Hercules’ 3D Prophet 4500 graphics card

Can I get back to you later about that poly?
Our man Dissonance wrote up a nice explanation of the Kyro’s “what you see is what you draw” approach to 3D rendering. This approach is known as deferred rendering, and it’s quite different from the more common method, immediate mode rendering. I suggest you see Dissonance’s write-up for a visual representation of the differences between the two approaches, but I will make a clumsy attempt to sum them up.

Immediate mode
Immediate mode renderers, including nearly every other 3D graphics card you can buy, draw three-dimensional scenes like the Pentagon processes its budget. They chew through oodles of bandwidth and processing power to get where they’re going, then throw out the extra stuff at the end. Immediate mode chips process all the polygons in a scene, apply shading and textures, then send the pixel hurtling down the pipeline with a Z value, or depth information, attached to it. This depth information can use nearly as much bandwidth as the color information for the pixel. Generally these days, a pixel will have 24 bits for Z data and 32 bits for color and transparency info. At the end of this process, the graphics chip starts drawing the scene. Only then does it use the Z information, stored in a Z-buffer, to determine whether one pixel overlaps with another.

Take, for instance, the fundamental object that defines all 3D game scenes: crates. (This I learned from Old Man Murray.) Odds are, any 3D scene will have a number of crates, and probably some barrels, too. Some of those crates and barrels will be behind others, so that crate A obstructs our view of barrel B. An immediate mode renderer processes things in whatever order they come down the pipe. It may draw barrel B in its entirety, then draw crate A in front of it. Only when it comes time to draw the actual pixels for crate A will the chip determine that barrel B shouldn’t be visible in the scene—and by then, it’s already done all the work to draw barrel B. Only the completed scene is sent to the display, so all you’ll ever see is crate A.

In other words, sending a polygon to an immediate mode renderer is like handing Rosie O’Donnell a gift certificate to Luby’s.

This process of drawing pixels that will be obscured by others is known as overdraw. Overdraw is the scourge of efficiency in real-time 3D graphics, and it feeds the number-one performance problem for graphics cards these days: memory bandwidth bottlenecks. Memory bandwidth limitations are the reason many newer 3D graphics cards aren’t much faster than their predecessors. An old GeForce DDR, for instance, will outrun a brand-new GeForce2 MX 400, because the older card’s 128-bit DDR memory interface gives it the edge.

Deferred rendering
To avoid overdraw and better use resources, the Kyro II chip takes things in a different order than immediate mode renderers. Heck, it takes a very different approach altogether. The Kyro uses tile-based rendering, segmenting the display into small sections and processing each portion in turn. Busting the screen into tiles gives the Kyro several advantages. Among them:

• Hidden surface removal — The Kyro chip maintains a list of polygons for each tile, and it determines early in the rendering process which polygons in the tile will be visible and which will be occluded. Making this determination for an entire scene at once would be difficult, but breaking the scene down into chunks makes it feasible.
• Deferred rendering — Only after the removal of hidden surfaces are the remaining pixels shaded and textured. In this way, overdraw is dramatically reduced. That means there’s less shading work (this form of shading is also known as lighting, or the L in T&L) and much less bandwidth used pulling texture data from the graphics card’s memory.
• On-chip operations — The Kyro has an on-chip “tile buffer” in which it stores the entire contents of the tile. The chip can then perform a number of operations on those pixels, like pixel blending and layering on additional textures, without accessing the card’s frame buffer (or Z-buffer) memory. As if the memory bandwidth savings alone weren’t enough, the Kyro’s makers claim this arrangement allows them to use lots of internal precision in such operations, improving image quality.

Taken together, these benefits are formidable. It’s hard not to think the deferred rendering approach will have a prominent place in the future of real-time 3D graphics.

