Introduction
Our bombshell report exploded onto the screens of our millions of faithful readers earlier this week as we informed everyone of the industries latest 3D powerhouse, ATi’s Radeon. The massive report required hours upon hours of last minute testing and writing to bring you the full scoop. However, because we know you want the latest information at earliest possible time, a few stones were left unturned. Well, we’re back after investigating a few more matters as well as elaborating on some of our earlier points. We’ll go over the addition of 64MB GeForce2 results, added resolution to our test suite, and detail the results of our overclocking the Radeon.
Details About Shipping Radeon Cards
Before we get into the benchmarking details, I’d like to share a bit of information with you that only reached us after the previous Radeon article. After I heard rumors saying that ATi’s yields of Radeon chips running at 183 MHz are not quite ideal, I asked ATi and found out that the Radeon cards with 32 MB memory have the core clocked at 166 MHz and the memory at 166 MHz only, giving ATi the chance to use the more available and less expensive 6 ns DDR SDRAM for those cards. You can imagine that this will make the 32 MB cards significantly slower than the versions with 64 MB.
I also have to correct myself for saying that all current Radeon cards lack the video in/out, because this is not true. The 64 MB cards are coming with all of ATi’s video in/out goodies, while the 32 MB versions don’t.
Please keep those two things in mind when shopping around for a Radeon card. The ones with only 32 MB are definitely a stripped down version of the 64 MB cards, with less performance and less features.
Radeon Card | Memory | Core Clock | Video Features | Suggested Retail Price |
64 MB DDR | 5.5 ns DDR SDRAM | 183 MHz | Video In/Out | $399 US |
32 MB DDR | 6 ns DDR SDRAM | 166 MHz | N/A | $279 US |
32 MB SDR (Shipping in September) |
6 ns SDR SDRAM | 166 MHz | N/A | $199 US |
64MB Monstrosities
Initially we intended on bringing our heavyweight challengers to the table when reviewing the Radeon but due to some unfortunate last minute issues, the 64MB GeForce2 didn’t make it. This problem was quickly fixed shortly after the release of the first article later that afternoon thanks to ELSA who was kind enough to quickly ship me a board (special thanks to Kim and Jennifer). Tom did find time in the first review to spot-check several important benchmarks with his GeForce2 64MB board to verify that our findings were consistent even when we compared the Radeon to the NVIDIA’s best offering but we obviously wanted to have an update that included the 64MB GeForce2 results for you to see yourselves. This is no longer an issue as not only are we providing standard 64MB GeForce2 results but also added resolutions and overclocked GeForce2 64MB scores. The addition of these tests should further support Tom’s initial findings. Let’s take a look at the data.
Test Setup
Graphics Cards and Drivers | |
Radeon DDR 64MB | 4.12.3044 |
GeForce2 GTS 32/64MB GeForce DDR 32MB |
4.12.01.0532 |
Voodoo5 5500 | 4.12.01.0543 |
Platform Information | |
CPU | PIII 1GHz |
Motherboard | Asus CUSL2 (bios 1000 BETA 013) |
Memory | Crucial PC133 CAS2 |
Network | Netgear FA310TX |
Environment Settings | |
OS Version | Windows 98 SE 4.10.2222 A |
DirectX Version | 7.0 |
Quake 3 Arena | Retail version command line = +set cd_nocd 1 +set s_initsound 0 OpenGL FSAA set to 4X Super Sampling |
Expendable | Downloadable Demo Version command line = -timedemo |
Evolva | Rolling Demo v1.2 Build 944 Standard command line = -benchmark Bump Mapped command line = -benchmark -dotbump |
MDK2 Demo | Downloadable Demo Version T&L = On Texture Quality = Max setting MipMap = On Trilinear Filtering |
Benchmark Results – Quake 3 Arena
In our first 16-bit test you’ll notice that the NVIDIA based products clearly dominate this test. The Radeon might not come in first place for frame rate but it does well until 1280×1024 where it begins to fall. Keep in mind that most folks who use this resolution would use 32-bit color mode.
We’ll see the transition of first place occur at 1024×768 where the Radeon takes a small lead and widens its margin slightly as resolution rises. This is where the benefit of having Hyper Z comes in.
Benchmark Results – Expendable
16-bit color is a mode that wastes much of the Radeons potential as you can see. The card does decently in 16-bit mode but isn’t nearly enough to take the top performance title.
The transition period for the Radeon begins at 1152 this time around. Even though it starts to take the lead very late, the margin in the end is respectable at the highest setting possible in Expendable.
Benchmark Results – MDK2 Demo
In 16-bit color this benchmark is owned by the GeForce based chips. I’m sure we’ll see more interesting results when the color depths rise a bit more. Until then, the Radeon has to hide its head.
ATi likes using MDK2 as a benchmark and if you look at the results you can see why. Starting at 1024x768x32 Radeon is able to take the lead, to finally have a 10% margin over GeForce2 GTS 64 MB at 1600x1200x32.
Benchmark Results – Evolva
Although the chart might make things look worse than they are for Radeon, the card performs very well. GeForce2 boards still dominate this benchmark at this color mode. Even the VD5 5500 works some magic and offers very respectable results.
Leaping over into 32-bit, I was surprised to see that the Radeon hasn’t crept into the lead sooner than it did. Regardless the GeForce2’s and the Radeon stick side by side for most of the tests.
