Notice ~ 

Are you searching for our italian Community  Forumzone.it?
Main Menu
Menu

Show posts

This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.

Show posts Menu

Topics - ggab

#1
QuoteATI to be taken over by Intel or AMD?, Hostile takeover possibility

Andrew Lee, an analyst with TD Securities who specializes in ATI, has said that ATI is the possible target of a stock takeover, according to a report at the National Post (subscription required). ATI's stock saw a much-needed boost at the end of the week, fueling speculation that ATI is possibly being courted or even targeted for a hostile buy. The stock hovered around CAN$14.50 for most of the week, but after a dip Thursday to $14.00, the stock shot up to 15.77, up 8.6 percent for the entire week. It had reached a peak of $15.90, which is still a far cry from the $25 range enjoyed before June, when the stock started to tank. Shares hit their 52-week low on July 7, at $13.90.

Takeover rumors started in June, when the company's third quarter results were less than impressive, and were followed by weak projections for the fourth quarter. Rumors indicate that both AMD and Intel are interested in ATI, as are Broadcom and Texas Instruments.

ATI is in a vulnerable position because delays in their next-gen hardware have left many investors expecting to see NVIDIA dominate the rest of this year, including the lucrative holiday spending season. Nevertheless, ATI has announced that the R520 will be entering full production sometime in the fall, and their deal with Microsoft for the Xbox 360 should earn them some scratch. Indeed, while no one expects ATI to tank in the near future à la 3dfx, the market's lack of interest in ATI is apparent, and the time is right for buying stock if you expect to see ATI recover next year. And if you happen to be a large corporation with loads of cash, you might want to buy an awful lot of that stock.


Source: http://arstechnica.com/news.ars/post/20050709-5077.html?9799


anything is possible, many thought the same thing when the rumours that Nvidia was going to buy 3Dfx. and when Intel bought DEC Alpha. time will tell

Thoughts?
#2
What a great tutorial Patrick Demski has done! :)
http://www.swwisa.net/wever/V5-OC%20Tutorial/V5-OC%20Tutorial%20-%20Main.html



Overclocking The 3dfx Voodoo 5500

Article by Patrick Demski - 01/15/2003



Over a year ago I built my first computer system from the motherboard up. Since I wanted a high-performance box that could also run Red Baron 3D, I had to go with a 3dfx graphics card. I chose the Voodoo 5500 because I had one in my current system & was familiar with its configuration and performance tweaks. I also decided that since I was already elbow-deep in hardware tinkering with the rest of the system, I'd take the time to mod the V5 before installation so that it could be reliably overclocked.

Despite the age of this particular piece of hardware, there are still a lot of people running it in their boxes because they need Glide support — and it's still readily available on eBay. For anyone interested in trying to get the most out of it, here's an article about the mods I made to my card and the results that I got.


Overclocking?

Over the last few years I've read a lot of good stuff and some BS on the subject of overclocking CPUs, buses, vid cards, whatever. Just so we're all starting on the same page, here's a quick overview of overclocking in general.

Overclocking anything in a PC means that you're going to increase the clock speed of the device beyond manufacturer's spec so that the length of one processing cycle will decrease — and hopefully you can push the clock speed fast enough that real-world performance is noticeably improved. The clock of any digital device is just a generator chip and a crystal oscillator. The clock generator chip can usually be stepped faster or slower through a simple software interface (more on this for the Voodoo 5500 later). Stepping the clock frequency is the easy part — but getting the device to function reliably at increased speed is the trick.

Every electronic device is spec'd by the manufacturer to operate at a stated clock frequency. In the case of the V5, this is 166 MHz. The spec clock frequency is determined by the circuit design of the device itself, the bus and other external devices to which it's connected, the semiconductor fabrication process used to manufacture the device and various marketing considerations. Combinations of these factors impact any attempt to overclock a device. Looking at how some of these apply to the Voodoo 5500 in particular:


1. The Circuit Design

The V5 is a dual-processor video card with 64 Mb of memory. The dual graphics processors on the V5 are a 3dfx design called the V-100 processor. The card design was 3dfx's final attempt to trump nVidia in the high-performance 3D graphics accelerator market (actually, there was a prototype Voodoo 6000 with four V-100 processors that never got to production).

Users of earlier 3dfx products are familiar with an architectural feature called Scanned-Line Interleaving (or SLI). This allowed the connection of two Voodoo cards in a single PC so that they functioned as one, high-performance unit. An SLI configuration requires a special cable. The V5 architecture is essentially like two separate 3dfx 32 Mb cards that are SLI-connected — but without the cost and inconvenience of two physical cards and a cable. This means that the ability to OC any given V5 depends on both of the processors and the on-card SLI bus operating compatibly at the higher clock speed — three major processing components instead of the one found on a single-processor vid card. No matter how many neat mods you apply to the card, you may have a dog that operates OK in-spec, but doesn't OC at all. Another card with the same mods will allow you to crank it up by 30 MHz. It's just the luck of the draw.

