|
Comment #11
Dear Adrian
I've read your article on cache latency with much interest and I would be
grateful if you could supply some information about overclocking a Celeron 366:
I currently have a celeron 366 that o'c to 458 fine. However, it does lousy at
550 (crashes all the time) if the 2nd level cache is enabled, regardless of
temperature and core voltage. When cache is disabled, the computer functions
fine but my frame rates really suck (with the ole Riva TNT) in games like Quake2
and NFS4. I am wondering if :
1) Would it work with a lower cache latency?
2) Is there any software that changes the latency in Windows95? (My Abit BX6
won't let me fiddle with cache latency). If so, where can I download this
software?
3) What is the best latency?
Thanks for your time!
Les Way
Hello Les,
As I mentioned in my article, manipulating the cache latency
certainly offers the possibility for overclockers to further overclock
processors with stubborn L2 caches. Logically, increasing the cache
latency slows down the cache but enables it stably to run at a higher
speed. So, I would advise you to try increasing the cache latency
instead of lowering it.
As for a software L2 cache latency manipulator, try H. Oda's
excellent WCPUL2 utility. You can get the latest version from his
website.
Again, as I mentioned in the article, the L2 cache latency of 14
is the best for the Intel Celeron.
Adrian Wong
Adrian's Rojak Pot
http://www.rojakpot.com/
http://www.adriansrojakpot.com/
Comment #12
Hi, I read your article about cache latency, and i think you should have used a P2 or P3, because the L2 Cache run half the speed of the core, then you would see much more performance hits between the different cache latency levels, maybe 5 to
10%. I find it logical that the Celerons' full speed L2 Cache don`t react much on the L2 cache latency, because it runs so fast already.
Don`t you agree?
Great site by the way!
Love It!
Mats Larsen
Norway
Hello Mats,
I agree that the performance difference between the different cache
latencies of the Celeron is minimal at best because of the high speed
they run at. Unfortunately, I only have a Celeron to work with.
Otherwise, it would be interesting to test the Pentium II and III as
well to see the actual amount of variances the different cache latencies
would have on their external L2 caches.
On a good note however, Rick
Dumont has offered to run the same tests for the Pentium III. So, we
may soon see the results of L2 cache latency on the Pentium III's
performance.
Adrian Wong
Adrian's Rojak Pot
http://www.rojakpot.com/
http://www.adriansrojakpot.com/
Comment #13
Hey guy,
I found your article pretty interesting, but I have to say that with my own experience and others
I'm sure that L2 lat. over 8 essentially disables the cache! I'm not sure if the same holds true for the
Celeron. if it does, then how did you come up with those results? I'm currently running a
PIII 600
oc'd to 672 stable with an Alpha cooler, but it only boots at that speed w/ L2 at 1.
It does run faster but not for long periods, although I can play Q3 at 702 but when
I go back to Windows and the Internet, it gets errors. anyway just thought
I'd throw my 2 cents in.
Steven Henri
Hello Steven,
I'm not sure about the Pentium II or III because I have never been
able to afford either. But as far as I'm aware of, the Celeron can run
with all the cache latencies. That's how I was able to get all those
results.
Increasing the L2 cache latency theoretically slows down the L2 cache
and according to ABIT's manual, the L2 cache will only fail if it had
been set to run too fast. So, I wonder why increasing it over 8 should
disable the cache. Just like increasing the CAS latency of the SDRAM
from 2 to 3, increasing the L2 cache latency should decrease the L2
cache's performance and improve stability at overclocked speeds.
As we found out in the article, increasing the latency does not slow
down the L2 cache because of synchronization and possibly ECC checking
issues. However, the benefit of increased stability at a higher latency
should still hold true.
Adrian Wong
Adrian's Rojak Pot
http://www.rojakpot.com/
http://www.adriansrojakpot.com/
Comment #14
Hi,
I just have an observation about your L2 latency article. Have you considered doing the same tests using a P2/P3 processor? The reason I ask
this is because these processors use L2 cache running at half the processor
clock rate instead of full rate like the Celeron you tested. It would be
interesting to see how the cache latency affects these processors in comparison.
Keep up the good work with the guides!
Drew Cowan
Inside Sales
Onvia.com
Hello Drew,
Yes, I most certainly have. Unfortunately, I do not have the
resources (the Pentium II and III processors) to run those tests.
Luckily, Rick
Dumont has offered to run those tests using his Pentium III system.
