Original Link: https://www.anandtech.com/show/836



It's been two months since we first previewed the Radeon 8500 and in that short period of time a lot has happened.

Since then, ATI has officially launched their 7x00 series of cards - including the 7500 we had a look at in August. ATI has also been hard at work sorting out the issues with their flagship 8500 that we ran into two months ago. The Radeon 8500 and 7500 have also received increases in clock speed from the original specs they announced. This could indicate either higher yields on the cores or an even stronger desire to compete with NVIDIA after the release of their Titanium line of cards.

But by far the most surprising thing throughout this whole situation has been NVIDIA's reaction to the 8500. While we concluded our preview by saying that, assuming history did not repeat itself, the Radeon 8500 had a lot of potential, NVIDIA obviously saw a dangerous amount of potential in that card. We have never seen NVIDIA so concerned with a competitor's product launch before. Their paranoia was definitely well founded as the Radeon 8500 seemed impressive when we first looked at it; now, with even more memory bandwidth and a higher fill rate, is the Radeon 8500 NVIDIA's worst nightmare?

ATI Draws the Line

A confusing product line is the easiest way to lose sales regardless of how great your products are. ATI's introduction of the four-digit Radeon nomenclature is intended to give everyone, at a single glance, an idea of how powerful cards are within a particular family. Unfortunately, with differences in memory types and the presence (or lack thereof) of a full fledged hardware T&L engine, ATI's new naming system can be a bit frustrating to more educated buyers and even more so to developers. Let's take a look at the past and present Radeon cards and see how they compare to one another:

 
ATI's Radeon Line: Past & Present
 
8500
7500
7200
7000
DDR
LE
SDR
VE
Core (Number of Transistors)
R200 (60M)
RV200 (30M)
Rage6C (30M)
RV100
Rage6C (30M)
Rage6C (30M)
Rage6C (30M)
RV100
Manufacturing Process (circuit width in microns)
0.15
0.15
0.18
0.18
0.18
0.18
0.18
0.18
Rendering Pipelines
4
2
2
1
2
2
2
1
Texture Units per Pipeline
2
3
3
3
3
3
3
3
Core Clock Speed (MHz)
275
290
166
183
166/183 (oem/retail)
148
166
183
Memory Clock Speed
275
230
166
183
166/183 (oem/retail)
148
166
183
Memory Bus
128-bit DDR
128-bit DDR
128-bit SDR
64-bit DDR
128-bit DDR
128-bit DDR
128-bit SDR
64-bit DDR
Memory Bandwidth (GB/s)
8.8
7.4
2.7
2.9
5.3/5.9
4.7
2.7
2.9
Hardware T&L Supported
Y
Y
Y
N
Y
Y
Y
N
Programmable Pixel & Vertex Shaders
Y
N
N
N
N
N
N
N
HyperZ
Y
Y
Y
Y
Y
N*
Y
Y
HyperZ II
Y
N
N
N
N
N
N
N
HydraVision
Y
N
N
N
N
N
N
Y
AnandTech Review

As you can see, the Radeon 7000 is essentially the Radeon VE without its dual display capabilities. This is actually the most irritating to developers since the chip does not have hardware T&L support, making it impossible to recommend that gamers purchase any Radeon or GeForce card to run their games.

The Radeon 7200 is a 64MB version of the Radeon SDR, and the rest of the lineup should be familiar to you already.



NVIDIA's Fears

While it was their Riva 128 that put NVIDIA on the map, it was their GeForce series of cards that made them a leader in the industry. The GeForce line, from the original 256 up to the current GeForce3 Ti 500, has been the most successful series of graphics cards we have ever seen. It is very clear that the vast majority of gamers use NVIDIA cards because of their performance and very reliable drivers. Looking at the AnandTech Community alone, out of 4911 members that have filled out their System Rigs page, over 3000 of them use NVIDIA cards -- that's over 60%. With all that NVIDIA has going for themselves, why on earth would they worried by a meager Radeon 8500?

 
ATI
NVIDIA
STMicro
 
Radeon
Radeon 7500
Radeon 8500
GeForce2 Pro
GeForce2 Ti 200
GeForce2 Ultra
GeForce3
GeForce3 Ti 200
GeForce3 Ti 500
Kyro II
Number of Transistors
30M
30M
60M
25M
25M
25M
57M
57M
57M
15M
Manufacturing Process (circuit width in microns)
0.18
0.15
0.15
0.18
??
0.18
0.15
0.15
0.15
0.18
Rendering Pipelines
2
2
4
4
4
4
4
4
4
2
Texture Units per Pipeline
3
3
2
2
2
2
2
2
2
1
Core Clock Speed (MHz)
183
290
275
200
250
250
200
175
240
175
Memory Clock Speed
183
230
275
200
200
230
230
200
250
175
Memory Bus
128-bit DDR
128-bit DDR
128-bit DDR
128-bit DDR
2 x 64-bit DDR
2 x 64-bit DDR
2 x 64-bit DDR
128-bit SDR
Memory Bandwidth (GB/s)
5.8
7.4
8.8
6.4
6.4
7.4
7.4
6.4
8.0
2.8
Special Features
HyperZ
HyperZ
HyperZ II
SmoothVision
   
