Why doesn't Intel move to 3-5 year desktop CPU product cycles?

[Fjodor2001]

Since Intel currently is improving CPU performance by ~8% per year, why not move to a 3-5 year product cycle for desktop PCs? I mean, for most people it's pointless upgrading for anything less than atleast a ~50% performance improvement. Also, at current TDP levels further TDP decreases are not that important for desktop PCs, unless you're building a fanless AIO system or similar.

Moving to a 3-5 year product cycle would save a lot of development costs. Also it is not so strange actually, since that's the product cycle time Intel used to have earlier when going from 286->386->486->Pentium. And still, then the CPU performance improvements per year were actually higher than currently.

For reference on which Intel CPUs were released what year, see:

http://www.intel.com/pressroom/kits/quickrefyr.htm

 

[Zodiark1593]

As desktops and laptops share the same chips (just differing voltages and clock speeds), it would make very little sense to do otherwise.

 

[ShintaiDK]

Because your wishes is not the majorities wishes.

They should also stop releasing new cars I guess. Might have a new car release every 10 or 20 years then

 

[Remobz]

Because your wishes is not the majorities wishes.

They should also stop releasing new cars I guess. Might have a new car release every 10 or 20 years then

I need a car now

Funny..lol!

 

[VirtualLarry]

Aren't they effectively doing something like this, with Broadwell being BGA-only? With "Haswell refresh" scheduled for LGA1150 next year instead of broadwell in LGA form-factor?

 

[Arachnotronic]

Aren't they effectively doing something like this, with Broadwell being BGA-only? With "Haswell refresh" scheduled for LGA1150 next year instead of broadwell in LGA form-factor?

Winner, winner, chicken dinner.

 

[pcunite]

The current refresh system works because we don't all buy at the same time. I just put together an Ivy system because I don't like testing motherboard BIOS revisions.

 

[blackened23]

Because OEMs want a new product on a yearly basis. Remember, the vast majority of the market is not an enthusiast that upgrades every 6-12 months, this is something to keep in mind.

 

[Fjodor2001]

Aren't they effectively doing something like this, with Broadwell being BGA-only? With "Haswell refresh" scheduled for LGA1150 next year instead of broadwell in LGA form-factor?

Yes, I guess so. The question is if that's an intentional strategy that will be repeated going forward.

The current refresh system works because we don't all buy at the same time. I just put together an Ivy system because I don't like testing motherboard BIOS revisions.

That's transitional. If they moved to 3-5 year product lifecycles, we'd all be synced within 3-5 years. Also, since there would be fewer motherboard models, there would be more developers per motherboard model to sort out bugs, leading to more stable motherboards.

Because OEMs want a new product on a yearly basis. Remember, the vast majority of the market is not an enthusiast that upgrades every 6-12 months, this is something to keep in mind.

If the majority are not enthusiasts, then they only upgrade when they feel there's an actual need (e.g. because they feel they'll get a substantial performance improvement). So then why do OEMs want a new product on a yearly basis? The concept of "yearly models" should primarily attract enthusiasts.

 

[AustinInDallas]

Im sure that multibillion dollar corporation never thought of that. You should e-mail the ceo

 

[moonbogg]

Im sure that multibillion dollar corporation never thought of that. You should e-mail the ceo

Hey, the OP can't have the answers to all the worlds problems, because I have them.

Problem: Intel doesn't care about enthusiasts and people who aren't enthusiasts are stupid.

Heres the answer to the problem:


.

 

[Haserath]

They most likely still make more money on yearly releases than stopping them entirely.

 

[Lonyo]

How would it save development costs?
You end up spending 3~5 years designing potentially bigger improvements that potentially could end up getting you nowhere?
Improving step by step with small increments is probably more cost effective than trying for a 3 year overhaul/significant improvement, because you will go for bigger gains and have bigger failures.

Incremental improvements are more consistent and predictable than trying for a long overhaul redevelopment. Most of the time people do both, they have incremental upgrades annually, as well as overhauls less often, such as the Core architecture from Netburst, or AMD and NV with their GPU architectures.

 

[Fjodor2001]

Im sure that multibillion dollar corporation never thought of that. You should e-mail the ceo

Ergo, every corporation is already doing everything perfect. No corporation needs to change.

