How much would soldered-on CPUs reduce the cost of motherboards?
In as much as memory modules seem to only work for one generation
of motherboards, how much would soldering on, say, 1GB of RAM
reduce costs? (By "reduce cost", I mean as compared to the total of
an equivalent motherboard with non-soldered-on CPU/memory).

Tony

"tony" <(E-Mail Removed)> writes:
>How much would soldered-on CPUs reduce the cost of motherboards?
>In as much as memory modules seem to only work for one generation
>of motherboards, how much would soldering on, say, 1GB of RAM
>reduce costs? (By "reduce cost", I mean as compared to the total of
>an equivalent motherboard with non-soldered-on CPU/memory).

I once asked a person who worked for Intel why they didn't move in
the direction of bringing a large block of memory onto the CPU chip
as a separate silicon die, sort of like the two die in the PPro,
but that wasn't providing main memory in that case.

For each generation of CPU and each market level I think you could
make a claim that there is a generally accepted level of memory.
If this were incorporated into the CPU package then the length of
the traces to the memory would be drastically shorter and thus
faster, they would be free to have wider paths to memory if that
were economical, the whole embarrassing issue of RIMM vs SDRAM might
have been hidden inside the package and the motherboard mfgrs could
have ignored most of that, even "specialty" versions of memory
might then be developed to handle latency, caching and overlap.

Intel would then make the profit from the sale of this memory in
the part, rather than having someone else get that. Furthermore,
this might dramatically reduce demands on additional external
memory bandwidth and thus the chipset folks could benefit from
this. And once most of the wide high bandwidth traffic to the
outside memory goes away you have lots more bandwidth to use for
the other parts of the system.

On Fri, 24 Mar 2006 15:41:18 -0600, Don Taylor <(E-Mail Removed)> wrote:

>"tony" <(E-Mail Removed)> writes:
>>How much would soldered-on CPUs reduce the cost of motherboards?
>>In as much as memory modules seem to only work for one generation
>>of motherboards, how much would soldering on, say, 1GB of RAM
>>reduce costs? (By "reduce cost", I mean as compared to the total of
>>an equivalent motherboard with non-soldered-on CPU/memory).
>
>I once asked a person who worked for Intel why they didn't move in
>the direction of bringing a large block of memory onto the CPU chip
>as a separate silicon die, sort of like the two die in the PPro,
>but that wasn't providing main memory in that case.
[snipped]

Is it safe to assume that whatever response you received was sufficient to
make the rest of your post moot?

It's easy to cobble up concepts when they are divorced from reality...

/daytripper

tony wrote:
> How much would soldered-on CPUs reduce the cost of motherboards?
> In as much as memory modules seem to only work for one generation
> of motherboards, how much would soldering on, say, 1GB of RAM
> reduce costs? (By "reduce cost", I mean as compared to the total of
> an equivalent motherboard with non-soldered-on CPU/memory).
>
> Tony
>
>
Any cost savings would be minumal. It might even be much more expensive.


It would be necessary to maintain a stock of boards with each different
supported processor and speed. Given that the processor tends to be more
expensive than the basic motherboard, this increases the inventory costs
from the motherboard manufacturer to the system integrator.

Configurations (what processor to use and speed) would need to be known
when the motherboard is manufactured, not at system integration time.

Also it would be impossible to upgrade processors or motherboards
without doing both at the same time.

If a component fails, motherboard or processor, the replacement cost is
now both, not one or the other.

Heatsinks would need to be redesigned.

For memory, the same is true. Just multiply the number of
processor/speed combinations by the different amounts of memory that
systems might be sold with. Now you have a stocking nightmare.

If you want to be able to upgrade the memory, a socket is needed anyway.

Consider one motherboard, Pentium and Celeron processors, say, 4
different speeds for each. Add 4 different amounts of memory. If they
are all soldered on the board...
1 board
x 2 cpus
x 4 speeds
x 4 memory capacities
------
32 items to stock

Now as separate items

1 board
+ 8 cpus
+ 2 memory stick types
-----
11 different items to stock


Now also figure that sockets are quite cheap in the high volumes they
are purchased.

