Click here to print this article.
Re-Printed From SLCentral
Hercules 3D Prophet 4500 64MB
Admittedly, I've been looking forward to testing out a video card based on the Kyro II for some time. I have the sad burden of admitting my past, where I had a poor history of predicting the winner in the graphics card market. Once upon a time, I trumpeted the values of the FireGL line from Diamond. The gaming performance of those cards fell flat, as they were designed for raytracing and 3D Studio Max. Even earlier than that, I believed in the performance capability of my S3 ViRGE VX card. It ran Tomb Raider and Descent, but it's poor Direct3D support quickly left it in the dust. I even held the banner for the ill-fated PowerVR once.
Those who know their history will be quick to tell me that if I think the PowerVR chipset was a failure, then I should scoff at the notion of the Kyro II, which is actually a grandson to the original PowerVR chipset. At one time, I would have thought those people to be right. I laughed at Sega when they chose the PowerVR2 chipset to power their Dreamcast, saying that they'd never have a 3D powerhouse using a chip like that! Sega and NEC proved me wrong, of course, and showed the world that the PowerVR chipset was not a foolish venture, as it made some of the most beautiful games of the day come alive, such as Jet Grind Radio, Ecco: Defender of the Future, and Soul Calibur.
The Kyro II has a lot of competition to live up to, and big shoes to fill. NVIDIA has long kept a corner on the market with their bigger-better-faster-more video cards that always seemed to push graphics to "ludicrous speed" and offered features that left people asking, "Why bother?" When the GeForce2 came out and supported a hardware-based texturing-and-lighting (T&L) engine, people said, "Why did they bother with a T&L engine if there aren't any games that support it?" Today, there still aren't too many games that support hardware T&L other than the notable Quake III Arena, but they are coming. In fact, when I read previews describing the potential performance of the Kyro II, those very same people remarked "The performance of this card will suffer due to its lack of hardware-based T&L." Turnabout is fair play, no?
STMicro designed the Kyro II with ambitious goals - The processor was intended to be a budget solution that didn't *act* like a budget solution. For one, the Kyro II chipset acts as the antithesis of the Voodoo3. Whereas the Voodoo3 would reduce precision to 24-bits and downsample to 16-bit color on output, the Kyro II performs all internal calculations at 32-bit precision, and outputs its results appropriately. Thus, even if a game is using only 16-bit rendering, the Kyro II applies 3D lighting and shading effects using 32-bit color calculations, increasing image quality in such situations.
However, the key to the Kyro's performance is in its use of HSR, or hidden surface removal. The simple explanation is this: Most 3D cards take the geometry data and arrange the scene without evaluating it beforehand. The Kyro II instead optimizes its rendering performance by removing any polygons that aren't visible in the current frame. This saves memory bandwidth and performance cycles by making fewer polygons to be textured and lightmapped and drawn.
There are upsides and downsides to this technology. If you play games like Tribes 2 or space shooters, then HSR isn't going to do much for you because you don't have surfaces hidden very frequently. On the other hand, in games like Unreal Tournament and Quake III Arena (map-dependent), entire walls of polygons are often rendered unnecessarily. By taking these rendering shortcuts, the processor doesn't have to do as much work, and can thus render frames faster.
The original PowerVR used similar time-shaving techniques to save on rendering time, employing a technique known as tile-based rendering. While a lot of hype was built around the launch of the original product, the driver implementation was difficult to implement. Few vendors latched onto the product, and thus only a few cards were ever made with the chip, the most notable brand being VideoLogic. Tile-based rendering lived on, however, and matured in later versions of the PowerVR chipsets. In fact, it eventually led to the development of HSR. Egaer to renew their name in the industry, STMicro returned to the retail market with the Kyro II, hoping to restore faith in its tile-based rendering and introduce the world to HSR. Stepping in to defend the honor of the Kyro II is none other than Hercules, recently one of the most noteworthy manfacturers on NVIDIA-based video cards.
