Rumour: Bulldozer 50% Faster than Core i7 and Phenom II.

[Phynaz]

The technological barriers that must be surmounted in order to implement FinFET make implementing HKMG look trivial. I will be quite impressed if anybody else in the industry is shipping FinFET enabled IC's before 2017.

I had to bold the above, because I think what you just wrote hit me just as hard - and maybe harder - than the original announcement.

I mean making metal gates look trivial?

Yikes!

 

[Nemesis 1]

Really? Did Penryn go to 47.5W from 95W Conroes? Power was down slightly, but not by 50%.

YA really unless Intel is lieing . Which I very much doubt. Intel says on the 22NM tri-gate That at = clocks IB would consum 50% less power than equal clocked SB.Also if you read AT recent article NOTE it says Intel says IB at beginning of 2012.

 

[Nemesis 1]

2-3 yrs is generous. Those who work with this stuff are saying it'll be more like 5 yrs.

Look at the gap between Intel's debut of HKMG and that of AMD/GloFo. (It will be nearly 4yrs)

The technological barriers that must be surmounted in order to implement FinFET make implementing HKMG look trivial. I will be quite impressed if anybody else in the industry is shipping FinFET enabled IC's before 2017.

The primary challenge and biggest issue is simply money. It takes money, ridiculous amounts of it, to run the sort of development program that Intel operated in order to bring their FinFET's out of the laboratory and into a production environment (and we've yet to see if it is actually manufacturable, 22nm isn't released yet for a reason).

It is one thing to say "IBM and GloFo will likely have FinFET at 14nm" and quite another for IBM and GloFo to conjure up the billions and billions of R&D dollars that are necessary to get it done in 5yrs, let alone the money it would take to do it in 2-3yrs.

Intels next design is something all together differant than Tri-gate Intel and QinectQ


IDC can you say if IBMs FinFet can have multi sources and drains like Intels TRI gate . Also can IBM grow its fins for greater performance as Intels TRI-Gate can . I would assume that They could grow the Fin . But I have yet to see a Fin Fet design with multi source/drain . I have also yet to see anyone else produce a graph of a FinFet Using more than 2 gates . I have seen the 4 sided design drawn on paper but failed to see Who was working on said design .

 

[Lonbjerg]

Well ok, could you provide a source of that usage for today? Because as I said, I've never heard those numbers in usage from anyone but you in this thread and according to what you wrote "all companies use those numbers for their products".

I would like to see his sources too...

 

[jimbo75]

Gotta love the way the intel fanboys are scrambling to declare the 2600K "midrange" all of a sudden. Yeah you know what is coming.

 

[Vesku]

Gotta love the way the intel fanboys are scrambling to declare the 2600K "midrange" all of a sudden. Yeah you know what is coming.

It does seem an interesting stance to take considering it is the current top end of SB and makes the current Intel Extreme chips look even less appealing than it did pre-SB.

I've been "meh" on whether Zambezi will compete with current SB in single threaded stuff but more and more it seems like it just might. Any bets on the standard review suite of software being refreshed by many review sites shortly after Bulldozer launch?

 

[lol123]

Gotta love the way the intel fanboys are scrambling to declare the 2600K "midrange" all of a sudden. Yeah you know what is coming.

It's not exactly "all of a sudden" since Intel has been clear all along that Sandy Bridge would first appear at the low to mid-end and that SB-E on Socket 2011 would not appear until much later. The fact that with the current gap in the line-up, i7-2600K (with a newer architecture than i7-990X) happens to be the fastest choice for most applications does not make it Intel's high-end offering - it's not priced that way (i7-990x still costs 3 times more, and i7-970 twice as much) and it's not roadmapped that way. As I've already said, many high-end features (above all triple-channel memory and the higher number of PCI-E lanes) are lacking on the Socket 1155 platform precisely because that's the way Intel has segmented its desktop CPU line-up.

 

[Vesku]

Is SB-E suddenly scheduled to launch with Bulldozer now? Like it or not, 2600K is the top tier Sandybridge atm.

