Anyone here?

[The Lone Gunman]

Anyone at all?? :-(

 

[Robert Myers]

Anyone at all?? :-(

The AMDroids who usually kept the group going are all in therapy.

A perusal of my google groups home page suggests that there are many
others dropping by to see if anyone has anything to say.

It's not as if there were nothing to talk about, actually.

Robert.

 

[willbill]

The AMDroids who usually kept the group going are all in therapy.

A perusal of my google groups home page suggests that there are many
others dropping by to see if anyone has anything to say.

It's not as if there were nothing to talk about, actually.

Robert.

brilliant response robert.

as an owner of two recent AMD server
mobos and CPUs, it gets my smile

bill

 

[chrisv]

The AMDroids who usually kept the group going are all in therapy.
8)

A perusal of my google groups home page suggests that there are many
others dropping by to see if anyone has anything to say.

It's not as if there were nothing to talk about, actually.

I can give a 6-month update on my Conroe overclock experience - zero
problems to report. (Using 1333 MHz FSB on an Intel D975XBX to
overclock my E6400 to 2.67 GHz.) Intel Rulz! 8)

 

[Gnu_Raiz]

brilliant response robert.

as an owner of two recent AMD server
mobos and CPUs, it gets my smile

bill

I got a chuckle as well, ever since hardware.chips.intel got
(absorbed?) by .chips things haven't been the same. I am still waiting
for that 100+ post about Barcelona to cheer me up if it ever happens
which is doubtful.

Being called an AMDdroid was a nice refresh, I just thought I was a
conscience consumer. The only thing that really uses my 2nd core on
my opty 165 is RC5-72. For some reason going SLI or Xfire just to get
a few more FPS in Oblivion, or Quake 4 doesn't really benefit me much.
I am happy that Intel finally seen the light and decided to help out
the *BSD's by offering wireless documentation and more support.

Gnu_Raiz

 

[gaffo]

The AMDroids who usually kept the group going are all in therapy.

A perusal of my google groups home page suggests that there are many
others dropping by to see if anyone has anything to say.

It's not as if there were nothing to talk about, actually.

Robert.

huh?

there is nothing to talk about until next summer and k8l is released.

until then.............boring...........

--

 

[Robert Myers]

huh?

there is nothing to talk about until next summer and k8l is released.

until then.............boring...........

Think of all the energy that went into whether hyperthreading was any
good or not. Now we've routinely got two actual CPU's, and not a word
about whether it does any good or not, for one thing. We're living in
a brave new world of computing, and no one seems to notice.

Along the same lines, memory bandwidth should be a bigger deal.
Anybody notice anything? Anybody care?

Robert.

 

[willbill]

Think of all the energy that went into whether hyperthreading was any
good or not.

it wasn't, which is why it no longer exists

Now we've routinely got two actual CPU's, and not a word
about whether it does any good or not, for one thing.

better than Intel's hyperthreading

also depends on the ap(s) yer running

We're living in a brave
new world of computing, and no one seems to notice.

that's probably because yer an optimist?

bill

 

[David Kanter]

it wasn't, which is why it no longer exists

No offense, but if multithreading is such a bad idea, how come EVERY
single MPU vendor, except AMD uses it?

IBM - POWER5/6, CELL, Xbox360 chip
Sun - Niagara I, II, Rock
Intel - Montecito, P4, future designs
Fujitsu - upcoming SPARC64 design

Gosh, if multithreading sucks, all these well paid engineers must be
real morons. It's a good thing you came along and cleared that up for
us.

better than Intel's hyperthreading

Better in what sense? It provides higher performance, and uses more
die area.

DK

 

[Tony Hill]

No offense, but if multithreading is such a bad idea, how come EVERY
single MPU vendor, except AMD uses it?

Multithreading in general is often good, Hyperthreading, as
implemented on the P4 at least, is rather ho-hum at best. There WERE
some applications were it really benefitted, but there were at least
as many applications that were negatively impacted by Hyperthreading.
Most of the time it ended up being about even.