One other circuit design factor worth mentioning is the video memory. 3dfx went with standard SDRAM from a variety of manufacturers. SDRAM in general does not OC well at all. I can't confirm this definitely, but I've read claims that the SDRAM on a V5 from certain manufacturers will take OC'ing better than others — in particular, Hyundai memory modules. This info originates with reports of overclocking the Voodoo 3000, so I'm not sure how applicable it is the the V5. In any event, the Voodoo 5500s with Hyundai memory seem to all be very early production units. If you're buying a Voodoo 5500, try to get one with memory modules from Hyundai (can't hurt, might help) — I've also seen cards with Toshiba and Hitachi memory, and supposedly these are not as good for overclocking. The board described in this article has Hyundai modules and, for whatever it's worth, I experienced no problems getting it to run OC'd.

2. The Semiconductor Fab Process

Electronic devices — chips — are manufactured in a facility called a fab, and each fab supports one or more processes. Processes are a synthesis of the kinds of materials used to construct circuit elements (CMOS, gallium arsenide, silicon germanium), and the discrete manufacturing steps that achieve a minimum possible element size. So, you'll see semiconductor fabrication processes referred to, for example, as 0.25 CMOS. This means that it's possible, using this process, to fabricate CMOS circuit elements as small as 0.25 microns in size. Common sizes are 0.35, 0.25, 0.15 and most recently 0.13 and even 0.10 micron.

Each process specs an operating max frequency for their circuit elements. Pushing past that frequency results in either an electrical or a thermal-mode failure. An electrical-mode failure is due to the device having significantly different electrical characteristics at the higher frequency than it does within its normal operating band. In other words, you can only increase the clock speed on a V5 to a certain maximum before one or more electrical circuit elements fail, stopping the operation of the chip where it's located. I mentioned earlier that the Voodoo 5500 uses SDRAM for video memory & SDRAM doesn't OC very well — it's a good bet that the max operating frequency of a V5 is going to be limited by the memory rather than any other component. There's nothing that can be done externally to increase this maximum, it's just physics in action. In general, the smaller the geometry — the higher the frequency that components fabbed in a particular process can support.

The other failure mode is thermal. Circuit elements give off heat when they operate and if that heat is not carried off quickly enough, individual elements will fail — either temporarily or permanently. Running any device at a higher frequency generates more heat than normal. It's easily possible to experience a thermal mode failure long before reaching the maximum operating frequency.

That's what overclocking is all about. The goal of any OC'ing technique is to ensure that the circuit does not experience a thermal breakdown before hitting the max possible operating frequency.

There's one other problem with operating electronics outside of their normal frequency range. Over the lifetime of a chip, it's circuit elements will switch untold bazillions of times. This normal operation causes the materials that the chip is made from to slowly change characteristics until the chip eventually fails completely. Consumer semiconductors use fabrication processes that are designed to allow a 10-year lifespan under normal operating conditions. Operating at higher-than-spec frequency tends to accelerate the "aging process" resulting in early failure. This applies even to chips that are well-cooled and run at reasonably low temperatures. It's impossible to say how much earlier for any given board, but OC'ing any electronic component measurably shortens its lifespan. For this reason, even if you're successful in getting your V5 to overclock, you should run it in this mode as sparingly as possible.

3. Marketing Considerations

Occasionally, a chip manufacturer will market products with an operating speed spec'd significantly lower than the device's capabilities (during those rare times when every fab employed is producing high-quality parts & good yield so you really don't have lesser-grade parts). There was a wave of interest in overclocking the early AMD Duron processors when people noticed that, with a few cooling mods, you could push the speed up by 30% of the spec'd clock frequency. Chip manufacturers do this to maintain the most profitable market segmentation between the low-end and high-end of their product line.

The Voodoo 5500 was supposed to reclaim the top of the vid card market from nVidia. They wanted screaming performance and you can take it to the bank that 166 MHz was the maximum speed possible — while still maintaining reliability in normal operation — for the production version of this product in stock configuration, . Don't expect to push it very far beyond this. In fact, the results that I got with the mods described in this article show that the V5 tops out at around 193-195 MHz — about a 16-17% bump above spec while maintaining stability. This figure was confirmed by comparing notes with some others who OC'd V5s. Although it's possible to push the speed even higher with certain applications, getting a Voodoo 5500 to run reliably for an extended period of time above 200 MHz is probably not possible.

OK, so, the goal of this whole exercise is to provide the Voodoo 5500 with best cooling possible, fixing some of its out-of-the-box design problems, so that we can push the clock to it's maximum possible frequency before hitting electrical or thermal failure. Here's the mods that I put on mine & the results that I got.


Step 1 — Remove The Stock HSF Hardware


When the Voodoo 5500 first appeared in mid-2000, it included the then-novelty of active cooling for the dual graphics processors. Unfortunately, the components and assembly technique that 3dfx used were so poor that it negated much of the benefit that could have been realized by doing it. So, this is the first thing that needs attention.


Fig 1


Fig 2

Figures 1 & 2 above show the stock V5. The 3dfx V-100 graphics processors are underneath the fans. The five large chips directly above the processors are two memory modules, the clock generator and then two more memory modules. I've already started things by pulling one of the fans off — just remove the four Philips-head screws in each corner of the fan, pull the power connector off the stakes on the board, and the fan comes right off. You can toss the fans — they'll be replaced with some much better units.

Now we come to the crowbar-and-blowtorch part of the mod. The aluminum heatsinks under the fans have to come off too, but they're glued on with what is supposedly thermal-conductive epoxy. Judging by the quality of the final assembly, though, I'm guessing that the plant manager decided to cut corners and this stuff is really Elmers Wood Glue — it's probably an excellent thermal insulator.


