 | |
Intel working on Z-RAM too
Hardware Chips Processor, cache, memory chips, etc. (comp.sys.ibm.pc.hardware.chips)
 |
| | Thread Tools | Search this Thread | Display Modes |
#21
 
| ||||
| ||||
|
Re: Intel working on Z-RAM too -
12-13-2006
, 02:39 PM
|
I think childishness is unique to Intel. LOL. Right...I think you have a lot to learn about the world then. All these companies have been childish. "native quad core"? |
irrational refusal to make use of something, just because another person
had it before you.
Quote:
|
Other times it's something like trying to come up with a Hypertransport work-alike, called CSI, which still somehow manages not to be compatible between all of its processor families, What does 'compatible between all CPU families' even mean in that sentence? Did you seriously expect IPF and x86 to fit into the same sockets? Or is the comment in relation to Xeon MP vs. DP? I guess I'm not sure what the complaint is. |
Yes, of course I expected IPF and x86 to fit into the same socket,
because that's exactly what Intel has been saying for years it was
trying to achieve!
Now before you start reaching into your bag of Intel apologist tricks.
I'm sure you're now going to say, "I've had press releases from Intel
for years, and they never said that". And then you'll challenge us to
produce links which point to Intel ever saying it was trying to make
Itanium and Xeon fit into the same motherboards. Well, let me just stop
you short right there, we've played that crap before, and we're just not
going to do it anymore. A lot of people around here can confirm that's
exactly what they've heard coming out of Intel's mouth before too.
Anyways, both IBM and Sun have announced that they're going to be making
their own Power and Sparc processors compatible with the Opteron socket
as part of Torrenza initiative. I think Cray has even said they'll try
to get their vector processor compatible with Opteron. So if two or
three outside companies are confident they can get their incompatible
architectures compatible with AMD's, then why would Intel have trouble
getting two of its own internal architectures to work with each other?
Quote:
|
Why would Intel want to license HT? They'd need to develop their own cache coherency protocol anyways, so using different (and better) lower layers isn't a big deal. Also, it isn't clear to me that HT is suitable for the entire range of applications that Intel is targeting (that being said, I know of no evidence that it is unsuitable). |
Opteron platform than anything in Intel's planbook. So far, we got GPUs,
vector processors, two different RISC processors, DSPs, etc. You can't
get much more diverse than that.
Quote:
|
Now it's all of this effort to recreate Z-RAM on bulk silicon. They are still going to have to add some SOI ontop of the bulk silicon to get this to work, now what's the point of going through all of this modification of bulk silicon to act like SOI, when SOI itself is around? And if they've announced they're going to SOI anyways, this tech will only have limited usage before they convert to SOI. They announced they'll be going to FD-SOI at some point in time. That could be 32 or 22nm for all we know... I really see no good reason for Intel to convert to PD-SOI if they are just going to be using it for a relatively short period of time. There needs to be a substantial ROI there. Also, the benefits of PD-SOI are less for smaller geometries than larger ones...so while it might represent a 1/2 generation in terms of switching speed at 130nm, it could be 1/3 or 1/4 generation at 65 or 45nm (I don't have hard numbers at my finger tips right now). |
it was Intel that was the only one talking about going to PD-SOI, if
they ever went.
Yousuf Khan
#22
| |||
| |||
|
|
David Kanter wrote: I don't think Return On Investment is even in Intel's vocabulary. They've been swimming in monopoly money for so long that they're probably just now figuring out ROI. |
corporations do. You just have to remember, the folks who make the ROI
calculations at Intel actually have the best information possible. You
and I don't.
Quote:
|
AMD produces at most 1/4 the silicon Intel does. IBM's even lower volume. This means that Intel generally is going to be more sensitive to increases in cost of their processes than AMD... Or exactly the opposite. Fabs are fixed cost outlays, the more you produce past your break-even point, the less you have to consider cost. |
processing steps, which slow down production, SOI wafers are incredibly
fragile. etc. etc.