 

[Fjodor2001]

How would it save development costs?
You end up spending 3~5 years designing potentially bigger improvements that potentially could end up getting you nowhere?
Improving step by step with small increments is probably more cost effective than trying for a 3 year overhaul/significant improvement, because you will go for bigger gains and have bigger failures.

Incremental improvements are more consistent and predictable than trying for a long overhaul redevelopment. Most of the time people do both, they have incremental upgrades annually, as well as overhauls less often, such as the Core architecture from Netburst, or AMD and NV with their GPU architectures.

Not necessarily. See the car industry for example. They develop a completely new platform every 5 years or so. In-between it's just minor updates, changing the design of the tail lights or whatever.

Every larger redesign will have fixed costs associated with it.

Maybe we're seeing something similar take place with Intel CPUs. We got Haswell, then we'll just get a Haswell refresh as a minor mid-life kicker in 1-1.5 years (keeping uarch and process node), until we get the next major redesign in 2-3 years or so.

 

[Ajay]

Well, it looks like they well may switch to a two year cycle on thw desktop and annual updates for mobile. Maybe in another 10 years you'll get yout wish.

 

[Roland00Address]

If the majority are not enthusiasts, then they only upgrade when they feel there's an actual need (e.g. because they feel they'll get a substantial performance improvement). So then why do OEMs want a new product on a yearly basis? The concept of "yearly models" should primarily attract enthusiasts.

Because it is not about you, it is not even about the consumers.

Because new models help keep cashflow going and inventory turn over for the OEM.

When you have a 1 or 2 years of a model then it is hard for a oem/supplier (HP) and a reseller (Best Buy) to conduct business. The reseller will hold off ordering excess inventory, instead trying to negotiating a better deal from the supplier. They reseller (Best Buy) may still sell the computer year long for $999 but they are trying to buy it from the supplier for $750 instead of $800. They reseller (Best Buy) has negotiating power with the supplier (HP) by holding off purchasing inventory, furthermore they can further negotiate with the supplier (HP) by offering to buy from another supplier (ASUS) for $725 unless the supplier is willing to sell at $750. The second supplier (ASUS) may be willing to cut the reseller (Best Buy) a break for they are trying to get new market share.

A new model gives HP more negotiating power with Best Buy. Even if the only thing changed by HP is a higher model number (2700k instead of 2600k) the average person does not know this and will buy what the sales person tells them, their IT guy tells them, or based off price.

There is a reason why all the major OEMS (HP, Dells, Acer, Asus, etc) have a 3 month product cycle it helps them respond not just to the consumer marketplace but also it is a new renegotiating cycle with all their resellers.

-----------------------------

Now everything changes when the OEM is not a commodity but instead a name, a name that customers will pay extra for. A company like Apple or Bose actually get to dictate the terms to a reseller like Best Buy. They can only market said products a certain way, it must have its own aisle, and you can only use Apple signs etc.

It goes even further than that with some OEMS. The OEM may even state that the price of said item may not drop below a certain amount under MSRP (for example you can state $98 or $99 instead of $99.99 but never $89) or that if it drops below a said price you can not advertise the sale price. Some OEMs will allow you to state a price "is too low to show" but some of the most anal OEMS will require you to state "Major Brand Name" and then the price in printed advertisements and when you arrive at the store you see the product is a Samsung TV and not some crappy Coby.

Break the rules on OEMS that have leverage with their resellers and the OEMs will cut off supply from the reseller.

-----------------------------

This is how we get said nonsense

8800gt->9800gt->gt240->gt330 being the exact same card (with a die shrink happening during the process of the 9800 and the gt240 generation).
same thing with the
8800gts 512->9800gtx->gts150->gts250 (there was a minor mhz bump on some of these cards.)

Nvidia used the same core for 5 generations for the OEMs such as HP wanted nvidia to do so (note I am not trying to single Nvidia out, all the part suppliers do so for the OEMs will pay the suppliers to do so, I am using nvidia for the 8800gt comes to memory for it was such a good card and we didn't see a true successor to it for so long until the gtx460.)

 

[Fjodor2001]

Maybe in another 10 years you'll get yout wish.

Where have I stated what I "wish"? I'm just discussing the options available as the pace of desktop CPU improvements slows down.

What do you "wish" for yourself by the way? Monthly upgrade cycles with only a new product number and box color?

