rampant speculation - Haswell may see the convergence of desktop and laptop cpus

SB has a die shrink from 45nm to 32nm and they are spending the power savings on integrated components, mostly the GPU, and the higher clock speeds. They still managed to go from 130W TDP to 95W TDP (desktop/performance cpus) because of better turboboost / tdp management.

IB is supposed to work on power conservation and by Haswell the dis size is 16nm. That's half the size. conservatively, the TDP for desktop performance cpus should drop below 70W.

70W is only 5W (edit: oops! 15W) higher than extreme performance SB mobile processors.

Anyone else think we are nearing the end of differentiating CPUs on mobile/desktop/workstation and moving to just mobile/workstation ? By 2013/2014 barebones laptops might come back, and with a vengeance, because mobo companies will have a compelling reason to breech the market, and enclosures aren't that patentable.


best summation:
think about it, there is no 32nm 4 core right now. so there is a drop for 130w/45nm to ??/16nm. If they doubled the cores (to 8), they could still halve the tdp, with somewhere around 20% leftover die space for wider buses, caches, more gpu stuff, and 10% leftover wattage for clock speed (ie, they could do the same design in 16nm with 8 cores and a 65W TDP and a turboboost clock of 4.2Ghz, with double the floorspace for the gpu than in sb). that's
assuming they didnt want to improve the design over the next few years

if they put work in it, they can really push the clock rate.. but you need liquid nitrogen to get past 5Ghz even on the Celerons that dont have any of the large caches and things that limit clock rates. There's just not much head room in clock rate after SB.. 25-50% probably

So the TDP for the performance desktop CPU will approach today's performance laptop's TDP. So, why keep extra lines in the factory? That's real money.

 

its an interesting thought
but we should also consider the cooling systems. if something doesnt change drastically by haswell, then a desktop will still be more o/c'able than a comparable laptop.

 

oo yeah, the cooling systems and the video cards (on length if not on wattage) will still differentiate desktop versus mobile. But at the CPU level, it will be gone. And once that goes, the mobos are almost obliged to follow (there is a different chipset to support different cpus currently .. all you need is the standard interface cpus to work in a mobile TDP and a generic enclosure to start whiteboxing laptops)

ps, trvelbug -- thats an awesome lapotop you have

 

TDP is only one factor, and extreme mobile CPUs are 45W, not 65W. I doubt that the desktop market will merge with the laptop market that soon as desktops can always add more cores for more "power".

 

Thanks, I was wrong but the distance is only 15W TDP, not 25W TDP (your 45W above). ref: Sandy Bridge (microarchitecture) - Wikipedia, the free encyclopedia

 

The Core i7 EE's are 55W. He's still wrong, but I just felt like I should point that out. (EDIT: I'm late!)


Also, Intel's probably not going to keep reducing TDP; it's much more likely that they're going to take advantage of lower power consumption and heat production to bump up clock speeds.

 

The fault with your thinking is that the SB processors coming out are replacing LGA 1156, not LGA 1366. The processors with 130W TDP are the ones on LGA 1366. The TDP has not gone down with SB at all, remaining unchanged from LGA 1156 to its successor LGA 1155.

 

Sockets and mobo footprint will mean mobile CPUs will always have their own sockets.

The actual silicon design is the same.

Intel found the comfortable levels of power consumption on the desktop and notebook, CPUs in their current form will always scale with voltage as will power consumption.

Unless there is a dramatic shift the speed will always be limited by power consumption.

 

the wattage, the TDP, the clock rate, these these are corelated at the same scale. If you are gaining 33% die in nm, then you are gaining 33% in (clock | wattage). You'd have to expect Haswell to clock over 8Ghz to make up the difference, and 12Ghz at 11nm.

its just speculation but I think the design is too complex to let the processors scale past 5Ghz .. at least much past that -- air-cooled SB might get close to that mark with overclockers. after that you have to build out more cores. But that is hidden progress because software is not up to step with intel already, and they're just into 8 cores at the workstation level. (Can you imagine intel trying to go the other way, eg. to sell 64 core 16nm chips to mainstream users? the only software that can take advantage of that today is industrial in nature. also the furtherest they have proven they can reach so far is 48 core, the interior bus gets bogged down and becomes the clock-limit factor)

As others have pointed out here, TDP and power consumption are intimately related. Why keep a differentation at the processor level when its the same on the inside, and it costs real money to run extra lines in the factory?