Fig 3

I tried to take a picture to show what this looks like, but it was hard to do with the camera that I was using. In Figure 3 you can just see a white line of glue between the heatsink and the processor — what doesn't come through clearly is the fact that the heatsink has a nice air gap between it and the processor because they laid so much glue on during assembly. The next job is to get those heatsinks off.

Here's how. Get a 2-inch wide putty knife, like you'd use to scrape paint or apply spackle to a wall. Stand the board on a hard table surface, on it's top edge (the long edge without the connector strips) so that the processors are facing away from you. Carefully insert the tip of the putty knife into the gap between the heatsink and the processor (the pathetic thing is that there's actually plenty of room to get a putty knife into this space without fighting it). Push firmly on the handle of the knife to seat it in the gap. Now take a big screwdriver or a small hammer and give the handle of the putty knife two or three sharp raps (if you're using a screwdriver, hit the knife with the side of the screwdriver handle — not the blade). The heatsink should pop right off. If it doesn't, twist the putty knife slightly to see if it'll pry loose, and if not, seat the knife again and give it another whack. Repeat for the other processor.

Here's what you should have now.


Fig 4


Fig 5

Now we're getting somewhere, but we've still got to get that crud off of the processors. It's hard as stone. Fortunately there's an easy way to get rid of it. For this next operation, you'll need a small can of acetone, a couple of clean, soft rags and an X-Acto or a razor knife.

Take a soft rag and soak part of it with acetone (Acetone is a widely used industrial solvent. It will dissolve many other substances and is highly volatile & highly flammable. These characteristics make it extremely useful and extremely dangerous to handle and use). Use it to apply the acetone to the gunk on top of one of the processors — get plenty of acetone onto the gunk, but don't get it onto anything else. You may inadvertently get a little onto the green printed circuit board around the processor which will cause the surface to dull. That's not a problem but if you get too much on this area, it can expose the bare metal of the traces causing electrical shorts or open circuits — either way, you'll destroy the card. So don't do it. Work carefully and neatly and you'll have no problems.

Like I was saying, soak the gunk on top of the processor with acetone and wait about 5-10 seconds. Then, take your X-Acto or razor knife and carefully shave off the topmost layers of gunk — you'll find that the acetone has softened the stuff 'til it's like putty. Keep scraping until you hit a point where the gunk is stone-hard again. Then repeat the acetone-wait-scrape process. Eventually you'll get down to the point where there's just a thin transparent film of gunk left on top of the processor. When you get to this point, stop scraping and just clean it off completely with your acetone rag. Repeat the whole process for the other processor too. You're all done and can break for a beer when your Voodoo board looks like Figure 6.


Fig 6


Step 2 — Applying The New Video Processor HSF Units

Now that you've got your V5 cleaned up, you're ready to start adding the new parts. The first thing to do is to attach the new coolers onto the V-100 graphics processors. Figures 7 & 8 show the units that I selected.


Fig 7


Fig 8

These coolers are made by a company called Tennmax, and you can order them directly via their Web site at www.tennmax.com. They're called the Stealth V3 Cooler and they're not cheap — $20 a unit and you obviously need two of them for a V5, so the cost is probably roughly half or better of what you paid for a used V5 in the first place. There are other coolers on the market for this kind of application — do a Google search on "video chip coolers" — and you can probably get away with something at half the price of these two

The Tennmax units worked perfectly on the V5 (no surprise there — Tennmax designed them for the Voodoo series) and even the mounting pins line up perfectly with the holes on the Voodoo 5500. They've got a nice stacked-plate design — the total cooling area is achieved by interconnected aluminum plates and the fan pulls cooling air in across those plates. They really work, too. You can feel them get warm when the board has been under heavy load. The entire Tennmax HSF assembly takes up little more room than the size of the cooling fan. On a tightly-packed motherboard, this is an important consideration. Anyway, I paid my money and I've been very pleased with the results — highly recommended.

Notice the gray area on the base of the unit shown in Figure 7. HSF manufacturers call this stuff phase change material — it's meant to eliminate the need for applying thermal compound to the chip and HSF base before mounting.

Whenever you mount an HSF assembly to a chip, you've got to use something to fill in the tiny irregularities and scratches on both components. Otherwise, the trapped air in these voids will act as a thermal insulator and the HSF will not function to it's full design efficiency. The theory behind PCM is that, when the chip heats up under load, the material undergoes a phase change from solid to almost-liquid and this fills in all of the scratches and air gaps providing a perfect medium for heat transfer between chip and HSF. Every manufacturer claims to have "a secret formula that enables near-perfect heat transfer . . . yada yada yada" — yeah, maybe. I'm an Arctic Silver guy myself, so this stuff always gets scrapped off and I never know if it works as claimed or not. For whatever it's worth, I've read other people's comments that say that the Tennmax PCM works pretty good. If you don't feel like messing around with thermal compound like I'm about to describe, go right ahead and skip forward to the part where I talk about mounting the units on the V-100 chips. Most manufacturers who OEM these kinds of parts will take advantage of the pre-applied PCM to eliminate an extra assembly step, so it has to work at least approximately as well as using thermal compound.