 

[lol123]

Is SB-E suddenly scheduled to launch with Bulldozer now? Like it or not, 2600K is the top tier Sandybridge atm.

Nobody has said anything to the contrary. Please consider reading the discussion in its entirety.

 

[jimbo75]

It's not exactly "all of a sudden" since Intel has been clear all along that Sandy Bridge would first appear at the low to mid-end and that SB-E on Socket 2011 would not appear until much later. The fact that with the current gap in the line-up, i7-2600K (with a newer architecture than i7-990X) happens to be the fastest choice for most applications does not make it Intel's high-end offering - it's not priced that way (i7-990x still costs 3 times more) and it's not roadmapped that way. As I've already said, many high-end features (above all triple-channel memory and the higher number of PCI-E lanes) are lacking on the Socket 1155 platform precisely because that's the way Intel has segmented its desktop CPU line-up.

What exactly does the higher number of PCI-E lanes do for gulftown? Nothing I can see.

What about all the new stuff SB brings that Gulftown lacks?

No wait I get it, Gulftown is "high end" because it renders SB to bits even though it loses everywhere else. Funnily enough, Thuban rendered Nehalem to bits but was still a total failure because it cant compete anywhere else. The hypocrisy of intel fanboys is staggering.

 

[lol123]

What exactly does the higher number of PCI-E lanes do for gulftown? Nothing I can see.

What about all the new stuff SB brings that Gulftown lacks?

No wait I get it, Gulftown is "high end" because it renders SB to bits even though it loses everywhere else. Funnily enough, Thuban rendered Nehalem to bits but was still a total failure. The hypocrisy of intel fanboys is staggering.

Gulftown is Intel's high-end because it's priced in the top range of their desktop CPU line-up and Intel has so far not released a Sandy Bridge CPU at the same market point, preferring instead to bring Sandy Bridge to the lower-end first with Socket 1155. There have been times where Celerons made excellent performance CPUs thanks to their overclockability but that didn't make them high-end or even mid-range processors all of a sudden. This has nothing do with AMD's release of Bulldozer so why you're calling people Intel fanboys is inexplicable. I'll just assume that what you're doing right now is trolling.

 

[Vesku]

Nobody has said anything to the contrary. Please consider reading the discussion in its entirety.

"I agree that the i7-2600K is a mid-end CPU, even though it has stellar performance and with the current line-up arguably has the best performance among Intel CPUs. However, remember that the Socket 1155 platform still lacks many features that Gulftown on Socket 1366 has and which SB-E on Socket 2011 will also have. "

It WILL be mid-end, right now it is high-end. Just as AMD x4 BE and x6 BE is their current high-end even though Bulldozer should be launching soon. If anyone had asked me what to get for a high-end enthusiast computer starting from SB launch to today I'd have said the 2600K with a top end motherboard.

 

[Tuna-Fish]

2-3 yrs is generous. Those who work with this stuff are saying it'll be more like 5 yrs.

Yea I agree. 2-3 years wasn't really realistic. It was a low end estimate of the time Intel will hold the absolute crown of semiconductor engineering. Again. Well, at 32nm GloFo got pretty close to them for all of 6 months -- that's better than anyone has done for the past few years.

I agree it can take a long time, but it comes across like no one else can ever have a similar technology. AMD and others will be at a disadvantage no doubt, but lets not think this is the "magic" that instantly kills the competition.

I don't think AMD will vanish because of it. More likely, Intel will do as before and let AMD set price floors, and try to collect as much of the profits as they can.

It's not just their process that makes their current processors better, process is just a part of it. I will say it certainly helps, but you can have the better process and still have a bad processor.

Yes. But other than Atom, Intel doesn't really do very bad processors any more. The finfets give Intel so much baseline advantage in speed and power consumption, that there might well not be enough room left in the microarch for AMD to make it up, even if AMD suddenly gets the optimal perfect designs. Intel processor designers still have to do their jobs, but it's like getting to compete in a marathon with a bicycle.