 

You seem to be forgetting about the addition of more cores. Intel has 6- and 8-core Sandy Bridge desktop CPUs planned (though no specific models have been announced; the proof can be found in the Wikipedia link you posted if you scroll down a bit). Those are going to be the Extreme desktop CPUs with 130W TDPs. With Ivy Bridge and Haswell, Intel is phasing out desktop dual-cores. As nice as it would be, this convergence is just never going to happen. Intel's Extreme CPUs will most likely always have a 130W TDP unless they hit some type of wall with the number of cores they can put on a die soon, along with a limit to what else they can put on the CPU and a clock speed wall. Realistically speaking, there's absolutely no chance this will happen within only five years (11nm), let alone the release of Haswell.

 

I certainly can see something that happened with professional graphic cards. 3D games pushed the graphic cards so much that the technology inverted and Nvidia and ATI think about game cards first and then add the professional flavour and not the inverse. If laptops keep increasing their market and desktops continue loosing, CPU makers will go there despite initial inertia.

 

if they did that, they would have to have the software keep pace. software is slow .. heck, there's still 32bit versions of windows 7. if today they tried to make even 8 cores mainstream, with everything else the same, they would succeed in building a more powerful machine, with the same TDP, that would rock.

But it would only rock in 4U format because it's hard enough to utilize 4 cores (thats 8 threads, remember) today. How do they market a chip that technically aught to be faster but user experience says its the same as the one you already own?

ps> I'd wager a bet that intel decides to pass on 6 and 8 core non-workstation sandy bridge cpus.

 

think about it, there is no 32nm 4 core right now. so there is a drop for 130w/45nm to ??/16nm. If they doubled the cores (to 8), they could still halve the tdp, with somewhere around 20% leftover die space for wider buses, caches, more gpu stuff, and 10% leftover wattage for clock speed (ie, they could do the same cpu in 16nm with 8 cores and a 65W TDP and a turboboost clock of 4.2Ghz, with double the floorspace for the gpu than in sb). that's assuming they didnt want to improve the design over the next few years

today:
mainstream desktops: pci2, hdd, 1Gz ddr3, 2-4 core, 650W systems
perf desk: pci2, hot gpu, ssd + hdd, 1.3-1.6Ghz dual-ddr3, 4 core, 650-1000W systems
laptop: pci2, hdd or ssd, 1Ghz ddr3, 2-4 core, 220W systems

2011:
mainstream desktops: pci2, hdd, 1600 dual-ddr3, 2-4 core, 400W systems
perf desk: pci3, hot gpu, ssd + hdd, 1800 dual-ddr3, 4 core, 600+W systems
laptop: pci2, hdd or ssd, 1600 dual-ddr3, 2-4 core, 200W systems

2012:
pci3 all around
4-core all around
mainstream desktops: hdd or ssd, 2Ghz dual-ddr3, 266W systems
perf desk: hot gpu, ssd + hdd, 2.5Ghz dual-ddr3, 480+W systems
laptop: ssd, 2Ghz dual-ddr3, 150W systems

2013:
mainstream desktops & laptops: ssd, 2.5Ghz dual-ddr3, 4 core, 220W systems
perf desk: hot gpu, ssd + hdd, <3Ghz tri-ddr3, 6-8 core, 400+W systems

2014:
8 core all around
ssds all around
lightpeak or something like sata 12Gbps internal
pci4?
ddr4 -- you could differentiate quad perf dekstops, tri-channel desktops, and dual-channel laptops or something .. but why bother?

2015:
?? gonna have to move past pci3 by here for sure


just a generic prediction based on the trends

 

Then you would lose that bet since they already have 6 and 8 core desktop sandy bridge processors planned. Not to mention the fact that they already have 6 core desktop nehalems.

 

This. Again, look at the Wiki link you posted. If you don't see it, I'll show you a pic.

Furthermore, they also plan to release 12- and even 16-core CPUs in the future, and yes, they will hit consumer desktops someday, probably around the same time as they hit servers. We aren't suddenly going to start seeing desktops drastically reducing in power consumption. It's nice to speculate, but your speculations are totally baseless.

Also, I'd REALLY like to know where you got that 220W laptop power consumption rating from, especially since my midrange laptop is running off of a 90W power adapter now.

 

Yes. Also the OP's estimates for desktop computer power consumption, like everything else here, was made up out of thin air.

 

well my desktop wattage was based on System Builder Marathon, December 2010: $1000 PC : Let's Give Core i3 A Chance

my performance desktop wattage for today is based on my current desktop machine.

my laptop wattage is based on the batteries .. a battery that is rated for 100v-240v~ 2.2A-1.0A 50-60Hz (just a sample rating i found when I googled) is outputting 220-240W.

But the system wattage is different than the CPU. once the CPUs power and TDP become comparable, it is just plain more expensive to intel to continue to run seperate lines.

