Nvidia Corp. formally launched the GeForce4 graphics accelerator Tuesday night, adding improved antialiasing and multimonitor support in addition to significantly enhanced overall 3D performance.
The GeForce4, as previously reported, was accompanied by first shipments of the NV17, now dubbed the GeForce4 440 Go. Both chips were unveiled at a press conference in San Francisco, supported by seven add-on card manufacturers and six PC OEMs.
Ever since the launch of the original GeForce graphics accelerator, Nvidias engineering efforts have been driven by the desire to allow audiences of CGI films like Shrek to view the same types of effects in computer games. "It is now true that crystal-clear graphics lie at the intersection of art and technology," said Jen-Hsun Huang, chief executive of Nvidia Corp., Santa Clara, Calif.
The GeForce4 will ship in six different flavors within two families: the GeForce Ti 4600, 4400, and 4200; and the GeForceMX 460, 440, and 420. The Ti or "Titanium" series of processors all feature the NV25; the MX family is actually based upon a desktop version of the NV17 core, which was announced at last falls Comdex. Also, the GeForce4 Ti family of products use 128-Mbytes of DDR frame buffer memory, while the NV17-based MX cards use only 64 Mbytes.
The GeForce Ti 4200 was announced, but will ship last, and its specifications have not been finalized, said Stephen Sims, the GeForce4 Ti product manager for Nvidia.
By including the NV17 core inside the GeForce4 brand, some observers said the company was hiding the NV17s shortcomings, including the lack of a programmable pixel shader--a feature that the GeForce3 included. Analysts differed over the issue.
"Theyre calling (the NV17) an MX product; all of Nvidias MX products have reduced capabilities compared to the flagship model, whatever line theyre in," said Dean McCarron, principal, at Mercury Research Inc., Cave Creek, Ariz. "That said, the GeForce 4 MX is turning in performance figures significantly beyond anything in the GeForce3 family, save the GeForce3 Ti. And so I can see the argument either way: on one hand, it is lacking features, on the other, it is offering performance well beyond the GeForce3."
"It isnt really in the same family," said Peter Glaskowsky of MicroDesign Resources Inc., Sunnyvale, Calif., of the NV17/GeForce4 MX. "I guess you could say theyre transitioning to where Titanium means you get all the features, and MX means the one you can afford.
"You could say verily that its a change in naming schemes, where a digit indicated a product generation and it doesnt any more," Glaskowsky said.
An Nvidia spokesman said in an emailed statement to ExtremeTech that the NV17 did not fit into either the GeForce2 or GeForce3 product line, because of its mix of advanced performance and features.
"While there is no perfect solution to naming and branding, calling NV17 anything other than a GeForce4 would have been doing the product and its customers a disservice," the Nvidia spokesman said. "The combination of the MX and Ti families with the model number differences (3-digit vs. 4-digit), provides adequate separation between the two products, allowing each to stand on its own, in its own class."
The new chip may also give Nvidia needed separation between itself and rival ATI Technologies Inc., Thornhill, Ontario, which announced its own 128-Mbyte frame buffer versions of the Radeon 8500 and 7500 family this week.
"They needed this part, and they needed it now," Glaskowsky said of Nvidia. "(The ATI) Radeon has been claiming the high end. Nvidias had the performance advantage, thats true. But ATIs been comparing itself to the fastest GeForce3."
The heart of the GeForce4: nFiniteFX II
The highlight of the NV25 or GeForce4 Ti series is the nFiniteFX II engine, which can process 136 million vertices per second, with 10.4 Gbytes/s of memory bandwidth available to the fastest GeForce Ti 4600 part. The GeForce 4 can process 1 trillion operations per second, the approximate number-crunching ability of 10 original Infinite Reality workstations from Silicon Graphics, according to Tony Tamasi, senior director of Nvidias graphics processor business.
Glaskowsky said he suspected the part is DirectX 8.1 compliant, although Nvidia officials did not confirm this.
Both the NV17 and NV25 also feature AccuView, an antialiasing technique which Nvidia says improves on the 4X sampling rate used by other components. Antialiasing smoothes the edges of lines, eliminating the "jaggies" that come from drawing a diagonal line across a pixilated CRT screen. The GeForce Ti 4600 processes 4.8 billion antialiasing samples per second; the GeForce 4 Ti 4400, 4.4 billion samples/s.
"One of the things we wanted to do with Accuview was provide antialiasing so good that youd always want to leave it on," Tamasi said.