Image quality
The Kyro II’s implementation of deferred rendering produces unimpeachable results. Using the card, you’d never know that what’s going on inside the chip is so radically different from everything else. That’s no mean feat, considering the Kyro II must process API calls intended for conventional graphics chips and handle them seamlessly.

Standing on deck, you’d never know it was a three-legged dwarf rowing the boat.

The image quality is on a par with any of its more conventional rivals, and in many cases, the Kyro II’s output looks cleaner and sharper to my eye. Imagination Tech’s claims about higher precision for pixel blending rings true. I swear I can see the difference at times; transparencies just look cleaner.

Serious Sam looks amazing on the 3D Prophet 4500

The PowerVR hype machine spends a lot of time talking about “internal true color” with reference to 16-bit color rendering. The basic gist of it is this: the chip renders everything internally at 32 bits, even when it’s using a 16-bit color video mode. But why would anyone ever use 16-bit color mode on this thing? The Kyro sports loads of pixel-pushing power, so why bother?

Anyhow, the Kyro’s output is quite nice. Compliments to the dwarf.

The usual suspects, minus a few
Beyond the radical new approach to 3D rendering, the Kyro II endows the 3D Prophet 4500 with most of the features you might expect to see in a new graphics card. Texture compression, dot-product and environmental bump mapping, and anisotropic filtering are all there. Conspicuous by its absence: a hardware transform and lighting (T&L) engine. The Kyro II chip has to rely on the host CPU and good drivers to handle T&L chores. We’ve noted above that deferred rendering saves on some lighting work, though, so that may not matter so much. (Is T&L necessary? Wumpus asked the same question a while back.)

The chip also lacks even a primitive form of the pixel and vertex shaders found in next-gen graphics chips like the GeForce3 and the upcoming ATI R200. Like all previous 3D chips, the Kyro II will be left high and dry once DirectX 8 applications (and their OpenGL counterparts) hit the market in force. That’s a ways off, however, and presently, a GeForce3 card costs three to four times what the 3D Prophet 4500 does. Save your money now, and you can upgrade to something better when the time comes—maybe even a Kyro 3.

The flat stuff
The 3D Prophet 4500 works just about like any other AGP card when it comes to basic text and windowed displays. I have noticed a couple of things that stand out, though—one good and one bad.

First, the bad news: the 3D Prophet 4500 has one of the dimmest signal outputs of any graphics card I have (and I have a lot of ’em). Plug it into a monitor adjusted for another card, and white background looks gray, the colors washed out. It’s not the end of the world; crank up the brightness and contrast on a good monitor, and the problem disappears. In the grand scheme, it’s a subtle difference. But not everyone—and especially not everyone buying a budget video card—has a really good monitor.

I mentioned this concern to Hercules, and they swapped out my test card for a new one. The second card’s output was just like the first’s. This is probably an issue with the Kyro II chip’s integrated RAMDAC and with the initial brightness and contrast settings in the Kyro drivers. I suggested to Hercules that brightness and contrast sliders in their drivers might help mitigate the problem, and right afterwards, I installed the latest Kyro drivers. Lo and behold, they’d just incorporated a gamma slider. The gamma adjustments help, but contrast controls would still be a welcome addition.

In all, the 4500 is a decent enough 2D display card, but those of you who will accept only the highest quality signal output (especially those with gigantic monitors) would do well to consider a Radeon instead.

Second, I gave the 4500’s DVD playback a workout, to see how it stacked up. I was surprised both by the quality of the playback and by its extra-low CPU utilization. On a 1.33GHz Athlon, the system reported between three and fifteen percent CPU use with CyberLink’s PowerDVD software. I’m not clear on the finer points on the Kyro II’s DVD decode assist capabilities, but it’s as good as anything else I’ve seen—especially in terms of CPU use.

In all, the Kyro II chip has the basic, non-3D graphics functions covered quite well. No “gotchas” reared their ugly heads during my testing.

