Accelerate Your Macintosh! Radius Tempest Video card Preview

Note: This article was written in 1997. I no longer have this card and Radius left the graphics card business years ago. I do not know the source of any driver updates for this card. All the links I had in this original article no longer work and were removed. The only link I have for Radius is www.radiusvintage.com. That URL is not responding now however (as of 10/15/2001).
For reviews of far faster and more modern Mac graphics cards, see the articles/reviews at the Video articles page.

Intro:

(Original 1997 article text follows)
Interest in Macintosh 3D video cards seems to be at an all-time high, especially with the many affordable, performance oriented cards that have recently been announced. I've received a very large number of requests for a review of the Radius Tempest 2D/3D video card lately, and Radius was kind enough to send a card for review, along with a pre-production set of drivers. I expect performance to increase as Radius finalizes the drivers.

As with all my video card tests to date - the card was tested in a Apple 8500, with a PowerLogix Pro 233 card set to 50mhz bus, 200mhz cpu (to emulate the average 200mhz system). The system configuration included:

128megs of 60ns Ram (two 64meg Dimms)
PowerLogix 1 meg Rapidcache
OS 7.6.1
Quickdraw 3D 1.5.1
Quicktime 2.5
Speed Doubler 2.03
LibMoto FPU extension
512k Disk cache
No Virtual Memory
Monitor for test Card:
Apple Multiple Scan 20
8500 Video monitor:
Mitsubishi Diamondtron 21"

Retaining this configuration allows direct comparison to the previous tests of the ATI RagePro, and Imagine 128/Twin Turbo M8 reviews and scores.

Observations:

The Tempest shows a high level of quality in construction. The Graphics Processor is heatsinked and the board layout was very clean. For a full-size picture (65k jpeg image) of the Tempest card - click here. The Tempest has some pretty impressive hardware onboard including:

3Dlabs PERMEDIA Processor
3Dlabs GLINT Delta pipeline preprocessor
8MB High Speed SGRAM

Read the specifications on the Permedia and GLINT delta (links above) and you'll see this is some of the most impressive hardware ever to grace a $399 SRP Mac Video card. With that said, I have to admit I was somewhat disappointed in the performance - especially the 2D performance. With the support hardware the Tempest has this must be a driver issue. An example of this was the fact the Gerbils 3D demo application did not seem to want to use texture mapping at the 1024x768, thousands color mode - even though the card has 8 megs of Sgram. Lowering the resolution to 832x624 resulted in texture mapping appearing on the intro Gerbils screen. Again, the driver set I have is a first effort - and surely will be optimized further in the future.

While on the subject of drivers, the Tempest has two levels of rendering quality, selectable via software. All tests were done on the standard (default install) renderer.

Note: The Tempest has only a Apple style 15-pin monitor jack. Owners of standard VGA monitors will need an adapter.

2D Performance:

All 2D tests were run at 1024x768, thousands (16-bit) color depth. In general MacBench scores were a little disappointing, as the RagePro and 8500 built-in video outperformed the Tempest in 2D.
Radius Tempest - MacBench 4.0 results, 1024x768, thousands colors, 75mhz refresh

As the graphs show, the built-in 8500 video does fairly well at the standard graphics tests with the 50mhz bus setting, which is standard for most 200mhz cpus (as this 8500 was configured). In general 2D speed of the Tempest was not as good as the 8500 video, based on Macbench and Norton SI 3.5 Video tests. In actual use the difference was less noticable than the scores indicate. Radius also stated the card/driver is not optimized for the 8-bit images that MacBench uses.
I also ran Norton Utilities v3.5 System Info Video test - again at 1024x768, thousands colors with the following results:

Tempest = 136
RagePro = 238
8500 Video = 132

Application Tests:
At 1024x768, thousands colors, Zooms in Photoshop were were noticably slower than either the Imagine 128 or Twin Turbo, and about equal to the 8500 built-in video. Scrolls and general operations with web resolution images in photoshop seemed at least as fast as the 8500 video however.