Both cores also boast nView, a multimonitor capability that will allow the graphics processor to drive up to 16 displays. Every Nvidia graphics processor from now on will include the nView capability, Tamasi said. Additional features include z-correct bump mapping, used to smooth the interaction of bump-mapped surfaces with other polygons, an integrated MPEG decoder, a TV encoder, and two RAMDACs to drive two separate flat-panel displays. The NV25 or GeForce4 Ti also includes second-generation z-culling, which only draws the polygons the user actually sees.
The chip boasts 63 million transistors, and is fabricated on Taiwan Semiconductor Manufacturing Co.s 0.15-micron process.
"Its a beast—what can we say," Tamasi said.
The GeForce Ti 4600 uses a 300-MHz core clock frequency and a 325-MHz memory interface (effectively 650-MHz using DDR memory); the Ti 4400 uses a 275-MHz core and 275-MHz memory interface, Sims said. As of now, the Ti 4200 will use a 250-MHz core clock rate and a 250-MHz memory interface; again, since the Ti 4200s specs have yet to be finalized, they may change, Sims said.
Memory is supplied by Samsung, Micron Technology and Infineon, according to Jon Peddie, principal at Jon Peddie Research in Mill Valley, Calif. The supply of the high-speed memory may be a gating factor in the supply of GeForce4 graphics cards, but the potential for shortages is small, he said. "Remember, prices increases in memory have historically been only temporary," he said. "They go up, level off, and come back down."
Nvidia claims the GeForce4 will be over twice as fast as the GeForce3 Ti 5000. The GeForce4 chip will be faster using Accuview antialiasing than the GeForce3 using aliased graphics, executives said. Reviews of the chip were placed under embargo until Wednesday morning, when ExtremeTech will also publish a performance analysis of the component.
Is there a need for such performance? Analysts differed over whether any game truly could be said to consume the power of the existing GeForce3; Bob Merritt of Semico Research Corp., Phoenix, said he didnt think so. Analyst Peddie said id Software Inc.s Return to Castle Wolfenstein pressed the chips limits.
Huang said a forthcoming family of chips—possibly called the GeForce5—will support AGP 8X, due out later this year. The nForce chipset will also receive a GPU core, not necessarily the GeForce4, Huang said. "Wait until you see the MCP-2 (Media and Communications Processor)," Huang said. "It will unleash an experience youve never seen before."
Finally, the Quadro workstation line is due for a refresh, almost certainly with the GeForce4. That chip will likely be announced within six weeks, and will probably feature a slower clock rate, trading raw performance for more accurate per-pixel processing, a source said.
Prices and cards
Nvidias suggested price for the GeForce4 Ti 4600 is $399; for the GeForce 4 Ti 4400, $299; and the GeForce4 Ti 4200, $199. Pricing for the GeForce 4 MX460, MX 440 and MX 220 will likely be about $179, $149, and between $99 and $129, respectively. All cards will include two VGA connectors for multimonitor support and most will include an Nvidia-designed cooling unit.
Visiontek, Gurnee, Ill., will have its MX440 and MX420 boards on retail shelves on Feb. 6, although Nvidias Huang said rival boards may appear several days later. Visiontek will also offer the GeForce Ti 4400 and 4600; the Ti cards will be available at the end of the month. Pricing was not immediately available.
PNY Technologies Inc., Parsippany, N.J., will offer the Verto GeForce 4 Ti 4600, Ti 4400, MX 440, and MX 420 for the suggested Nvidia prices. Each card will include, respectively: Star Wars: Starfighter; i/o gaming glasses; the game, "Ballistics"; and 10 700-Mbyte CD-Rs.
Asustek will release six cards: the V8460 GeForce4 Ti 4600; the V8440 GeForce4 Ti 4400; the V8170Pro GeForce4 MX 460; the V8170DDR GeForce4 MX 440; and the V8170SE GeForce4 MX graphics card. Pricing, availability, and bundles have not been announced.
EVGA.com, a smaller player in the card market, said it will release a line of e-GeForce MX cards, complete with a custom cooling solution.
Microstar International Inc. (MSI) announced the G4MX440 graphics accelerator, but declined to offer pricing and availability at press time. Gainward and Leadtek have also signed on to use the chip, according to Nvidia.
The mobile version of the GeForce4 440Go will be offered in the Toshiba Satellite 5005-S507 notebook this month, according to Nvidia. PC OEMs using the chip, which include Compaq, Gateway, Hewlett-Packard, MicronPC.com, and Toshiba, will have systems available within 60 days.
Clarification: Visionteks GeForce4 MX cards, not the Ti cards are scheduled to be on retailers store shelves on Feb. 6, according to Visiontek executives. CompUSA, however, says on its web site that it expects the cards to ship Friday, Feb. 15.