For more details, and talks in past semesters, consult the full schedule of talks.
Past topics can (and should) be repeated occasionally. In addition, here are some topics people might like to hear about:
To give a talk, please contact swig@math.arizona.edu.
If you intend to use your computer for extended periods of time or plan to do a lot of graphics work, you will want to carefully consider your chioce of monitor and video card. Inadequate video hardware can cause eye strain, headaches, and nausea.
Monitors range in price from under $300 to several thousand dollars, and most video cards cost under $300. Given this fact, your best bet may be to find a monitor that meets your needs and budget, and then choose a video card that matches the monitor's specifications. Driving a low quality monitor with the world's best video card (or vice versa) doesn't really do you much good; the two go hand in hand.
There are two screen technologies commonly in use in monitors: shadow mask and Trinitron. Trinitron monitors can be sharper, brighter, and generally yield a higher image quality, all other things being equal. Trinitron monitors are normally more expensive than their shadow mask counterparts.
Some newer shadow mask monitors realize an image quality that rivals similar Trinitron monitors -- be on the lookout for these. Also, Sony's patent on Trinitron technology expired, and so other companies can now produce Trinitron monitors. Some of these are substandard, and you'll want to avoid them.
Screen size can be an important factor. For example, if you plan to spend a lot of time previewing TeX documents, you will want a larger screen so that you can see more of the document at once. When reading monitor spec sheets, remember that there is a distinction between the physical size of the screen and the maximum viewable image size on the physical screen. Quoted screen sizes are normally measured along the diagonal, with an implied 4:3 aspect ratio. For example, the CTX 1785XA 17" monitor has a 16" (diagonal) viewable area -- about 89% of the physical screen area is viewable. Contrary to popular cybermyth, monitor size is not a robust indicator of relative machine speed (or monitor quality, for that matter).
The quality of the electronics that drive the monitor's picture tube have a profound effect on the displayed image. To an extent, this quality can be ascertained from the monitor's technical specifications as described below.
When buying a monitor, it is a good idea to go to a store that has a range (in price/quality) of monitors on display, side by side. Let your eyes be the final judge.
Price will usually limit the class of monitor you buy. You should be aware that monitor prices tend to fall more slowly than do CPU prices. In other words, there are very few "steals" in the monitor world -- you get what you pay for.
The horizontal sync rate is basically the number of lines per second that the monitor must draw at a given resolution and refresh rate. This is mainly important for Linux users: if you exceed the monitor's rated HSync specs, you can burn out the monitor's circuitry. Really. Fire, sparks, etc. Dark screen follows.
Some monitors can only sync (ie, display) at a fixed set of horizontal and vertical frequencies. The complement of this set consists of what are called "multisync" monitors -- they can sync to any valid video signal in a given range. Buy a multisync monitor.
The monitor's bandwidth is, roughly speaking, the rate at which it can draw the pixels on the screen that make up the image you see. To analyze the bandwidth figure, which is usually quoted in MHz, we need to discuss display resolution.
You computer's video card renders the displayed image as a series of dots, called pixels. To the video card, the screen is simply a matrix of pixels, each with its own color and intensity. The display resolution is simply the dimensions of this matrix. During every vertical refresh period, the card shoots the pixel data to the monitor in a left-to-right, top-to-bottom sequence.
What resolution should you choose? Part of the answer depends on the operating system you are running; the other part depends on the size of the monitor's screen. Many operating systems provide the following standard display resolutions:
To see how the monitor size constrains the usable display resolution, consider the letter "x". To draw an "x" on the screen, the computer colors pixels in a manner appropriate to produce something that looks like "x". In a given font, the "x" has fixed size in pixels. As you raise the display resolution on a fixed physical screen size, the "x" will become smaller. At some point, it will become so small that you will need to raise the font size to read the text without squinting. Unless you have specific need of extremely high resolution, you have gone nowhere: your window is the same size (on the screen) as it was in a lower resolution.
The following table lists some (ad hoc) "natural" display resolutions versus monitor sizes.
| Screen Size | Typical Resol | Max Resol |
| 14" or 15" | 800x600 | 1024x768 |
| 17" | 1024x768 | 1280x1024 |
| 19" or 20" | 1280x1024 | 1600x1200 |
As a rule of thumb, 1280x1024 or better is real nice for previewing TeX documents. Less than 1024x768 can get real frustrating.