So, let's keep our fingers crossed and wait for the results! :)
Adrian Wong
Adrian's Rojak Pot
http://www.rojakpot.com/
http://www.adriansrojakpot.com/
Comment #15
Just wanted to say that if you run a L2 cache latency above 8 on a Pentium III
(100MHz) you will disable the L2 cache.
Not sure about the Celeron/Pentium II/Pentium III E or B or EB. Does anyone know this?
Brant Robinette
Hello Brant,
Many people have been saying the setting a cache latency of over 8
disables the cache in the Pentium II and III. I don't own either so can
you tell me if the system's performance actually dropped a lot or a
utility told you that the L2 cache was disabled?
Adrian Wong
Adrian's Rojak Pot
http://www.rojakpot.com/
http://www.adriansrojakpot.com/
Comment #16
I also ran a lot of tests on L2 cache latency and here's what I
found. As I stepped down the L2 in 1ns increments, I gained roughly .5% performance all the way to
3ns. 2ns and 1ns just plain crashed. I tested with WinBench 99, Winstone
99, and 3DMark 99 Max. Running each test twice at the given speed and restarting between each test (yes
I am insane). I have left my system at the 3ns mark just cause it hasn't hurt it yet and 2.5% gain in performance couldn't hurt any.
Lord Nagash
Hello Lord Nagash,
I think you meant 1 clock cycle increments, not 1ns. But thanks for
your contribution! :-)
BTW, may I know what processor you are running?
Adrian Wong
Adrian's Rojak Pot
http://www.rojakpot.com/
http://www.adriansrojakpot.com/
Comment #17
Adrian,
Congratulations on an excellent article on a topic not really covered
anywhere else on the Net.
The one thing which would add a lot to it is the maximum CPU frequency
achievable for each Latency.
Also, the same benchmarks, but with a PII with its not-on-die Cache.
Mehran Khazra
Hello Mehran,
It's pointless for me to test the maximum CPU frequency achievable
for each latency with my processor because each and every processor is
different. Even if you use the processor of the same speed grade and
from the same batch, it's likely to behave differently from my processor
and every other processor as well.
Theoretically, increasing the L2 cache latency will enable the cache
to run at a higher frequency with greater stability. But the
effectiveness of this method will depend a lot on the processor itself.
All anyone can do is test to see if increasing the L2 cache latency will
enable him/her to force the stubborn L2 cache to behave properly.
If you really want to quantify the effectiveness of such a method,
the task will require many, many processors of different types and speed
grades and thousands of benchmark runs. Thus, it's
impossible for me to do so since I only have a lowly Celeron processor. :)
I agree, Mehran, Pentium II benchmarks would certainly be interesting
since its cache bandwidth is lower than that of the Celeron. So, any
changes in cache performance should be reflected more in its performance
than they have with the Celeron's. In the near future, I doubt I'll be
able to afford a Pentium II processor just for these tests. Don't worry,
however, because a reader, Rick
Dumont, has offered to benchmark the L2 cache latency performance of
his Pentium III.
Adrian Wong
Adrian's Rojak Pot
http://www.rojakpot.com/
http://www.adriansrojakpot.com/
Comment #18
Hello Adrian,
Just like to say thanks for making the article about the L2 cache
latencies.. I had trouble getting my computer stable when I overclock my celeron 500 to 563.. it's now rock solid at latency 14:) Well, it's
either that or the thermal goo I put in place of the pad thingy.. oh yeah, I was reading your article
about AGP sidebanding and it's effect on running the AGP bus faster than 66MHz.. well, my Asus AGP-V3800
TNT2 is running just fine with sidebanding mode on (160MHz core and memory, NVidia v3.65 reference drivers) at 75MHz.. so guess it just
depends what card you have:) Anyway, keep on doing all them wonderful tests.. you test a lot of stuff nobody even thinks about testing:)
Take care,
J-Sea (Jarkko Seppanen)
...Useless Invention: Parachute that opens automatically, upon impact.
Well Jarkko,
Glad to know that increasing the L2 cache latency is helping you
stabilize your processor! :-)
About the AGP sidebanding article, I believe I stated that AGP
sidebanding will only work up to 83MHz. Beyond that's impossible, as far
as I know.
I run tests on such subjects because I find them really interesting
and I intend to keep up my work as long as I can. So you can expect more
of such articles from me in the near future! :)
Adrian Wong
Adrian's Rojak Pot
http://www.rojakpot.com/
http://www.adriansrojakpot.com/
|
