<!–#set var="article_header" value="NV18 & NV28:
NVIDIA Chips with AGP-8x Flavor” –>
Introduction
Simply put, the graphics card market is divided into two large segments. One is the retail segment, which provides its customers, the do-it-yourselfers, with a single card packaged in a colorful box along with some nice extras like games. The other is the much more lucrative OEM market for off-the-shelf PCs. This second segment is definitely the more interesting one for card and chipmakers alike. After all, when a company like Dell or Gateway decides to use a certain card in one of their product lines, it means an order of several thousand cards right there. This is exactly the market NVIDIA is targeting with its refreshed GeForce4 chips NV18 and NV28. AGP 8x is the new buzzword in the OEM market and therefore the next “must-have” feature as well. After all, 8x is twice as much as 4x – right? Alright, frown if you will, but this kind of argument may well win over some first-time buyers looking through their home-order brochures.
The MX440-8x now uses the newer and faster BGA memory. Although NVIDIA’s reference board only used a passive heatsink to cool the chip itself, the cardmakers will surely not pass up a chance to visually “tune” their cards by including an impressive looking fan.
It is hardly surprising, then, that only the two best-selling cards in NVIDIA’s line-up, the budget GeForce4 MX440 and the slightly pricier Ti4200, are being treated to the newer and faster bus interface. The new versions of these chips will be called MX440-8x (NV18) and Ti4200-8x (NV28). Rumors about these revisions have been making their way through the net for a while now – ever since ATi’s introduction of the Radeon 9000 PRO, to be exact. Suddenly, those companies that chose ATi’s chips were able to cash in on the advertising hype around AGP 8x, leaving NVIDIA’s partners stuck at “only” AGP 4x. Like it or not, in the world of OEM sales this is a clinching argument.
Now NVIDIA is trying to close the gap – and just in time for the holiday season, too. A bit early? Not really, considering the contracts with OEMs are inked a long time before the actual systems using the card will be sold.
The new Ti4200-8x reference board shows now outward change from its predecessor. Only the name on the chip itself gives the version away.
Beyond the switch to AGP 8x, NVIDIA has also cranked up the clockspeeds of the MX440 -8x. Instead of running at 270MHz (GPU) and 400MHz DDR (memory), the new cards are set to 275/500MHz. The Ti4200-8x remains unchanged at 250/500MHz for the 64MB version. Interestingly, the memory of the reference cards supplied by NVIDIA for testing ran slightly faster, namely at 512MHz (NV18) and 514MHz (NV28). Officially, both cards are being advertised as using a memory clockspeed of 500MHz, though.
Introduction, Continued
NVIDIA GeForce4 Ti4200 |
NVIDIA GeForce4 Ti4200-8x |
NVIDIA GeForce4 MX440 |
NVIDIA GeForce4 MX440-8x |
ATI Radeon 9000 Pro | |
Chip Technology | 256-bit | 256-bit | 256-bit | 256-bit | 256-bit |
Process | 0.15 Micron | 0.15 Micron | 0.15 Micron | 0.15 Micron | 0.15 Micron |
Transistors | 63 Mio | 63 Mio | – | – | – |
Memory Bus | 128-bit DDR | 128-bit DDR | 128-bit DDR | 128-bit DDR | 128-bit DDR |
Memory Bandwidth | 8,2 GB/s | 8,2 GB/s | 8,8 GB/s | 8,8 GB/s | 8,8 GB/s |
AGP Bus | 1x/2x/4x | 1x/2x/4x/8x | 1x/2x/4x | 1x/2x/4x/8x | 1x/2x/4x/8x |
Memory | 64/128MB | 64/128MB | 64MB | 64MB | 64/128MB |
GPU Clock | 250 MHz | 250 MHz | 270 MHz | 275 MHz | 275 MHz |
Memory Clock | (514/500) MHz 64MB (444) MHz 128MB |
(514) MHz Official: 500MHz |
200 (400) MHz | 256 (512) MHz Official: 500MHz |
MHz 275 (550) MHz |
Speicher | SD/BGA 3,3-4ns | SD/BGA 3,3-4ns | SD 4-5ns | BGA 3,6ns | SD |
Vertex Shader | 2 | 2 | – | – | 2 |
Pixel Pipelines | 4 | 4 | 2 | 2 | 4 |
Texture Units Per Pipe | 2 | 2 | 2 | 2 | 1 |
Textures per Texture Unit | 4 | 4 | 4 | 4 | 6 |
Vertex S. Version | 1.1 | 1.1 | – | – | 1.1 |
Pixel S. Version | 1.3 | 1.3 | – | – | 1.4 |
DirectX Generation | 8 | 8 | 7.1 | 7.1 | 8.1 |
FSAA Modi | MultiSampling | MultiSampling | MultiSampling | MultiSampling | SuperSampling |
Memory Optmizations | LMA II | LMA II | LMA II | LMA II | Hyper Z II |
Display Outputs | 2 | 2 | 2 | 2 | 2 |
Chip Internal Ramdacs | 2 x 400 MHz | 2 x 400 MHz | 2 x 400 MHz | 2 x 400 MHz | 2 x 400 MHz |
Chip External Ramdacs | – | – | – | – | – |
Bits per Color Channel | 8 | 8 | 8 | 8 | 8 |
Special | – | – | TV Encoder On-Chip | TV Encoder On-Chip | TV Encoder On-Chip FullStream |
Now weґre sure there are some readers out there mournfully looking at their now “outdated” AGP 4x system. But is the twofold increase in speed really necessary, not to mention – tangible? To be blunt: no. Even future applications are unlikely to require a bus interface that is faster than AGP 4x. 8x is more of a future investment than a current requirement.
AGP 8x uses pins that were previously unused. Visually, there is no difference to the 4x connector.