   Important Note: Less is more here. To create a heat exchange interface between the top surface of the chip and the base of your HSF, you either use the phase change material, or you apply thermal compound, but don't do both! Smearing thermal compound on a PCM pad will just create a big mess and will actually transfer less heat than use of either alone.

OK, I'm going to get rid of the PCM and use Arctic Silver thermal compound instead. This is the best way that I know of to get maximum heat transfer from the chip to the HSF. Arctic Silver is the industry standard for high-grade thermal compound. You can get it from almost any Web site that sells computer cooling parts. I got mine from www.2cooltek.com when I ordered several other items used for this project. BTW, Arctic Silver manufactures a few different thermal compounds. At the time I did this project I used Arctic Siver II. Shortly after it was finished, they released Arctic Silver III. I have no idea what the difference is between the two versions, but if I was doing this today I'd get a tube of Arctic Silver III. I've still got plenty of AS II left over, though, and since it's got a near-infinite shelf life I'll probably be using it for years to come.

Removing the PCM is easy — just start at one edge and carefully scrape it off with your fingernail. It's got a consistency like the gummy adhesive that's used to tack advertisements into magazines. You can see in Figure 9 that it rolls up into a neat ball. Don't use any kind of tool to do this — the base of the Tennmax Stealth V3 is just a thin aluminum plate and if you scratch it, you'll have to polish it with fine grit sandpaper (like 600 and 1200 grit) to get it smooth again.


Fig 9

When you've got the PCM pad removed, wipe the base completely clean with a little acetone on a clean rag. You're ready to apply thermal compound to the V-100 chips and the HSFs and mount them to the board.

All thermal compounds are mostly silicon grease. Silicon grease is dielectric (a nonconductor of electricity) and doesn't change with age. It never "dries", you can lift off a heatsink that you put together five years ago and the grease will have the same consistency as the day you applied it. Arctic Silver takes this one step further by adding microfine particles of pure silver to the silicon grease. Silver is one of the best elements for conducting heat and Arctic Silver is measurably superior to other thermal compounds. One other point — buy the smallest tube available (the 3 gram size from 2cooltek is enough to treat a pile of chips).

If you've never applied thermal compound before, I strongly suggest that you spend some time reading the articles on the Arctic Silver Web site at www.arcticsilver.com. It's not hard to do, but like anything else, it will be a whole lot easier if you understand what you're trying to accomplish before starting. Here are a few general guidelines — then go over to arcticsilver.com and read their stuff too.

   1. Arctic Silver thermal compound is approximately like thick silver paint. It's very easy to get it all over everything — fingers, clothes, tools, areas of the board where you don't want it — just by being careless. It can be wiped off completely with acetone, but better to work carefully to make sure it doesn't start spreading everywhere. Once you've finished applying thermal compound, be sure that you've got clean hands before you do anything else — or you may find yourself smearing silver on everything you touch.

   2. You aren't going to use very much at all. A chip with an uneven surface (like the V-100s on your Voodoo card) will require a "thick" coating of Arctic Silver. A thick coating is about as thick as a piece of common laser printing paper. On a very high quality chip package like an AMD or Intel microprocessor, you can apply a layer so thin that you can read the manufacturer's printing on the surface of the package through it. Thinner layers are better — a too-thick layer of thermal compound will actually retain heat instead of conveying it to the HSF for dispersal.

   3. Arctic Silver is not a conductor, but the combination of silver particles and silicon grease can act as a capacitor if it bridges two points with a voltage across them. This can produce strange side-effects or complete failures. Never get thermal compound onto chip leads, pc board traces or anywhere else that conducts electricity. If you get Arctic Silver on any of these areas, clean it off immediately with a Q-Tip and some acetone.

OK, assuming that you know what you're doing, here's what your V5 should look like just before you attach the Tennmax coolers.


Fig 10

My own technique for applying Arctic Silver is to put a small amount at the edge of the chip furthest away from me, then smear it over the surface of the chip by pulling it towards me with the edge of a paper business card. After the whole surface is buttered up, you can smooth it out with a few final wipes, and you're good to go.

Mounting the Tennmax coolers couldn't be easier. Start by applying a little Arctic Silver to their bases too. In this case, you should wipe almost all of it off — you want to apply just enough so that the fine scratches in the surface of the aluminum pad are filled in with thermal compound. I don't have a picture of this, but you get the idea. When you've got the units prepared like this, you're ready to mount them to the board.

The Tennmax coolers come with plastic attachment pins, one for each of the diagonal ears that are part of the base of the cooler. You're going to push these pins through the corresponding mounting holes in the Voodoo 5500 board, then just continue pushing evenly on both pins simultaneously until the Tennmax unit is firmly attached to the V-100 chip. In Figure 11, I shamelessly ripped off pictures of these mounting pins from the Tennmax Web site. You can clearly see how these things work.


Fig 11

So, here's a summary of what to do next.

   1. Position one of the Tennmax coolers directly above one of the V-100 graphic processor chips. Make sure that the mounting ears of the cooler are lined up with the corresponding holes on the Voodoo card (their are two holes per processor on diagonal corners). Start with the cooler rotated just slightly, so it's at an angle to the edge of the chip.