I'm really hoping that T-Ram really works out for AMD. (with, for some reason, no equivalent on-chip storage breakthrough for Intel.) That would give them their own manufacturing-related advantage, and caches that don't consume power when they are not being queried, and that are more than twice as dense as what Intel can muster, should get them a bit closer to competition. But to really get how much finfets help Intel -- this still wouldn't be as big an advantage.

 

[jimbo75]

You are cracking me up now. GF is not standing still, and intel has yet to prove that it can viably mass-produce FinFET's.

 

[Phynaz]

What would you accept as proof?

 

[nonameo]

You are cracking me up now. GF is not standing still, and intel has yet to prove that it can viably mass-produce FinFET's.

I'm sure they can. I don't think they'd announce something like that without some level of internal assurance.

 

[jimbo75]

Yeah it's not like they announced Larrabee was due imminently. Over and over. Oh wait.

 

[Tuna-Fish]

Yeah it's not like they announced Larrabee was due imminently. Over and over. Oh wait.

They could manufacture larrabee. It just sucked.

Frankly there are one or two things that Intel does fiendishly good, and in most of the others they just barely stumble from catastrophe into other.

Process tech is their core business, it's what they do. It's the reason they collect nearly all of the profits in x86. They claim they can manufacture finfets, I believe them. When they claim they can produce competent GPU's, (or the software to run them), I reserve judgement until I see them.

Similarly, while I have no issues believing that AMD has superior graphics, or a processor architecture that claws back the difference between Phenom and SNB, I'll believe that GloFo is catching up to Intel when I see it.

 

[gdansk]

Any bets on the standard review suite of software being refreshed by many review sites shortly after Bulldozer launch?

:hmm: I'm willing to bet they will. But I was hoping they would bias toward the future, with a much stronger multi-threaded emphasis.

 

[Idontcare]

IDC can you say if IBMs FinFet can have multi sources and drains like Intels TRI gate . Also can IBM grow its fins for greater performance as Intels TRI-Gate can . I would assume that They could grow the Fin . But I have yet to see a Fin Fet design with multi source/drain

Of course they can make multiple source/drain contacts, that part is not novel. If you can make one fin then you can make an array of fins all connected by the same gate.

You are asking about the "width" of the transistor, right? And the analog to xtor width in a finfet?

I have also yet to see anyone else produce a graph of a FinFet Using more than 2 gates .

I have seen the 4 sided design drawn on paper but failed to see Who was working on said design .

Samsung has patented a 5-sided finfet. TSMC has fabricated and published work on their Ω-Gate finfet (the channel looks just like an Ω symbol). Quad-gate is also referred to as "all around gate" and nanowire Finfet.

http://inst.eecs.berkeley.edu/~ee130/fa07/lectures/BJT_lec20-1.pdf

 

[Nemesis 1]

So does any of these companies have working proto types? Carbon nanowire I think we skipped ahead one . Intel has one in between if she bares fruit.

In December 2005, researchers at Intel and QinetiQ announced the development of a new, ultra-fast, yet very low power prototype transistor using indium antimonide (chemical symbol: InSb) that could form the basis of microprocessors and other logic products beginning in the second half of the next decade. The prototype transistor is much faster and consumes less power than previously announced transistors.

85nm Gate Length Enhancement and Depletion mode InSb Quantum Well Transistors for Ultra High Speed and Very Low Power Digital Logic Applications
Enhancement and Depletion mode InSb Quantum Well Transistors for High Speed and Low Power Logic Applications
File type/Size: PDF 716KB

Intel Makes Transistor Breakthrough Using New Materials
File type/Size: PDF 1.4MB


http://download.intel.com/technology/silicon/InSb_press_tech_presentation.pdf

85nm InSb Quantum-Well Transistors with Ultra High Speed Performance and Very Low Power Dissipation

File type/Size: PDF 770KB

Read an overview
File type/Size: PDF 167KB

of Intel's activities in nanotechnology from Intel fellow Robert Chau that appeared in the May 2005 issue of Nanotech Briefs.
You can also explore the following presentations and publication links to learn more about our transistor and nanotechnologies

 

[RobertPters77]

Getting back to BD and Llano prices.

