J
John
[snip]
I have Barton with 512Mb but running Win98SE (and Linux mostly) Is there
similar a setting for Win98SE?
I have Barton with 512Mb but running Win98SE (and Linux mostly) Is there
similar a setting for Win98SE?
D
Dave Hull
Just one question. Probe reads that you have a CPU with 512Mb. Does
MB or KB?
No, but if I make the change under windows XP, and it acknowledges it, then
and I know that the processor has 512 KB of L2 cache, then it seems to be
a good bet that windows is operating properly and the "verifier" is not
operating
correctly. The verifier checks for installed memory, not what Windows
detects, no?
Dave
MB or KB?
No, but if I make the change under windows XP, and it acknowledges it, then
and I know that the processor has 512 KB of L2 cache, then it seems to be
a good bet that windows is operating properly and the "verifier" is not
operating
correctly. The verifier checks for installed memory, not what Windows
detects, no?
Dave
P
Peter A. Stavrakoglou
Daniel Yates said:
A good free tweaking utility similar to Microsoft's "TweakUI" is
"FreshUI" that you can get from http://www.freshdevices.com/. The
program allows this setting you speak of to be changed within the
gui - along with a lot of other settings.
G
Gene Puhl
P
Pond Scum
Daniel Yates said:
This is utter nonsense, and obvious to anyone who knows even the slightest
bit about how a CPU operates. Enabling the L2 cache in an Athlon XP, or
Pentium 4 is all-or-nothing deal. You can't only use half of the L2 cache.
There are no bits in the cache control registers that control how much
of the cache to enable, they just enable or disable the whole thing.
People whose knowlege of computers is so limited that they write such
ridiculous and uninformed statements should really find something else to do
with their time, since they're not helping anyone by spreading their
misinformation.
That registry entry is read by the OS and used to control the placement
of virtual memory pages in physical memory to reduce L2 cache collisions.
This Microsoft Knowledgebase article:
http://support.microsoft.com/default.aspx?scid=kb;en-us;183063
explains it, and also states that it is only useful on processors with a
direct mapped L2 cache, where you might see a 2% improvement in performance.
The L2 cache on a Barton is 16-way set associative, a P4 is 8-way, so they
are already going to have fewer collisions, since there are multiple
locations in the cache where data at a particular address in physical
memory can be cached. A direct-mapped cache is 1-way set associative, so
data at a particular address in physical memory can only reside in one
possible location in the cache. This is why scattering pages in physical
memory can reduce collsions in the direct-mapped L2 cache and improve
performance, but has little or no effect if the CPU has an 8-way or 16-way
set associative L2 cache.
To find a processor which used a direct-mapped L2 cache you'd probably have
to go all the way back to the Pentium. Pentium II, AFAIK always had an
integrated 4-way set associaitve L2 cache.
-Jonathan (e-mail address removed)
"The great thing about the Internet is that it allows anyone anywhere to
publish anything regardless of their education, age, or experience. The
bad thing about the Internet is that it allows anyone anywhere to publish
anything regardless of their education, age, or experience."
--from the video "Don't Get Caught in the Web"
S
Strontium
Love that reply. Too true. /me thinks back to the K6-2 days of L1....yuck.
-
Pond Scum stood up at show-n-tell, in
[email protected], and said:
-
Pond Scum stood up at show-n-tell, in
[email protected], and said:
S
Strontium
PS, I remember when L2 was onboard (mainboard). Good old days...NOT.
-
Pond Scum stood up at show-n-tell, in
[email protected], and said:
-
Pond Scum stood up at show-n-tell, in
[email protected], and said:
A
AJ
So XP detetecting and listing the size of the L2 cache is just for
fun? Yes, reports so far indicate that Sandra has trouble detecting an
improvement in performance after the registry change.
S
Strontium
Sandra is crap, in the first place 
Play with your tinkertoys....
-
AJ stood up at show-n-tell, in (e-mail address removed),
and said:
Play with your tinkertoys....
-
AJ stood up at show-n-tell, in (e-mail address removed),
and said:
D
Daniel Yates
Gene Puhl said:
I think they were thinking mroe on the lines of only one in the UK, which to
the best of my knowledge it is.
Daniel
M
Martin Eriksson
Pond Scum said:
<snip>
Well well.. I figured. Most of the registry "tweaks" in Win2000/XP etc ar
just for show. For example all those "make your modem faster" tweaks..
hoho.. lol. rofl.. yeah.
/M
P
Pond Scum
AJ said:
If Windows XP actually even does anything with that value, the benefit is
limited to systems with direct-mapped L2 caches, which means Pentium and
Pentium MMX processors.
