Comment #101

Hi!

I've downloaded all your BIOS guide. I think is great. However I want to discuss a topic about the 'PNP aware OS' setting:

I've got a Shuttle HOT-591P motherboard (VIA Apollo MVP3) with an AGP card and an old SB16 ISA card. The remaining slots are empty. When I switch 'PNP aware OS' to YES (I'm using both MS-DOS and W98SE) the sound card stops working under DOS because the IRQ and DMA settings go nuts. Thus, I have set up 'PNP aware OS' to NO, and the card works just fine under DOS (IRQ and DMA matches their settings, that is, IRQ 5 and DMA

On the other hand, Windows seems to work just fine no matter the setting is.

Thank you very much and keep up the good work.

X

Hello X,

You should only set PnP OS Installed to Yes only if all your operating systems support Plug and Play. DOS does not, which is why your sound card stops working if PnP OS Installed is set to Yes. So, if you are going to use your sound card or similar PnP cards in real mode DOS, you will have to set PnP OS Installed as No. Then the BIOS will handle the allocation of resources, instead of the operating system.

Hope that helps you some! :)

Adrian Wong
Adrian's Rojak Pot
http://www.rojakpot.com/
http://www.adriansrojakpot.com/

Comment #102

Hi!

I've been reading the comments and I want to clarify an issue with video shadowing on modern video cards. Most modern video cards have BIOS ROM sizes that are larger than 32 KB, so let's say, the BIOS ROM is C000-CBFF in size, thus enabling video shadow, just only copies the portion C000-C7FF to RAM, leaving the rest in its original location, making the system highly unstable (I have tried this with a-no-so-modern S3 ViRGE card -> AGP 4MB, with a 48KB Video ROM) and it locked up the system all the time. If you have the imperative need to have video shadow enabled, YOU HAVE TO COPY THE ENTIRE RANGE, so, you have to enable one or several consecutive regions:

+ Enable Video BIOS Shadowing at C000-C7FF (32 KB) and
+ Enable the remaining portions, for instance, C800-CBFF and so on until all video bios is shadowed.

That's why is advisable to DISABLE all video shadowing on modern systems (PCI cards and newer AGP cards). Video shadowing even made no sense for older ISA cards if you could tweak your ISA BUS CLOCK and the card supported it. I made this sometime ago with an old 386 DX and gained about a 60% percent in video access. And I had video shadowing disabled.

Thank you again for your page. It rocks!

X

Hello X,

That makes lots of sense! Thanks! I'll be sure to include this in the guide.

Adrian Wong
Adrian's Rojak Pot
http://www.rojakpot.com/
http://www.adriansrojakpot.com/

Comment #103

Hi,

Regarding your BIOS guide, you suggested Enabling SMART capability even if you do not plan on using it. I did enable it, as it was disabled by default when I got my system. However, I pinpointed a problem that I was having to the enabling of SMART capability.

For some reason after I enabled it, whenever I used my system to do something network or internet related (Downloading files off the server in my apartment, or playing a network game), my system would spontaneously restart. I disabled SMART, and everything works well again. I love your guide, you are doing a great job, but I just thought that addressing this might help some people who may have the same problem.

Regards,

Johnathan P. Dinan

Hello Johnathan,

Thanks for that tip! I'm not sure if SMART is what's actually causing such problems but there's always such a possibility since SMART is designed to send its data through a network. SMART may be sending packets of data through the network when there's nothing monitoring / receiving them. This may have caused the spontaneous restarts. So, I'll just add it into the guide just in case other people face similar issues.

Adrian Wong
Adrian's Rojak Pot
http://www.rojakpot.com/
http://www.adriansrojakpot.com/

Comment #104

AGP Aperture Size setting with 3dfx Voodoo3 and Banshees

Dear Adrian,

These two cards do not utilize AGP Texture swapping according to 3dfx's specs. Since there are no textures being swapped across via the AGP bus, no system RAM is utilized for storage (everything in handled locally by the onboard video RAM). So in this case what should I put as the AGP Aperture settings? Can any value do?

Steven Phua

Hello Steven,

Both cards do not AGP texturing. Therefore, the AGP Aperture Size has no meaning for those cards. You can set it to any value you like but it will have no effect at all.

Hope that helps you some! ;)

Adrian Wong
Adrian's Rojak Pot
http://www.rojakpot.com/
http://www.adriansrojakpot.com/

Comment #105

My Asus P2B-F offers these two options with USWC being faster if the video card supports it but dangerous because your system may not boot if it doesn't. Will it really offer a speed boost? My system boots fine with it on but I am wondering whether it is a waste. Great guide by the way!