   2. Lower the cooler until it makes contact with the V-100 Arctic Silver layer. As it makes contact, turn it slightly so that the edge of the cooler is now square with the edge of the chip. This will eliminate any small air pockets in the thermal compound. Push the cooler firmly onto the chip while making sure that the plastic pins fall into their holes. Be careful as you do this — they don't call it thermal grease for nothing. The cooler will slide all over the top of the chip package if you push it off-center.

   3. Use your thumbs to push down on the plastic pins until they lock tightly. Push both pins evenly and be careful not to get the cooler cocked unevenly across the top of the chip package.

If you've done everything correctly, here's what it'll look like.


Fig 12


Fig 13


Fig 14

And that's it for the processors. All that's left is to heatsink the memory and clock generator chips.


Step 3 — Heatsinking The Memory & Clock Generator Chips

   Unlike the graphics processors, the memory and clock generator chips don't have conveniently-placed holes that we can use to anchor a heatsink. So, we'll need a different approach — or two. TIme to get acquainted with some other ways to stick heatsinks onto chips.

   BTW, unless I tell you otherwise, everything on this page — heatsinks, thermal epoxy, frag tape — can be had from www.2cooltek.com.

   OK, let's take this one side of the card at a time, front first. Across the top of the card are four memory chips and the card's clock generator. I decided to use two 20 X 30 mm finned aluminum sinks, each one covering two memory chips, and a single 10 X 20 mm sink for the clock generator. The only prep these heatsinks needed was lapping with several grades of fine grit wet-or-dry sandpaper, finishing with 2000 grit wet for a near-mirror finish. I lap on a plate of glass, and the finished heatsinks are flat to within several ten-thousandths of an inch.

   When I mounted the Tennmax HSFs on the graphics processors, I used Arctic Silver thermal compound — since the HSFs got pinned right through the board there was no need to permanently glue the units onto the processors. In fact, you should never use the following techniques on a processor chip since you'll always use a HSF that's removable.

   Anyway, recall that I mentioned that thermal compound never "dries", it always stays the same as the day you applied it. To mount the heatsinks on the memory and clock chips, we're going to switch to a thermal epoxy (or thermal adhesive), also made by Arctic Silver. As you can see in Figure 15, Arctic Silver Thermal Adhesive is a two-part epoxy. You mix equal quantities and you have about 10 minutes of work time before it begins to set up. I usually let it sit all night to finish curing completely. Once it's cured, you'll never get those sinks off of the chips without risking serious damage so be sure you've got everything just exactly the way you want it before it hardens. Here's a few tips on getting through this step.

       1. Make sure that you're all set to work quickly once you've mixed up a batch of epoxy. You won't have enough time to chase down a tool or think about what to do next.

       2. Check clearances inside your box before you glue anything down — remember, this is permanent. Put the vid card into its AGP slot, then gently place the heatsinks on top of the chips exactly where they'll go when they're glued down. Check to see if you're hitting anything (when I do the back of the card, you'll see where I had to correct a problem just like this).

       3. Just like thermal compound, thermal epoxy can have a capacitive effect if it bridges two points with a voltage across them. Only apply the adhesive to the mounting surfaces, never get it on a chip lead, trace, etc. Unlike thermal compound, thermal epoxy can't be wiped off after it's cured and it's even difficult to clean off completely when it's still pliable — if you get it where it doesn't belong and it's causing strange side-effects or instabilities, the card is permanently damaged and is probably worthless.

       4. Wipe both the chip surfaces and the base of the heat sinks off with acetone before gluing them together.

       5. To avoid the situation described in 3, keep two things in mind while applying it. First, you don't need a lot — a very thin layer will work best (remember, the thicker the layer, the worse the thermal conductivity). Never apply a layer thicker than a sheet of paper, just like with thermal compound. This will prevent the epoxy from oozing out between the two mating surfaces. Second, when you apply the epoxy, only apply it to the chip surface and lay on a stripe that stays about 1/16th of an inch off of the chip leads. Then you simply press the heatsink onto the chip, position it properly and hold it for a minute. That's it — you should have no epoxy spreading out from between the heatsink and the chip, and when it cures completely it'll be permanently mounted.


Fig 15

Figure 16 shows the front side of my V5 with all of the sinks mounted.


Fig 16

Once you've got the front of the board finished, the very last thing is to get the back done too. The back side of a Voodoo 5500 has four memory chips on it, just like the front. It also has another area that is often totally neglected when OCing — the board area directly beneath the graphics processors. Everybody knows you have to hang a big honkin' cooler onto the top of the processor, few people realize how much heat flows out of the bottom of the processor, heating up the epoxy printed circuit board and keeping the processor running warm. This is a very easy problem to fix on a vid card and we'll tackle it here.

You can see this area in Figure 17 (similar for both processors). The square open area ringed by discrete circuit elements is where we need to apply a heat sink. Fortunately, this area is just a fraction larger than 1/2 X 1/2 inch square — fortunate because it allows you to easily fabricate a small heat sink that fits exactly.


Fig 17

In Figure 18 you can see the two small heat sinks I made. I took an old video processor heat sink that I had laying around from another project and, as it turned out, the sink was designed so that if I quartered it I would have four pieces almost exactly 1/2 inch square with 5 X 5 pins on each quarter piece — perfect! After lappng the chip flat (easiest to do this before you cut it into small pieces), I used a Dremel tool with a thin cutting disc to do the slicing. A little clean-up with some sandpaper and the new heatsinks are good to go.