How does its performance fare? In the testing following I pit the Hercules 3D Prophet 4500 against an ATI All in Wonder RADEON and another Hercules card, the 3D Prophet II GTS 64MB, based on the GeForce2 chipset. I originally opted to have the Kyro II overclocked as another "card" to compare results to, but you'll find out more about that later. It should also be noted that the Prophet 4500 comes in two flavors - with and without TV-out. Hercules supplied us with the 4500 sans TV-out. The TV-output equipped model should cost you about $10 more.
The tests were conducted in both Windows 98 and Windows 2000 using DirectX 8.1 and the newest official drivers available. For the Kyro II, the drivers used were the official Hercules v7.103 drivers. The GeForce2 was run using NVIDIA WHQL-certified reference 12.41 drivers, and the ATI card was using the drivers-only installation, v5.13.01, build 3102. Tests were conducted using MadOnion's 3DMark2000 and 3DMark2001, eTestingLabs' 3DWinBench 2000, and Quake III Arena.
The testing system was an Athlon 900 on an Abit KT7A-RAID with 256 MB of RAM and an IBM Deskstar 75GXP 20GB on the ATA-100 controller. The most recent official 4-in-1 drivers from VIA were installed (4.32), along with the SBLive! 5.1drivers, and the appropriate video card drivers. Let's begin.
Here, we see that the Kyro II seems to perform about half as well as a GeForce2. Oddly enough, as I was watching the benchmarks for 3DMark2000 go by, the framerates on the Kyro II weren't all that much lower than those of the competitor cards. Next, we have the same tests in Windows 2000.
The GeForce2, using v12.41 WHQL drivers, didn't much like the idea of switching to 32-bit color mode. It would cause a memory dump every time 3DMark2000 tried to switch it over. This seems to be a bug in 3DMark2000, however, because 3DMark2001 ran in 32-bit color with this configuration.
Here, the Kyro II exhibits much better benchmarking performance, and comes up to speed with the Radeon. It still doesn't match up to the raw power in the GeForce2 however...
3DMark2001 is a much more difficult test to run than 3DMark2000. 3DMark2001 adds a lot more detail and newer technologies into the equation, resulting in much lower benchmarks for older cards. This explains the much lower scores across the board.
The results from this test almost didn't seem worth talking about. Here, the Kyro II gets blasted by the other competitors, and then it doesn't even show up on the other tests. 3DMark2001 would constantly lock up the system, forcing a reboot. I re-ran the 640x480 benchmark five times, and managed to finish it three, giving me the composite score below. I couldn't even get the benchmark to complete with any higher color depths or resolution.
I had similar crashing problems with 3DMark2001 in Windows 2000, but this time they plagued the GeForce2. The tests would consistently lock up, usually during Dragothic #2. I finally got the tests to run, oddly enough, in 1024x768, but I couldn't get any results with 640x480.
The Kyro II refuses to give up, posting low scores, but at least showing up on the board. Interesting to note is that the Kyro II's performance delta between resolutions is very minimal. The Radeon's scores decrease by as much as 25% when going from 640x480 to 1024x768, but the Kyro II gives up very little ground, losing only a few points.
Due to some Windows driver problems orienting around my monitor, Windows flat-out refused to go to 1280x1024 on my NEC MultiSync XV17+, even though it is capable of handling that resolution. Nonetheless, I did manage to get some testing scores for the Kyro II in 1280x1024. With the Kyro II installed, Windows didn't seem to have a problem with 1280x1024, oddly enough. Since none of the other cards would run in 1280x1024, I decided to keep this data separate, but include it here. I've also included the 1024x768 data from above, so that you'd have something to compare it to:
If you'll recall from before, the Kyro II's performance started out low, but it didn't give out as quickly as the competing cards did. While the Radeon lost as much as 25% of its performance as resolution went up to 1024x768, the Kyro II barely budged. When increasing the resolution further, the Kyro II still managed to hold up under pressure and return respectable scores.
If you've read this far expecting the 3DWinBench2000 scores, then give up now. The Kyro II wouldn't run 3DWinBench2000 at all. I attempted to contact eTestingLabs regarding this anomaly, but I received no reply from them at the time of the posting of this review.
But how does it Quake?