Lightspeed Memory Architecture
HRAA
Deferred Renderer

It turns out that the Radeon 8500 isn't all that meager. In fact, the part has a greater fill rate and more memory bandwidth than NVIDIA's GeForce3 Ti 500. ATI's SMOOTHVISION AA is also supposedly better when it comes to visual quality and theoretically should incur a lesser or equal performance penalty to that of NVIDIA's Quincunx AA. And although support for Pixel Shader 1.4 isn't really a strong point of the Radeon 8500, John Carmack has already commented on the chip's future performance in Doom 3 saying that it will be able to render the game's many textures in as few as a single pass compared to the multiple passes that the GeForce3 would have to take.

There are some very valid causes for NVIDIA to be worried about the Radeon 8500, but none of their worries has any merit if ATI cannot get over their biggest problem: drivers.



Today's Radeon 8500

ATI was very rushed in the release of the Radeon 8500 for the main reason that they had a very small window of opportunity to succeed with this part. Immediately following NVIDIA's Titanium launch, ATI could assume that they wouldn't face any other competition until next year. If the Radeon 8500 could outperform the GeForce3 Ti 500 then ATI could have a performance leading part throughout the end of this year, hopefully gaining some additional clout among gamers; a segment in which they have trailed behind NVIDIA.


Click to Enlarge

For the past few weeks ATI has been shipping cards to retailers all over the globe. There have been a few delays that kept some cards from shipping out until recently, and ATI has held back review samples until this week after numerous delays which is why you see a delayed review here on AnandTech and on other websites. The delay doesn't bother us though since we're pleased just to be given the opportunity to test it, but we felt you should have some sort of an explanation as to why this article took so long to appear. ATI has actually taken the review rollout of this card very seriously and has distributed close to 360 Radeon 8500 boards worldwide to the press and analysts. For comparison purposes, the only launch recently that had such great exposure was AMD's launch of the Athlon back in 1999 with close to 400 review systems.

The card we were provided with was a sealed retail card clocked at the 275/275MHz (core/mem) frequency that all retail 8500 cards will be shipping. The board ships with three Valve titles; Team Fortress, Counter-Strike, and Half-Life: Uplink. Also included in the box is a DVI to 15-pin analog VGA connector to make use of the second video output which enables ATI's HydraVision multi monitor support.

The performance you will see in this review today is the performance you will get out of the card when you purchase it which was not the case when we first looked at the 8500.

The specifications and a discussion of the features of the chip can be found in our original Radeon 8500 article here.

Charisma Engine II

Pixel Tapesty II

HyperZ II

Video Immersion II

HydraVision

TRUFORM

SMOOTHVISION



ATI can play the driver game too

NVIDIA clearly spoiled the preliminary introduction of the Radeon 8500 back in August with the release of their Detonator 4/XP drivers. We were told by NVIDIA that these new drivers would not only improve performance, but that they would be made publicly available the very same week we tested with them. Obviously, that didn't happen, and it ended up taking another month before the drivers were released. The performance gains were tangible, but the drivers weren't fit for release when NVIDIA provided them to the press and honestly shouldn't have been used. Hindsight being 20/20, we made a promise to ourselves that we would not allow any further performance enhancing drivers to be used in our video card reviews unless we could make the drivers publicly available to our readers immediately.

When ATI approached us regarding the final review of the Radeon 8500 they mentioned that they would be providing us with two sets of drivers; the first set would be the WHQL retail drivers (v 7.60) that would be bundled with every card and the second set would be a special "performance" driver that they wanted us to test with. Obviously we had issues with this so we tested under both and agreed to publish benchmarks from the "performance" driver only under the stipulation that ATI would release this performance driver to the public by the end of day today (10/17/2001). ATI was actually planning on releasing this driver in a couple of weeks but they graciously agreed to move that release date up in order to comply with our requests.

This performance driver is not WHQL certified (v 7.60.04) but it actually didn't change performance in any of the games we tested in. The only benchmarks this special driver had any effect on were 3DMark 2001 and Villagemark; both Direct3D benchmarks and both of which are more theoretical tests than real world tests.

Despite this we definitely applaud ATI for their compliance with our request as they truly didn't have to. Also according to ATI they will consistently offer driver improvements for the Radeon 8500 much more so than they have for any other product.