The chip specs
The Kyro II chip’s specifications are an inversely gaudy display of hardware efficiency, an orgy of austerity—from the low transistor counts to the cheapo memory to, well, everything else. It starts with the chip itself, which amounts to just a small section of a GeForce3 chip, transistor-wise.

At 15 million transistors, the Kyro II is barely even there. It’s half the size of a Radeon, and under a third the transistor count of a GeForce3 chip. The pipeline specs are similarly puny…

In terms of nominal fill rate, the Kyro 2 is seriously wimpy. These numbers are almost entirely theoretical, but the Kyro’s peak texel fill rate is half that of its nearest competitor, and a quarter that of a GeForce2.

However, the Kyro II has the ability to deliver eight-layer multitexturing in a single rendering pass thanks to its tile buffer. (The GeForce3 uses a similar trick to deliver four layers per pass.) Adding extra texture layers in this way takes clock cycles, but avoids the big performance penalty that would come with additional rendering passes. By contrast, a GeForce2 would have to render a scene four times in order to lay down eight textures per pixel. Most current games don’t apply nearly that many textures, but even at four textures, the Kyro II is—wait for it—much more efficient.

As for memory bandwidth…

The Kyro II matches the GeForce2 MX (and the MX 400, which isn’t on the chart) in memory bandwidth, but trails well behind the more common DDR solutions. The chip’s low-rent spec means the Kyro II will have to be madly efficient just to keep pace with the competition.

Our testing methods
As ever, we did our best to deliver clean benchmark numbers. All tests were run at least twice, and the results were averaged.

The test system was built using:

Processor: AMD Athlon processor – 1.33GHz on a 266MHz (DDR) bus

Motherboard: Gigabyte GA7-DX motherboard – AMD 761 North Bridge, Via VT82C686B South Bridge

Memory: 256MB PC2100 DDR SDRAM in two 128MB DIMMs

Audio: Creative SoundBlaster Live!

Storage: IBM 75GXP 30.5GB 7200RPM ATA/100 hard drive

The system was equipped with Windows 2000 SP1 with DirectX 8.0a. No, we didn’t use Win9x/ME for this test, and no, we don’t regret it. For Microsoft operating systems, Win2K is the present and the future. If a product can’t perform well in Win2K, it deserves to be counted as a poor performer.

For comparative purposes, we used the following video cards and drivers:

• Hercules 3D Prophet 4500 64MB with 7.114 drivers
• 3dfx Voodoo 5 5500 64MB with 1.04.00 drivers
• ATI Radeon 64MB DDR with 5.13.1.3132 drivers
• NVIDIA GeForce2 GTS 64MB (Asus AGP-V7700) with 11.01 NVIDIA reference drivers
• NVIDIA GeForce2 Ultra 64MB (NVIDIA reference card) with 11.01 NVIDIA reference drivers
• NVIDIA GeForce3 64MB (NVIDIA reference card) with 11.01 NVIDIA reference drivers

We used the following versions of our test applications:

• 3DMark 2001 Build 200
• Quake III Arena 1.17
• Serious Sam v1.00
• SPECviewperf 6.1.2
• VillageMark 1.17
• Vulpine GLMark 1.1

The test systems’ Windows desktop was set at 1024×768 in 32-bit color at a 75Hz screen refresh rate. Vertical refresh sync (vsync) was disabled for all tests. Most of the 3D gaming tests used the default or “normal” image quality settings, with the exception that the resolution was set to 640×480 in 32-bit color.

All the tests and methods we employed are publicly available and reproducible. If you have questions about our methods, hit our forums to talk with us about them.

VillageMark
VillageMark was developed by the makers of the Kyro architecture to show of their product’s advantages. To do so, the 3D scene in VillageMark include lots and lots of overdraw. The test also lays down three textures per pixel—just enough to force a GeForce2 into multi-pass rendering.

Nefarious.

As you might expect, the Kyro II mops the floor with everything else out there in this test. And not just a little bit.