Yes, in the immortal words of...well, someone important, how does it Quake? Surprisingly, after the beating the Kyro II took in 3DMark200X, it rallied from behind to come near the top of the Q3A scores. For the following tests, I used four.dm_66 in timedemo mode using Q3A's default performance settings for Fastest, Normal, and High Quality. For Maximum, I increased the model complexity to maximum and bumped up the resolution to 1024x768. Q3A was running version 1.29f, just released as I went to perform the testing. I had preferred to fall back to 1.27, but I couldn't find it anywhere since everyone was so hopped up over 1.29f.
Here we see the Kyro II holding its own. While it wasn't able to best the Radeon, it came very close to it. When the options were maxed out in Q3A, however, the Kyro II was able to keep its cool much better than the Radeon, posting nearly 40 frames per second.
For the Windows 98 Q3A testing, the scores were neck-and-neck. To balance things out some, I decided to put the Kyro II's Full-Screen Anti-Aliasing to the mark, and I've posted some scores from that as well.
No, you're not seeing things. The Kyro II actually posted higher scores in Fastest mode with FSAA turned on. I reran the benchmark several times and the scores were consistently higher with FSAA turned on. Unfortunately, the same case cannot be made for FSAA at higher resolutions. Then again, the higher the resolutions get, the less important FSAA becomes anyway.
If you listen to the benchmarks alone, then it doesn't look too good for the Kyro II. The problem may lie in that the benchmarks are not very HSR-friendly. MadOnion's benchmarks generally carry an equal number of outdoor and indoor scenes, offering potential for the card to perform its HSR magic. Something still didn't work out right, though, and the scores forecast the Kyro II's doom. If you look at the Q3A scores, though, things don't seem to be so bad. The Kyro II performs just fine alongside its brethren, and the other cards have been on the market for over a year, giving plenty of time for driver optimizations and maturity.
Since benchmarks don't tell the full story, I elected to give the Kyro II some general playtesting. Instead of just benchmarking the Kyro II in Quake III, I started playing through a few tiers using it. Just to make things interesting, I cranked up the engine to maximum quality and ran the game at 1280x1024. The results were nothing short of surprising. The graphics were smooth and seamless, and didn't feel as if they dipped a bit below 45 FPS. This didn't seem to make any sense at all, but the more I played it, the more things seemed unbalanced in the ruling of the video card's performance. I tried out other titles, like Unreal Tournament and Trespasser. Unreal Tournament seemed to perform equally well. Trespasser, on the other hand, failed to run. It seems to have fallen victim to the Kyro II's immature drivers.
In watching the performance of the benchmarks, I frequently saw the Kyro II stutter and jitter in places, yet it turned out comparable scores to the other cards. The game performance of the 3D Prophet showed no signs of this, however, leading me to believe that the benchmarks don't accurately measure the capabilities of a card using HSR algorithms. When employed in actual practice, the games tended to respond very well to the card, running just as fast as they did on the other cards.
Full-Screen Anti-Aliasing is also available on the Kyro II, as you've noticed with the benchmarks I gave in Q3A a page before. The Kyro II's implementation is questionable, however. 3DMark2001, which contains the option to run tests in FSAA, would not allow the option to be set when using the Kyro II. Even if I turned on the "Force FSAA" options buried within the control panel, the card still would not run with FSAA turned on. Furthermore, when running Q3A and Alice with FSAA turned on, the games ran okay, but the keyboard and mouse response seemed to lag by almost a half-second. Even casual gamers would scoff at such an occurrence, let alone twitch gamers who live and die by their mouse's refresh rate.
Above and below, you'll see some images below which depict the image quality of the Kyro II's laggy FSAA. Unfortunately, the JPEG compression somewhat destroyed the image quality, but you can still see areas where FSAA is taking effect. In particular, notice the surface edges on the rocket launcher, and also the stone molding in the lower-left corner at the bottom of the archway. The FSAA quality of the card seems to do really well. If only the performance were so pleasing…
The Kyro II chipset runs normally at 175MHz, and the 3D Prophet attaches 64 MB of 5 nanosecond RAM to the card, leading one to believe that the card's design leaves a decent amount of room for overclocking. Testing this out proved otherwise, as I was only able to take the card up to 181MHz while stable, using PowerStrip 3. I tried it out at 185MHz first, but began to see some shearing and artifacting, so I slowly stepped the card down.