3D Performance:

Walker 1.1 has become somewhat of the Macbench of 3D tests - and is truly a torture test of 3D card performance. Some of the scenes have a very high number of polygons - much more than the average game or model would contain. Here the Tempest gave a very good showing - equal or better to the RagePro and in general the highest scores I've seen in Walker. As with most other video cards, some scenes showed severe artifacting (black triangles), which may be the result of the driver or a QD3D 1.5.1 bug. Literally every video card I've seen has exhibited this behavior in this test.
The scores below are the minimum frames per second observed during two spins around each scene. It was interesting to note that after several spins - the Tempest framerates increased slightly (about 1-2fps), as if the data was being buffered. I have not noticed this effect with other video cards.
Walker 1.1 Scores (1024x768 - thousands colors)
Corridor Scene (49k polygons)

Tempest = 3.53 fps
Rage Pro = 3.53 fps
Twin Turbo M8 = 2.45 fps
Imagine 128S2 = 2.67 fps
8500 Video = 2.45 fps (50mhz bus)

Exhibit Scene (20k polygons)

Tempest = 9.09 fps
Rage Pro = 8.57 fps
Twin Turbo M8 = 5.45 fps
Imagine 128S2 = 6.67 fps
8500 Video = 5.29 fps (50mhz bus)

Staircase Scene (25k polygons)

Tempest = 10.59 fps
Rage Pro = 9.68 fps
Twin Turbo M8 = 7.74 fps
Imagine 128S2 = 6.15 fps
8500 Video = 6.15 fps (50mhz bus)

Atrium Scene (16k polygons)

Tempest = 14.55 fps
Rage Pro = 13.12 fps
Twin Turbo M8 = 7.27 fps
Imagine 128S2 = 10.91 fps
8500 Video = 7.27 fps (50mhz bus)

As the scores show, even the initial Tempest driver release shows 3D performance is slightly better in Walker than the RagePro, and up to 2X faster than the 8500 video even when run in a 50mhz bus system.

Gerbils 1.03 Performance:
Frame rates ran from a minimum of mid-60's fps to 130+ fps. As mentioned previously, the initial Gerbils logo screen did not show texture mapping at 1024x768 resolution. Lowering the video mode to 832x624 did show the logo with a texture map. This may be a texture memory allocation bug in this driver version. Performance was literally identical at both resolutions however.

3D Application Performance:
Modeling in Infini-D 4.0 was very fast, and shaded previews were handled at least as well as the Imagine 128 Series 2 using the software renderer. Using the HW (Hardware) renderer did allow movement of shaded objects without reverting them to wireframe, but resulted in a much less fluid movement as a result (lag while the other views wireframes were updated). Even closing some of the other view windows did not help. Overall I was disappointed in the performance with the hardware renderer - although I could move shaded 3D objects, the lag in display updates made it a frustrating experience when attempting to place objects in a scene (example - tutorial 6 and 7). Again, this may be a driver issue (or a Infini-D 4.0 one) that will be improved upon in the future.

Using the standard renderer in Infini-D 4.0 also resulted in a less pleasing shaded view of objects. There was a distinct lack of transparency in the bottle in Lesson 7 of the tutorial for instance using the HW renderer. Given time, I intend to retest using the better quality rendering extension. Based on the lack of textures shown in the Gerbils logo at 1024x768, I intend to retest Infini-D at a lower resolution as well.

Game Performance:
Currently the driver set does not support Rave games like Quake. Launching Rave Quake resulted in a "unable to find Rave card" message. However I could run standard Quake, which with flames on, at 640x480 screen size produced a Timedemo demo1 score of 10.7fps (about 1/2 the speed of the RagePro and Power3D Rave tests). It didn't seem nearly that slow in actual game play, but texture artifacts (sparkling) was present in most areas.

Movie Performance:

Movie playback performance was about equivalent to the 8500's built-in video. The following shows the results of the standard MacBench 4.0 Video playback battery of tests run at 1024x768, thousands colors:

As done in the Imagine 128 vs TT review, I also ran a "custom" video playback test at full screen (1024x768, thousands colors) using the std Cinepak2 320x240 movie blown up to full screen. The movie was copied to ram to eliminate any effects of the 4X cdrom on playback speed. The results are shown below in frames played and max frame rate:

As you can see, movie playback performance is slightly less (but not noticably so) for non-scaled movies than the 8500 built-in video. However when scaling to full screen is done, the 8500 video is about 30% faster.

