YKhan

AMD to allow per-core overclocking for Phenom? - The Tech Report
"There will be a twist, though: a screenshot posted by the site
suggests the new Overdrive will enable discrete overclocking for each
of a Phenom X4 processor's four cores. In the screenshot, a chip is
running with multipliers of 11X, 12X, 12.5X, and 11.5X for its four
cores, achieving respective speeds of 2.93GHz, 3.2GHz, 3.33GHz, and
3.07GHz."
http://techreport.com/discussions.x/13511

nobody@nowhere.net

On Thu, 01 Nov 2007 09:26:59 -0700, YKhan <yjkhan@gmail.com> wrote:

>AMD to allow per-core overclocking for Phenom? - The Tech Report
>"There will be a twist, though: a screenshot posted by the site
>suggests the new Overdrive will enable discrete overclocking for each
>of a Phenom X4 processor's four cores. In the screenshot, a chip is
>running with multipliers of 11X, 12X, 12.5X, and 11.5X for its four
>cores, achieving respective speeds of 2.93GHz, 3.2GHz, 3.33GHz, and
>3.07GHz."
>http://techreport.com/discussions.x/13511

What would be the L3 speed then? I'm assuming it runs at full CPU
clock under normal conditions, but if the cores run each at its own
clock speed, there would be no such thing as "CPU clock", unless we
take the average. I'm afraid the average is not much more meaningful
than average patient temperature in a hospital - go average infectious
ward with the morgue, and the result may just be close to the norm :-)
Does it synchronize then to the fastest, or the slowest, or just runs
at stock speed? Can it run asynchronously from the cores? If yes,
wouldn't there be a performance hit that takes away all the advantages
of overclocking each core to its limits and then some? Too many
questions, and not that much in terms of possible advantages.
All this is aside from the usual official stance of chip makers that
does not condone running the chip outside of the spec - unless there
was a change of heart that I didn't notice.
As for the photo...well, Photoshop can produce amazing results in
skilled hands.

NNN

Yousuf Khan

nobody@nowhere.net wrote:
> What would be the L3 speed then? I'm assuming it runs at full CPU
> clock under normal conditions, but if the cores run each at its own
> clock speed, there would be no such thing as "CPU clock", unless we
> take the average. I'm afraid the average is not much more meaningful
> than average patient temperature in a hospital - go average infectious
> ward with the morgue, and the result may just be close to the norm :-)
> Does it synchronize then to the fastest, or the slowest, or just runs
> at stock speed? Can it run asynchronously from the cores? If yes,
> wouldn't there be a performance hit that takes away all the advantages
> of overclocking each core to its limits and then some? Too many
> questions, and not that much in terms of possible advantages.
> All this is aside from the usual official stance of chip makers that
> does not condone running the chip outside of the spec - unless there
> was a change of heart that I didn't notice.
> As for the photo...well, Photoshop can produce amazing results in
> skilled hands.


As I understand it, the L3 cache runs at the external clock speed, not
at any of the internal core speeds.

Yousuf Khan

krw

In article <rcgpi31t6r1s225hchgkt43b6grtcu9dfr@4ax.com>,
mygarbage2000@hotmail.com says...
> On Thu, 01 Nov 2007 09:26:59 -0700, YKhan <yjkhan@gmail.com> wrote:
>
> >AMD to allow per-core overclocking for Phenom? - The Tech Report
> >"There will be a twist, though: a screenshot posted by the site
> >suggests the new Overdrive will enable discrete overclocking for each
> >of a Phenom X4 processor's four cores. In the screenshot, a chip is
> >running with multipliers of 11X, 12X, 12.5X, and 11.5X for its four
> >cores, achieving respective speeds of 2.93GHz, 3.2GHz, 3.33GHz, and
> >3.07GHz."
> >http://techreport.com/discussions.x/13511
>
> What would be the L3 speed then? I'm assuming it runs at full CPU
> clock under normal conditions,

As Yousuf said, L3s generally run at the memory interface speed. The
L3 (or at least its controller) generally is part of the memory
interface/BIU.

> but if the cores run each at its own
> clock speed, there would be no such thing as "CPU clock", unless we
> take the average. I'm afraid the average is not much more meaningful
> than average patient temperature in a hospital - go average infectious
> ward with the morgue, and the result may just be close to the norm :-)

Well, if one CPU is dead... I guess you have a point. ;-)

> Does it synchronize then to the fastest, or the slowest, or just runs
> at stock speed? Can it run asynchronously from the cores? If yes,
> wouldn't there be a performance hit that takes away all the advantages
> of overclocking each core to its limits and then some? Too many
> questions, and not that much in terms of possible advantages.

L3s are generally large with only two or four ports, usually time
multiplexed, running at the I/O or memory speed.

> All this is aside from the usual official stance of chip makers that
> does not condone running the chip outside of the spec - unless there
> was a change of heart that I didn't notice.

As long as the manufacturer does it, it's not "over clocking". ;-)
This allows them to run each core at its maximum, rather than that of
the slowest.

> As for the photo...well, Photoshop can produce amazing results in
> skilled hands.

Marketeering is no different. They're just featuring a liability.
;-)

--
Keith

Rick Jones

In comp.sys.intel krw <krw@att.bizzzz> wrote:
> As long as the manufacturer does it, it's not "over clocking". ;-)
> This allows them to run each core at its maximum, rather than that
> of the slowest.

I'm sure the CPU schedulers would _love_ that

rick jones
--
a wide gulf separates "what if" from "if only"
these opinions are mine, all mine; HP might not want them anyway...
feel free to post, OR email to rick.jones2 in hp.com but NOT BOTH...

krw

In article <fgr1ft$im9$1@usenet01.boi.hp.com>, rick.jones2@hp.com
says...
> In comp.sys.intel krw <krw@att.bizzzz> wrote:
> > As long as the manufacturer does it, it's not "over clocking". ;-)
> > This allows them to run each core at its maximum, rather than that
> > of the slowest.
>
> I'm sure the CPU schedulers would _love_ that

Why would they care? They'd *BETTER* not assume any timing
relationship.


--
Keith

Bill Davidsen

YKhan wrote:
> AMD to allow per-core overclocking for Phenom? - The Tech Report
> "There will be a twist, though: a screenshot posted by the site
> suggests the new Overdrive will enable discrete overclocking for each
> of a Phenom X4 processor's four cores. In the screenshot, a chip is
> running with multipliers of 11X, 12X, 12.5X, and 11.5X for its four
> cores, achieving respective speeds of 2.93GHz, 3.2GHz, 3.33GHz, and
> 3.07GHz."
> http://techreport.com/discussions.x/13511
>
Not really new, Intel Core2 allows the cores to run at different speeds
AFAIK, the /proc/cupinfo shows varying speeds under light load, moving
all to full speed under heavy load.

The newest Intel CPUs, Penryn tech, seem to have the ability to o/c a
single core under light load, on the theory that overall power use will
stay within design limits, and single thread performance will be better.
Since I've only seen this mentioned for the "extreme" CPU, I can't guess
if it will also be on something the average person might buy.

--
Bill Davidsen
He was a full-time professional cat, not some moonlighting
ferret or weasel. He knew about these things.