This only makes sense if there are very few configurations and high
volume to justify the effort. (For example, the card on a disk drive.)

craigm

"craigm" <(E-Mail Removed)> wrote in message news:0A_Uf.978$(E-Mail Removed)...
> tony wrote:
>> How much would soldered-on CPUs reduce the cost of motherboards?
>> In as much as memory modules seem to only work for one generation
>> of motherboards, how much would soldering on, say, 1GB of RAM
>> reduce costs? (By "reduce cost", I mean as compared to the total of
>> an equivalent motherboard with non-soldered-on CPU/memory).
>>
>> Tony
> Any cost savings would be minumal. It might even be much more expensive.
>
>
> It would be necessary to maintain a stock of boards with each different supported processor and speed. Given that the
> processor tends to be more expensive than the basic motherboard, this increases the inventory costs from the motherboard
> manufacturer to the system integrator.
>
> Configurations (what processor to use and speed) would need to be known when the motherboard is manufactured, not at system
> integration time.
>
> Also it would be impossible to upgrade processors or motherboards without doing both at the same time.
>

Whoaaa! The assumption was that the processor-upgrade game is over and
that speed hasn't been "important" since it has been measured in MHz.

> If a component fails, motherboard or processor, the replacement cost is now both, not one or the other.

But that rarely happens. Why impose on all pieces for the exceptional case?

> Heatsinks would need to be redesigned.

They are continually being redesigned already. Since the motherboard is
being redesigned to have a soldered-on CPU, this really is a moot point.

> For memory, the same is true. Just multiply the number of processor/speed combinations by the different amounts of memory
> that systems might be sold with.

I have a feeling a large percentage of the buyers would choose the standard
configuration. Let the one's having special requirements pay the price of those.
Of course, having soldered-on boards doesn't mean the other kind have to end.
It's just a furthering of the "integrated board" concept (onboard audio/video/LAN).

> Now you have a stocking nightmare.

On the contrary, I have a feeling the demand for "high-end" boards with all the
"replaceable" components would subside. How many people buy separate
preamplifier/amplifier/tuner anymore? Most buy integrated receivers.

> If you want to be able to upgrade the memory, a socket is needed anyway.

Memory would be harder to do as soldered-on, but not impossible. For the
past few years, if you put a Gig in most of the boards, they would be fine.
The problem is where that is overkill.

[contrived, unlikely manufacturer scenario omitted]

> This only makes sense if there are very few configurations and high volume to justify the effort.

Now you're beginning to understand.

Tony

On Fri, 24 Mar 2006 19:17:01 GMT, "tony" <(E-Mail Removed)> wrote:

>How much would soldered-on CPUs reduce the cost of motherboards?
>In as much as memory modules seem to only work for one generation
>of motherboards, how much would soldering on, say, 1GB of RAM
>reduce costs? (By "reduce cost", I mean as compared to the total of
>an equivalent motherboard with non-soldered-on CPU/memory).

You CAN reduce costs but soldering chips on motherboards, and this is
indeed done in some situations. However processors have long been a
differentiating point between many systems, and soldering chips onto
the system board complicates things a lot. As a general rule,
soldering chips onto your motherboard makes sense if you've got a
fairly specific application in mind (ie home theater computers
perhaps?) where a "good enough" level of processor performance is
sufficient and it's other features that really determine the worth of
the system, but otherwise you're probably better off with socketed
chips.

Now, as to how much money it actually saves? Well that's a tough one
to judge. The cost of the socket itself is pretty small, all things
considered. The real advantage would be that you could make a
somewhat cheaper and simpler package for the processor. However if a
company needs to make two versions of the chip, then this advantage is
rather lessened. So where does this leave us in the end? I really
don't know, but I suspect that you're looking at less than a $10
savings on components, possibly less than $5. Hence the reason why it
only really makes sense on the low-end of things at this time.

Of course, as the nature of computer's evolve, this may change. If
you look at gaming consoles you'll see that the processors are pretty
much always soldered onto the board. If, as many people have been
professing, gaming consoles and PCs are well along the process of
merging into one, then we may start to see a lot more "PCs" with
soldered processors.

The story is pretty much the same with integrated memory. Soldering
memory on-board would provide an even greater benefit here from
up-front cost savings and performance points of view, but also poses
even greater restrictions in terms of customizing of common PCs.
There's also the issue of warranty support. Where CPUs have very low
failure rates, memory has fairly high failure rates. Having to
replace a whole motherboard because a single memory chip has gone bad
can really eat into your up-front cost savings.

-------------
Tony Hill
hilla <underscore> 20 <at> yahoo <dot> ca

tony wrote:
>
[...]
>
> I have a feeling a large percentage of the buyers would choose the standard
> configuration. Let the one's having special requirements pay the price of those.
> Of course, having soldered-on boards doesn't mean the other kind have to end.
> It's just a furthering of the "integrated board" concept (onboard audio/video/LAN).

Yes, that would be fine for a bread-and-butter computer with a low or middle
grade CPU for internet and common tasks. In practice it's difficult to make a
substantial CPU upgrade on an old board; if at all possible it may need BIOS
upgrade. And due to power considerations, you couldn't always upgrade a Nothwood
P4 to a Prescott P4.