The memory should theoretically have been able to handle 200MHz while remaining within tolerance specifications, so the core is most likely at fault. This is understandable, given the method of the core's calculations and operations. The card does not rely as much on memory bandwidth as it does on algorithm strength, so overclocking the card to 185 should not have generated any satisfying results. As it was, 181MHz only gave about 1.5 FPS more than usual in all settings, so I elected not to represent this data in the comparisons. I have recently been reading reports on the Internet of people receiving newer cards that would clock as high as 195MHz, so your mileage may vary.
For those who are into extreme cooling and case modding, the Kyro II emits very little heat. Original reference designs called for only a minimal heatsink on the processor, but Hercules elected to outfit all their 3D Prophet 4500 cards with a heatsink-fan combo in the fashion of Thermaltake's Blue Orb. While likely to be unnecessary, it does show an effort on the part of Hercules to ensure the life of the chip should other cooling systems prove inadequate. It should also be noted that some reviews of video cards I've seen have faulted the manufacturer for failing to provide RAM sinks on the memory, claiming that they were out to save a buck. While that may be true, in this case the RAM sinks are nowhere near necessary, and I believe that Hercules decided against using them in order to pass the savings on to the customer.
And this brings us to the central idea behind the Kyro II. The chipset is essentially designed to be a wolf in sheep's clothing, giving intense performance at a budget price. Other chipsets have made this dubious claim before (Rendition's Verite, ATI Rage IIc, anything with the word "Savage" in the title), but the Kyro II really manages to live up to this claim. It performs well in today's games, and leaves me reason to believe that it will do so in the future. It may not support the newest of features, but, it does support quite a few, and those it does support, save FSAA, it supports very well. The pundits that wish to point out the Kyro II's lack of hardware T&L need only wait, however, as STMicro has already stated that the forthcoming Kyro III will feature a hardware T&L engine.
NVIDIA's drivers have the edge on the Kyro II by several years as well. NVIDIA, using a similar core design in all its processors since the TNT, has had the advantage of refining a single driver core set for years, while the Kyro II has only had drivers for a few months now. The Kyro II drivers need some definite improvement. When trying to benchmark the cards, they frequently reported errors while trying to change to a given video mode, when the system was already running in that video mode. The FSAA lag also exhibits a problem that should probably be corrected within a driver revision or two. Given a little training, this upstart could really set out and accomplish things.
Hercules retails the 3D Prophet 4500 on their website at $149.99 and the 3D Prophet 4500XT at $159.99 (The XT model features TV-out), but searching some of the vendors at iBuyer.net gives prices on this card as low as $130, making it an reasonably affordable alternative to the NVIDIA cards. Unfortunately, the card had a lot more value a year ago when it was in development. The presence of the GeForce3 has pushed down the prices on the GeForce2, making the difference between this card and a GeForce2 much less, if anything at all.
Pros & Cons
In the end, if you're into bleeding-edge triple-digit framerates, go somewhere else to blow your cash. If you're debating between the purchase of a GeForce2 MX and this card, though, the immediate performance of this card can certainly win you over. The GeForce2 MX can push more frames than this one, and thus has more brute-force power, but this card can do pretty good at any resolution and color-depth, whereas the GeForce2 MX will wimp out as the going gets tough. The GeForce2 MX is also well streamlined, which means that any weak element in your system will likely hinder its performance. With the recent pricedrops in the GeForce2 GTS, however, the decision of buying this card becomes a difficult one. If you've read about the recent flap concerning NVIDIA's marketing tactics against the Kyro II, however, you'll start to wonder if there's a reason NVIDIA's afraid of what this chip can do.
In that light, this chip seems very well designed, but its alternative rendering techniques make it look bad in benchmarking, and the driver quality is questionable. I honestly feel that in actual practice and usage, the chipset does exactly what it was intended to do, and does it very well.
The 3D Prophet 4500 may not get the performance marks, but it works very well nonetheless. If you're in need of a good budget upgrade that won't get held back by your processor, then look no further. You might want to keep your eye on it for a little while, though, to make sure things start working properly on the driver side.
Rating: 7/10 SystemLogistics
Re-Printed From SLCentral