI, I am inclined to think it would be that direction...they could go to FinFETs, but I doubt that would make any sense given their production partners.

Everything at 14nm will be FinFET including Samsung, TSMC, GF, IBM, Intel.

Looks like only IBM is doing SOI at 20nm. AMD is cutting too much money from GF for them to ramp 20nm SOI. Bulk is the choice for 20/28nm.

those 14nm nodes will be for mobile socs (low power, low power high perf, ultra lp etc). afaik, only glofo has one (allegedly) working high performance node and it's 28nm. the only customer for that hpp node is amd (steamroller, excavator etc).
chances are, if amd doesn't make enough money from kaveri, they might have to optimize their apus for low power nodes or wait for (a long time) glofo to prepare smaller hpp nodes. glofo is making much more from their mobile processes. single, low earning customer vs multiple profitting customers and 'arm steering'.

 

[Cazalan]

A phone chip is one with a phone-class TDP rating. If you can introduce more performance than competence in that thermal package that is welcomed.

And what do most Tegra 3 and Tegra 4 chips go into? Tablets and their hand held gaming platform Shield. These are clearly more than just phone-class TDP parts. They're a family of parts. There is a Tegra 4 and a 4i. The 4i is the one going into phones. The Tegra 4 is going into Surface 2 and Shield.

http://www.techradar.com/news/phone-and-communications/mobile-phones/nvidia-s-tegra-3-to-run-30-smartphones-in-2012-1082199

http://www.digitaltrends.com/mobile/zte-geek-announced-again-becomes-the-worlds-first-tegra-4-smartphone/

And how many of those phones made it to production?

NVidia only lists 5 Tegra 2 phones, 2 Tegra 3 phones, and 1 Tegra 4 phone. NVidia continues to lose money on Tegra because their sales are so low.

http://www.nvidia.com/object/tegra-superphones.html

They turn down the clocks to get into phone-TDP range, and boost the clocks for tablet-TDP range. Just like Apple does with their A6/A7 parts for iPhone/iPad. That makes them tablet-TDP parts too.

 

[Cazalan]

Man i am sick of reading about Arm, Does anyone here even think Arm is going to become mainstream just curious.

With over 2 billion arm cores shipped per year they're already very mainstream.

Unless you mean mainstream for a PC replacement. That will depend on the OS. Microsoft doesn't seem to be making many strides with their Windows 8 RT. Almost all of those tablets have been dumped for Windows 8 Pro (x86) systems.

Apple has the best chance of making that happen, if they start replacing their Mac Books and Mac Pros with ARM chips and ship OSX 11 for ARM. I think they're still several years out from doing that but they may surprise. They got the money to do whatever they want.

 

[juanrga]

Everything at 14nm will be FinFET including Samsung, TSMC, GF, IBM, Intel.

Looks like only IBM is doing SOI at 20nm. AMD is cutting too much money from GF for them to ramp 20nm SOI. Bulk is the choice for 20/28nm.

Only to add that TMSC will make FinFet at 16nm and what GF itself feels about bulk process for 20nm:

It does not look like GF will go full bore into FD-SOI, and they will offer a 20 nm bulk/HKMG/planar product. GF will not go FinFET until 14 nm, but they feel their 22/20 nm bulk, planar process will be the right combination of performance and cost. IBM and Samsung are also pursuing this particular technology in their own Fabs. This looks in fact to be the primary process at this node, and only client demand will shift production from bulk silicon to FD-SOI.

I mean nobody would be perplexed if AMD carrizo is 20nm bulk for instance.

A phone chip is one with a phone-class TDP rating. If you can introduce more performance than competence in that thermal package that is welcomed.

And what do most Tegra 3 and Tegra 4 chips go into? Tablets and their hand held gaming platform Shield. These are clearly more than just phone-class TDP parts. They're a family of parts. There is a Tegra 4 and a 4i. The 4i is the one going into phones. The Tegra 4 is going into Surface 2 and Shield.

http://www.techradar.com/news/phone-and-communications/mobile-phones/nvidia-s-tegra-3-to-run-30-smartphones-in-2012-1082199

http://www.digitaltrends.com/mobile/zte-geek-announced-again-becomes-the-worlds-first-tegra-4-smartphone/

And how many of those phones made it to production?

NVidia only lists 5 Tegra 2 phones, 2 Tegra 3 phones, and 1 Tegra 4 phone. NVidia continues to lose money on Tegra because their sales are so low.

http://www.nvidia.com/object/tegra-superphones.html

They turn down the clocks to get into phone-TDP range, and boost the clocks for tablet-TDP range. Just like Apple does with their A6/A7 parts for iPhone/iPad. That makes them tablet-TDP parts too.

Well, those are enough phones to sustain my claim that Tegra 3/4 are phone-class TDP rated. There is no AMD Temash phones for instance.

 

[juanrga]

Meanwhile, I updated and uploaded my article on Steamroller/Kaveri. The final version with acknowledgements

http://juanrga.com/en/AMD-kaveri-benchmark.html

There are many corrections and improvements over the draft that presented here weeks ago.

I continue assuming that SR will be a bout a 20% faster than PD. However, some rumours suggest that could be up to a 40% faster.