The MS Knowledgebase article dates back to Windows NT, when contemporary
CPUs would benefit from scattering pages in physical memory. It's hard to
say whether WinXP looks at that registry key and does anything with it,
or if it is just a leftover from NT. The minimum recommended CPU for XP
(300MHz Pentium family or compatible) is going to have at least a 4-way
set associative L2 cache, so it isn't going to benefit from scattering
pages. Without access to the kernel source code, the only way any of us
are going to know whether XP does anything with that value is to 1) find an
old system that would benefit from the page scattering (233MHz Pentium MMX
w/512kb direct-mapped external L2 cache), 2) install XP on it, 3) write
some benchmark programs that intentionally try to create cache line
collisions in the L2 cache that page scattering would alleviate, and 4)
measure the performance with varions settings of SecondLevelDataCache.
-Jonathan (e-mail address removed)
P
Pond Scum
Martin Eriksson said:
It can be useful to know how big the L2 cache is. For performance reasons,
a programmer might adjust the size of chunks of data their code processes
to insure that the chunks fit entirely in the L2 cache. Or, you might need
to know the L2 cache size to know when to perform software prefetching using
the SSE2 instructions. But this is quite different than claiming that
a registry entry controls how much of the L2 cache is enabled.
Since that KB article was written, Intel has extended the CPUID instruction
so it now reports back the size of all the caches (TLB, L1, L2, L3) and
their associativity. On a modern system, most pieces of code that need to
know cache sizes will use the CPUID instruction, not some little-known
registry entry.
-Jonathan (e-mail address removed)
A
AJ
..........................................................
It can be useful to know how big the L2 cache is. For performance
reasons, a programmer might adjust the size of chunks of data their
code processes to insure that the chunks fit entirely in the L2 cache.
......................................................................snip
WOW! When you do something you really go all the way. That has to be
the single biggest back pedal in the history of usenet.
It can be useful to know how big the L2 cache is. For performance
reasons, a programmer might adjust the size of chunks of data their
code processes to insure that the chunks fit entirely in the L2 cache.
......................................................................snip
WOW! When you do something you really go all the way. That has to be
the single biggest back pedal in the history of usenet.
E
Ed Light
I got such a tiny increase in cpu benchmarks that I suspect the l2 cache
does run but isn't quite optimized.
does run but isn't quite optimized.
P
Pond Scum
AJ said:
The original post on the registry value, and its effect on performance and
use of the L2 cache was complete and total garbage, period. My follow up
answers the unasked question "Is there any situation where it would be useful
to know the size of the L2 cache?" The two hypothetical situations I
described are COMPLETELY UNRELATED to how the NT kernel uses the L2 cache
size.
If you weren't so ignorant, you'd have realized that. Now that I've explained
it in terms simple enough for an idiot like yourself, do you understand now?
-Jonathan (e-mail address removed)
A
AJ
I'm starting to see what you mean about anybody being able to publish
anything!
D
Daniel Yates
Pond Scum said:
Hey! I only posted it because I saw it in a hardware mag and thought it
would be worth sharing. And guess what - if you change the value in the
registry yet the system detects the value of the L2 cache anyway then there
is no harm done to anyone is there? Have you actually tested the setting? I
find a lot of you "people who seem to know what you are on about" take a
whooooooole lot on trust when it comes to what Microsoft would have you
beleive Don't forget, XP is the opertating system that they claimed never
needed to restart after an application install!! That was a laugh. But hey,
I think you are more than likely right about the L2 cache. But what does
that then tell you? It tells me that MS are doing nothing but rehashing th
same things over and over, hence all the defunct entries in the registry.
Daniel
D
Dave Hull
If you weren't so ignorant, you'd have realized that.
Maybe this is true, but...
Why does ignorant, which we are all guilty of being in some areas, equate to
being foolish or stupid? You might know about computers, but does that
makes people who don't retarded in some sense? Jonathan, how is it that
people can be fools for not being born with the knowledge necessary to
understand computers and software?
I'm sure that you have a red herring stored for just this occasion, but,
please, try to address the question without anything superfluous.
Cheers,
Dave
Maybe this is true, but...
Why does ignorant, which we are all guilty of being in some areas, equate to
being foolish or stupid? You might know about computers, but does that
makes people who don't retarded in some sense? Jonathan, how is it that
people can be fools for not being born with the knowledge necessary to
understand computers and software?
I'm sure that you have a red herring stored for just this occasion, but,
please, try to address the question without anything superfluous.
Cheers,
Dave
E
Ed Light
Daniel Yates said:
Personally, I found that it boosted benchmarks by 1% to 5%, except for one
minus result. I think it causes Windoz to optimize for the cache size.
Folks, stop arguing with the bigot and benchmark! I've filtered him out,
myself.