You don't want to apply thermal epoxy directly to the pc board — that matrix of holes you can see in the area where we're going to put the heatsink are vias going all the way up to the bottom of the graphics processor. Not a good idea to get epoxy pushed up inside those holes. You need to get some frag tape for this last step. I used to know who named this stuff frag tape and why, but I've forgotten (probably in the archives at www.hardocp.com). It's actually made by 3M and they call it just plain old Thermal Tape. The stuff is double-sided thermally conductive plastic tape. It doesn't conduct heat as well as thermal compound or thermal epoxy, but it works well enough for an application like this (in fact, if you're a little squeamish about using thermal epoxy on the memory chips, go right ahead and frag tape everything. It won't be quite as good thermally, but it beats bare chips by a long shot).

To apply the frag tape and mount the heat sink to the board, first cut a piece of tape from your strip that's a little bigger than the base of the heatsink. Peel off the paper backing from one side and stick it onto the base of the heatsink. Press the heatsink firmly onto a smooth surface and make sure that you don't have any air bubbles trapped beneath the tape — if you do, work them off the edge with your fingernail or peel off the tape and start over with a fresh piece. Now take some fine-grit sandpaper and sand the tape along the edges of the sink, cutting right through the paper backing and the tape. Do this all around the base of the heat sink and you'll have it covered perfectly with frag tape. Finally, peel off the remaining paper backing and press the heatsing onto the square area on the back of the card. Press hard and hold it for moment. That's it — they're mounted. One nice thing about frag tape, you can remove it later if you need to.


Fig 18

Figure 19 shows the back of the V5 with all of the heatsinks mounted. Notice the big sink that's furthest away from the camera — the fins are ground down at an angle. When I did a test fit of the board in my box, I noticed that this sink would hit one of the system's RAM modules and I'd never get the V5 to seat properly. Because I caught it before the chip got mounted to the Voodoo, the fix was easy. I just ran those fins over some 220 grit sandpaper until they were cut down far enough, then I finished up with finer grits to polish out the scratches and burrs. Since the sink is aluminum, it's soft as butter. In five minutes the chip was ready for mounting and cleared the RAM module just fine.


Fig 19

And that's it for the hardware mods to this card. Figures 20 and 21 are just a few more shots of the finished product. In Figure 20 you can see an edge-on view of the V-100 heatsinks — when I've got the card OC'd to the max, they become very warm, especially the passive unit on the back of the board

You can also see the heatshrink tubing that I used to route the fan wires. Instead of using the stock power stakes from the original fan, I just ran the new fan wires into a single female power connector. You can barely see it, but notice how this connector is plugged into the Voodoo 5500's onboard power socket. You need to supply power to the card even if you're not running fans off of the power stakes. The card won't boot it's BIOS if it's not powered. So, you run the fans into this plug and the plug itself taps off of a Y-cable that's part of a larger circuit used to run the other case fans.


Fig 20


Fig 21


Software & Performance Benchmarks

   I don't have the time to provide all of the details I had originally planned to include in this article about software, configuration settings and perfomance benchmarks. It'll just have to wait for another day (or month). But to get you started, here's some key info.

   1. Not all software will work OC'd even when the board is stable. I was supremely disappointed to discover that after all this work, Red Baron 3D won't run overclocked by even a few MHz, much less at the full potential of the board. It's probably due to timing constraints that are hardcoded into the RB3D graphics engine. But on the upside, just about everything else I've tried not only runs, it runs very well with noticable performance benefits. I've had IL-2 running at 1024 X 768, 2X FSAA, with all detail options maxed out and it's as smooth as oil. I've also run Falcon 4.0 with the last Super Patch & gotten similar results. Other people have told me that Quake III and other games built on the Quake engine run great on an OC'd Voodoo 5500, but I haven't tested them myself. In general, I don't regret the time and money spent modding the card — every so often I need the extra frames and it's good to know that it's available.

   2. I'm running the stock 3dfx drivers, version 1.04.00, the last fully-released version before 3dfx went belly-up. My OS is Win 2K and Win 98 (I've got a removable drive system) although I'm running more and more Win 2K these days since it doesn't seem to have any problem supporting anything new. When I run IL-2 I use the Wicked GL minidriver for OpenGL to correct deficiencies in the original drivers, but it seems that these have disappeared from the Web recently so I'm not sure what a new user should do for 3dfx OpenGL support (the original drivers have some severe limitations).

   3. You'll need two more add-on products to get the card working well in OC'd mode.

       — Gary Peterson's Voodoo5 Overclocker, version 1.2, dated November 12, 2000. You can get it here, from www.voodoofiles.com. This is the best utility I know of to enable and control overclocking on a Voodoo 5500. When you install it, you can go to the Display properties in the Windows Control Panel and select Advanced; you'll see a new properties tab that will allow you to set the Voodoo clock speed by moving a slider control. Don't know if it'll work with Win XP, seems to work with all earlier Win OSs.