So, how to make use of all of this? Let's say you want a 17" monitor and that you want a 1280x1024 resolution with a 75 Hz vertical refresh. To compute the required bandwidth, you can use the formula:
Bandwidth = 1.35 * HRes * VRes * VRefresh
The magic number 1.35 is to take the horizontal and vertical blinking regions into account. These regions appear as the dark border at the edges of the screen. For Linux people who roll their own display modes: setting the blinking region correctly is a religious issue but is essential to obtaining a stable display. Setting the width of the HSync and VSync pulses and placing them correctly with respect to the blinking regions is even deeper magic.
In general, the allowed bandwidth and sync rates increase with monitor size (and cost), so as to accomodate the larger screen area. However, this is not always the case: there are 21" monitors out there that only support 1024x768! Remember: with a monitor, you get what you pay for. Be on the lookout for horse traders.
A monitor, due to its price, is not something you want to upgrade. Therefore, you should spend as much money as you can on your initial purchase. RAM and disk prices have fallen dramatically in recent years, while monitors have remained relatively constant (and expensive). When making your initial purchase, do not skimp on the monitor! Toss out some RAM or disk and buy more of that later when you have more money.
To be concrete, consider the following (local) February 1997 costs:
| What | Cost |
| 32 MB EDO RAM | $180 |
| 2.1 GB EIDE HD | $259 |
| CTX 1785XA 17" | $999 |
versus the February 1998 costs (same local vendor):
| What | Cost |
| 32 MB EDO RAM | $89 |
| 4.3 GB EIDE HD | $259 |
| CTX 1785XA 17" | $625 |
The price per unit of storage (RAM/disk) dropped by 50% in the last year, while the same model of monitor only dropped by a third. The moral is: if you don't need 128 MB of RAM and 16 GB of disk in the short run, buy a better monitor and pick up the storage later.
Color depth refers to how much color information the video card stores per pixel of screen image. This determines both the number of simultaneously displayable colors and the number of distinct displayable colors.
All SVGA cards can display 8 bits per pixel, for example. This means that a maximum of 256 distinct colors may be displayed simultaneously. Each of these 256 colors can be selected from a palette of 16,777,216 colors. If your card can display 16 bits per pixel, a maximum of 65536 colors may be displayed simultaneously. Each of these colors is encoded as 5 bits of red, 6 bits of green, and 5 bits of blue. If you can display 24 bits per pixel, 16,777,216 colors can be simultaneously displayed, and each pixel is encoded as 8 bits per R,G,B.
Having more than 8 bits per pixel can be very nice: the JPEG's you view in netscape are beautiful. There is a downside, however: you will need extra video RAM to store the extra data, and rendering will be slower as there is more data for the CPU to send to the video card (but see the discussion of AGP below).
As stated earlier, the video card considers the screen to be a matrix of pixels. These pixels are stored in high speed memory known as video RAM. The amount of video RAM you need depends on both the number of pixels at a given display resolution and the color depth chosen.
For example, if you want a resolution of 1024x768 at 8 bits per pixel (1 byte per pixel), you will need 1024x768x1 bytes or 768 kB of video RAM. Almost all cards come equipped with at least 1 MB of VRAM now, so this mode shouldn't be a problem. If you want 1280x1024 at 16 bit per pixel, you'll need 1280x1024x2 bytes or 2.5 MB of VRAM. Cards usually come equipped with 1, 2, 4, 6 or 8 MB of VRAM, so you'll need a 4 MB card. Finally, if you want 1600x1200 at 32 bits per pixel, you'll need 1600x1200x4 or 7.32 MB of VRAM -- so you'll have to get an 8 MB card.
Buy all the video RAM you're likely to need when you purchase the video card. Some vendors will tell you that you can upgrade later. In practice, however, it is almost impossible to get your hands on VRAM for a card after the fact (because the video card manufacturers keep changing it around).
Maximum dotclock is to the video card what bandwidth is to the monitor. Bandwidth measures how fast a signal the monitor can track; maximum dotclock measures how fast a signal the video card can produce. For a given resolution and refresh rate, you compute the required dotclock according to the formula given above.
