iXBT Labs :: Computer Hardware In Detail

Platform

Video

Multimedia

Mobile

Other

Brief Retrospective Review of 3D Accelerators in Three Tests




Time flies by, IT industry is rapidly developing, especially 3D accelerators. Their power and capabilities have markedly improved over the past 5 years. 5 years ago, in 1998 we got the unexampled 3dfx Voodoo2. It was a separate video card supporting 3D modes only. It had to be installed together with a 2D card. 

At that tome the market could offer few combo 2D/3D accelerators: NVIDIA Riva 128 and ATI RAGE/II/Pro. There were some other attempts to combine 2D and 3D even in a single chip, but they were unpopular, in contrast to the successful 3dfx Voodoo Graphics (it was also a separate 3D accelerator). Another noticeable solution was Matrox M3D on the first PowerVR product. Let me quote an abstract from a once popular review (actually the first one) from Reactor Critical written by Konstantin Martynenko AKA Kenguru, founder of this site: 

"Finally I got Voodoo. Permedia 2 and Riva 128 based cards were not new on the market, but PowerVR was a new-comer. It had to come to the scene anyway due to the fame of Matrox, whose products are brought to Moscow by many computer firms. Matrox released a PowerVR PCX2 based card and gave it a plain name of m3D. 

Briefly, PCX2 is the second-generation integrated chip of PowerVR architecture developed by NEC and Videologic. NEC makes chips, and Videologic produces cards of the Apocalypse family. The m3D was the first clone of Videologic Apocalypse 3Dx. Apocalypse 3Dx and m3D have absolutely identical capabilities. 

The m3D is a daughter PCI card which interacts with a 2D card without cables. Data between the m3D and a 2D card are delivered by a PCI bus. I'd been searching an appropriate candidate for long while, and finally chose m3D just because I didn't want to throw away my Millennium. In this aspect m3D looks like Voodoo (you don't have to refuse a 2D card), but Voodoo has some disadvantages, one being so grave that it brought to zero all advantages of a good video card. 

When I opened a huge retail box I found in there a mouse-size card, - I haven't seen a smaller one before. It incorporated one big chip PCX2, and two smaller ones - 4 MB SDRAM memory. The memory of m3D is fully dedicated to textures; it supports neither a frame buffer nor a Z one. The m3D uses memory of the main 2D card for the frame buffer. It looks like overlay for a hardware MPEG decoder. The window area is not of the 2D card; it belongs to another card, i.e. m3D, which renders pictures in this window. We can suggest that the m3D can work actually in any modes of the 2D card. But it's not that simple. 

The matter is that overlay is performed in the digital form, i.e. m3D logically disposes of a part of the 2D card's video memory and carries out there its calculations. It creates some limitations for color depth. The color depth of Windows 95 desktop must coincide with that used in the m3D. It's simple to arrange it. The m3D works with that color depth which is set in Display Properties. For example, if 24bit truecolor is set, the m3D will utilize 24bit RGB color. It also concerns full-screen games which change the resolution! Hence the first problem. If you prefer to work in truecolor, and play in hicolor (because of performance), you will have to change the desktop's color depth every time. 

... 

Resume. PCX2 is a very good chipset, and m3D an excellent card. If you are looking for a new card and don't want to lose the old one, go with the m3D." 

Do you feel nostalgic? Isn't it the naivety that people have when excitedly start something new and feel, in every fiber, good prospects of this sphere? We all started something... Sometimes I reread my first reviews to get relaxed because they make me smile... :-) 
"I haven't seen a "live" Diamond Monster 3D II, but I'd like to see how Diamond rearranged the chips so that they heat less (it was the official reason to delay the Diamond's 3Dfx Voodoo2 accelerator). I have a special fan in the PC case for all PCI cards (I had to use it for Riva 128 video cards as well, as they quickly raise their temperature). When I switched it on, the tests were flawlessly passed several times, and Forsaken DEMO was run some 15 times. 

... 

Canopus's new card Total 3D V128 has marginally lost to Diamond Viper V330 in all parameters. It's a complicated matter since both cards are built on the same chipset Nvidia Riva128, but Viper V330 does not support TV input/output, while Total 3D V128 does. It's possible that some excessive logic elements have affected the performance. 

... 

After we published our first review and Reactor Critical their diametrically opposite one, we retested the card together, and I must say that in Incoming game when movement is very slow some objects (in particular, mountains) in some scenes move disproportionately, and it creates the effect called "jelly" by Reactor Critical. However, this negative effect doesn't make visible artifacts and leaves playability at a rather high level. We suppose that it's caused by inaccuracy of calculations in case of irregular types of variables. The same effect is noticeable in Quake2. 