Hell, you heard Del say that there was a huge internal fight over
whether to use SOI or not. They aren't using SOI for everything, just
the high-end stuff.
SOI adds substantial costs, it would also force Intel to totally rework
their circuit stuff, etc. etc. And frankly, SOI is worth less and less
performance at every node.
Quote:
|
Going forward Intel does have some problems, I think they are focusing too much on their current competition (AMD) and not enough on the future competitors (cell phone guys). This last comment applies just as much to AMD though... Intel has sold off all of their cell phone businesses now, so AMD is their only competitor worth worrying about. |
only needs to worry about AMD, and vica versa, you are sadly mistaken.
How many people in China own PCs? How many people own cell phones?
When I was in Tanzania, many of the adults had cell phones. PCs? Not
a chance. When cell phone processors get powerful enough, AMD and
Intel will have to worry; since it takes 5 years to do anything, that
means they should probably have started thinking about this issue a
year ago or so.
DK
#23
| |||
| |||
|
|
David Kanter wrote: Right...I'm sure they did. Perhaps you mean in the Northstar design that John Borkenhagen worked on. I agree it was an early form of Multithreading, but: It is Switch on Event Multithreading (SoEMT), not Simultaneous multithreading (SMT). So, no, IBM did not do SMT before. A friend has one of the critical patents that Intel references. And...that doesn't change the fact that SMT != SoEMT. DK As Keith knows, there was a heated debate within IBM over the merits of SOI. And today the IBM Fab manufactures both bulk and SOI. The ASIC products, for example, are bulk. The microprocessors and game chips are SOI. And yes, IBM cares about yield. |
Quote:
|
Northstar had one form of multithreading, the one that could be added economically. It was enough of a win that it was carried forward for several generations of processors. |
not SMT, and Keith should know better than to try to foist that fallacy
on us.
SoEMT is much less complex for a variety of reasons. It doesn't let
you exploit unused issue slots, but it does let you get a lot of
overlap for memory operations, which quite frankly, for servers (where
northstar was aimed) is the key point.
Quote:
|
Neither IBM nor Intel has a monopoly on smart or stupid. Neither does AMD either. |
are far less reasonable in their assumptions.
DK
#24
| ||||
| ||||
|
|
Del Cecchi wrote: David Kanter wrote: Right...I'm sure they did. Perhaps you mean in the Northstar design that John Borkenhagen worked on. I agree it was an early form of Multithreading, but: It is Switch on Event Multithreading (SoEMT), not Simultaneous multithreading (SMT). So, no, IBM did not do SMT before. A friend has one of the critical patents that Intel references. And...that doesn't change the fact that SMT != SoEMT. DK As Keith knows, there was a heated debate within IBM over the merits of SOI. And today the IBM Fab manufactures both bulk and SOI. The ASIC products, for example, are bulk. The microprocessors and game chips are SOI. And yes, IBM cares about yield. There we have it...SOI is not for every product. |
How much ASIC work does INTC do?
Quote:
|
Northstar had one form of multithreading, the one that could be added economically. It was enough of a win that it was carried forward for several generations of processors. Absolutely. I'm not denigrating SoEMT; I'm just pointing out that it's not SMT, and Keith should know better than to try to foist that fallacy on us. |
I-stream then)?
Quote:
|
SoEMT is much less complex for a variety of reasons. It doesn't let you exploit unused issue slots, |
the same cycle. Therefor unused issue slots *DO* go unused. Since
the P4 has such a piss-poor issue queue perhaps this doesn't
matter.
Quote:
|
but it does let you get a lot of overlap for memory operations, which quite frankly, for servers (where northstar was aimed) is the key point. Neither IBM nor Intel has a monopoly on smart or stupid. Neither does AMD either. I'm inclined to agree with that; sadly it seems like other folks here are far less reasonable in their assumptions. |
extraneous crap to try to "prove" your point.