However, ATI will definitely have to work hard if they plan on coming close to NVIDIA's ability to develop, execute and release drivers.



The Test

Windows 98/2000/XP Test System

Hardware

CPU(s)

Intel Pentium 4 2.0GHz
Motherboard(s) ABIT TH7-II RAID
Memory

256MB PC800 Mushkin RDRAM

Hard Drive

IBM Deskstar 30GB 75GXP 7200 RPM Ultra ATA/100

CDROM

Phillips 48X

Video Card(s)

ATI Radeon 7500 64MB DDR
ATI Radeon 64MB DDR

NVIDIA GeForce3 Ti 500 64MB DDR
NVIDIA GeForce3 64MB DDR
NVIDIA GeForce3 Ti 200 64MB DDR
NVIDIA GeForce2 Ti 200 64MB DDR
NVIDIA GeForce2 Pro 64MB DDR

Hercules 3D Prophet 4500 64MB SDR

Ethernet

Linksys LNE100TX 100Mbit PCI Ethernet Adapter

Software

Operating System

Windows 2000 Professional Service Pack 2

Video Drivers

ATI Radeon 8500 Drivers v7.60.04
ATI Radeon 7500 Drivers v5.13.01.3276
NVIDIA Detonator 4 v21.85
Hercules 3D Prophet 4500 8.162a Drivers



Theoretical Performance

In all of our CPU and chipset reviews we start out with a few theoretical benchmarks; more specifically, how does the product we're looking at perform when the pesky real world variables are silenced. Things such as software incompatibilities, drivers, etc... take a back seat to the raw performance of the product; in this case, the Radeon 8500.

For the theoretical performance of the Radeon 8500 we chose two benchmarks: 3DMark 2001 and VillageMark. We'll start with 3DMark 2001.

 
Multitextured Fill rate (MTexels/s)
Triangle Rate (MTriangles/s)
EMBM (fps)
DOT3 (fps)
Vertex Shader (fps)
Pixel Shader (fps)
Point Sprites (MSprites/s)
ATI Radeon 8500 (7.60.04)
1806.8
9.7
98.8
91.8
85.4
99
23.1
ATI Radeon 8500 (7.60 WHQL)
1806.7
10
94.5
79.8
81.4
77.8
23.1
ATI Radeon 7500
1137.5
2.8
98.7
54.6
45.4
NS
0.2
ATI Radeon 64DDR
736.2
1.7
71.4
42.5
35.8
NS
0.2
NVIDIA GeForce3 Ti 500
1579
3.7
116.2
124.2
54.3
90.2
17.7
NVIDIA GeForce3 Ti 200
1165.6
2.7
95.4
93.5
43.1
67.4
13
NVIDIA GeForce3
1331.5
3.1
104.3
107.2
47.6
76.8
14.8
NVIDIA GeForce2 Ti 200
626.3
3.7
NS
46.5
38.8
NS
8.5
NVIDIA GeForce2 Pro
620.9
2.9
NS
45.6
38.5
NS
8.4
STMicro Kyro II
343.2
1.2
78.8
37.4
15.6
NS
0.6

The first column of numbers is the multitextured fill rate of the cards compared here and in this case, the Radeon 8500 does incredibly well. With a 35% fill rate advantage over the GeForce3 and a 14% advantage over the GeForce3 Ti 500 the potential for the Radeon 8500 to succeed is definitely there. You should also note that the fill rate of the Radeon 7500 is virtually identical to that of the GeForce3 Ti 200, but we already know from previous investigations that the 7500 is no match for the GeForce3 Ti 200.

The Radeon 8500 again flexes its muscle in the high polygon count triangle rate test, offering close to 3x the triangle throughput of the GeForce3 Ti 500.

The highest performance with Environment Mapped Bump Mapping enabled lies with the GeForce3 Ti 500 and interestingly enough, the Radeon 8500 is no faster than the Radeon 7500 in this test. This could be the first sign of the flagship's current limitations. It is also worth noting that performance with DOT3 bump mapping enabled is improved tremendously with the 7.60.04 drivers.

If we are to believe that 3DMark 2001 is the first and only true DX8 test available today, then we can conclude that the Radeon 8500's Vertex Shader is considerably stronger than the GeForce3 Ti 500's unit. It seems like it would take a NV core with dual vertex shaders to offer the same performance as the Radeon 8500; is it a coincidence that the NV2A core used in the Xbox (and supposedly NV25 derived) has dual vertex shaders? The same huge advantage isn't extended to pixel shader performance but the Radeon 8500 does hang on to the lead.