There's a chip on the card called a "RAMDAC". Some cheaper cards use cruddy RAMDAC chips. At 8 bits/pixel, all will be well, but above 8 bits/pixel, the maximum dotclock will drop significantly. The best way to avoid these beasts is to buy name-brand video cards. Reputable manufacturers include ATI, Diamond, Matrox, and and many others. If someone says "it's just like a Diamond", you should say "well then give me a Diamond for the same price -- you keep that one." If you're tempted by a deal, make sure you have a manual for the card and can get (at last some of) the following information about the manufacturer:
There are a number of different PC bus architectures in use. Each type of bus slot requires a card with a specific interface. Here's how the bus types relate to video cards:
| Bus | Comments |
| ISA/EISA | The Industry Standard Architecture (ISA) and Extended ISA (EISA) busses are standard for all Intel architectures; virtually all system boards have a number of ISA/EISA slots. (E)ISA video cards are very inexpensive; however, their performance is poor -- especially if you have an i486 or higher CPU. This is primarily because the bandwidths of the ISA and EISA busses are 8 and 33 MB/sec, respectively. |
| VLB | The Vesa Local Bus (VLB) is used only on i486 machines. If you have a 486, you will probably want to get a VLB video card because the Local Bus has a much higher throughput (up to 132 MB/sec bursts) than the (E)ISA bus. |
| PCI | The Peripheral Component Interconnect (PCI) bus is the standard for Pentium and Pentium Pro machines. The PCI bus a much higher bandwidth (132 MB/sec) than either the (E)ISA or VLB architectures. Most currently available video cards are designed to work with the PCI bus. |
| AGP | The Accelerated Graphics Port (AGP) bus architecture is the up-and-coming video card standard for Pentium II machines. The AGP graphics slot has a bandwidth of 528 MB/sec and is independent of the PCI bus. This means that bus contention between high speed disk and network controllers and the video card is greatly reduced by the AGP design. Also, AGP boards can do texture mapping directly out of system memory, instead of having to download texture bitmaps from main memory into the card's (limited) video RAM. |
The main thing to note is that the graphics card shares the PCI bus (and its 132 MB/s bandwidth) with all other PCI devices -- including your hard disk controllers (for HDD and CDROM devices) and ethernet card.
The AGP design connects the bandwidth-hungry video card closer to the CPU, freeing up the PCI bus for other things:
More information about AGP can be found at the Intel AGP page.
Special features covers a spectrum of possibilities. For example, a card may support hardware alpha blending (primitive transparency in 3D rendering) in 32 bit-per-pixel mode or hardware texture mapping (mapping textures onto 3D polygons). Some ATI cards have a built-in TV tuner. In all cases, you'll need to determine what drivers are available for your operating system with respect to these special features, as well as what available software makes use of the available drivers.
A reasonable buying strategy is as follows: decide on the screen resolution and vertical refresh rate you'd like to run and pick a monitor that meets these specifications (and that you're otherwise happy with in terms of size and image quality). Then, decide on the color depth you'd like to have at each resolution to figure out the amount of video RAM you'll need. Then, pick a card with an appropriate dotclock limit and all of the other features you'd like.
| Model | Video RAM | Max Dotclock | Bus | Price |
| Diamond Stealth 3D 2000 Pro | 2 MB | 170 MHz | PCI | $73 |
| Matrox Millennium II | 4 MB | 250 MHz | PCI | $219 |
| Diamond Monster 3D II | 4 MB | 130 MHz | PCI | ~$250 |
| ATI All-In-Wonder Pro | 8 MB | 200 MHz | PCI | $323 |
| Matrox Millennium II | 8 MB | 250 MHz | PCI | $329 |
| Model | Technology | Size | VSync Range | Bandwidth | Price |
| Optiquest Q53 | Shadow Mask | 15/13.8 | 50-90 Hz | 110 MHz | $230 |
| CTX 1785XA | Trinitron | 17/16 | 50-120 Hz | 135 MHz | $596 |
| Princeton Graphics EO90 | Shadow Mask | 19/18 | 50-150 Hz | 210 MHz | $700 |
| Nokia 445 XPro | Shadow Mask | 21/20 | 50-150 Hz | 270 MHz | $1450 |
| Nanao TX-D7S | Trinitron | 20/18.8 | 50-160 Hz | 210 MHz | ~$1900 |
| CPU Performance | Graphics Performance | ||||
| System | SpecInt95 | SpecFP95 | Texture Fill | Poly/sec | System Price |
| Pentium II 333 MHz | 12.8 | 9.1 | 90 MPix/sec | 3 M | $4,000 |
| SGI O2 200 MHz R5K-SC | 5.4 | 5.7 | 42 MPix/sec | 473 K | $20,000 |
| SGI O2 195 MHz R10K-SC | 9.3 | 8.9 | 44 MPix/sec | 507 K | $30,000 |
| SGI Onyx2 InfiniteReality | 11.4 | 19.1 | 194-776 MPix/sec | 11 M | $300,000 |
The PC graphics specs (texture fill rate and polygons/sec) are for the 3Dfx Voodoo2 chipset -- this chipset will be available in the Diamond Monster 3D II when it comes out. The M3D-II is an add-on card that provides 3D acceleration to your existing video card.