       — Ruud Stijger's Voodoo4+ Runner, version 0.74b, dated March 25, 2001 (be sure to get this date rather than the earlier v. 0.74b, especially if you're running Win 2K). You can get it here, from www.glideunderground.com. This utility will add an icon to your systray and will allow you to custom-configure all of the major V5 settings for each game that you run. You can launch your games from the icon and V4+ Runner will switch all of the Voodoo setings to exactly where you want them before loading the game, then switch them back to your system defaults when you exit it. Not really necessary for overclocking, but extremely useful if you run any flavor of Voodoo card — eliminates all of those visits to the 3dfx Hub and having to set a bunch of parameters that you can never remember to get your game to run right.

   Hope this stuff helped. Good luck!!!
#3
about: emulating "Hardware Frabe Buffers (Effects) Emulation" in a Glide3 Wrapper.


Zeckensack is the author of a well known Glide wrapper:

http://www.zeckensack.de/glide/


this is related to Nintendo64 emulators (especially video pluging):
HWFBE allows plugin to use video memory for auxiliary frame buffers. This greatly speed ups frame buffer emulation and improves image quality. Frame buffer effects like motion blur now work as smooth as on real console and in high resolution. This feature also allowed me to emulate many things which are hardly possible to emulate with usual frame buffer emulation, e.g. dynamic shadows. HWFBE is fully supported only by Voodoo4/5 and partially by Voodoo3 and Banshee.


Zeckensack's post (asking for help):

The readme just says that HWFBE requires an extension. It doesn't say which one. Likely candidates (grabbed from a Voodoo 3's extension list) are TEXTUREBUFFER, SURFACE and TEXUMA.
Unfortunately none of these is described in the Glide3 SDK docs I have lying around. SURFACE is mentioned briefly, to the extent that I know I won't be able to implement it (it deals with DirectDraw surfaces).
I have full docs for CHROMARANGE, TEXCHROMA, TEXMIRROR, FOGCOORD and PALETTE6666. What the others are supposed to do, I can only guess.

edit:
Found it in the Glide64 sources. It's TEXTUREBUFFER. However, I still don't understand exactly what this extension is supposed to do.

Theory as follows:
grTextureBufferEXT specifies a region in texture memory (w the usual small_lod, large_lod, aspect_ratio and format arguments) as the current render target. Rendering commands now write into this space instead of the frame buffer.
To "turn it off" one calls grRenderBuffer.
Correct?



ggab says: does anybody know something about this? i'll apreciate it a lot :-)
#4
good news :) (great work LEO! ;))
Quote
* Status: No Playable (yet)[8D] [only in "wireframe mode]
* The subject does not come anything simple
* Evos 1.0x.xx WIP

~Raziel64
link:
http://www.voodoo3dfx.com/Foro/viewtopic.php?p=2505#2505
#5
nice review
http://www.hardwarezone.com/reviews/video/v3-3500TV/v3-3500tv.html

from http://www.falconfly.de/voodoo3-3500.htm

###WIN 9x###

22 Nov 2000
3dfx Voodoo3 3500 V1.06.00rc1   8907k ZIP Non-US Version, Release Candidate 1           

29 Nov 2000
3dfx Voodoo3 3500 V1.06.00rc1   8943k ZIP US Version, Release Candidate 1           

FalconFly's Tip -
When using a Voodoo3 3500 with the TV Application, dedicated Voodoo3 3500 Drivers are required...
Be sure to choose the correct regional Driver! Above Non-US Versions will NOT function with US based Systems, and can cause significant errors!

1[?]"significant errors" which ones?
2[?] what is "US based Systems" ? a 3500 with NTSC support? Non-US are Pal-N / Pal-M ?
3[?] which is the best driver (with VisualReality support) avaible for win9x ?
4[?] and AmigaMerlin 2.9 (does it has a TV tuner Application)?

5[?] is this the Lastet VisualReality for Win9x?
20 Jun 2000
VisualReality V1.05.00b4   
7499k ZIP
US Version TV Software, Beta4 Release

###Win 2k/XP###

06 Dec 2000
VisualReality V3.10.06   12611k ZIP  use for above Driver       

14 Sep 2003
AmigaSport V3.0 WinXP/Win2K -= Latest Release =- 6416k ZIP
AmigaMerlin Team's Voodoo3-3500/TV Driver
updated Glide, improved TV out
experimental NTSC to PAL-M and PAL-N Suppor

###Alternative TV Software###
30 Sep 2003
DScaler V4.18b 1957k ZIP Beta Release   
Source : http://deinterlace.sourceforge.net

FalconFly's Tip -
The 3dfx VisualReality TV Software is known to experience Problems on some Systems, mostly under Win2k/XP.
The DScaler Software is a highly recommended Alternative in this case.
Many Versions exist, but above were found recommended & tested as suitable for common use.
Visit DScaler Project page for Details and support...

6[?] "to experience Problems on some Systems"
Which ones, are there still there?
7[?] and for winXP, best config? (which TV tuner Software and How do u configure it...)