This test’s results do show us something more than a really, really slanted result in a really, really slanted test (although that’s mainly what they show). They demonstrate the advantage of the Kyro II’s deferred rendering architecture, but they also show how the competition is modifying the conventional rendering approach to gain back some ground. Both the Radeon and GeForce3 chips can lay down three textures without additional rendering passes, and both chips employ a series of tricks—most famously dubbed HyperZ in the case of the Radeon—to reduce overdraw and use memory bandwidth more efficiently. (For more on these techniques see here for the Radeon’s HyperZ and here for the GeForce3.) The GeForce2 Ultra has a 1.5GB/second memory bandwidth advantage over the Radeon, for instance, but the Radeon edges it out.

The point is this: what the Kyro’s doing is, to one degree or another, all the rage in 3D these days. Everybody wants to reduce overdraw and lay down more textures per rendering pass. The Kyro II does it best, but the conventional competition is moving in the same direction.

Serious Sam
Now that we’ve got all of that theoretical mumbo-jumbo out of the way, let’s look at the first real-world test of the 3D Prophet 4500 in action. Note that Serious Sam is very much an application benchmark, not a proper synthetic or theoretical benchmark. The game makes use of the capabilities of each video chip to the best of its abilities, adjusting to each chip’s features as it can. That said, Serious Sam offers a useful real-world comparative test.

It also lets us plot performance over time, so we’re not just graphing averages. In my GeForce3 review, the Serious Sam results plotted frame rate averages in one-second increments. Here, we’ve used five-second increments, because the game’s default test increments changed in never versions. I’ll have to dig into it and figure out how to use one-second increments, which are more interesting, next time around. However, these will do for now.

At 640×480, it’s all about good drivers and polygon throughput. The 4500’s performance tracks close to the Radeon’s, raising the question of whether there’s any real advantage to the Radeon’s T&L engine. The NVIDIA cards run in a pack above the rest, showing off their polished drivers and, perhaps, their T&L engines.

At an intermediate resolution, real playability is the goal. Here, the 4500 distances itself from the Radeon and tops the GeForce2 GTS, as well. Only the high-priced NVIDIA cards are faster.

Super-high resolutions are all about fill rate, and the Kyro II’s deferred rendering scheme gives the 4500 some advantage here. With their fast memory and quad pixel pipelines, the GeForce2 Ultra and GeForce3 cards still have the lead.

These results show one thing clearly: the Kyro II chip is for real. With only 175MHz SDRAM and a couple of pixel pipelines, this little chip can run with the big dawgs.

Quake III Arena
Let’s see if the 4500’s performance carries over to Quake III.

At low res, things stack up pretty much like they did with Serious Sam.

At 1024×768, with all of Q3A’s image quality options maxed out, the 3D Prophet 4500 runs at over 85 frames per second. Not a bad showing at all.

Once more, when we’re dealing with an extremely high resolution, the Kyro II’s peak fill rate maxes out right where the GeForce2’s does. The Kyro II’s relative performance will depend on the amount of overdraw in a given scene, but generally, it’s on par with the GeForce2 GTS.

3DMark 2001
The 2001 rendition of 3DMark measures performance with an eye toward future games, not present ones. The test uses DirectX 8, pushes gobs of polygons, and checks for advanced features like dot-product bump mapping and pixel shaders. This test ought to push the Kyro II a bit, even though the chip handles current games well.

It would seem the cards without on-chip T&L hang out in the basement in 3DMark 2K1. A closer look at the results will help clarify why that is.

Fill rate
I’m afraid 3DMark’s synthetic fill rate tests don’t adequately account for deferred rendering architectures, but the numbers are interesting, so we’ll have a look.

In the single-texture test, the 3D Prophet 4500 delivers almost exactly on its 350 Mpixel/second pixel fill rate. None of the other cards even come close to theirs.

