|
Comment #1
This is for the enabling Video RAM Caching? That was good for your setup... but what if your CPU has a bandwidth of a Celeron
567MHz w/ 4.5GB/sec (567MHz * 64-bit )... and your graphics card's bandwidth is 135MHz w/ 2.2GB/sec (135 *
128 bit)... the CPU's bandwidth is 2X more... Do you think it is worth enabling then in your opinion? I know this is more than a question than a comment...
Costa Vlachos
Hello Costa,
It's not a question of how much faster the L2 cache is over the video
RAM. No matter how big that difference is, it can never be as big as the
performance difference between the L2 cache and the system RAM.
Remember, the system RAM (PC100)'s bandwidth is only 0.8GB/s, a
paltry number compared to the video RAM's bandwidth of 2.2GB that you
suggested.
The question, rather, should be - should I use my limited amount of
L2 cache to bolster my system RAM's low bandwidth (0.8GB/s) or use it
instead to boost the video RAM which is already 2.75X faster than your
system RAM. As the benchmark results have shown, enabling this feature
will improve the graphics card's performance but there will be a drop in
the overall system performance. So, I guess I'll stick by my opinion
that it's a waste of time to enable Video RAM Cacheable.
Adrian Wong
Adrian's Rojak Pot
http://www.rojakpot.com/
http://www.adriansrojakpot.com/
Comment #2
Hi there Adrian,
First time I saw one of your reviews (sorry, sorry but the Net is too big you know), well, now that
I saw almost all of them, I have 3 words for your ego: they are great!!!
I saw your article on Video RAM Caching and I found it very interesting, now,
I have two plus one question to make:
1- I have a K6-2 300MHz running at 350MHz (3.5 x 100MHz) on a DFI
P5BV3+ rev. B motherboard with 2MB cache L2 (my cache L2 is external and accessed at the same speed of the FSB , so,
100MHz), 3x32MB(96MB) PC100 DIMM and a Savage4 32MB AGP2X. Now, with the amount and
speed of cache L2 I have, do you recomend to enable or disable video ram caching?
2- Did you noticed any major visual fps slowdown comparing games with the two settings?
3- I know this is a bit out of content, but why on every Win9x info program (e.g. CPUid) do
I have the cache L2 amount and latency disable (just like if I haven't) and
DOS programs show me the 2MB? Any idea??
That's all, thanks in advance and keep 'reviewinginging' :))
P.S.: Also thanks to Erik from Savage
World for posting the link to your article and sorry if you find this E_mail tooooooo long.
Marco Gera @ MAIA
Hello Marco!
I haven't had any opportunity to test the Video RAM Cacheable
feature on a system's that's heavy on L2 cache like yours. However, I'll
try to evaluate its effectiveness from a theoretical point of view.
- Now, your L2 cache's running at 100MHz. That means it has a
bandwidth of 100MHz x 64-bit = 0.8GB/s. Of course, its lower latency
means it's still much faster than the PC100 SDRAM you use. I'm not
sure how fast the SDRAM on your Savage4 graphics card runs but I
will just leave it at 125MHz. That gives it a bandwidth of 125MHz x
128-bit = 2.0GB/s! So, I guess you can see that it's pretty
pointless for you to enable the Video RAM Cacheable feature
when your L2 cache's bandwidth is already much lower than your video
RAM's.
- Nope, as I stated in the article, there were no major drop in
framerates. But the enabled setting was consistently slower.
- I'm not sure about this. Perhaps your copy of CPUID does not
support your processor and motherboard. Try updating it to the
latest version.
Hope that helps you some! :)
Adrian Wong
Adrian's Rojak Pot
http://www.rojakpot.com/
http://www.adriansrojakpot.com/
Comment #3
Hello,
I just read your article on Video RAM caching. I'm glad you ran all the tests,
since I sure wouldn't want to :)
However, I have one concern about the article:
To the best of my old and confused memory, Video BIOS shadowing and caching
really only mattered if the program you were using was controlling the video
card via BIOS calls. This mainly applied to DOS text programs. If the video
card was being accessed directly by a custom driver (i.e. Windows) then the
video BIOS wasn't used AT ALL. And therefore it should have no difference.
Of course, I could be wrong, so I went ahead and disabled the Video BIOS
cacheable option anyway. After all, I don't have any DOS programs I want to
speed up :)
Mark Kundinger
Hello Mark,
I think there's some confusion here. The article is about the caching
of the video RAM by the L2 cache, not the caching of the video BIOS.