Benchmark Results – Evolva, Continued
To make things really fun for our test subjects we’ve toggled Bump Mapping on to see how it shakes things up. As I expected in 16-bit mode, the Radeon shys away from the pack as it lingers behind. The one thing that isn’t so hot is the fact that the GeForce2 cards are able to hit a high resolution at a reasonable frame rate while the Radeon runs out of steam at about 1024 resolution.
Things get better for the Radeon but the race remains a very tight one.
Performance Summary
There isn’t much to add onto our previous articles’ analysis aside from three things I noted. The first was that there are many playable 32-bit resolutions at 1024+ that users can take advantage of when running a Radeon. Our initial testing had only shown that the Radeon dominated 32-bit high resolution testing but at that high of a setting, the frame rate was still rather low. With our added results you can see that the resolutions 1152x864x32 and 1280x1024x32 are more playable.
The Radeon doesn’t have the greatest performance in 16-bit color but by no means is it a slow performer. People have been hammering at this and I find it ridiculous that a 90 FPS result is considered “poor” and not labeled as “not as good as” when compared to the GeForce2 results. I agree that the GeForce2 dominated the 16-bit area but that advantage only makes for good benchmark victories and not better game play. Besides, most people who want to play at higher resolutions are also playing in 32-bit color, following the simple principle of ‘either all or nothing’.
ATi’s Radeon is without a doubt a good solution and is probably the card of choice to purchase for a new high-end gaming card, however, not enough to justify replacing a GeForce2 solution unless the owner is really into using FSAA or needs some of the fancier video options the Radeon can provide. The performance gains aren’t drastic enough for me to advise someone to swap out with their GeForce2 board. Caution should be taken here as money could be wasted making a hasty decision.
Overclocking
After hours experimenting with a beta version of Powerstrip and our sample Radeon board, we had come to the conclusion that the core and memory speeds were locked in some strange way. Although we could alter the settings in our utility, Powerstrip, the setting would either return to the core speed to the memory setting or not change but perform as if it had. This was discovered after finding some very disturbing results where the overclocked hardware performance showed no gains. After speaking further with ATi, it indeed turns out that this is true and that the core and memory clock are currently synchronized. Upcoming drivers may change this but for the time being, we are limited to this.
The most stable setting found for my early Radeon board was 200MHz graphics core and 200MHz memory speed. Anything above this caused severe visual issues. Extra steps were taken to cool the graphics chip down but it was maintaining a steady 45 C temperature during overclocked testing. 45 C was the observed temperature of the stock clocked part under heavy loads. Without being able to have separate core and memory speeds, I was unable to determine for sure which of the two was holding us back from our 200 MHz barrier.
Since we had added the nearly 10% overclocked Radeon, I figured we’d add our results from a slightly overclocked GeForce2 64MB board. This should give the Radeon a bit more of a challenge. Let’s have a look at our results.
Overclocking Results – Quake 3 Arena
Here we have a comparison of the fastest settings possible for each of our test boards. 16-bit color in Quake3 shows us no surprises, as the overclocked GeForce2 is a monster.
With High Quality settings turned on the Radeon improves its standing but the GeForce2, with slightly increased memory settings, flexes some major muscle. The stepped up GeForce2 64MB card proves that once again that NVIDIA needs to get with it and put faster memory on their hot rod graphics chip. Radeon’s HyperZ doesn’t seem to buy much in Q3. A normally clocked Radeon 64 MB scores pretty much the same results as a GeForce2 GTs 64 MB card with 366 MHz memory clock.
Overclocking Results – Evolva
Hammering the card for fill-rate and T&L performance at the same time apparently is just too much for the Radeon as the GeForce2 card pounds it. Will things change in 32-bit color mode?
Things get better for the Radeon once 32-bit color comes into play again but Evolva still seems to have shown us that the Radeon has a tough time dealing with the raw power of the GeForce2’s massive fill-rate. The 10% overclocked memory on the GeForce2 helps tremendously and almost keeps the Radeon from taking top honors at the top of our resolution scale.
If you stop to consider how the Radeon does before and after you see a 64MB GeForce2 board running overclocked memory, you start to really value Hyper Z and wonder just how long until NVIDIA comes up with some type of solution for this memory problem. We’re talking a real solution. Once that happens, things can get really scary for ATi.
Wrap Up
Without a doubt ATi has a reason to still be standing tall after further proof that it’s now shipping a leading edge product capable of matching and at times surpassing the industries best. The things to keep in mind here is that the Radeon may be our overall choice for best consumer graphics solution but by no means is it able to run over the competition from NVIDIA. The GeForce2 series of cards provide formidable performance and when the 64MB GeForce2 board is brought into the picture, things are extremely close. Even at the higher color and resolutions, the Radeon barely squeaks by at times with narrow victories. I don’t think either solution is a bad choice for a serious graphics upgrade but I still cannot see myself advising anyone to drop his or her GeForce2 for this product unless they greatly value the other Radeon specific benefits. Hopefully this addition will add to what you already know from the initial write-up and assist you in future decisions if you happen to end up in the market for this caliber of a graphics solution. Thanks to ATi and NVIDIA, you’ll probably end up having to make a tough choice between the two competing chips but without fierce competition like this, our choices wouldn’t be nearly as interesting.