Hence I believe that few end-users would ever upgrade the CPU. The problem may be
more acute for the system builders as there may be a delay between motherboard
and processor releases, and they would like to use the most cost effective
processor version at the time of build.

[...]
>
> Memory would be harder to do as soldered-on, but not impossible. For the
> past few years, if you put a Gig in most of the boards, they would be fine.
> The problem is where that is overkill.

My laptop from 1999 came with 32MB soldered on the motherboard plus two SODIM
sockets. Hence the odd amount of 288MB in total.

[...]

tony wrote:
> "craigm" <(E-Mail Removed)> wrote in message news:0A_Uf.978$(E-Mail Removed)...
>
>>tony wrote:
>>
>>>How much would soldered-on CPUs reduce the cost of motherboards?
>>>In as much as memory modules seem to only work for one generation
>>>of motherboards, how much would soldering on, say, 1GB of RAM
>>>reduce costs? (By "reduce cost", I mean as compared to the total of
>>>an equivalent motherboard with non-soldered-on CPU/memory).
>>>
>>>Tony
>>
>>Any cost savings would be minumal. It might even be much more expensive.
>>
>>
>>It would be necessary to maintain a stock of boards with each different supported processor and speed. Given that the
>>processor tends to be more expensive than the basic motherboard, this increases the inventory costs from the motherboard
>>manufacturer to the system integrator.
>>
>>Configurations (what processor to use and speed) would need to be known when the motherboard is manufactured, not at system
>>integration time.
>>
>>Also it would be impossible to upgrade processors or motherboards without doing both at the same time.
>>
>
>
> Whoaaa! The assumption was that the processor-upgrade game is over and
> that speed hasn't been "important" since it has been measured in MHz.
>
>
>>If a component fails, motherboard or processor, the replacement cost is now both, not one or the other.
>
>
> But that rarely happens. Why impose on all pieces for the exceptional case?
>
>
>>Heatsinks would need to be redesigned.
>
>
> They are continually being redesigned already. Since the motherboard is
> being redesigned to have a soldered-on CPU, this really is a moot point.
>
>
>>For memory, the same is true. Just multiply the number of processor/speed combinations by the different amounts of memory
>>that systems might be sold with.
>
>
> I have a feeling a large percentage of the buyers would choose the standard
> configuration. Let the one's having special requirements pay the price of those.
> Of course, having soldered-on boards doesn't mean the other kind have to end.
> It's just a furthering of the "integrated board" concept (onboard audio/video/LAN).
>
>
>>Now you have a stocking nightmare.
>
>
> On the contrary, I have a feeling the demand for "high-end" boards with all the
> "replaceable" components would subside. How many people buy separate
> preamplifier/amplifier/tuner anymore? Most buy integrated receivers.
>
>
>>If you want to be able to upgrade the memory, a socket is needed anyway.
>
>
> Memory would be harder to do as soldered-on, but not impossible. For the
> past few years, if you put a Gig in most of the boards, they would be fine.
> The problem is where that is overkill.
>
> [contrived, unlikely manufacturer scenario omitted]
>

Contrived? Look at the number of different PC configurations offered by
Dell or HP.

>
>>This only makes sense if there are very few configurations and high volume to justify the effort.
>
>
> Now you're beginning to understand.

You are defining a market that doesn't exist today (at least, in the
US). There may be a marketplace where this makes sense in other areas,
but you would have to find lots of buyers.