We haven't noticed either any holes between textures or a "rainbow". 

That was all about Matrox G200... We even had a meeting with Konstantin (on the right on the photo below) to reach a consensus because the  views were entirely different. 



It's taken from the materials of 1997-1998. 

But I'm not going to turn today's article into recollection about former favorites. It's very interesting to compare at least once performance of the former leaders and famous 3D accelerators on modern applications with graphics cards of latest generations. Obviously, far not all old 3D accelerators are able to work on modern platforms, and first of all, because of the AGP 3.0 standard which cuts off all cards unsupporting 1.5V power supply. It covers almost all 3dfx video cards (when the company started making cards itself). 

As you might know, we have a museum of computer equipment, and many video cards get there as exhibits (they remained after tests, or they were gifts from friends; there are quite rare things). Unfortunately, we don't have all 3D accelerators for PC in our museum, but the list of cards is pretty lengthy. Fortunately, many cards are still operable and can be tested. 

Our 3Digest monthly covers 32-34 graphics cards. The list of them is regularly updated, and accelerators, which have no support, or drivers for them are rarely released, or monthly retesting makes no sense, or drivers do not boost performance or cope with bugs, go to the archive. Today we decided to fish some cards out of the archive and test together with modern cards which are still covered by 3Digest. But it's impossible to test all the samples from the archive because: 

  1. Many accelerators are not able to work anymore (for example, GeForce256 based cards), and some of them are installed in our office computers; 
  2. Due to problems with 1.5V on AGP most former leaders can't be tested as all our testbeds have mainboards with such limitation (i7205, nForce2). 

But do not worry, - our second Retro review will deal with all the accelerators on the old platform based on i440BX and 233MHz Pentium II. It will be an opposite picture: performance of modern accelerators on an old platform. 

Below is a list of the accelerators we tested and wanted to test but couldn't because of the AGP limitation (the red cross on the photos means that a given card is dead for new platforms). Some of them were described in our 3Digests (the link will be given in this case). 

1998

i740 4MB





When first rumors about this accelerator reached the market, it was predicted more fame and twice better performance than even that of Riva 128 had... The reality was more prosaic. But this accelerator found its niche. It was backed by Intel... 

The chip is presented by Diamond Multimedia's card. 

NVIDIA Riva 128 4MB





ASUSTeK's video card. 

3dfx Voodoo2 SLI 24MB





The dual accelerator serves a base for Quantum3D's Obsidian2 X-24. Why dual? - Because we tested the cards in 1024x768, which is not supported by the single Voodoo2. 

NVIDIA RivaTNT 16MB





The chip is represented by ASUSTeK's card. 

1999 

NVIDIA RivaTNT2 M64 32MB





MSI's card. This is actually one of the rarest cases when TNT2 M64 works on modern platforms. 

NVIDIA RivaTNT2 Ultra 32MB





Hercules's card. It doesn't work on i7205. 

3dfx Voodoo3 16MB





This is a flagship of the Voodoo3 line - Voodoo3 3500TV. But all these cards are only AGP 1.0 certified! 

3dfx Velocity100 8MB





That was the favorite product of summer 1999! It was a cheap but effective card... Now it's only an exhibit. 

S3 Savage4 32MB





This is an 3D idol of a myriad of budget-concious users! It was a large aggressive community that combated those who didn't want to use drivers mixing, S3Tweak utility and regularly reflash new BIOS versions to make performance satisfactory. One can still compose legends on the Savage... The forefather of S3TC texture compression technology (it appeared yet in Savage3D in 1998, but at that time S3TC drew no attention at all). 

Matrox G400 32MB





Matrox's card. 

ATI RAGE 128 PRO 32MB





At that time the Canadian company made cards alone, that is why this chip comes on its own card. 

S3 Savage 2000 32MB





It was a significant product: after the tough criticism of Savage4 its fame was fading away but the Savage2000 pushed it up again, and that card was tweaked multiple times. But it didn't help S3 as a GPU maker. The Savage4/2000 floated just a little after 1999 thanks to the drivers mixing. 

2000

ATI RAGE MAXX (2 x RAGE 128 PRO) 64MB





Born dead. While the card performed pretty well in Windows 98, it failed in Windows 2000 (not to mention Windows XP) because of lacking drivers (ATI then admitted impossibility  to write drivers for NT systems). Later, the support for Windows 9x was withdrawn. 

NVIDIA GeForce2 GTS 32MB





AOpen's card represents this chip. 