Stability/Reliability:

There were no application or system crashes of any kind during any of the tests. The drivers appear rock solid from my initial tests.

Summary:

Given the impressive hardware present on the Tempest, I was somewhat disappointed in the performance. Given Radius' track record I'm sure this will be improved in future driver versions. I look forward to new drivers from Radius, and if I am allowed to retain the card until they arrive I'll update this page to reflect the results.

Base System Notes:

The base system used to test the Radius Tempest^TM was our standard PowerMac 8500, set to 200mhz cpu, 50mhz bus speed, with 128megs of matched memory (60ns), standard built-in video with 4 megs Vram, running OS 7.6.1. Disk cache was 512k, Speed doubler 2 and LibMoto were active (Speed Doublers faster disk access was disabled). Quicktime 2.5 and Quickdraw 3d 1.5.1 extensions were also enabled (as done in prev. testing).

Cache Used :

PowerLogix 1 meg RapidCache (p/n: RC1MB02)
(in my system the standard RapidCache runs as fast a bus speed as the Ultra model, and costs less).
PowerLogix PowerBoost Pro 233 (set to 50mhz bus/200mhz cpu)

Benchmark Notes:

All tests were run using MacBench 4.0. (Link expired - Macbench was removed from the web and is no longer available).

= Availability =
Cyberian Outpost had stock on the Tempest at $387.99 at the date this was published.
(This card however is no longer made)

= Other Site Topic Areas =
Systems | CPU Upgrades | SCSI | IDE | Firewire | Cache | Video | Forums | FAQ

= Top of Page =

= Back to WWW.XLR8YOURMAC.COM =

Copyright © , 1997.
No part of this sites content is to be reproduced in any form without written permission. All brand or product names mentioned here are properties of their respective companies.
Users of this web site must read and are bound by the terms and conditions of use.