Sure they do. And they will call them their workstations class machines, just like the 12 cores that they make today. You are right, eventually they will trickle those highly multi-core systems to the mainstream, but don't expect to see 12-16 core desktop lines in the same year they roll out the 16core workstation cpu.

Again, if they just double the number of cores in the next two years, and double the Gpu cores from sb, they still will have more than twice the wattage to play with if they send their engineers on a two year vacation.

if they put them to work, they can really push the clock rate.. but you need liquid nitrogen to get past 5Ghz even on the Celerons that dont have any of the large caches and things that limit clock rates. There's just not much head room in clock rate after SB.. 25-50% probably.

The engineers have to do something with die-space increase, and the company has to come up with some valid reason to spark sales. Trying to lead a m, migration to laptops is a great motivation.

 

You are confusing possible power output with actual power output. Taking TDP at face value as power consumption is pretty bad itself, but this is even worse. Look at the power consumption benchmarks for the systems they build. Their $400 gaming PC is what I would call pretty mainstream. Even overclocked, and running furmark and prime95 to load the computer more than could ever be experienced normally, it was still only drawing 237 watts from the wall.

But there is no reason to believe that this will ever happen in the foreseeable future. Why would Intel want to sacrifice more cores and higher clock speeds to have one single consumer line? And I really don't think there are any monetary savings to be had. Intel has continued to diversify it microprocessor product lines. It is making many different varieties of silicon and putting them in more different types of packaging than it ever has before.

 

How does making desktops and laptops use the same socket give them something to do with the extra die space?

 

Wait who is confusing power output with possible power output? I just jotted a note about typical (and hypothetical) systems, with their wattage -- its natural to list the PSU's wattage as the system wattage, regardless of actual draw.

By extension, the same is true of laptops with their batteries.

yes, TDP is not power, in fact it is ambient radience of lost power, but there is a constant relation between power in and TDP out, relative to the AND/OR/NOT/XOR gates in the IC (the gates that lose a signal generate heat), and it's area. The design of the CISC instruction abstractions is pretty well defined, and the result is that TDP follows wattage relatively tightly.

The savings isnt so big but with margins every little bit counts. Intel won't pass on making 16, 32 and 64 core CPUs. They just will call those "workstation cpus" and put them on lines that generate 1-4% of their total CPU output. The desktop and mobile lines would continue to generate the bulk of their CPUs .. and because these are the lines with the highest output, any savings in the margins here is important... it translates into higher profit per chip and cheaper per-chip prices at the stores.

The mobile lines of today will also drop down to half their current TDP, but that will be repurposed for ultra-low watt and (maybe 2015-ish) even core i-series in talbets and cell phones. Yeah, I said it: a core i5 cell phone.

hhm, Intel's i-core heading for mobiles, what with all of Apple's i- foos are cellphone or tablet sized devices ... one wonders what kind of notes they pass amungst themselves in their strategic planning meetings.

 

Yes and the higher TDP threshold of desktops means that they will always have faster chips.

Also you are not going to see many LGA slots in a laptop since they are less secure than pin ones. The number of pins is desirable to keep low due to the footprint of the chip/socket and a lower power draw = less power/ground pins needed.

 

Please, won't see the first post again? I recapped some of my points there and Im afraid you might have missed the part about clock speeds being limited by the cache and the instruction set.

That's interesting but I'm not entirely sold on it. You are saying that the current sockets, which are µPGA-988 pin-outs, are as small as they can get? I'd like to see a reference to that because I've never seen anything about the actual number of G's that the pins can take.

They came up with those PGAs at the same time as the LGA 1156, and the 150 connector difference has to be just the power difference. It seems like nothing (other than a possible worry about the physical strength of the pins) is to stop them from doing a 2000 pin PGA socket to match the 2000 pin LGA socket they'll probably want for Haswell.

 

But the hex core Intel CPUs and AMD CPUs are some of the best clockers out there on air and under extreme conditions.

The only reason we don't have 32nm quads is it's a great differentiation feature for the hex and the limited capability they have to produce them.

What I mean about the sockets is look at the base. You have pads on desktop and pins on laptops. The pin count is also different. Putting 2000 pins on the base of a CPU is ****ing hard, thats why they have moved away from that.

If cache speeds or similar became a problem they would just move them onto their own clock domain. AMD's level III cache resides on the uncore, along with the memory controller which is why overclocking the uncore on them (AKA North bridge) yields such high gains.

 

This is complete bologna, for both desktops and laptops. You are completely making this up. And if you aren't, whoever you are taking a cue from has it wrong. Way wrong. So if you run a laptop on battery vs its wall power supply, it all of a sudden is of a system of a different "wattage?" Or the "wattage" of my desktop is only dependent on the specified maximum power output of the power supply? Nonsense.