In any case, you are right. If the video BIOS is not used, it is not
cached but it will definitely be shadowed, if video BIOS shadowing was
enabled. But as you said it yourself, the video BIOS isn't used by Win9x
programs at all. Only DOS software will need to use it. However, if at
any time the BIOS is accessed, the L2 cache will cache it even though
it's very unlikely that it will be needed later. This deprives the
system RAM of L2 cache bandwidth needlessly.
So, I guess that's a good reason why the video BIOS shouldn't be
cached. As for video BIOS shadowing, it's pointless to shadow it since
video BIOS calls are rarely made. The memory used by the shadowing of
the video BIOS would serve you better by joining the memory pool
instead. For more information on the caching and the shadowing of the
video BIOS, check out the BIOS
Optimization Guide.
Hope that helps you some! :)
Adrian Wong
Adrian's Rojak Pot
http://www.rojakpot.com/
http://www.adriansrojakpot.com/
Comment #4
I noticed in your article that 32MB textures were as high as you went in your test with 3DMark. You should rerun the 3DMark portion of your tests using 3DMark 2000. Set your AGP aperture size to 256MB and video caching to on and you will see a major improvement when running 64MB textures, especially when
using 32-bit color and 32-bit textures with a resolution of 1024 x 768 or higher. That is the only reason I can see to have video caching on.
It is nice to see people writing articles about such out-of-the-way topics such as this. Most people pay no attention to important BIOS settings.
R00dDaWg
Hello ROOdDaWg!
Thanks for that tip there! I'll be sure to check out 3DMark 2000 as
soon as I can and run those tests that you recommended.
Adrian Wong
Adrian's Rojak Pot
http://www.rojakpot.com/
http://www.adriansrojakpot.com/
Comment #5
Hello,
I've been through your article about "Video RAM Caching" and I really
liked it. I wish you could expand your tests so you could provide more info on other configurations of CPUs/Motherboards and Graphics cards.
I noticed that for all the tests you are using an overclocked system at
112MHz FSB. In the past I also did some testing with my system using FSB beyond the 100MHz. To be more precise I did some benchmarking
using the 3D Mark 99 Max. So, I found out that whenever the AGP was running beyond the standard AGP Frequency (66MHz) I was getting some
performance hits to any test that had to do with AGP transfers. You can check it yourself by running the 3D Mark 99 Max test on your system at
100 and 112 MHz FSB (assuming a 2/3 AGP Ratio). What are your findings about Video RAM
Caching under such conditions?
My system is based on:
ASUS P2B
Celerom 300A @ 450MHz
288MB of crappy CAS2 PC100 SDRAM
Diamond Viper 770 Ultra Revision D (it has the AGP4x support jumper)
Quantum Fireball EX 12.7
SBLive
etc.
George Gagaudakis
LOL! :)
George, how I wished I had the resources to do just that! I already
consider myself very lucky to be able to run those tests using my single
Celeron-based system. It's all thanks to my parents, who sacrificed
their hard-earned money so that I can build this system. Without their
help, I wouldn't even be able to build this system and run such tests. I
really wish I have the resources that other hardware webmasters have but
at this moment, I don't see how that will be possible. :-(
Hmm... I have never noticed such a performance hit while overclocking
the AGP bus. In fact, as the benchmark results from the AGP
Sidebanding vs. AGP Overclocking guide show, with mipmapping
enabled, the overclocked AGP bus performed about the same as the
standard AGP bus and when mipmapping was disabled, the overclocked AGP
bus was much faster than the standard AGP bus.
The results of the video RAM caching benchmarks should not be
affected by the overclocking of the AGP bus because the purpose is to
test only one variable factor - the Video RAM Cacheable feature. Thus,
all other factors have been kept constant so they will not affect the
results of the tests. Whether overclocking the AGP bus actually reduces
its performance or not, the results for this guide would still be the
same - enabling the feature will improve the performance but there will
be a larger drop in overall system performance.
Hope that answers your question! :)
Adrian Wong
Adrian's Rojak Pot
http://www.rojakpot.com/
http://www.adriansrojakpot.com/
Comment #6
Hi Adrian,
There might be more impact of this feature (Video RAM caching) using old DOS games and applications: Maybe it's worth to check this - maybe
with Duke Nukem 3D or something similar...
Greetings from Munich/Germany!
Manfred / Lini
P.S.: Nice site featuring some good articles by the way - I just wish my
ISP had a better connection in your direction... <sigh>
Hello Manfred,
DOS games could benefit from such a feature since they (mostly) don't
support linear framebuffering. However, since all graphics cards these
days have very fast VGA cores, there's little need for such an
enhancement. After all, DOS games hardly require much graphics
acceleration.
Adrian Wong
Adrian's Rojak Pot
http://www.rojakpot.com/
http://www.adriansrojakpot.com/
|