 

[Fjodor2001]

Because it is not about you, it is not even about the consumers.

Because new models help keep cashflow going and inventory turn over for the OEM.

When you have a 1 or 2 years of a model then it is hard for a oem/supplier (HP) and a reseller (Best Buy) to conduct business. The reseller will hold off ordering excess inventory, instead trying to negotiating a better deal from the supplier. They reseller (Best Buy) may still sell the computer year long for $999 but they are trying to buy it from the supplier for $750 instead of $800. They reseller (Best Buy) has negotiating power with the supplier (HP) by holding off purchasing inventory, furthermore they can further negotiate with the supplier (HP) by offering to buy from another supplier (ASUS) for $725 unless the supplier is willing to sell at $750. The second supplier (ASUS) may be willing to cut the reseller (Best Buy) a break for they are trying to get new market share.

A new model gives HP more negotiating power with Best Buy. Even if the only thing changed by HP is a higher model number (2700k instead of 2600k) the average person does not know this and will buy what the sales person tells them, their IT guy tells them, or based off price.

There is a reason why all the major OEMS (HP, Dells, Acer, Asus, etc) have a 3 month product cycle it helps them respond not just to the consumer marketplace but also it is a new renegotiating cycle with all their resellers.

-----------------------------

Now everything changes when the OEM is not a commodity but instead a name, a name that customers will pay extra for. A company like Apple or Bose actually get to dictate the terms to a reseller like Best Buy. They can only market said products a certain way, it must have its own aisle, and you can only use Apple signs etc.

It goes even further than that with some OEMS. The OEM may even state that the price of said item may not drop below a certain amount under MSRP (for example you can state $98 or $99 instead of $99.99 but never $89) or that if it drops below a said price you can not advertise the sale price. Some OEMs will allow you to state a price "is too low to show" but some of the most anal OEMS will require you to state "Major Brand Name" and then the price in printed advertisements and when you arrive at the store you see the product is a Samsung TV and not some crappy Coby.

Break the rules on OEMS that have leverage with their resellers and the OEMs will cut off supply from the reseller.

-----------------------------

This is how we get said nonsense

8800gt->9800gt->gt240->gt330 being the exact same card (with a die shrink happening during the process of the 9800 and the gt240 generation).
same thing with the
8800gts 512->9800gtx->gts150->gts250 (there was a minor mhz bump on some of these cards.)

Nvidia used the same core for 5 generations for the OEMs such as HP wanted nvidia to do so (note I am not trying to single Nvidia out, all the part suppliers do so for the OEMs will pay the suppliers to do so, I am using nvidia for the 8800gt comes to memory for it was such a good card and we didn't see a true successor to it for so long until the gtx460.)

That was a long text! If I interpret you correctly, what you're really saying is that Intel might be changing to a product cycle like I described earlier after all:

"See the car industry for example. They develop a completely new platform every 5 years or so. In-between it's just minor updates, changing the design of the tail lights or whatever.

Maybe we're seeing something similar take place with Intel CPUs. We got Haswell, then we'll just get a Haswell refresh as a minor mid-life kicker in 1-1.5 years (keeping uarch and process node), until we get the next major redesign in 2-3 years or so."

I.e. Intel will only make a major redesign every 2-3 years for desktop CPUs (instead of each year as before with the tick/tock strategy), and in-between just have cheap very minor upgrades to keep OEMs happy so they still can pursue the strategy you described in your comment.

 

[Ajay]

Where have I stated what I "wish"? I'm just discussing the options available as the pace of desktop CPU improvements slows down.

What do you "wish" for yourself by the way? Monthly upgrade cycles with only a new product number and box color?

Hey, no need to be hostile; I certainly wasn't being so.

I'm fine with buying an 8 core Haswell-E (if they don't suck) and holding onto to that for five years, then I'll see what's out there. If clock speeds stall or drop, as a recent quote by Mark Bohr seems to indicate - a 22 or 14nm CPU may be the end of the line. We could be waiting 20 years for some new materials and processes to become affordable to bring back performance to workstations, since folks will probably have maybe one AIO for serious work and no desktops by then (IMHO).

 

[Roland00Address]

"See the car industry for example. They develop a completely new platform every 5 years or so. In-between it's just minor updates, changing the design of the tail lights or whatever.