Thanks

Dylan Nash

Hello Dylan,

USWC or Uncacheable Speculative Write Combination improves performance for Pentium Pro systems (and possibly other P6 processors as well) with graphic cards that have a linear framebuffer (all new ones do). By combining smaller data writes into 64-bit writes, it reduces the number of transactions required for a particular amount of data to be transferred into the linear framebuffer of the graphics card.

As the Asus manual has probably noted, it may cause problems like crashes, video problems and an unbootable computer. Unfortunately, the only way to know if your card supports it is to test it yourself.

Since I do not have such a setting in my own BIOS, I can't test it and tell you the degree to which it will improve your system's performance or if it really improves it at all! Like above, the only way to know is to test it yourself.

PS. I have tested FastVid, a utility which supposedly does the same thing (Check out the old article - The Phoenix Project). The test results showed that it does more harm than good with an Intel 440BX motherboard.

Adrian Wong
Adrian's Rojak Pot
http://www.rojakpot.com/
http://www.adriansrojakpot.com/

Comment #106

Hi again,

Regarding comment #89 again (I won't let you off the hook that easily :-)...

BTW, I forgot to mention something in my first comment: I think your guide is much better than the manual to my ABIT MB, and I haven't found anything better on the 'net, so it is appreciated. But there are some areas where it needs to be improved, IMHO, and I'm trying to help where I can.

Also, I work as a programmer and have been writing programs for more than ten years (as a hobby or as work), with quite a bit of assembler programming. I'm also interested in computer architecture, and I've read a bit about it. In all, I think I know quite much about how a CPU works on a fairly low level, even if I'm not an expert on the subject.

AW> You are correct to say that if the BIOS isn't used, it will not be
AW> cached at all. But during bootup, the BIOS will be used and thus,
AW> will take up some cache bandwidth. Similarly, changing refresh
AW> rates will access the video BIOS and thus cause the BIOS to be
AW> cached. This are all the needless caching that we want to avoid.

Oh? So you're saying caching is needless now? I wonder then, why did the Athlon designers decided to include 128 kB L1 cache and up to 8 MB L2 cache? (I am a bit sarcastic here... :-)

AW> Note that caching the BIOS will not result in a faster booting time
AW> because after the BIOS is read, it will then be cached for future
AW> use. Thus, only the next BIOS access will be faster but because
AW> there will usually be no BIOS access after that, there will be
AW> absolutely no performance improvement from the caching of the BIOS.

The BIOS isn't read "just once". It contains "programs" that are executed during boot, and the CPU reads those programs piece by piece (a few instructions at a time), making many thousand accesses in total (maybe hundreds of thousands).

A cache is useful for three main reasons:

1)   To the CPU, every single RAM access is slow (and ROM accesses are even slower), whereas cache accesses are fast.
2)   Programs tends to spend lots of time in small areas of code (I've seen the figures 80% of the time in 20% of the code). This is caused by loops, functions etc., where the same code fragment is executed many times (and, consequently, the same bytes are read many times). Caches are designed to exploit this.
3)   The cache always reads entire "lines" (e.g., 16 bytes) - if possible using more efficient burst reads - making it a kind of read-ahead caching.

The two last reasons works together to give the cache a high hit rate, in the range of 90 to 95% (in general). And all three reasons work together to make the CPU run faster, because most accesses can be satisfied by a quick look in the cache (L1 or L2), rather than accessing the slow RAM - or even slower BIOS ROM. So, even if the BIOS isn't used that much, when it is used the cache often make things faster.

In the end, the difference in boot speed might be neglible, so one could argue if this is worth arguing about. :-) I'm just trying to get the facts right here, that's all. As Jonathan states in comment #82: "You
must realise there is no point in having a guide if the information in it is wrong."

AW> If I'm not mistaken, because of a linear framebuffer, that section
AW> of the video RAM will appear to be part of the system memory and
AW> thus cacheable. The purpose, I guess, would be to increase the
AW> transfer rate between the system RAM and the video RAM. But like I
AW> wrote in the guide, it works much better in theory than in
AW> practice. In fact, it slows down the overall performance of the
AW> system by wasting L2 cache bandwidth on something that doesn't need
AW> caching as much as the system RAM.

No, the purpose is to make CPU accesses to the video RAM faster, which isn't the same thing (again, the cache is used by the CPU only). The interesting question in this case is: what is meant by "video RAM"? I get the impression it doesn't apply to the video modes normally used by Windows. If so, how could it affect speed when running Windows applications?

About the only test where this could make a clear difference, AFAICS, is the DDSpeed tests, since they test raw CPU to video RAM access speed. But the variations there are as large (or small) as in other tests, that stresses other things, indicating that the setting has no effect at all.