>
> Tony
>
>

"craigm" <(E-Mail Removed)> wrote in message news:y4aVf.1511$(E-Mail Removed)...
> tony wrote:
>> "craigm" <(E-Mail Removed)> wrote in message news:0A_Uf.978$(E-Mail Removed)...
>>
>>>tony wrote:
>>>
>>>>How much would soldered-on CPUs reduce the cost of motherboards?
>>>>In as much as memory modules seem to only work for one generation
>>>>of motherboards, how much would soldering on, say, 1GB of RAM
>>>>reduce costs? (By "reduce cost", I mean as compared to the total of
>>>>an equivalent motherboard with non-soldered-on CPU/memory).
>>>>
>>>>Tony
>>>
>>>Any cost savings would be minumal. It might even be much more expensive.
>>>
>>>
>>>It would be necessary to maintain a stock of boards with each different supported processor and speed. Given that the
>>>processor tends to be more expensive than the basic motherboard, this increases the inventory costs from the motherboard
>>>manufacturer to the system integrator.
>>>
>>>Configurations (what processor to use and speed) would need to be known when the motherboard is manufactured, not at
>>>system integration time.
>>>
>>>Also it would be impossible to upgrade processors or motherboards without doing both at the same time.
>>>
>>
>>
>> Whoaaa! The assumption was that the processor-upgrade game is over and
>> that speed hasn't been "important" since it has been measured in MHz.
>>
>>
>>>If a component fails, motherboard or processor, the replacement cost is now both, not one or the other.
>>
>>
>> But that rarely happens. Why impose on all pieces for the exceptional case?
>>
>>
>>>Heatsinks would need to be redesigned.
>>
>>
>> They are continually being redesigned already. Since the motherboard is
>> being redesigned to have a soldered-on CPU, this really is a moot point.
>>
>>
>>>For memory, the same is true. Just multiply the number of processor/speed combinations by the different amounts of memory
>>>that systems might be sold with.
>>
>>
>> I have a feeling a large percentage of the buyers would choose the standard
>> configuration. Let the one's having special requirements pay the price of those.
>> Of course, having soldered-on boards doesn't mean the other kind have to end.
>> It's just a furthering of the "integrated board" concept (onboard audio/video/LAN).
>>
>>
>>>Now you have a stocking nightmare.
>>
>>
>> On the contrary, I have a feeling the demand for "high-end" boards with all the
>> "replaceable" components would subside. How many people buy separate
>> preamplifier/amplifier/tuner anymore? Most buy integrated receivers.
>>
>>
>>>If you want to be able to upgrade the memory, a socket is needed anyway.
>>
>>
>> Memory would be harder to do as soldered-on, but not impossible. For the
>> past few years, if you put a Gig in most of the boards, they would be fine.
>> The problem is where that is overkill.
>>
>> [contrived, unlikely manufacturer scenario omitted]
>>
>
> Contrived? Look at the number of different PC configurations offered by Dell or HP.

But that is under the "old" model where MHz rules and people buy into the
marketing strategy of upgradeability/replaceability. I don't think anyone is
likely to bite on "oh, and your motherboard allows upgrading of the CPU"
spiel anymore.

>>>This only makes sense if there are very few configurations and high volume to justify the effort.
>>
>>
>> Now you're beginning to understand.
>
> You are defining a market that doesn't exist today (at least, in the US). There may be a marketplace where this makes
> sense in other areas, but you would have to find lots of buyers.

On the contrary, I think the market for replaceable/upgradeable CPUs has been
dead for quite some time. It's just that the buyer hasn't had the choice to choose
the more integrated offering.

Tony

"Tony Hill" <(E-Mail Removed)> wrote in message news:(E-Mail Removed)...
> On Fri, 24 Mar 2006 19:17:01 GMT, "tony" <(E-Mail Removed)> wrote:
>
>>How much would soldered-on CPUs reduce the cost of motherboards?
>>In as much as memory modules seem to only work for one generation
>>of motherboards, how much would soldering on, say, 1GB of RAM
>>reduce costs? (By "reduce cost", I mean as compared to the total of
>>an equivalent motherboard with non-soldered-on CPU/memory).
>
> You CAN reduce costs but soldering chips on motherboards, and this is
> indeed done in some situations. However processors have long been a
> differentiating point between many systems, and soldering chips onto
> the system board complicates things a lot. As a general rule,
> soldering chips onto your motherboard makes sense if you've got a
> fairly specific application in mind (ie home theater computers
> perhaps?) where a "good enough" level of processor performance is
> sufficient and it's other features that really determine the worth of
> the system, but otherwise you're probably better off with socketed
> chips.

Not very convincing.

> Now, as to how much money it actually saves? Well that's a tough one
> to judge. The cost of the socket itself is pretty small, all things
> considered. The real advantage would be that you could make a
> somewhat cheaper and simpler package for the processor. However if a
> company needs to make two versions of the chip, then this advantage is
> rather lessened. So where does this leave us in the end? I really
> don't know, but I suspect that you're looking at less than a $10
> savings on components, possibly less than $5. Hence the reason why it
> only really makes sense on the low-end of things at this time.

$5 or $10 indeed doesn't sound like much. But maybe in volume (Dell)?

> Of course, as the nature of computer's evolve, this may change. If
> you look at gaming consoles you'll see that the processors are pretty
> much always soldered onto the board. If, as many people have been
> professing, gaming consoles and PCs are well along the process of
> merging into one, then we may start to see a lot more "PCs" with
> soldered processors.

Especially if small form factor becomes standardized.

> The story is pretty much the same with integrated memory. Soldering
> memory on-board would provide an even greater benefit here from
> up-front cost savings and performance points of view, but also poses
> even greater restrictions in terms of customizing of common PCs.
> There's also the issue of warranty support. Where CPUs have very low
> failure rates, memory has fairly high failure rates. Having to
> replace a whole motherboard because a single memory chip has gone bad
> can really eat into your up-front cost savings.

Yes, that would be bad. Hasn't memory become fairly robust these days though?

Tony