While we have not seen extensive use of Point Sprites in current games, the potential is there. ATI's waterfall demo takes advantage of the Radeon 8500's powerful point sprite capabilities.



Toying with Theory (continued)

Next we have VillageMark, a benchmark originally developed by the folks at PowerVR to show off the benefits of a deferred based rendering system. In order to do this there is an extraordinary amount of overdraw (rendering of pixels that aren't displayed to the screen) which doesn't affect cards like the Kyro II since hidden pixels are not rendered at all.

VillageMark has the potential to be a great measure of the efficiency of things such as ATI's HyperZ II and NVIDIA's Visibility Subsystem but in order to make sure that it's a good benchmark of occlusion culling subsystems (discarding unused pixels) we tried disabling HyperZ II on the Radeon 8500 to see what sort of a performance hit the card took under VillageMark:

Radeon 8500 - Z Occlusion Culling Performance
VillageMark - 1024x768x32
HyperZ II Enabled

HyperZ II Disabled

113

59

|
0
|
23
|
45
|
68
|
90
|
113
|
136

It's clear that VillageMark would serve just fine as a HyperZ-Mark considering the 91% increase in performance seen when enabling HyperZ II is noticeably greater than the real world performance we saw resulting from enabling HyperZ on the old Radeon from way back when. Although the "performance drivers" had little effect on the real world tests we should mention that these scores (taken with the "performance drivers") were improved 35% by the new driver. Normally we would conclude that these new drivers work better with the Radeon 8500's HyperZ II but then we would have seen a similar boost in all games not just these Direct3D theoretical tests.

Now that we know how useful VillageMark is as a theoretical benchmark let's take a look at how well the GeForce3 fares without its precious Visibility Subsystem:

GeForce3 - Z Occlusion Culling Performance
VillageMark - 1024x768x32
Visibility Subsystem Enabled

Visibility Subsystem Disabled

63

38

|
0
|
13
|
25
|
38
|
50
|
63
|
76

For the GeForce3 you've got to remember that NVIDIA's Visibility Subsystem is much more than just Z-Buffer compression and Z-Occlusion Culling, it includes NVIDIA's Crossbar memory architecture. The way we "disabled" the Visibility Subsystem was by running a GeForce2 at 200/460 (core/mem) which gave us the same theoretical fill rates as the GeForce3 without the enhancements of the Visibility Subsystem. Because there are other memory bandwidth enhancements at work outside of those that ATI implements we decided to see how the performance changed when switching to 16-bit color where memory bandwidth isn't as important. It is worth it to note that there is no performance difference between the 8500's 16-bit and 32-bit scores.

GeForce3 - Z Occlusion Culling Performance
VillageMark - 1024x768x16
Visibility Subsystem Enabled

Visibility Subsystem Disabled

52

42

|
0
|
10
|
21
|
31
|
42
|
52
|
62

Again we see a healthy boost from the Visibility Subsystem but you'll notice that the boost isn't nearly as great as it was with 32-bit color enabled indicating that a significant part of the performance improvement is due to NVIDIA's Crossbar Memory controller and not the other features of the Visibility Subsystem. In both of these cases the GeForce3 is not able to realize the same performance gains that the Radeon 8500 does when HyperZ II is enabled.

There was little doubt that the Radeon could hold its own against the GeForce3 and even the new Titanium line on paper, but where it really matters is in the real world. It's time to step out of Pleasantville...



Black & White

A very different type of game than what you're used to seeing in these reviews, Black & White places you in control of a civilization as their god.  This Direct3D game relies on a high enough frame rate that navigating around your worshippers' villages isn't choppy.  The user has a third person perspective of the world.

The game was set to 1024 x 768 x 32/1600 x 1200 x 32 with the preset maximum detail configuration used.  We used fraps to take the average frame rate during the intro scene with the family and their child journeying towards the water (the movie was skipped).  Note that the latest version of fraps does take an actual average frame rate so this is not an estimate.

Black & White will not start with fraps running in the background so you must first start B&W, switch back to the desktop to run fraps, and then switch back to B&W.

Black & White
Maximum Detail - 1024x768x32
NVIDIA GeForce3 Ti 500

NVIDIA GeForce3

NVIDIA GeForce3 Ti 200

ATI Radeon 8500

NVIDIA GeForce2 Ti 200

NVIDIA GeForce2 Pro

ATI Radeon 7500

STMicro Kyro II

ATI Radeon 64DDR

64

58

53

51

44

43

40

33

29

|
0
|
13
|
26
|
38
|
51
|
64
|
77

When we first looked at the Radeon 8500 it was being outperformed by the GeForce2 Pro; with final silicon and drivers it can now claim victory over the GeForce2 Pro but it unfortunately cannot declare the same over any card with GeForce3 in its name.