Maybe we're seeing something similar take place with Intel CPUs. We got Haswell, then we'll just get a Haswell refresh as a minor mid-life kicker in 1-1.5 years (keeping uarch and process node), until we get the next major redesign in 2-3 years or so."

I.e. Intel will only make a major redesign every 2-3 years for desktop CPUs (instead of each year as before with the tick/tock strategy), and in-between just have cheap very minor upgrades to keep OEMs happy so they still can pursue the strategy you described in your comment.

It is my belief Intel is going to move to a longer product cycle for desktops not because they can save money in such a way (at least directly) but instead so their new factory capacity will go first to laptops, laptop chips have higher aosp and thus higher margins, and mobile which have great margins and much greater volume and thus higher total profit then desktop chips.

There is very little performance increase you can get with desktops now a days via increasing mhz speeds. If you increase mhz speeds you increase tdp once you go past 3 ghz. Thus to improve speeds while keeping the same tdp you need an improved architecture or to increase the core count. Yet the same R&D for laptop chips also is the same R&D for desktop chips for they are the same core but with just increase mhz (and thus increased tdp and increased voltages). So you don't save R&D money by slowing down desktop production.

What you do save by slowing down desktop production is more spare capacity on the newest process node and you can make your desktop chips on the old process node. This new spare capacity will be then focused on mobile and laptop chips.

---------------------------------

How this all translates to model numbers and satisfying oems. My prediction Intel will not be increasing cpu speeds with a haswell refresh instead they will be gpu turbos for their integrated graphics. We might see an intel hd 4700 with 1450 mhz instead of 1350 for max turbo. The cpus will get new names (possible i5-3335 instead of i5-3330) to satisfy the oems So in effect we get something like this

Late 2014 possible 2015 at the exact same time.
14nm
Mobile Airmount (Atom's Silvermont successor),
Laptop Broadwell (Haswell Successor),
Desktop All in One Chip Broadwell (Haswell Successor)
22nm
Desktop Normal (Haswell Refresh with faster integrated graphics)

Note the Desktop All in One Chips will get a refresh this is because with All in Ones the OEMS like using laptop chips or the low powered variants of desktop chips. OEMs are also willing to pay more for better graphics and Intel can recoup higher AOSP by having Iris Graphics inside said all in one instead of nvidia or ati radeon dedicated cards. The OEMs will be fine with switching with Iris graphics for it will allow cheaper costs (2 chips vs 1), less points of failures, thinner design, better cooling, etc.

 

[ShintaiDK]

^^ We have "almost" a winner.

The highest margin is servers. yet they essentially get last.

Factory nodes goes to those that is most important to secure. For 14nm it will be the mobile broadwell line and airmont. Desktop and servers will wait.

 

[Smoblikat]

They would lose business if people do what I do. As soon as a new gen is out I upgrade to the previous gen stuff, the example is my sig.......we have the HD7000 series and Haswell, and I am using HD6000 series and IVB.

 

[Roland00Address]

^^ We have "almost" a winner.

The highest margin is servers. yet they essentially get last.

Factory nodes goes to those that is most important to secure. For 14nm it will be the mobile broadwell line and airmont. Desktop and servers will wait.

I agree the highest margins are servers. They get last though not because of the margin but because of all the certification and such that is necessary for servers. This takes time and thus makes it impossible to be the lead candidate for a new process node.

Remember the sandy bridge sata port bug (2 years ago) that caused a billion dollar recall for Intel. Intel purposefully goes slower with servers to make sure such a thing never happens with mission critical server parts. The only reason why intel can ask for such margins on server parts is that they are able to be mission critical.

 

[Ajay]

^^ We have "almost" a winner.

The highest margin is servers. yet they essentially get last.

Factory nodes goes to those that is most important to secure. For 14nm it will be the mobile broadwell line and airmont. Desktop and servers will wait.

Well, some of that, with servers, is validation time; that, and the fact that for traditional servers, the upgrade cycle is longer. That's changing with some companies like Google and Facebook, that are looking for fast, cheap clusters and it appears that they upgrade faster, but that may only be for expansion - I haven't looked into it enough. The server side of the business may be the last best hope for enthusiast's as well.

Otherwise I agree, good post Roland00Address.