Secrets revealed on Video card fine tuning?

Resources, guides, discussions for modding your system.

Secrets revealed on Video card fine tuning?

Postby atang1 » Sat Feb 15, 2003 10:43 pm

Monitors are slow because of the 1/30 second to complete a picture. LCD flat panels are even slower due to tin oxide liquid crystal speed to change state(transparent or opague).

The video card chipsets vary due to the scheme of video accelerator for processing and rendering. The specification gets very complex, today. The concept however, is just using video instructions from cpu to do vector video processing by using a video processor and much memory to cache stages of computation and data transfer to the monitor. Some times also cache previous frame(that we want to eliminate the ghost frames).

So, in the bios, you can adjust some video basic input and output system, including shadowing video bios. Then in the operating system software, Windows control panel can set the limit on color and resolution. The color and resolution is limited by the final size of memory resided on the video card.

For instance, 2 mb of memory on a PCI video card can do true color of 32 bits and 800z600 resolution. If you change to 1024x768 resolution then the color will be limited to 16 bits. You ran out of memory on the video card to do 32 bit color. This also tells you that 2 mb is one frame of video. It tells you that you do not need two transfers from your memory to your monitor. Single burst of data transfer is the fastest data transfer you can have, therefore the fastest performance for your video presentation.

On the other hand, if you use less memory per frame, and the rest to cache another frame' you are wasting your resources. If the next frame can not be squeezed into the leftover free memory, then data transfer will be delayed by the residual memory transfer before monitor will get the complete picture.

Large dram memory on the motherboard is also a problem when you have too much cached video files. Memory of cached files must be cleared frequently to speed the computer operation(eliminating overhead of reading long FAT tables full of unimportant files).

The control panel, monitor setup, can help you speed up your video card. Next, how to adjust the bios for fastest video operation, depending on what vintage your bios was written?
atang1
 

Postby atang1 » Sun Feb 16, 2003 11:29 am

Earlier bios are very simple, shadow the video bios on your video card. Then you are allowed to cache the video ram in your L2 cache. The read and write of the video to the monitor is speeded up.

Then came the accerator video card. The memory has to be increased to 2 mb for the color and resolution combination. Your monitor still can not go beyond 1024x768 except for graphics workstations. The 2 mb per frame is true for windows operating system regardless of PCI or AGP video cards. Gaming requirement of 32 mb is 16 frames of pictures ready to go to the monitor without delay. Data streaming if you will. So, whether we need 32 mb of frame size is questionable, and size determination should be experimented for fastest speed of computation.

ISA video cards gave way to 33 mhz PCI video cards. PCI video cards give way to 66 mhz AGP video cards. Now 1x, 2x 4x(low voltage) give way to 8x AGP video cards. But still AGP FSB frequaency is 66 mhz. 8x is merely an internal multiple of the 66 mhz frequency.

Initially, the bios will allow you to set the aperture size. The dram memory size that AGP can share with other memory usage. Then, the frame buffer size can be set in the bios. Now that DDR is used on the video cards, slewrate and driving strength becomes important adjustment to keep operations smooth and fast.
atang1
 

Re: Secrets revealed on Video card fine tuning?

Postby atang1 » Mon Feb 17, 2003 6:27 pm

atang1 wrote:Monitors are slow because of the 1/30 second to complete a picture. LCD flat panels are even slower due to tin oxide liquid crystal speed to change state(transparent or opague).

The video card chipsets vary due to the scheme of video accelerator for processing and rendering. The specification gets very complex, today. The concept however, is just using video instructions from cpu to do vector video processing by using a video processor and much memory to cache stages of computation and data transfer to the monitor. Some times also cache previous frame(that we want to eliminate the ghost frames).

So, in the bios, you can adjust some video basic input and output system, including shadowing video bios. Then in the operating system software, Windows control panel can set the limit on color and resolution. The color and resolution is limited by the final size of memory resided on the video card.

For instance, 2 mb of memory on a PCI video card can do true color of 32 bits and 800z600 resolution. If you change to 1024x768 resolution then the color will be limited to 16 bits. You ran out of memory on the video card to do 32 bit color. This also tells you that 2 mb is one frame of video. It tells you that you do not need two transfers from your memory to your monitor. Single burst of data transfer is the fastest data transfer you can have, therefore the fastest performance for your video presentation.

On the other hand, if you use less memory per frame, and the rest to cache another frame' you are wasting your resources. If the next frame can not be squeezed into the leftover free memory, then data transfer will be delayed by the residual memory transfer before monitor will get the complete picture.

Large dram memory on the motherboard is also a problem when you have too much cached video files. Memory of cached files must be cleared frequently to speed the computer operation(eliminating overhead of reading long FAT tables full of unimportant files).

While the single frame size concept is interesting, the color bits actually slows down the pixel transmission speed. Changing back fro true color 32 bits bcak to 16 bits, video 2D pixels doubled in speed, but video 3D pixels sped increased 5 times.

The control panel, monitor setup, can help you speed up your video card. Next, how to adjust the bios for fastest video operation, depending on what vintage your bios was written?
atang1
 

Postby atang1 » Thu Feb 20, 2003 9:04 am

How fast is 10ns sdram dimm compare to edo simms at 60ns in data transfer rate(MB/s)?

This is rather interesting. that I continue to use Winetune98. The Wintune results show a minimum of differences. Sdram at 315 MB/s and EDO at 275 MB/s.