ATI honestly wasn't aiming at the low end of the GeForce3 line with the Radeon 8500 but that's where the card falls in line today.

Black & White
Maximum Detail - 1600x1200x32
NVIDIA GeForce3 Ti 500

NVIDIA GeForce3

NVIDIA GeForce3 Ti 200

ATI Radeon 7500

NVIDIA GeForce2 Ti 200

STMicro Kyro II

NVIDIA GeForce2 Pro

ATI Radeon 64DDR

32

28

25

18

17

17

16

12

|
0
|
6
|
13
|
19
|
26
|
32
|
38

Unfortunately the Radeon 8500 would cause the system to reboot at 1600 x 1200 x 32 and thus is absent from these benchmarks. We would have expected the Radeon to be even more competitive with the GeForce3 Ti 200 and possibly even the GeForce3 here because of its memory bandwidth advantages.



Serious Sam

As Croteam is hard at work on the followup to Serious Sam we continue to use it as a benchmark since the game makes use of many newer graphics features. For our tests we set all of the graphics options to their maximum values with the exception of ansiotropic filtering which was left disabled.

Serious Sam
Maximum Settings - 1024x768x32
NVIDIA GeForce3 Ti 500

NVIDIA GeForce3

ATI Radeon 8500

NVIDIA GeForce3 Ti 200

ATI Radeon 7500

NVIDIA GeForce2 Ti 200

NVIDIA GeForce2 Pro

ATI Radeon 64DDR

STMicro Kyro II

91.2

86.9

74.7

73.8

61.4

57.1

56.4

47.8

41.2

|
0
|
18
|
36
|
55
|
73
|
91
|
109

Rewinding back to August the GeForce3 held a 25% advantage over the Radeon 8500, today that advantage is significantly reduced but not eliminated. Again we see that the Radeon 8500 is very competitive with the GeForce3 Ti 200 but not much beyond that.

Serious Sam
Maximum Settings - 1600x1200x32
NVIDIA GeForce3 Ti 500

NVIDIA GeForce3

ATI Radeon 8500

NVIDIA GeForce3 Ti 200

ATI Radeon 7500

NVIDIA GeForce2 Ti 200

NVIDIA GeForce2 Pro

ATI Radeon 64DDR

STMicro Kyro II

55.8

48.7

47.9

42.8

37.1

27.7

27

24.8

17.3

|
0
|
11
|
22
|
33
|
45
|
56
|
67

At 1600 x 1200 memory bandwidth is king giving the Radeon 8500 more in common with the older GeForce3 than the Ti 200. The performance of the card still isn't up to par with what we'd expect unfortunately.

Serious Sam
Multitextured Fill Rate
NVIDIA GeForce3 Ti 500

NVIDIA GeForce3

ATI Radeon 8500

NVIDIA GeForce3 Ti 200

ATI Radeon 7500

NVIDIA GeForce2 Ti 200

NVIDIA GeForce2 Pro

ATI Radeon 64DDR

STMicro Kyro II

609.08

541.95

498.55

455.02

381.02

291.18

284.81

256.35

176.28

|
0
|
122
|
244
|
365
|
487
|
609
|
73

Another beauty of the Serious Sam engine is the ability to measure fill rate from within the game. Compare these figures to the 3DMark 2001 figures and you'll see why we placed those in the "theoretical" benchmarks section. The Radeon 8500 falls in between the GeForce3 and the GeForce3 Ti 200 both of which have a lesser theoretical fill rate. These figures are resolution independent meaning that they could very well be driver limited; ATI has expressed their displeasure with the current status of OpenGL drivers of the Radeon 8500 which could explain some of the poor performance here.



Quake III Arena

One of the most reliable gaming benchmarks (its results are always repeatable and rarely vary) is Quake III Arena and it's based on an engine that is still in the news. We used the 1.29g patch and the video options set to the default High Quality settings. The resolution was adjusted and demo 'four' was run in timedemo 1 mode.

Quake III Arena
High Quality - 1024x768x32
NVIDIA GeForce3 Ti 500

ATI Radeon 8500

NVIDIA GeForce3

NVIDIA GeForce3 Ti 200

NVIDIA GeForce2 Ti 200

NVIDIA GeForce2 Pro

ATI Radeon 7500

ATI Radeon 64DDR

STMicro Kyro II

223.7

207.5

205.2

187.3

130.6

127.3

123.1

87.0

71.6

|
0
|
45
|
89
|
134
|
179
|
224
|
268

In our original preview, Quake III Arena was one of the only benchmarks where the Radeon 8500 came close to the GeForce3 in performance. In spite of this there was still a 15% performance delta but now the Radeon 8500 comes out ahead. The GeForce3 Ti 500 is in its sights but just too far for now. The Radeon 8500 is finally performing closer to where it should but still under what we'd expect.