AW> I believe in the Optimization of The Intel Celeron 450A and Diamond
AW> Viper V330 System, the first sentence of my conclusion went "Well,
AW> it's pretty clear that the Video RAM Cached feature and FastVid do
AW> more harm than good.". While the differences may not be much, note
AW> that they were consistent and thus cannot be explained away as mere
AW> variances.

Consistent? Not quite. In 25% of the tests (excluding Final Reality and the overall 3DMark test), caching seemed to make the system "faster" or did no harm. And in some 3DMark tests, the actual frame rate was the same, even if 3DMark claimed caching made the system slower. If you
include these in the "no harm" category (not unreasonable to do, I think), you suddenly have almost 40% of the tests where caching did no harm or even improve performance.

I ran the D3D test (as found on your site) a couple of times here, taking the average of the results from run 1 and 2 (A), as well as run 3 and 4 (B). The result:

Test                Average A       Average B
Fill rate                74.60                74.63
Polygons          1165.10            1153,20
Intersections        20.07                20,06

The differences between the averages are about the same as in the Phoenix Project for the same test, in one case even noticably larger. And I haven't touched the video RAM caching option in between. I just ran the test four times in a row. This suggests that these "mere" variations does make up for all the differences. Try yourself.

You do say in the testing procedure that most tests were run twice and then averaged. Why do that if there are no variations? And averaging two values won't eliminate any variations, just reduce them (halve them, I think).

(Side note; when caching was slower, even by very little, you say on the Phoenix pages that it, e.g., "scored slightly lower than that of the standard system". But when caching was faster by equal amounts, relatively speaking, you say it "didn't do anything at all". I sense a bias here... ;)

Magnus Holmgren

Hello Magnus! :)

About the "needless caching", I meant the caching of the video BIOS. It's really needless since it's not constantly in use and thus would not benefit much from a cache. The L2 cache's bandwidth would be better used to cache the system RAM instead.

If what you say about the BIOS during boot-up is true, then you could be right - caching the BIOS could speed up booting up speed. I'll have to run some tests to confirm that later. However, it's questionable the reduction in boot-up time would be anything significant. Also, the booting up time is only a fraction of the time we spent on our computers. So, a faster booting time might not be worth it if it reduces overall system performance.

About the caching of the video RAM, it involves the caching of the video RAM at A0000h-AFFFFh region via the L2 cache. You can get more details about it in my recent guide - Video RAM Caching. In any case, the benchmark tests I ran showed that it actually decreased overall system performance even though it increased the performance of the graphics card. This goes to show that while a particular subsystem may benefit from a BIOS feature, it does not necessarily mean the entire system will as a whole benefit as well.

Actually, the test results were consistent. If a series of three results (before averaging) were all consistently higher or lower than an opposing series of three results, I would call that consistent results. The standard benchmarking procedure of running three separate tests and averaging the results were done to obtain a more accurate result, not a more consistent one. If the Video RAM Cached feature (see Video RAM Caching) consistently causes the system to perform even 0.1% slower in the various benchmarks, then I would say that it decreases performance and should not be enabled.

Caching is supposed to improve performance of the subsystem it acts upon. In most cases, it does and the whole system benefits. But in some cases, like in the case of the caching of the video RAM by the L2 cache, it means the sacrifice of L2 cache bandwidth to boost the bandwidth between the video RAM and the CPU. To be honest, caching the video RAM does improve the performance of the graphics card a little. But the drag it caused on the overall system performance was greater than its benefit to the graphics card. While the difference may be minimal at best, I doubt it's in anyone's best interest to slow down their systems (even by a little bit) just because everyone believes caching will improve performance.

Adrian Wong
Adrian's Rojak Pot
http://www.rojakpot.com/
http://www.adriansrojakpot.com/

Comment #107

Adrian, I have this question about the SDRAM configuration in BIOS.

The BIOS in my Asus P3B-F lets me adjust the SDRAM Configuration. The options are: By SPD, 125Mhz (8ns), 143Mhz (7ns), or Manual (set CAS latency, RAS to CAS delay, etc. yourself).

My RAM is a PC100 Samsung GL (not the very good Samsung).

I've been running at 100Mhz bus speed all this time. Then I tried selecting the 125Mhz, and 143Mhz setting in the BIOS. Stable in both settings. I know this doesn't necessarily mean my RAM can run at 125Mhz & 143Mhz.