The Kyro II doesn’t deliver any more performance with multitexturing than with a single texture. By contrast, all of the other cards post significant gains. Again, though, it’s hard to see how this test reflects the 3D Prophet 4500’s real-world performance—or even its theoretical performance, taking overdraw into account.

Poly throughput
Here’s where the T&L-less chips seem bound to stumble. The polygon throughput tests in 3DMark 2001 are brutal. Can our test system’s 1.3GHz Athlon help the Kyro II keep up with it’s T&L-equipped competition?

Nope. Well, it kept up with the Radeon just fine, but the NVIDIA cards are far out ahead.

Add more lights, and things even up a bit. This time around, the Kyro II even beats out the Radeon—a testament to the quality of the Kyro’s drivers.

Finally, we have the vertex shader test, which doesn’t really test full, DirectX 8-style vertex shaders, or only the GeForce3 card would complete the test.

The Radeon and its T&L unit finally manage to pull away from the software-only crowd. In these synthetic tests, at least, the 3D Prophet 4500’s lack of on-chip transform and lighting capabilities puts it at a minor disadvantage.

Bump mapping
The Kyro II chip supports both dot-product and environment-mapped bump mapping techniques. Only the GeForce3 and Radeon chips share this distinction. Here’s how its performance stacks up using both bump-mapping methods.

Not too bad, all things considered. However, I honestly have to wonder whether these specific, hard-wired bump mapping effects won’t give way to more advanced pixel-shader effects in future games. Current games, on the other hand, only rarely employ bump mapping.

Game tests
Now for the game tests. 3DMark 2K1’s game tests are very high-poly affairs. The Kyro camp has made the argument that newer, more complex games increase overdraw, heightening the need for deferred rendering solutions. These game benchmarks should test that argument.

The 3D Prophet 4500 holds up well enough in the low-detail version of the test, but it absolutely chokes on the high-detail version.

Dragothic looks like it ought to have loads and loads of overdraw, but here again, the Kyro II chip can’t quite keep up.

The Lobby test ought to be a decent indicator of how these cards will run Max Payne, the game on which this test is based. The Hercules card keeps up with the pack in low-detail mode, but again stumbles badly when the detail level rises.

Before we move on, I should mention that some folks have called something to my attention about DirectX 8, which 3DMark 2001 uses, and the Kyro II. At present, it seems the Kyro II’s drivers don’t support DirectX 8 fully. As a result, some of the card’s 3D acceleration features, present in the hardware, aren’t used by 3DMark 2001. Should Hercules offer an updated driver for the 3D Prophet 4500, the card might perform better in 3DMark 2001. An update to DirectX with specific provisions for the Kyro II and its current drivers might also help performance.

It’s the eternal struggle. Right now, the Kyro drivers are behind the curve.

Vulpine GLMark
Vulpine GLMark uses OpenGL to test many of the same features 3DMark tests in Direct3D. It has snazzy water effects on the GeForce3, uses lots of polygons, and has a creepy looking chick in spandex, just like 3DMark 2000 did. Oddly enough, the 3D Prophet 4500 would not run the benchmark with “Advanced features” enabled. The option was greyed out by the program. In testing for the scores below, the other cards all used the “Advanced features” option, so they were probably doing more work.

Well, that wasn’t pretty. Let’s just move along. Nothing to see here. Move along.

SPECviewperf
Lets pull the fish out of water for a sec and watch it flop around. SPECviewperf measures performance in high-end OpenGL applications, the kind one would run on a workstation. The 3D Prophet 4500 is by no means aimed at such markets, so we’re not expecting miracles here. Then again, none of the other cards we’re testing are specifically aimed at the workstation market, either. Still, folks sometimes want to dabble in 3D modeling without ponying up for an expensive high-end card, so this will be interesting.

Did I mention the 3D Prophet 4500 isn’t aimed at the workstation market? Without T&L and without better optimized OpenGL drivers, the Kyro II chip withers under the force of a zillion polys.