Quake III Arena
High Quality - 1600x1200x32
NVIDIA GeForce3 Ti 500

NVIDIA GeForce3

ATI Radeon 8500

NVIDIA GeForce3 Ti 200

ATI Radeon 7500

NVIDIA GeForce2 Ti 200

NVIDIA GeForce2 Pro

ATI Radeon 64DDR

STMicro Kyro II

123.5

106.5

105.7

92.7

60.3

58.8

56.7

40.3

30.8

|
0
|
25
|
49
|
74
|
99
|
124
|
148

Showing the somewhat dated nature of Quake III as a benchmark, at 1600 x 1200 x 32 the memory bandwidth dependencies are not great enough to push the Radeon 8500 any further up the charts. In fact it falls slightly behind the GeForce3 and still out of reach of the Ti 500.



Wolfenstein MP Test

In the tradition of id releases, the upcoming Return to Castle Wolfenstein game was introduced in a publicly available Multiplayer Online Test. If you haven't tried the game yet we suggest downloading a copy and enjoying it yourself. The game is based on a modified version of the Quake III engine with an updated look to the graphics. We recorded two quick demos for this test by hopping on a random online server and going at it for a few; one outside on the beach with many players visible and some very big explosions and the other inside the compound with very few players. You can download the two demos here.

We tested with the built in "High Quality" settings enabled and texture filtering changed from Bilinear to Trilinear. To run the benchmark do the following:

1) Extract the demo files to c:\PathToWolfMPTest\demomain\demos\
2) Bring down the console by hitting the '~' key
3) Type 'timedemo 1' and hit enter
4) Type 'demo atdemo6' and hit enter to run the first demo, or 'demo atdemo8' to run the second demo

We chose atdemo8 for this comparison since it is more video card bound while atdemo6 is more of a CPU test.

Wolfenstein MP Test
atdemo8 - 1024x768x32
NVIDIA GeForce3 Ti 500

NVIDIA GeForce3

NVIDIA GeForce3 Ti 200

ATI Radeon 8500

ATI Radeon 7500

NVIDIA GeForce2 Ti 200

NVIDIA GeForce2 Pro

ATI Radeon 64DDR

STMicro Kyro II

142

139.4

135.8

129.2

115.7

114.1

112.3

89.4

84.5

|
0
|
28
|
57
|
85
|
114
|
142
|
170

Although it's based on the same engine as Quake III Arena, the standings do change a bit with the Radeon 8500 falling about 5% below the GeForce3 Ti 200. The only card that is able to hold a respectable lead over the Radeon 8500 is the GeForce3 Ti 500 at 10%.

Wolfenstein MP Test
atdemo8 - 1600x1200x32
NVIDIA GeForce3 Ti 500

NVIDIA GeForce3

ATI Radeon 8500

NVIDIA GeForce3 Ti 200

ATI Radeon 7500

NVIDIA GeForce2 Ti 200

NVIDIA GeForce2 Pro

ATI Radeon 64DDR

STMicro Kyro II

103.4

90.4

86.4

79

61.2

52.6

51.2

42.8

33

|
0
|
21
|
41
|
62
|
83
|
103
|
124

Remember how we said that Quake III Arena wasn't memory bandwidth intensive enough even at 1600 x 1200? The picture changes with Wolfenstein as the Radeon 8500's memory bandwidth advantage gets it above the GeForce3 Ti 200. Unfortunately for ATI, raw power isn't everything and they fall short of the gold yet again. At this point you should realize that there is a lot of untapped power in the Radeon 8500 but the only question is when will it be tapped, if ever.



UnrealTournament

Although at this point it's a better CPU benchmark than a video card benchmark, UnrealTournament as usual is a part of our standard test suite. We tested using the publicly available "Thunder" demo by Reverend.

We mentioned in our last review that ATI's drivers had a problem with UnrealTournament. The problem has not been solved in the latest driver build and basically results in vsync limited scores in spite of the fact that vsync is disabled in the drivers. This explains the poor performance of both the Radeon 8500 and 7500. ATI provided us with a fix for the solution by rebooting after disabling vsync but it did not work under Windows 2000.

Unreal Tournament
1024x768x32
NVIDIA GeForce3

NVIDIA GeForce3 Ti 200

NVIDIA GeForce3 Ti 500

NVIDIA GeForce2 Ti 200

NVIDIA GeForce2 Pro

STMicro Kyro II

ATI Radeon 8500

ATI Radeon 7500

ATI Radeon 64DDR

102.72

102.41

101.75

99.58

98.07

85.35

72.14

66.11

60.9

|
0
|
21
|
41
|
62
|
82
|
103
|
123

There's not much to say here except we'd like to see how the 8500 performs without the drivers limiting its performance.