Anyway, here's the complete list of TV features :-
FM Stereo and TV Tuner Feature Sets
Complete FM stereo tuner capabilities, including Seek and Scan
CD-quality FM stereo with dbx
125 channel cable-ready TV tuning capacity, including video capture, playback and recording
MPEG 2 encoding via software codec
Full motion video support via DirectDraw/DirectSound Support for NTSC, PAL and SECAM formats, as well as S-Video and Composite In/Out
DVD playback support via Intervideo WinDVD
Zoom, skin selection, and channel surf functions
Support for closed-caption and teletext with PAL versions
Real-Time MPEG-2 and .AVI Video Capture
Advanced MPEG-2 Video Editing Software
Output high quality video to TV/VCR
Stereo Audio with dbx support
Program your computer to record your favourite programs to your hard drive
Control Brightness, Contrast, Color and Tint for the perfect picture
Supports channel-blocking for Parental Control

8[?] the last one :D
all these features' list, are working both win9x & winXP?
it would be nice, to do a mini-tutorial (updated to the lastest files' releases :) )

thanks in advance, bye [8D]
Gabriel ;)
#6
<<<<<<my tests:>>>>>>   with lastest v2.0.9.171 & a voodoo3

### mesaFX 6.1.0.8 (hardware)





Report:
{*} Compiled vertex array support.*This extension improve OpenGL performances by used video memory for vertex transformation.
{*} Multitexture support.*This features speed up rendering complex rendering like lightmaps or environment mapping.
{*} No 8bit palette texture support.*This may leads performance loss on some old applications.
{*} No Fragment program extension found. *This feature enables per pixel programming (DX8 Pixel Shader) Some for actual or future OpenGL programs may requires this extension to run.
{*} No ICD Registry Entry. *The current OpenGL driver doesn't expose the SOFTWARE/Microsoft/Windows (NT)/CurrentVersion/OpenGLDrivers registry entry. Unable to detects driver version, driver revision name and filename.
{*} No S3TC compression support.*This may leads performances loss on some applications.
{*} No Stencil support *Your current desktop configuration DOES NOT support stencil buffering. Some applications may not run.
{*} No vertex program extension found*This feature enables vertex programming (DX8 Vertex Shader) Some for actual or future OpenGL programs may requires this extension to run.
{*} Texture Edge Clamp extension found. *This feature enables an important DX6 feature for texture mapping.
{*} Secondary color support.*This feature enables specular lighting on polygons.

{*} @ both mesa´s.

{only} Few texture units found.*This would slow down some applications using Fragment Program or extensive texture mapping.
{only} Small texture size support. *Some programs may not display large textures correctly.


### mesa 5.0.1 (software)





Report:
//remember {*} @ both mesa´s, so here too.
and...
{only} No Hardware support *You current desktop configuration DOES NOT support OpenGL Hardware Accelerated.
{only} No ZBuffer support *You current desktop configuration DOES NOT support Z-buffering.
{only} Occlusion test support *This extension provides an hardware accelerated culling for objects.

this mesa 5 software compiling is available here:

http://www.iespana.es/serchmx/Mesa501.zip

----------------------------------------------------------------
2 questions=

-> with glView, in v1.2:  GL_EXT_blend_subtract
this extension is not supported by voodoo3? why?

because in a voodoo5 IT IS supported.
and the % are better: (1 extension more in each Ogl versions)
1.1 --- 100%
1.2 -- 81%
1.3 -- 55%
1.4 -- 53%
1.5 -- 16%


-> this v1.2 SGI extensions:
GL_EXT_blend_color
GL_EXT_blend_minmax

that, aren´t supported by mesaFX right now... (well, there is one more: GL_EXT_blend_subtract  but it seems that with v5 are supported....)

can be or is possible to implement them, in future versions? (i wish... :) )

because some applications/games with OGL v1.2 that i tried with the mesa 5.0.1 by software runs OK (slow... but works), so if the 2 extensions left in mesaFX can be done via hardware or even with a software implemention, it will be so cool!! and has the same v1.2 at 100% as the mesa 5.0.1 (software) one.
#8
the link:
http://www.3dfxzone.it/enboard/pop_profile.asp?mode=goEdit

loads only partial, in Avatar Selection stop!, and all the following page doesn't load anymore...
so when i need to change some settings, i must use IExplorer :-(

so, with Internet Explorer everything is OK.

But i think the forum has same bugs (only in Profile?) with Opera 7.1x and newer... (still in 7.5 beta1).
#9
the first time u run the emu (0.99) and load a game with glide64 v0.5 video plugin, the screen has some artifact bugs: glide3x from amigasport3 and others glide3x i try, all the some bug, and only in XP happens, not in 9x.
Also i tried with the compatibility mode (to choose 98/me win2000, etc) and no matter i choose, all the same. :-(
The screen has some pieces of white boxes in some little areas (about 5*10 pixels in 4 places at the hole screen), there will be there until u press ESC and return game or exit that game and open again.
i'll draw an example, X are the white boxes that are in the screen, also there are flashing at about a refresh of 1 or 1,5hz, and the " . " points are normal pixels in screen, the image are correct, only affects the little rectangules in the screen that stay there forever, until you do the 2 tips above...
No image can't post, because with Screen capture "F12 key" doesn't appear in the bmp file.

------------------------------------------
|.........................................|
|....x....................................|
|.........................................|
|................................x........|
|..............x..........................|
|.........................................|
|.........................................|
|.......................x.................|
------------------------------------------
this is a very rare bug, somebody have this bug too?
my spec:
voodoo3 + amigasport3 video driver (from www.3dfxzone.it) + xp sp1 + athon 1700+ + via kt266a (DDR)

EDIT: i have change/update some info...