Being curious, I benchmarked the memory performance using Sisoft Sandra. The results kinda surprises me:

@125Mhz :
CPU/Memory bandwith = 232 MB/s
FPU/Memory bandwith = 220 MB/s

@143Mhz:
CPU/Memory bandwith = 282 MB/s
FPU/Memory bandwith = 260 MB/s

What do these values mean??

I doubt my memory can go to 7ns. But the BIOS setting is at 143Mhz now, and the SiSoft results are certainly showing a difference.

Can you explain this?

Michael Ting

Hello Michael,

You are right in that setting the BIOS setting of the SDRAM to 143MHz will not set your SDRAM to run at 143MHz. Your SDRAM will run at the same speed as the system bus, irrespective of just how high your SDRAM can actually run. However, faster SDRAM (those that can run at higher than 100MHz) will normally be able to run at 100MHz with faster timings - ie. lower RAS-to-CAS delay, lower CAS latency, etc...

So, what that BIOS option actually do is select the best SDRAM settings for the SDRAM speed that you have selected. Normally, you should leave the option as By SPD. That will allow your BIOS to query the SPD chip on the SDRAM module to determine the optimal settings for the module. However, optimal settings in this case means maximum, safe settings. The SDRAM DIMMs can probably run with faster timings without a problem.

That's what the other settings are for. Although your SDRAM has an access time of 10ns, you can set the BIOS to consider it as a 125MHz-capable DIMM. Then, the BIOS will set faster timings for the SDRAM operation. If the quality of your DIMM is good, it will be able to operate with the faster timings with no reliability issues at all. That will enable the SDRAM to work faster and increase the effective SDRAM bandwidth.

If you are feeling a little adventurous, you can try to force the BIOS to consider it as a 143MHz-capable DIMM in order to elicit even faster timings. That will further improve the performance of the SDRAM DIMM, at the possible expense of unreliable operation. But if you find that your system has become unstable, decrease the timings immediately.

Adrian Wong
Adrian's Rojak Pot
http://www.rojakpot.com/
http://www.adriansrojakpot.com/

Comment #108

I won't go and flatter you, because with all this praise, your head is liable to explode, and I wouldn't get to see any new things on this marvelous page. =^)

I just wanted to post an alternative (at least for Netscape users) regarding printing your pages. In comment #58, the problem of printing only colored text, but not the white text that comprises most of the page was detailed, and comment #64 gave one fix. Here's the alternative that doesn't disrupt the color scheme of a page:

Under the File>Print Setup menu, you need to check the following:
Black Text
Black Lines

and voila, you can print nearly any text on nearly any page - unless it's a picture.

Hope this helps.

Sincerely,

Alex Calara

Hey, thanks for the tip, Alex! :)

Adrian Wong
Adrian's Rojak Pot
http://www.rojakpot.com/
http://www.adriansrojakpot.com/

Comment #109

Hi Adrian,

I was wondering if you'll be doing a "Power Management" section for your BIOS tweak page? It'd be great for dumb-asses like me ... hehe :) Anyways, keep up the good work! -- a very informative site that I visit frequently for updated tweaks :)

Peter Cheung

Hello Peter,

I'll definitely include a Power Management section when time permits. Just hang in there in the meantime! :)

Adrian Wong
Adrian's Rojak Pot
http://www.rojakpot.com/
http://www.adriansrojakpot.com/

Comment #110

Great site! I am a tech junkie running a K3-400 @ 450 with 160MB PC100 on a Tyan Trinity 1598s with 2MB L2 (L3 actually). I often check back at ur site to see what opinions u and other folks are having about the BIOS. I agree with a lot of your findings but i think that those findings aren't always accurate for everyone because of the system ur using.

The Celery as I call it is not a powerhouse cache setup whereby the K6-x's are. My system performs better with both system and video BIOS cacheable enabled. Also the AGP Master 1 WS write and read "add" one clock tick to AGP write operations according to my BIOS guide. With both disabled, I get much much faster performance on my machine.

However I only leave AGP Master 1 WS read disabled for a happy medium. Leaving them both off my machine gets ahead of my connection in multiplayer games! As far as the enabled caching of system and video being faster that may have to do with the 4ns 2mb onboard L2 (L3) not sure what u think?

Keep up the great work!!!

David Pickard

Hello David,

You are right. Different chipsets have different optimal settings. This guide is based almost entirely on the Intel 440BX chipset. So, it may not be accurate for other chipsets. However, I doubt, even with a large and fast L2 cache, the system will benefit from caching the system and video BIOS. As I have noted, both BIOSes are hardly needed - only for legacy use. So, caching them will have little, if no use at all.

Adrian Wong
Adrian's Rojak Pot
http://www.rojakpot.com/
http://www.adriansrojakpot.com/