Unreal Tournament
1600x1200x32
NVIDIA GeForce3 Ti 500

NVIDIA GeForce3

NVIDIA GeForce3 Ti 200

ATI Radeon 8500

NVIDIA GeForce2 Ti 200

NVIDIA GeForce2 Pro

STMicro Kyro II

ATI Radeon 7500

ATI Radeon 64DDR

89.69

82.37

72.44

49.41

46.76

45.39

43.1

40.69

34.88

|
0
|
18
|
36
|
54
|
72
|
90
|
108



Max Payne

Short but sweet is the best way to describe Max Payne. The game will definitely leave you wanting more not only in terms of the game but in your hardware as well. We debuted this benchmark in our Detonator 4/XP and Radeon 8500 articles. The guys at 3D Center have great instructions on how to benchmark Max Payne (including how to use the AnandTech benchmark).

MaxPayne
1024x768x32
NVIDIA GeForce3 Ti 500

NVIDIA GeForce3

NVIDIA GeForce3 Ti 200

ATI Radeon 8500

NVIDIA GeForce2 Ti 200

ATI Radeon 7500

NVIDIA GeForce2 Pro

ATI Radeon 64DDR

STMicro Kyro II

80.98

75.57

71.13

68.27

55.88

54.08

53.55

38.29

26.72

|
0
|
16
|
32
|
49
|
65
|
81
|
97

The standings continue to remain the same under Max Payne. This is actually quite interesting because 3DMark 2001 is based upon the Max Payne engine and the Radeon 8500 does extremely well under that benchmark yet is unable to outperform the GeForce3 Ti 200 under Max Payne. Max Payne is also a Direct3D game so ATI's claims of poor OpenGL performance aren't valid here either.

MaxPayne
1600x1200x32
NVIDIA GeForce3 Ti 500

NVIDIA GeForce3

ATI Radeon 8500

NVIDIA GeForce3 Ti 200

ATI Radeon 7500

NVIDIA GeForce2 Ti 200

NVIDIA GeForce2 Pro

ATI Radeon 64DDR

STMicro Kyro II

53.25

45.57

40.54

39.76

26.78

25.74

23.9

19.41

12.18

|
0
|
11
|
21
|
32
|
43
|
53
|
64

At 1600 x 1200 the 8500 is able to come close to the GeForce3 once again, but it is still over 30% away from the GeForce3 Ti 500.



Driver Performance: Windows 98 vs. Windows 2000 vs. Windows XP

We first introduced this section in the Radeon 8500 Preview and we are bringing it back again to see if there are any performance issues caused by our decision to use Windows 2000 SP2 as the benchmarking OS of choice. If you look back at our preview you'll note that the performance gap between 9x/Me and 2000/XP was not as large as it once was.

Windows 98SE
Windows 2000 SP2
Windows XP
Serious Sam
ATI Radeon 8500
76.0
74.7
72.6
NVIDIA GeForce3 Ti 500
90.8
91.2
89.8
Quake III Arena
ATI Radeon 8500
203.7
207.5
205.6
NVIDIA GeForce3 Ti 500
211.4
223.7
217.9
Wolfenstein
ATI Radeon 8500
128.1
129.2
128.2
NVIDIA GeForce3 Ti 500
140
142
140.3
UnrealTournament
ATI Radeon 8500
95.34
72.14
70.66
NVIDIA GeForce3 Ti 500
106.39
101.75
102.3
MaxPayne
ATI Radeon 8500
75.49
68.27
57.28
NVIDIA GeForce3 Ti 500
80.60
80.98
80.11

The performance of both cards under Serious Sam, Quake III Arena and Wolfenstein thankfully does not change much as you make the transition from one OS to the next. The vsync problem with the Radeon 8500 drivers is no present under Windows 9x/Me allowing UnrealTournament to run much faster under that OS than Windows 2000. It is worth noting that the Radeon 8500 comes very close to the GeForce3 Ti 500's performance in that benchmark with the vsync issue eliminated.

Under Max Payne, performance has improved considerably since we originally looked at the card but the majority of the performance increases are under Windows 9x/Me. While it's arguable that 9x/Me is more of a gamer's OS choice, neither OS is very stable and the future of Windows is in the NT based OSes (2000/XP) so ATI and NVIDIA should be focusing on performance there. NVIDIA seems to be doing fine but ATI still has some room for improvement.