NVIDIA GeForce2 Pro 64MB





This GPU comes on Chaintech'a card. 

NVIDIA GeForce2 MX200 32MB





Gigabyte'a card. 

NVIDIA GeForce2 MX400 32MB





NVIDIA's reference card on GeForce2 MX which was overclocked up to the level of MX400 - 200/183 MHz. 

3dfx Voodoo4 4500 32MB





3dfx's card produced at STB's plant (3dfx bought this firm yet at the end of 1998). 

3dfx Voodoo5 5500 64MB





3dfx's card. 

ATI RADEON 32MB SDR





ATI's card. 

ATI RADEON 64MB DDR





The same. 

SIS 315 128MB





Joytech'a card. 

PowerVR/STM KYRO 32MB





Hercules card. 

2001

NVIDIA GeForce3 64MB





NVIDIA's reference card. 

NVIDIA GeForce3 Ti200 64MB





ABIT'a card. 

PowerVR/STM KYRO II 32/64MB







64MB card from Hercules, 32MB card from S/U/M/A/. 

Matrox G550 32MB DDR





This "new old" product ships on the Matrox's card. 

ATI RADEON 7500LE 64MB DDR





Jetway's card. 

ATI RADEON 7500 64MB DDR





ATI's card. 

ATI RADEON 8500 128MB DDR 3.6ns HYNIX (275/550 MHz)

ATI RADEON 8500 64MB DDR 3.6ns (275/550 MHz)

ATI RADEON 8500LE 64MB DDR 3.6ns (250/500 MHz)

ATI RADEON 8500LE 128MB DDR 4ns (250/500 MHz)

2002

Trident Blade XP 32MB





HIS card. 

S3 Savage XP 32MB





This is a demo card, i.e. a graphics card on the early silicon (T&L, for example, doesn't work here). Production of such cards weren't even started. 

PowerVR/STM KYRO II SE 64MB





Hercules's card. The same product. Actually dead. 

NVIDIA GeForce4 MX 440 64MB





eVGA'a card. 

NVIDIA GeForce4 MX 440 with AGP8x 64MB DDR 3.6ns (275/513 MHz)

NVIDIA GeForce4 MX 440 with AGP8x 128MB (280/520 MHz)

NVIDIA GeForce4 MX 440 with AGP8x 64MB GS (300/600 MHz)

NVIDIA GeForce4 Ti 4200 64MB DDR 4ns (250/513 MHz)

NVIDIA GeForce4 Ti 4200 128MB DDR 4ns (250/444 MHz)

NVIDIA GeForce4 Ti 4200 128MB DDR 3.3ns (250/550 MHz)

NVIDIA GeForce4 Ti 4200 128MB DDR 4ns (250/500 MHz)

NVIDIA GeForce4 Ti 4200 with AGP8x 128MB DDR 4ns (250/513 MHz)

NVIDIA GeForce4 Ti 4200 with AGP8x 128MB DDR 2.8ns (275/600 MHz)

NVIDIA GeForce4 Ti 4600 128MB DDR 2.8ns (300/650 MHz)

NVIDIA GeForce4 Ti 4600 GS 128MB DDR 2.8ns (310/680 MHz)

ATI RADEON 9000 64MB DDR 4ns (250/400 MHz)

ATI RADEON 9000 128MB DDR 3.6ns (250/400 MHz)

ATI RADEON 9000 Pro 128MB DDR 3.6ns (275/550 MHz)

ATI RADEON 9000 Pro 64MB DDR 3.6ns (275/550 MHz)

ATI RADEON 9500 64MB 3.6ns (275/540 MHz)

ATI RADEON 9500 128MB 3.6ns (275/540 MHz)

ATI RADEON 9500 PRO 128MB 3.6ns (275/540 MHz)

ATI RADEON 9700 Pro 128MB DDR 2.8ns (325/620 MHz)

ATI RADEON 9700 128MB DDR 2.8ns (275/540 MHz)

SIS Xabre 400 64MB DDR 3.5ns (250/500 MHz)

SIS Xabre 400 128MB DDR 4ns (250/500 MHz)

SIS Xabre 600 64MB DDR 4ns (315/630 MHz)

Matrox Parhelia 128MB DDR 3.3ns (220/550 MHz) (Retail)

2003

NVIDIA GeForceFX 5800 Ultra 128MB DDR2 (500/1000 MHz)

NVIDIA GeForceFX 5800 128MB DDR2 (400/800 MHz)

NVIDIA GeForceFX 5600 Ultra 128MB DDR (350/700 MHz)

NVIDIA GeForceFX 5200 Ultra 128MB DDR (325/650 MHz)

ATI RADEON 9200 64MB DDR 4ns (250/400 MHz)

ATI RADEON 9800 Pro 128MB DDR (380/680 MHz)

Tests

Testbeds: 

  • Pentium 4 3066 MHz based computer: 
    • Intel Pentium 4 3066 MHz; 
    • ASUS P4G8X (iE7205) mainboard; 
    • 1024 MB DDR SDRAM; 
    • Seagate Barracuda IV 40GB; 
    • Windows XP SP1; 
    • ViewSonic P810 (21") and ViewSonic P817 (21") monitors. 
    • all latest drivers (for the middle of April 2003). 