--
Keith
#25
| |||
| |||
|
| snip |
Quote:
|
Hell, you heard Del say that there was a huge internal fight over whether to use SOI or not. They aren't using SOI for everything, just the high-end stuff. |
<snip>
--
Keith
#26
| |||
| |||
|
|
No, the real reason is cross-licenses. You're thinking that Intel has access to ZRAM through AMD's access? Or that Intel is just implementing something that they might have gotten through another source? Intel definitely doesn't have access to ZRAM through AMD, it would have to get its own license from Innovative. The Intel-AMD cross-license only affects the instruction set. |
INTC has enough cash on hand to buy all the licenses in the world, or
at least all the licenses they need. Might cut into profit margin in
case of per unit fee, but shouldn't be an unsurmountable obstacle. If
I am not mistaking, INTC still gets a cut off every CPU AMD sells - or
at least used to ;-)
In other words, if ZRAM proves to be useful, you should expect INTC to
follow AMD in using it quite soon - about as soon as they did with
AMD64 - oh, pardon me, Intel doesn't do anything AMD..., they do
their own superior EM64T ;-)))))))))) They sure will find more
fitting, Intel-like name for ZRAM tech as well
NNN
#27
| ||||
| ||||
|
|
Yousuf Khan wrote: David Kanter wrote: I don't think Return On Investment is even in Intel's vocabulary. They've been swimming in monopoly money for so long that they're probably just now figuring out ROI. Very mature response. Intel certainly does have a concept of ROI; all corporations do. You just have to remember, the folks who make the ROI calculations at Intel actually have the best information possible. You and I don't. AMD produces at most 1/4 the silicon Intel does. IBM's even lower volume. This means that Intel generally is going to be more sensitive to increases in cost of their processes than AMD... Or exactly the opposite. Fabs are fixed cost outlays, the more you produce past your break-even point, the less you have to consider cost. Actually that's not really true. SOI wafers cost more, there are more processing steps, which slow down production, SOI wafers are incredibly fragile. etc. etc. Hell, you heard Del say that there was a huge internal fight over whether to use SOI or not. They aren't using SOI for everything, just the high-end stuff. |
which is a long way from "huge fight". Back in the day there was honest
disagreement over the magnitude of the power and performance difference.
And SOI is not just being used for "high end stuff". It just isn't
being used for Standard Cell ASIC chips.
Quote:
|
SOI adds substantial costs, it would also force Intel to totally rework their circuit stuff, etc. etc. And frankly, SOI is worth less and less performance at every node. |
measurements? Run 3D field simulations? Or was this Intel spin?
Remember sometimes spin is true and sometimes it isn't.
Myself, I really don't know the numbers for SOI advantage at 65 or 45nm
process nodes.
Quote:
| Going forward Intel does have some problems, I think they are focusing too much on their current competition (AMD) and not enough on the future competitors (cell phone guys). This last comment applies just as much to AMD though... Intel has sold off all of their cell phone businesses now, so AMD is their only competitor worth worrying about. Intel still retains their high-end Xscales. If you think that Intel only needs to worry about AMD, and vica versa, you are sadly mistaken. How many people in China own PCs? How many people own cell phones? When I was in Tanzania, many of the adults had cell phones. PCs? Not a chance. When cell phone processors get powerful enough, AMD and Intel will have to worry; since it takes 5 years to do anything, that means they should probably have started thinking about this issue a year ago or so. DK |
useful service to the average person, even in a LDC. What use is a PC
to the average person in a LDC?
You really ought to turn down the heat in your posts. Not everyone that
disagrees with you is evil or stupid.
Quote:
--
Del Cecchi
"This post is my own and doesn’t necessarily represent IBM’s positions,
strategies or opinions.”
#28
| |||
| |||
|
|
No matter how much you like or dislike INTC, one has to admit that INTC has enough cash on hand to buy all the licenses in the world, or at least all the licenses they need. Might cut into profit margin in case of per unit fee, but shouldn't be an unsurmountable obstacle. If I am not mistaking, INTC still gets a cut off every CPU AMD sells - or at least used to ;-) |
lots of money to pay license fees (so does AMD), but it simply won't do
it without trying to get around them.