2D Performance & Image Quality

We first introduced comprehensive 2D image quality tests in our GeForce3 Roundup where we actually subjectively compared the analog 2D output of GeForce3 cards at high resolutions. We concluded that most cards were similar at 1600 x 1200 while very few were clear at any higher resolutions. We repeated that comparison here for this Radeon 8500 review and decided to compare the 8500 directly to the VisionTek Xtasy GeForce3 Ti 500 card. At 1600 x 1200 @ 85Hz on a 21" Sony Trinitron FD the VisionTek and Radeon 8500 cards were very difficult to tell apart; the 8500 was arguably slightly better but it was very difficult to tell. At 1900 x 1440 @ 75Hz the two were again, very close in image quality but this time the Radeon 8500 was a bit clearer and it was definitely noticeable. At all lower resolutions the two were identical in 2D image quality.

With the release of the Detonator XP drivers NVIDIA has been claiming superior 2D performance under Windows XP, citing the use of an XP effects benchmark to illustrate superior performance when using Windows XP. In order to see how much slow down you'd truly incur when using one video card vs. another under Windows XP we went to a more real world test, Content Creation Winstone 2001. The only variable that was changed was the video card; again we had the VisionTek Xtasy GeForce3 Ti 500 vs. the ATI Radeon 8500. On our test bed the Radeon scored a 68.3 while the GeForce3 Ti 500 scored a 68.4; that's less than a 1% difference and definitely not worth making a big deal out of.

Compatibility

Obviously a big issue for ATI has been compatibility, as their drivers have not always been known for universal compatibility. Luckily times change and things are much different now, but we did run into a few compatibility issues and are able to put others to rest.

The first compatibility issue was the Black & White issue at 1600 x 1200 x 32, however the game worked fine at all other resolutions. Secondly, Madden 2002 would not run at all on the Radeon 8500. ATI is aware of this issue and is promising a fix in the next version of their drivers which is due out later this month.

While we tested all of the games on a Pentium 4 2.0GHz platform under Windows 2000, we verified compatibility with Athlon platforms and the Windows XP OS. The card ran fine on an EPoX KT266A motherboard with an AMD Athlon XP 1800+ and it also ran fine with the supplied drivers under Windows XP. We have not completed extensive compatibility or stability testing under either of those situations but it was important to let you all know that in our limited experience, everything has been working fine. The card and drivers we tested showed worlds of improvement over the original Radeon 8500 silicon and drivers we previewed just two months ago; there were no more random crashes or issues with games.

FSAA: A no-show

Unfortunately one of the most interesting features of the Radeon 8500 that we could take advantage of in current games, ATI's SMOOTHVISION AA, is disabled in the current Radeon 8500 drivers. In spite of the fact that there is a SMOOTHVISION AA panel under Direct3D, the only AA modes supported are the older and inefficient supersampling methods. ATI is promising SMOOTHVISION support in a newer driver revision due out later this month.



Final Words

We would be lying if we didn't mention that the Radeon 8500's "final" performance was a bit disappointing; we weren't expecting parity with the $199 GeForce3 Ti 200, we were expecting a GeForce3 Ti 500 killer. All of the specs pointed at a higher performing product, but in the end we are limited by what has been ATI's Achilles' heel: drivers. Luckily the majority of these driver issues are performance related and not compatibility/quality related, but it still means that if you purchase a Radeon 8500 today you won't be getting all that you're paying for; untapped potential is great if it is eventually going to be exposed.

This brings us to the question of exactly how "final" the Radeon 8500's drivers are. ATI has already stated on numerous occasions that they will continue to improve the Radeon 8500's drivers and with each step they will get closer to their goal of extracting every ounce of performance out of the R200 chip. But how long will it take? It won't be a week, not a month, and maybe not even a full quarter before we see the Radeon 8500 running at its full potential. And when the day comes that it is running as fast as it can without any drivers holding it back, what will NVIDIA be doing? You better believe that NVIDIA isn't sitting around idle while the Radeon 8500 begins to encroach on their territory. A company that is used to dominating the market now has a brand new part that is able to come within a few percent and even outperform two of their fastest GPUs. The Radeon 8500 isn't a performance leader part yet, but it very well could become one.

So for those buying today, is the Radeon 8500 a waste of money? Because of its aggressive pricing, you can find Radeon 8500 cards well below the MSRP of $299. Assuming a street price of $250, the Radeon 8500 can give you performance approximately equal to that of a $199 GeForce3 Ti 200 (street price of $179). The potential and additional features of the Radeon 8500 could justify the increase in cost but at this point, we'd recommend more of a wait and see strategy. If you need solid performance today at the best overall price, the GeForce3 Ti 200 is perfect for you. That may change as the Radeon 8500's drivers improve but for now it's just not worth it.

ATI has a short window of opportunity to make the 8500 a success; if they don't accomplish that, NVIDIA will definitely make sure that there is never this close of a call as long as they can help it.

Log in

Don't have an account? Sign up now