The latest Sisoft Sandra of 2003 standard final, which has just been reported by Pcworld, shows up more details and comparisons with other typical memory results.

With Sdram and EDO mixed, it shows 85 MB/s With EDO simms alone, it shows 140 MB/s. With Sdram alone, it shows 162 MB/s. And Sandra has four selectible results of faster memories for you to compare your results.

Considering EDO simms has only a bandwidth of 16 mhz on the drams but 66 mhz at data bursts; and the synchronous drams has a bandwidth of 66 mhz, the difference is amazing.

Why do I study these older technology? It tells us the difference of pipelined burst, or netburst technology of the recent Intel cpus(P4) or hypertransport of Alpha and AMD XP cpus or Rdram, and the synchronous parallel 32 bit data transfer of DDR. DDR may win by a slim margin of a photo finish(by a nose if you will) depending on frequency used.

Study the latest Sandra published results of different parts in a computer, and you know what to choose for your next dream computer. But did Sandra test the overall performance due to wait state fine tuning, and resource management of the parts to achieve the best performance? Not yet, and may never is the answer. You have to do it yourself or let Microsoft adjust everything, hardware bios and software timing loop for you, transparently.

This is the life, highest standard of living, someone does everything for you. You just have to pay thru the nose. Linux can never achieve that for you, not until some one like Microsoft takes over. Is Lindows or Lycons or maybe even Red Hat that some one, only Walmart(they have the buying power to dictate their needs) knows?
atang1
 

Postby atang1 » Sat Feb 22, 2003 5:55 am

Now that we know the relative memory speeds, video cards can be overclocked simply and especially with earlier PCI video cards.

The easiest way to overclock the PCI cards is to change the clock ratio of FSB and PCI bus. The originall specifications are based on 33 mhz 32 bit bus. This turns out to be 1/2 ratio for the original 66 mhz FSB. Then the chipset can do 75 and 83 mhz frequencies, and the selection for PCI bus became 33 mhz and 1/2 ratio. 1/2 of 75 and 83 mhz can be translated into percentage of increased speed of video cards.

So, which video cards can be overclocked? It is not really a mystery. The date codes on the semiconductor or the pcboard will give you an indication of the feasibility.Vintage 97-98 PCI video cards are overclockable to over 40 mhz PCI bus. Some video memeory are edo with 83 mhz capability. Older ones use 60 ns FPM memory. And same with vintage PCI modem cards, definitely overclockable.

There are , of course, software for video card overclocking or performance tuner. Perftune.exe is the performance tuner(219 kb). You can still download it from the internet free.

When it comes to AGP video cards, overclocking is tied to the FSB increase at a fixed ratio. This is the limiting factor sometimes to be able to overclock the cpu. Earlier vintage 4mb 1x AGP are seldom overclockable. Most of the yr2000 AGP video card will have a fair chance of slight overclocked capability. Look at the sdram date codes?

Tha newer video cards of course came with overclockable software to make things easier. However, my interest is to help people with older equipment to squeeze out a little more performance to get on the internet.
atang1
 

Postby atang1 » Wed Feb 26, 2003 9:41 pm

Terminology of features in bios to fine tune video card speed and power has been changing.

video ram caching, video buffer, and frame size are maybe one of the same.

Video aperture size in dram is for video data processing before sending the data stream to the video card.

They are differnt levels of cache. Almost like L1 and L2 cache for cpus. But they both resides in the dram, and has to have FAT table and protection mode.

Once you understand the level business and the size and power depending on application software needs. You know how to size them with respect to each other. You size up the frame so that video speed in MB/s reach a plateau. Then you size up the aperture until the same video speed reach a plateau. This will be the fastest video you can achieve. regardless of which video card.

Then you can adjust the video card by overclocking.
atang1
 

Postby atang1 » Sat May 10, 2003 8:15 am

To understand how to fine tune video card, is to understand the development of video cards.

Earliest video cards depended on 6845 chip. It is buffered at B8000 and all the address registers for cga standards are defined in Xt/at bios. Then came Ega and Vga buffered at A0000. The address registers werre changed for more capcbility. Vga started to use DAC(digital to analog converter), and PEL mask, address and data registers. We still use Vga to springboard to fancier video schemes and better performace.

In 1996, S3 developed the Trio64 chip, using a Vga bios which kick starts the 64 bit video system with EDO video memory runing at 83 mhz-60 ns. The PCI bus was runing at 33 mhz.

In year 2000, Agp(1x 66mhz) video cards used S3 Trio 3D/2X on board chip. Video memory uses 16 mb dram technology. 4 memory-10ns chips for 8 mb. Vga bios chip was ysed to boot on Vga video.

All this of course points out the overclocking capabilities of the PCI or Agp buses. The limitation is in the speed of the video memory against the bus frequency and the wait state in the video card between the video processor and the video memory. While it is easy to overclock the bus frequency, it is hard to change the wait state on the video card.

So, starting in 1998, people begin to sell Ramdac cards with 3d redition engines with larger memory to suplement the 2 mb PCI video cards. These accelerated cards made the migration path for today's video card designs. More data bits abreath. More memory for video signal processing. In the end, more video data processing slows down the video cards.

So, how do we fine tune these video cards be it old and less video processing or those later video cards with more speed and more data processing? That is the question.
atang1
 


Return to System Modification

Who is online

Users browsing this forum: No registered users and 2 guests