Antialiasing
Like the GeForce2 and Radeon, the Kyro II offers full-scene antialiasing using an ordered grid supersampling method. This method isn’t as fancy as the rotated grid techniques found on the Voodoo 5 and GeForce3, but it’s not too bad, either. I won’t delve into antialiasing theory here, since I’ve written about it fairly extensively right here, and the Kyro doesn’t add any new wrinkles to the game.

In theory, the Kyro II’s tile-based rendering scheme ought to offer some opportunities for super-efficient antialiasing schemes. However, I’m not aware of any especially neat tricks the chip uses for AA at present. Since antialiasing performance is all about pixel-pushing power, and since deferred rendering is all about high fill rate, one would expect the 3D Prophet 4500 perform well with antialiasing turned on.

The 4500 does offer one unique AA option. In 2X AA mode, the chip will let the user pick whether the samples should be grabbed from a double-height (vertical) or double-width (horizontal) source image. Other cards don’t allow this option, far as I know. I’ve included screenshot samples of both techniques below.

Non-AA reference images
The next few pages will show you exactly how the antialiasing output of the cards we’ve tested compares. I think you’ll see the Kyro II’s output is very similar to the Radeon and GeForce2.

Voodoo 5, no AA

Radeon, no AA

GeForce2, no AA

GeForce3, no AA

3D Prophet 4500, no AA

2X antialiasing

Voodoo 5, 2X AA

Radeon, 2X AA

GeForce2, 2X AA

GeForce3, 2X AA

3D Prophet 4500, 2X AA (Horizontal)

3D Prophet 4500, 2X AA (Vertical)

4X antialiasing

Voodoo 5, 4X AA

Radeon, 4X AA

GeForce2, 4X AA

GeForce3, 4X AA

3D Prophet 4500, 4X AA

Antialiasing performance
Now that you’ve spent the requisite time leaning into the monitor to detect subtle differences in antialiased images, let’s look at how the cards perform.

In 2X AA mode, the 4500 beats all but the high-end GeForce cards. The GeForce3, with a more economical AA implementation called multisampling, pretty much romps.

4X AA mode doesn’t bring many surprises. The Kyro II is again very competitive, but only the GeForce3 offers playable performance at 1024×768. The Radeon’s AA implementation is especially weak; it won’t even run the test right at 1024×768.

Texture filtering action
The Kyro II chip is supposedly capable of more advanced forms of texture filtering, including trilinear and anisotropic filtering. One of the things we do to test a card’s capabilities is run it through our “acid trip Quake” test, and see how it fares. If the card’s doing trilinear filtering properly, the colored bands in the pictures below should fade into one another gradually, like so:

The GeForce2’s trilinear filtering blends smoothly between mip maps

However, with trilinear turned on, the Kyro II chip’s output looks like so:

It looks like the 3D Prophet 4500 isn’t doing trilinear here

That screenshot was taken with texture compression enabled, and Kyro doesn’t appear to be doing trilinear filtering. However, turn off texture compression in Quake III, and trilinear filtering magically starts working:

Disabling texture compression magically enables trilinear

So what’s the story? Kristof Beets with PowerVR Technologies offers this explanation:

The KYRO chip supports 2 trilinear modes, slow and fast. The slow mode is used when S3TC is disabled and shows the nice color bands that you were expecting. The fast mode works differently and clashes with the system used by Quake to generate the color bands.

The fast system works as follows :

Rather than reading from 2 pre-generated MipLevels in memory the lower less detailed MipLevel is generated on-the-fly by the chip from the higher more detailed level. So rather than accessing 4 texels in 2 maps one accesses 16 texels in the upper map which allows the generation of the 4 high level and 4 low level values. Now because Quake colors the different MipLevels differently the color band effect fails since KYRO only access 1 miplevel to create the fast trilin effect.