VSync off in drivers, texture compression off in applications. Texture detail set to High Quality. 

Test applications: 

  • Return to Castle Wolfenstein (MultiPlayer) (id Software/Activision) - OpenGL, multitexturing, Checkpoint-demo, maximum test settings, S3TC OFF, the configs are available here
    •  

  • Serious Sam: The Second Encounter v.1.05 (Croteam/GodGames) - OpenGL, multitexturing, Grand Cathedral demo, test settings: quality, S3TC OFF 
    •  

  • Unreal Tournament 2003 Demo v.1077 (Final Release) (Digital Extreme/Epic Games) - Direct3D, Vertex Shaders, Hardware T&L, Dot3, cube texturing, default quality 




 
 


 
 



Conclusion

Why didn't we test the cards in 800x600? As you can see, most powerful accelerators are limited by the system resources yet in 1024x768; in 800x600 1/3 of the cards will have equal scores :-). Besides, it's 2003. Retro is a good thing, but gamers prefer 1024x768 and higher. 

Actually, all cards removed from 3Digest and simply obsolete accelerators (they are marked lilac on the diagrams) are at the bottom. Only some of them perform more or less well (for example, GeForce3, but it was taken out of production long ago, and vendors are selling out remains from stocks). 

This material is a once-only action, and I ask you not to wait for regular reviews of old graphics cards :-). 
 
 
 
 
 

Andrey Vorobiev (anvakams@ixbt.com
 

Article navigation:



  Most Popular Reviews More   RSS  

Intel Core i5-660/661 Processors

The long-awaited dual-core solutions for the LGA1156 platform.
Jan 09, 2010 · Processors: Intel

AMD Athlon II X3 425, 435 Processors

Three cores for the price of two.
Oct 30, 2009 · Processors: AMD

E-MU 0404 USB

A top-class external audio interface.
Nov 23, 2006 · ProAudio

Intel H55, H57 Chipsets

Integrated chipsets for lower-end Nehalem processors.
Jan 18, 2010 · Chipsets

Creative Sound Blaster X-Fi Surround 5.1

An external X-Fi solution in tests.
Sep 09, 2008 · Sound Cards
  Latest Reviews More   RSS  

ASUS M4A785G HTPC/RC Motherboard

A microATX board with a high-quality audio codec and a remote.
Feb 05, 2010 · Motherboards

Thermaltake Element G Mid Tower Case

Compact size, fan controller, and stylish lights.
Feb 03, 2010 · General Platform

ATI RADEON HD 5670 Graphics Card

A low-end solution with DirectX 11 support.
Feb 02, 2010 · Video cards: ATI GPUs

i3DSpeed, January 2010

Updated the list of benchmarks. Added test results for Radeon HD 5670.
Feb 01, 2010 · 3Digests

Intel Celeron E3300 Processor

A full-fledged low-end solution or a "socket stub?"
Jan 29, 2010 · Processors: Intel
  Latest News More   RSS  
Feb 09, 2010 Intel Introduces Itanium 9300 Processor
Feb 09, 2010 ATI Unveils Radeon HD 5570 Graphics Card
Feb 02, 2010 Microsoft Launches SideWinder X4 Keyboard With Advanced Anti-Ghosting Technology
Feb 02, 2010 Intel, Micron Introduce 25nm NAND
Feb 02, 2010 Daily Mailbox
Feb 01, 2010 Transcend Releases 4GB, 8GB DDR3 RDIMMs For Cloud Computing and Server Applications
Feb 01, 2010 BenQ Unveils Two 15.6-inch LED Displays
 
  Useful Links Get listed  

Computer Memory Upgrades

Apple Memory Upgrades

Platform  ·  Video  ·  Multimedia  ·  Mobile  ·  Other  ·  Compare Prices  ||  Feedback  ·  Advertise at iXBT Labs  ·  About us  ·  Affiliates  ·  Forum


52

Copyright © Byrds Research & Publishing, Ltd., 1997—2010. All rights reserved.