I think the cross-license has been said that neither Intel nor AMD pay
each other royalties on the instruction set, and they're free to use
each other's extensions on the instruction set, just so long as they fab
their own CPUs, and don't do it through outside fabs (or at least don't
take more than 20% of their output from outside fabs). That's not a
problem for Intel, of course, but AMD is probably metering its output
heavily now that they use Chartered for some of their output.
Quote:
|
In other words, if ZRAM proves to be useful, you should expect INTC to follow AMD in using it quite soon - about as soon as they did with AMD64 - oh, pardon me, Intel doesn't do anything AMD..., they do their own superior EM64T ;-)))))))))) They sure will find more fitting, Intel-like name for ZRAM tech as well |
Z-RAM or whatever Intel calls it when it brings it out, will be quite a
bit more difficult. As we can see, Intel has matched AMD's instruction
set (well, it is the same instruction set after all), but it still so
far hasn't countered Hypertransport and direct memory controller of
AMD's K8-series.
Yousuf Khan
#29
| |||
| |||
|
|
nobody (AT) nowhere (DOT) net wrote: |
Quote:
|
In other words, if ZRAM proves to be useful, you should expect INTC to follow AMD in using it quite soon - about as soon as they did with AMD64 - oh, pardon me, Intel doesn't do anything AMD..., they do their own superior EM64T ;-)))))))))) They sure will find more fitting, Intel-like name for ZRAM tech as well Well, the AMD64 instruction set was easy, it's just software. However, Z-RAM or whatever Intel calls it when it brings it out, will be quite a bit more difficult. As we can see, Intel has matched AMD's instruction set (well, it is the same instruction set after all), |
do another Win port for x86. The had the 64-bit port for Itanic
and that was it. ...but no one needs 64 bits. ;-)
Quote:
|
but it still so far hasn't countered Hypertransport and direct memory controller of AMD's K8-series. |
--
Keith
#30
| |||
| |||
|
|
Seriously, those folks that do the ROI calcs must cry themselves to sleep every night, at the lack of influence they have on the management over there. They save money by using old tech on one hand, but then they fritter it away by having to spend more money on R&D to get around the limitations on the other hand. I'm just amazed at the x-ray vision you seem to have into Intel. For |
Wall Street seems impressed with neither at the moment. Both have lost
about 25% of their value in the last twelve months.
I can throw around words like ROI, too. Fortunately, even though I've
been the CEO of a very small but profitable company, I never really had
to understand such things because there was someone there who did.
Maybe one of these days we'll be able to read a plausible accounting of
how Intel managed to fumble away so much, just as we are now able to
read plausible accountings of how IBM practically gave control of the
PC away.
Quote:
|
From what I can see, Intel is a Wall-Street driven company that cares very much about what Wall Street thinks, and Wall Street understands |
experience in this area than most engineers.
The only reason for wading through these discussions is that you'll
occasionally get a glimmer of what might really be going on from
someone who might be in a position to know. For my own interests, I'd
like to hear even plausible rumors about what's going on with process
technology, but I'm not all that interested in proposed insights as to
how well or poorly AMD or Intel is being managed.
In any case, IBM would have been the much better investment than either
Intel or AMD, having gained about 12% over the last twelve months. I
could go on from that to say something like, "A chastened IBM has
learned to learn from its mistakes," but what the hell would I know?
If I did, I'd probably be making a fortune on Wall Street rather than
puttering around wondering what _is_ going on. You _can_ find
non-engineers who say things like that, but most of them are earning a
living putting out opinions, rather than by staking actual wealth on
their "knowledge."
Robert.
|
« Previous Thread
|
Next Thread »
| Thread Tools | Search this Thread |
| Display Modes | |
Linear Mode
| |
Powered by vBulletin Version 3.5.3
Copyright ©2000 - 2008, Jelsoft Enterprises Ltd.
Copyright ©2000 - 2008, Jelsoft Enterprises Ltd.