Note: This article was written in 1997. I no longer have this card and Radius left the graphics card business years ago. I do not know the source of any driver updates for this card. All the links I had in this original article no longer work and were removed. The only link I have for Radius is www.radiusvintage.com. That URL is not responding now however (as of 10/15/2001). For reviews of far faster and more modern Mac graphics cards, see the articles/reviews at the Video articles page.
Intro:
(Original 1997 article text follows) Interest in Macintosh 3D video cards seems to be at an all-time high, especially with the many affordable, performance oriented cards that have recently been announced. I've received a very large number of requests for a review of the Radius Tempest 2D/3D video card lately, and Radius was kind enough to send a card for review, along with a pre-production set of drivers. I expect performance to increase as Radius finalizes the drivers. As with all my video card tests to date - the card was tested in a Apple 8500, with a PowerLogix Pro 233 card set to 50mhz bus, 200mhz cpu (to emulate the average 200mhz system). The system configuration included: 128megs of 60ns Ram (two 64meg Dimms) PowerLogix 1 meg Rapidcache OS 7.6.1 Quickdraw 3D 1.5.1 Quicktime 2.5 Speed Doubler 2.03 LibMoto FPU extension 512k Disk cache No Virtual Memory Monitor for test Card: Apple Multiple Scan 20 8500 Video monitor: Mitsubishi Diamondtron 21" Retaining this configuration allows direct comparison to the previous tests of the ATI RagePro, and Imagine 128/Twin Turbo M8 reviews and scores.
Observations:
The Tempest shows a high level of quality in construction. The Graphics Processor is heatsinked and the board layout was very clean. For a full-size picture (65k jpeg image) of the Tempest card - click here. The Tempest has some pretty impressive hardware onboard including: 3Dlabs PERMEDIA Processor 3Dlabs GLINT Delta pipeline preprocessor 8MB High Speed SGRAM Read the specifications on the Permedia and GLINT delta (links above) and you'll see this is some of the most impressive hardware ever to grace a $399 SRP Mac Video card. With that said, I have to admit I was somewhat disappointed in the performance - especially the 2D performance. With the support hardware the Tempest has this must be a driver issue. An example of this was the fact the Gerbils 3D demo application did not seem to want to use texture mapping at the 1024x768, thousands color mode - even though the card has 8 megs of Sgram. Lowering the resolution to 832x624 resulted in texture mapping appearing on the intro Gerbils screen. Again, the driver set I have is a first effort - and surely will be optimized further in the future. While on the subject of drivers, the Tempest has two levels of rendering quality, selectable via software. All tests were done on the standard (default install) renderer. Note: The Tempest has only a Apple style 15-pin monitor jack. Owners of standard VGA monitors will need an adapter.
2D Performance:
All 2D tests were run at 1024x768, thousands (16-bit) color depth. In general MacBench scores were a little disappointing, as the RagePro and 8500 built-in video outperformed the Tempest in 2D. Radius Tempest - MacBench 4.0 results, 1024x768, thousands colors, 75mhz refresh As the graphs show, the built-in 8500 video does fairly well at the standard graphics tests with the 50mhz bus setting, which is standard for most 200mhz cpus (as this 8500 was configured). In general 2D speed of the Tempest was not as good as the 8500 video, based on Macbench and Norton SI 3.5 Video tests. In actual use the difference was less noticable than the scores indicate. Radius also stated the card/driver is not optimized for the 8-bit images that MacBench uses. I also ran Norton Utilities v3.5 System Info Video test - again at 1024x768, thousands colors with the following results: Tempest = 136 RagePro = 238 8500 Video = 132 Application Tests: At 1024x768, thousands colors, Zooms in Photoshop were were noticably slower than either the Imagine 128 or Twin Turbo, and about equal to the 8500 built-in video. Scrolls and general operations with web resolution images in photoshop seemed at least as fast as the 8500 video however.
3D Performance:
Walker 1.1 has become somewhat of the Macbench of 3D tests - and is truly a torture test of 3D card performance. Some of the scenes have a very high number of polygons - much more than the average game or model would contain. Here the Tempest gave a very good showing - equal or better to the RagePro and in general the highest scores I've seen in Walker. As with most other video cards, some scenes showed severe artifacting (black triangles), which may be the result of the driver or a QD3D 1.5.1 bug. Literally every video card I've seen has exhibited this behavior in this test. The scores below are the minimum frames per second observed during two spins around each scene. It was interesting to note that after several spins - the Tempest framerates increased slightly (about 1-2fps), as if the data was being buffered. I have not noticed this effect with other video cards. Walker 1.1 Scores (1024x768 - thousands colors) Corridor Scene (49k polygons) Tempest = 3.53 fps Rage Pro = 3.53 fps Twin Turbo M8 = 2.45 fps Imagine 128S2 = 2.67 fps 8500 Video = 2.45 fps (50mhz bus) Exhibit Scene (20k