So what you see is trilinear but a fast implementation which accesses only one texture (the high detailed miplevel). Since the lower level is auto generated by assuming that the lower level is generated from the higher level using a 2 by 2 filter. Now in the case of Quake Color Bands the lower levels are different from the higher levels, they do not have the 2 by 2 filter relation and thus the Fast Implementation seems to fail. Real games with real texture contents do use this 2 by 2 pixel filter relation between the MipLevels and thus the fast implementation is technically correct.

I hope this helps you to understand the images you obtained from Quake.

So the Kyro II chip is doing trilinear filtering, but it doesn’t show up in our “acid trip Quake” test. No harm, no foul. Thanks to Kristof for clarifying this issue for us.

According to the specs, the Kyro II chip is also capable of anisotropic filtering with 16 texel samples, which is stronger than the degree of anisotropic filtering on the GeForce2. To my eye, it wasn’t working in Quake III, but Kristof assures me this problem will be rectified in the next driver release.

Sizing it up
Our test suite was not kind to the 3D Prophet 4500. In Quake III and Serious Sam, the 4500 held up quite well. The Kyro II chip was huffin’ and puffin’ in most of the rest of the tests. That’s not an unexpected result, because our test suite is heavy on polygons and next-gen 3D features. It’s intended to push graphics cards to their limits, expose their soft underbellies, and give ’em a tweak. At 175MHz, with a three-legged dwarf below deck rowing for all he’s worth, we knew the 3D Prophet 4500 had a few seams in its armor.

But let’s not overlook its strengths. In the tests that matter most to today’s gamers, at intermediate resolutions and in antialiasing modes, the 3D Prophet 4500 managed to match the GeForce2 with regularity and beat out the Radeon more often than not. Subjectively, it was faster than my 32MB GeForce2 GTS in all of the games I played. That’s a heck of an accomplishment for a card with such a miserly hardware spec.

More than that, this card inspires confidence. For the most part, it just works right, and its drivers never once rendered my PC unbootable (which is more than I can say for the Radeon) or caused a system crash. With its smaller transistor count, I also found the 3D Prophet 4500 caused less trouble for me in my PC, whose fast Athlon and full complement of PCI cards has left it a bit jumpy about power. Even with a nice 400W power supply, a GeForce3 or GeForce2 Ultra can make it fussy about booting up. With the 3D Prophet card, things were a bit less hairy. Not everyone will be affected by such problems with other graphics cards, but the Kyro’s teensy size and power requirements are a plus in certain situations.

Still, the 3D Prophet 4500 is most impressive because of the technology behind it. As a showcase for Imagination Tech’s PowerVR technology, it’s downright exciting. The fundamental ideas behind the Kyro’s tile-based rendering method are very solid, and the implementation ain’t bad, either.

As a video card, the 3D Prophet 4500 isn’t quite up to par with the mid-range cards from NVIDIA and ATI. That’s a shame, because if the Kyro II could be mated with the DDR memory and high clock speeds of its primary competitors, it’d blow them away—at least in terms of fill rate. As it stands, it’s a TNT2-spec card that sometimes shows its weaknesses. High-polygon scenes and programs that use the latest and greatest 3D features tend to strain the card. Yes, it’s much more efficient, but in worst-case scenarios, it just doesn’t have the oomph to overcome what finesse can’t.

This card is an excellent budget video card, however, and a great baseline for graphics accelerators. It’d make a GeForce2 MX 400 break out into cold sweats. So long as Hercules keeps the cutting the price with regularity—and with this hardware spec, that ought to be easy—the 3D Prophet 4500 will go on my recommended list for all of my cheapskate friends. I’d like nothing better than to see the same thing happen in the OEM market, to save me from that creepy feeling I get walking by the shelves full of 1.3GHz Pentium 4 systems with TNT2 M64 cards at Best Buy. With cards like this one around, there’s just no excuse for committing the TNT2 M64 sin, even for the most frugal PC manufacturers.