Somethings I noticed; Llano and Bulldozer are both competing against sandy. Although in different price ranges.

Llano is getting it's own socket(FM1) and it's meant as the low-end replacement for AM3 proper, So Llano will be fighting against Sandy in the sub 200$ level. And llano has to contest with Nvidia on the gpu front and it might take away sales from the OEMs like XFX/Sapphire/Asus etc. So either expect alot of OEMs jumping ship/or going out of business. Or more likely, The radeon 7k series to be a huge boost performance wise.


Now on to BD. The FX4110 is priced right at the i5-2500's level. I'm guessing it might retail between the 2500 and the 2500k. With it being overclockable out of the box it will compete with the 2500k head to head. Although the clocks are unknown at this time. I'm guessing around the 3ghz mark.

The FX6110 has no equal and is meant to be a step above the 4110. So it has nothing to compete against it. Meaning it's in it's own segment. The FX6110 is a Question Intel has no Answer for yet. Or Amd is anticipating Intel dropping the 2600's price and the FX6110 is already there to fight it on that level.

Now the 81XX's. I'm guessing Amd is planning ahead and Making the FX8k's competition to Sandy Bridge-E. If that's the case than Amd probably has a hit on it's hands with Bulldozer.

In short: Llano FM1 vs Sandy LGA1155. And; Probably a decent performance upgrade for Next gen radeons. Bulldozer is meant to compete with SB-E or at the very least take away early adopter marketshare.

 

[Idontcare]

So does any of these companies have working proto types?

Yes, pathfinding is a necessity for everyone in the industry. Intel just has to be one step ahead of the pack in this game owing to the fact they are one step ahead in the node cycle. That doesn't mean everyone else is waiting on pins and needles to take lead from Intel.

Both HKMG and FinFET xtor designs have been bandied about as leading candidates to replace traditional planar CMOS for ages. There is nothing novel here in terms of the industry of process tech R&D, what is unique is the aggressiveness in terms of the timeline for inserting these technologies into HVM environments.

But again I will remind that this is too is not a trophy that is solely Intel's year after year, they were slow to adopt copper, slow to adopt low-k, delayed their transition to immersion litho, and never got onboard with SOI.

Everyone in this industry makes calculated risks, performs their cost-benefits assessments, does their pathfinding.

But somewhere along this discussion here you lost me in terms of the relevance.

 

[podspi]

I like how this thread is swaying from giddiness over rumored Bulldozer performance to writing AMD's (and everybody else') obituary due to Intel's "new" manufacturing advantage -- one they've held for a very long time.

I understand that FinFETs are quite the accomplishment (from a technical standpoint) and that even if these were planar 22nm transistors this would be quite an advantage, but is this really anything more than Intel's normal advantage? (They've had HKMG forever, AMD and the army of ARM manufacturers are still around).

Personally, I expect Intel to continue designing perfectly competent processors on the best process on earth, leading to more profit than you can shake a gold-coated stick at. Same as they have been for a while. Am I missing something?

 

[RobertPters77]

Is it just me or does the A8-3550P seem abit on the expensive side?

I was thinking they would retail around the ~150$ (~100$ for CPU, ~50$ for the GPU value wise) since its performance is probably like a Phenom II x4 ~2.7ghz or so and 5570 discrete card alongsides it.


Also yeah... I actually kinda hopeing/expect the Bulldozer to kick SandyBridges behinde.

I think that the a8-3550 is what you're looking for. The P is probably a better version. Maybe BE version with unlocked multiplier for cpu+gpu?

An Athlon X4 640+ 5570 cost about 160 on newegg(100 cpu/60 gpu). So it's about right.