polygons) Tempest = 9.09 fps Rage Pro = 8.57 fps Twin Turbo M8 = 5.45 fps Imagine 128S2 = 6.67 fps 8500 Video = 5.29 fps (50mhz bus) Staircase Scene (25k polygons) Tempest = 10.59 fps Rage Pro = 9.68 fps Twin Turbo M8 = 7.74 fps Imagine 128S2 = 6.15 fps 8500 Video = 6.15 fps (50mhz bus) Atrium Scene (16k polygons) Tempest = 14.55 fps Rage Pro = 13.12 fps Twin Turbo M8 = 7.27 fps Imagine 128S2 = 10.91 fps 8500 Video = 7.27 fps (50mhz bus) As the scores show, even the initial Tempest driver release shows 3D performance is slightly better in Walker than the RagePro, and up to 2X faster than the 8500 video even when run in a 50mhz bus system. Gerbils 1.03 Performance: Frame rates ran from a minimum of mid-60's fps to 130+ fps. As mentioned previously, the initial Gerbils logo screen did not show texture mapping at 1024x768 resolution. Lowering the video mode to 832x624 did show the logo with a texture map. This may be a texture memory allocation bug in this driver version. Performance was literally identical at both resolutions however. 3D Application Performance: Modeling in Infini-D 4.0 was very fast, and shaded previews were handled at least as well as the Imagine 128 Series 2 using the software renderer. Using the HW (Hardware) renderer did allow movement of shaded objects without reverting them to wireframe, but resulted in a much less fluid movement as a result (lag while the other views wireframes were updated). Even closing some of the other view windows did not help. Overall I was disappointed in the performance with the hardware renderer - although I could move shaded 3D objects, the lag in display updates made it a frustrating experience when attempting to place objects in a scene (example - tutorial 6 and 7). Again, this may be a driver issue (or a Infini-D 4.0 one) that will be improved upon in the future. Using the standard renderer in Infini-D 4.0 also resulted in a less pleasing shaded view of objects. There was a distinct lack of transparency in the bottle in Lesson 7 of the tutorial for instance using the HW renderer. Given time, I intend to retest using the better quality rendering extension. Based on the lack of textures shown in the Gerbils logo at 1024x768, I intend to retest Infini-D at a lower resolution as well. Game Performance: Currently the driver set does not support Rave games like Quake. Launching Rave Quake resulted in a "unable to find Rave card" message. However I could run standard Quake, which with flames on, at 640x480 screen size produced a Timedemo demo1 score of 10.7fps (about 1/2 the speed of the RagePro and Power3D Rave tests). It didn't seem nearly that slow in actual game play, but texture artifacts (sparkling) was present in most areas.
Movie Performance:
Movie playback performance was about equivalent to the 8500's built-in video. The following shows the results of the standard MacBench 4.0 Video playback battery of tests run at 1024x768, thousands colors: As done in the Imagine 128 vs TT review, I also ran a "custom" video playback test at full screen (1024x768, thousands colors) using the std Cinepak2 320x240 movie blown up to full screen. The movie was copied to ram to eliminate any effects of the 4X cdrom on playback speed. The results are shown below in frames played and max frame rate: As you can see, movie playback performance is slightly less (but not noticably so) for non-scaled movies than the 8500 built-in video. However when scaling to full screen is done, the 8500 video is about 30% faster.
Stability/Reliability:
There were no application or system crashes of any kind during any of the tests. The drivers appear rock solid from my initial tests.
Summary:
Given the impressive hardware present on the Tempest, I was somewhat disappointed in the performance. Given Radius' track record I'm sure this will be improved in future driver versions. I look forward to new drivers from Radius, and if I am allowed to retain the card until they arrive I'll update this page to reflect the results.
Base System Notes:
The base system used to test the Radius Tempest^TM was our standard PowerMac 8500, set to 200mhz cpu, 50mhz bus speed, with 128megs of matched memory (60ns), standard built-in video with 4 megs Vram, running OS 7.6.1. Disk cache was 512k, Speed doubler 2 and LibMoto were active (Speed Doublers faster disk access was disabled). Quicktime 2.5 and Quickdraw 3d 1.5.1 extensions were also enabled (as done in prev. testing). Cache Used : PowerLogix 1 meg RapidCache (p/n: RC1MB02) (in my system the standard RapidCache runs as fast a bus speed as the Ultra model, and costs less). PowerLogix PowerBoost Pro 233 (set to 50mhz bus/200mhz cpu)
Benchmark Notes:
All tests were run using MacBench 4.0. (Link expired - Macbench was removed from the web and is no longer available).
= Availability = Cyberian Outpost had stock on the Tempest at $387.99 at the date this was published. (This card however is no longer made)
= Other Site Topic Areas = Systems \| CPU Upgrades \| SCSI \| IDE \| Firewire \| Cache \| Video \| Forums \| FAQ
= Top of Page = = Back to WWW.XLR8YOURMAC.COM =

Copyright © , 1997. No part of this sites content is to be reproduced in any form without written permission. All brand or product names mentioned here are properties of their respective companies. Users of this web site must read and are bound by the terms and conditions of use.