False. While there of course TDP and power consumption are positively correlated, power consumption does not follow TDP "relatively tightly." TDP is extremely rough, erred always on the side of lots of caution, and often just applied to a whole family of chips operating within a large variance of voltages and frequencies.

Wrong. It is their differentiation of different processor lines that nets them more money in the first place. While it is true that having all of their processors on a single socket might save them a penny on each processor they make, having their current differentiation allows them to sell their different processors at far more than the cost of having separate processor lines. They have BGA and socket mobile varieties, mainstream desktop, and performance desktop, not to mention their previous generation processors they are still pumping out for desktops and laptops. Now look at the prices of comparable chips from the different product lines. If the difference is even $1 between any of them, Intel is making, not losing, money off of their different processor lines.

Stop, just stop it please. You just never stop making stuff up with absolutely nothing to back any of it up. You are now well beyond the realm of "speculating" and squarely in the realm of "dreaming."

 

Oh god, I didn't even read that last part! lol, i5 in a cellphone.

Hey, I know how they'll use that "extra die space." They'll throw wifi and mobile broadband modules! That's how we'll get it into cell phones! Yeah, and all of that with a 2W TDP. Die shrinks do make miracles, don't they?

Seriously, your point about decreasing TDPs, the point that lead you to making this thread, was disproven a while back. Then you made that random timeline, which was DOA. (Well, unless you can explain how my 90W power adapter is supplying 220W of power.) Look, I know it's hard to admit that you're wrong sometimes, but really, you don't have a clue what you're talking about. I mean, your entire point depends on Intel suddenly changing it's market strategy. Actually, it depends on Intel and AMD doing it because Intel would fall behind AMD if they stopped trying to increase performance while AMD pushes ahead.

I'm sorry about being so harsh, but you really are just embarrassing yourself, and if you keep going you're just going to end up annoying people and making them want to tear your points to shreds, even if it means making you look like a fool. At the very least, admit that these are just things that you wish would happen in the future. Who knows? Maybe it will happen, but there is absolutely zero chance of it happening in four years.

 

Yeah, seems right. I didn't say current TDP performance chips wouldn't exist at all, I speculated that the "desktop" and "mobile" lines would merge because the desktop TDP is headed towards where the mobile TDPs are today. The high end "workstation" line stuff will be more clearly differentiated, and remain limited capacity runs.

Supposed to roll out LGA2011 in the second half of 2011, I think. -- just a heads up

The pin out for the laptop sockets are nearly the same number as the pin outs for the desktop sockets that were designed at the same time. -- It's a good argument in any case .. I'm just holding out hope.

At some point you reach the problem that heat in local areas on the chip is far greater than the heat over the remainder of the chip. The cpu cooling system is only expected to carry the TDP , that's the average heat from the whole chip .. when the difference between overall heat and local heat is too great, that local part of the chip can fail. This is why we don't have Terrahertz processors today, I guess.

So, regardless of why they hit ceilings, make a prediction, how high can they go with current clock rates? 5Ghz is my guess. My challenge is this, if you guess less than 12 Ghz, you either have to defend even more cores than 8 on desktop run-of-the-mill mainstream cpus, or come up with some other way besides just new cores, to spend that die, otherwise you have to lower the TDP by the same proportion as your Ghz is to 12.

 

Look, I cede -- people (at least the 3 of you) disagree, even though not for the same reasons. It wasn't billed as a certainty .. read the title. It's my rampid speculation.

 

Desktop processor TDP is NOT headed down. I don't know where you get this from other than your own imagination.

Stop making stuff up. Seriously. Just cut it out. Either that, or hand your computer engineering diploma back to the institution that granted it to you.

Aside from the Netburst fiasco, Intel's (and AMD's) processors have continuously gotten more and more powerful clock for clock over their predecessors. Intel's options to increase performance are not only to crank up the clock rate or add more cores.

 

Intel have stated that each feature should give a 1% performance increase for each 0.5% increase in power as a target.

Also there are no CPUs with 2000 pins, only 2000 pads. Pads are a bad idea for notebooks.

 

That seems to be too harsh. Intel said similar things as he does in that they are facing a contineous challenge of getting the heat out of the ever shrinking 'line width', they compared it with cooling a neuclear reactor.

 

No, what you are referring to is that CPU dies put out lots of heat on such a small area that cooling them is difficult. In terms of heat output per unit of area, CPU dies do rank up there pretty high. Around nuclear reactor, but below rocket nozzle. Or at least something like that. What roberto.tomas said has nothing to do with that and was entirely a creation of his own imagination. It is like he is making up computer engineering theories.