Hyperthreading certainly never lived up to the same performance that
we see from multithreading in IBM's Power chips or Sun's Niagara
chips.

Better in what sense? It provides higher performance, and uses more
die area.

MUCH better performance and surprisingly little increase in die area.
These days most processors are more then 50% cache by die area. Adding
hyperthreading to a single core chip might add 5% to the total die
size, dual core might add 25-30% to the die size.

In the specific case of Intel's single core P4 with hyperthreading to
dual-core Core 2 Duo the difference was even smaller. A single-corem
hyperthreading 65nm P4 chip with 1MB of cache is about 81mm^2. A
dual-core 65nm Core 2 Duo chip with 2MB of cache is 111mm^2. When you
factor in that their cache is running roughly 16mm^2 per MB, you end
up that the Core 2 Duo processor is only about a 12-13% larger die
then a single-core Hyperthreading P4 chip.

Of course, as IBM, Sun and others have demonstrated, multithreading
and dual core are not mutually exclusive.

 

[Robert Myers]

MUCH better performance and surprisingly little increase in die area.
These days most processors are more then 50% cache by die area. Adding
hyperthreading to a single core chip might add 5% to the total die
size, dual core might add 25-30% to the die size.

In the specific case of Intel's single core P4 with hyperthreading to
dual-core Core 2 Duo the difference was even smaller. A single-corem
hyperthreading 65nm P4 chip with 1MB of cache is about 81mm^2. A
dual-core 65nm Core 2 Duo chip with 2MB of cache is 111mm^2. When you
factor in that their cache is running roughly 16mm^2 per MB, you end
up that the Core 2 Duo processor is only about a 12-13% larger die
then a single-core Hyperthreading P4 chip.

Of course, as IBM, Sun and others have demonstrated, multithreading
and dual core are not mutually exclusive.
--

All problems with NetBurst came back to the same cause: power
consumption. Running NetBurst fast had a painful price in power
consumption, and I have a feeling that power consumption constraints
were why hyperthreading was so lame (although, in terns of non-cache
transistors and power consumption vs. performance, it was a wash in
server applications). The architecture was short on issue ports and
could only issue from one thread at a time. It isn't that intel
didn't have the wits to add more issue ports. It probably couldn't
have been fit into the power budget, which was a constant problem with
NetBurst. As you have correctly concluded, trying to draw conclusions
about multithreading from NetBurst is a flawed exercise to begin with.

Robert.

 

[David Kanter]

Multithreading in general is often good, Hyperthreading, as
implemented on the P4 at least, is rather ho-hum at best. There WERE
some applications were it really benefitted, but there were at least
as many applications that were negatively impacted by Hyperthreading.
Most of the time it ended up being about even.

Hyperthreading certainly never lived up to the same performance that
we see from multithreading in IBM's Power chips or Sun's Niagara
chips.

That has to do with the underlying microarchitecture. And I'd point
out that there are some workloads where HT provides up to a 50%
boost. The issue is that the target workloads for the P4 are
different than Niagara or POWER5/6.

MUCH better performance and surprisingly little increase in die area.
These days most processors are more then 50% cache by die area. Adding
hyperthreading to a single core chip might add 5% to the total die
size, dual core might add 25-30% to the die size.

No, SMT should add about 5% to the area of the core. CMP doubles the
size of the core.

In the specific case of Intel's single core P4 with hyperthreading to
dual-core Core 2 Duo the difference was even smaller. A single-core
hyperthreading 65nm P4 chip with 1MB of cache is about 81mm^2. A
dual-core 65nm Core 2 Duo chip with 2MB of cache is 111mm^2. When you
factor in that their cache is running roughly 16mm^2 per MB, you end
up that the Core 2 Duo processor is only about a 12-13% larger die
then a single-core Hyperthreading P4 chip.

I suspect the cache designs were different for C2D and the P4.

Of course, as IBM, Sun and others have demonstrated, multithreading
and dual core are not mutually exclusive.

Yes, and I expect that's what will happen in the future.

DK