Note: yes it is based in _marketing slides_, LOL, because no Kaveri sample has been shipped still. However, all the people I consulted agrees on that my estimations of performance look good. Real benchmarks will be included when the apu was ready.

Also I tweeted recently something that has been unnoticed in this forum. Since nobody more has reported this I will add here because is very important: AMD will be extending OGL to offer MANTLE-like performance

http://www.dsogaming.com/news/amd-aims-to-give-opengl-a-big-boost-api-wont-be-the-bottleneck/

 

[palladin9479]

Man i am sick of reading about Arm, Does anyone here even think Arm is going to become mainstream just curious.

Not in the DT world, not nearly enough software. That's the #1 reason, binary compatibility with existing software. Tablets and phones have taken over most communication aspects and those are perfectly fine for Arm. Even media consumption to an extent. The only reason Apple is even in the DT market is cause it jumped ship to x86 and thus can dualboot Windows / x86 Linux which gives it easy access to a ton of programs.

Seriously software accessibility is the #1 driver of the consumer market.

 

[juanrga]

Another game going 8 threads/cores

http://www.pcgamer.com/2013/10/08/watch-dogs-system-requirements-announced-specify-64-bit-os-and-minimum-6-gb-of-ram/

Recommended

* Processor: Core i7 3770 @ 3.5Ghz or AMD FX-8350 @ 4.0Ghz

 

[8350rocks]

EDIT: My understanding is that a migration to FD-SOI (UTBB most notably) is not out of the question for these parts. I couldn't get outright confirmation in the conversation; however, he said they were considering alternatives to bulk process and wouldn't state specifically which one. Based on the history they have with SOI, I am inclined to think it would be that direction...they could go to FinFETs, but I doubt that would make any sense given their production partners.

Everything at 14nm will be FinFET including Samsung, TSMC, GF, IBM, Intel.

Looks like only IBM is doing SOI at 20nm. AMD is cutting too much money from GF for them to ramp 20nm SOI. Bulk is the choice for 20/28nm.

Well, IBM uses GF as their production partner, along with STMicro's designs for UTBB.

If GF is doing SOI for IBM @ 20nm, then they could surely offer the same technology to AMD.

Additionally, FinFET @ 20nm and 14nm shows to be more expensive for the same performance based on the last SOI consortium meetings and the presentation from IBM/STMicro/GF.

So far as I know, the plan for GF is 28nm SHP is FD-SOI UTBB, 20nm SHP is FD-SOI UTBB, 14nm XM-Hybrid is some variety of FinFET using a hybrid FEOL with BEOL from the 20nm process.

I don't have to expound on the virtues of FD-SOI UTBB's back biasing capability...it's really a superior product in that regard. Also, IBM was discussing the possibility of UTBB FD-SOI all the way to something in the 7nm range. Much smaller and you could no longer use an insulator because you wouldn't have enough space.

 

[8350rocks]

LOL. This is the fifth time that you ask me if Android is a full blown desktop OS and this is the fifth time that I say you: NO

And why do you insist on confounding Android with some of the ARM OS that are ready or are being developed?

Ask this question rocks: do you really believe that the ARM supercomputer that is being developed will run Android?

Who is developing an ARM OS? M$? Red Hat? Apple?

No, none of those are even considering an ARM DT OS. Canonical has a mobile OS called Ubuntu Touch that will run on ARM. Their DT version isn't nearly as capable as the x86 version.

Apparently, you are caught in an endless loop:

while 2 != 3 {
ARM is RISC
RISC is "incomplete" // (first mistake)
ARM is "incomplete" // (second mistake)
ARM cannot run a complete DT OS // (third mistake)
ARM only can run Android // (misguided conclusion)
}

I take note that you didn't offer any answer about if you believe that the ARM supercomputer will run Android or not.

I never said ARM has to run Android, I said it won't be a viable DT PC architecture without a DT capable OS. Which it doesn't have, and one is not imminently on the horizon either.

Your loop seems to be:

If 0 != 1 {

ARM will replace x86 without a software environment

NVidia will magically become a player in ARM race, even though their Tegra projects are hemorrhaging money.

Seattle will magically rule the world, even though it's not built yet.

ARM has a DT capable OS, but no one can produce the evidence
}

Else {

The entire thread is wrong about ARM, and juanrga is right even though there is no real world numbers for Seattle, or a DT OS, or a software environment

}

Frankly, I am not taking speculation of Kaveri into account, only the benchmarks from the early ES's. Additionally, based on what I can project, we can make assumptions from the performance of PD and get rough ideas.

You are making assumptions attempting to extrapolate data from an ARM chip of a caliber not yet made. Additionally, the performance figures you quote are entirely speculation with no prior architecture example as a baseline. You sell this as though it is gold, and expect us to buy. Sorry, but if I am buying gold, I have to have it verified that I am getting what I am indeed thinking I am buying.

Can we get off the topic of ARM now?

 

[griptwister]

Is it possible that juanrga is using this topic just so that way we view his site?

Watch Dogs and 6Gb of RAM? Looks like I'll be cutting it close with 8Gigs.