While AGP 4x with its 266MHz offers a maximum theoretical bandwidth of about 1.1 GB/s, 8x allows for twice as much data to be pushed across the bus, namely 2.1GB/s. Additionally, there are now several new side-band addressing modes as well as an “isochronous operation” mode, which ensures maximum available bandwidth regardless of system load. This could prove a very valuable addition for video streaming and similar applications. Games, however, aren’t nearly as demanding, making this a moot point for gamers.
Introduction, Continued
Current games run just fine with AGP 4x. Even when limited to the slower AGP 2x, most cards don’t take much of a performance hit. There are several reasons for this. Aside from the design of a games 3D engine and its efficiency, the 3D performance of a graphics chip also plays a major role, as it has to be capable of processing all the data it is being fed. While the cardmakers would like us to think that insane amounts of polygons and huge textures need to be transferred via the AGP bus, that isn’t quite in line with reality. After all, a chip like the GeForce4 MXwith its theoretical peak polygon count of 34M triangles/s would never be capable of rendering a scene as complex as that in our example. This is also the reason why we haven’t seen such intricate scenes in any games yet.
Even offloading textures from a card’s on-board memory to system RAM AGP 8x isn’t much help, as its 2.1GB/s are nowhere near the bandwidth modern cards are able to sustain on their own. Take the GeForce4 MX440-8x or Ti4200-8x, for example. Each offers memory bandwidth around 8GB/s. The new Radeon 9700 PRO offers more than twice that – roughly 20GB/s. So we see that even if a game were to use such a large texture set, AGP 8x would be less help than hindrance, creating a bottleneck and throttling performance. Since cards with less than 64MB are a now dying breed, this scenario is unlikely to occur, even with a memory hog like FSAA enabled, and 128MB cards are as good as immune to memory constraints anyway. That kind of capacity will hardly be needed until the next generation of true DirectX 9 games with their extensive effects hits the shelves – and maybe not even then.
Okay, that covers the theory. To find out what AGP 8x offers in the real world, we conducted a few tests with the new NVIDIA cards and pitted them against their predecessors. Since the “old” MX is clocked lower than the new model, we raised its clockspeeds to identical levels to see how much of a performance advantage AGP 8x would have over AGP 4x.
GPU | Core | Memory |
GeForce4 MX420 | 250 | 400 |
GeForce4 MX440 | 270 | 400 |
GeForce4 MX460 | 300 | 550 |
GeForce4 Ti4200 (64MB] | 250 | 500 |
GeForce4 Ti4200 (128MB] | 250 | 444 |
GeForce4 Ti4400 | 275 | 550 |
GeForce4 Ti4600 | 300 | 650 |
GeForce4 Ti4200-8x (64&128MB] | 250 | 500 |
GeForce4 MX440-8x | 275 | 500 |
Official NVIDIA card clockrates in MHz.
All AGP 8x cards are fully downward compatible. If a mainboard “only” has an AGP 4x interface, the cards will recognize this and behave like any normal AGP 4x card. Currently, only VIA’s KT400, SIS’s 648, and NVIDIA’s nForce 2 chipsets offer AGP 8x functionality. Therefore, we ran our tests on a SIS 648 board with a Pentium 4 2.2 GHz.
Test Setup
Hardware | |
Processor | Intel Pentium 4 2,2 GHz (100 MHz) |
Memory | 2 x 256 MB, PC 266, CL2 |
Graphic Cards | NVIDIA GeForce4 Ti4200 NVIDIA GeForce4 Ti4200-8x NVIDIA GeForce4 MX440 NVIDIA GeForce4 MX440-8x ATI Radeon 9000 Pro |
Mainboard | SIS 648 Reference |
Drivers & Software | |
Graphics Driver | NVIDIA – v. 40.71 ATI – v. 02.3 |
DirectX Version | 8.1 |
OS | Windows XP Professional SP1 |
Benchmarks & Settings | |
Aquanox | Retail Version v1.17 |
Max Payne | Retail Version v1.05 Benchmark using ‘Shooting Alex’ |
Unreal Tournament | Demo v1.1 |
2003 Demo | Standard Benchmark |
Jedi Knight II | Benchmark using ‘jk2ffa’ |
3D Mark 2001 SE | Pro Version, Build 330 |
Max Payne
In Max Payne, the differences between the two revisions of the chips are so small as to be called nonexistent. Thanks to its increased clockspeed, the MX440-8x can now position itself much closer to the Radeon 9000 Pro. AGP 8x doesn’t seem to have any kind of influence on the scores.
Aquanox
The MX440-8x seems to benefit ever so slightly from the faster AGP 8x interface in Aquanox. Although they also came equipped with 64MB, this effect did not extend to the Ti4200 boards, however.
Jedi Knight II
Jedi Knight II seems to favor the 8x version of the MX400 over its older sibling as well, but again we didn’t observe this effect on the Ti4200 boards. The new MX’s higher clockspeeds allow it to pull just a bit ahead of the Radeon 9000 Pro in this game.
Unreal Tournament 2003 Demo – DM-Antalus
This test consisted of the Flyby test of the current demo of Unreal Tournament 2003. In addition to the average framerates, our graphs also show the minimum framerates for each card. Surprisingly, the Radeon 9000 Pro slips up quite badly in this test, offering abysmal performance and dropping as low as 10fps at 1024×768 – a full 3x slower than the MX440. For comparison, the Ti4200 pushes 60fps at the same settings, while the new 8x version stays just above the 70fps. At higher resolutions, the 8x Ti’s advantage disappears.
3D Mark 2001 SE b330
Where the 440-8x again pulls just ahead of its older version, both Ti4200 variants perform exactly the same. Even this rather theoretical (synthetic) benchmark AGP 8x doesn’t seem to offer much of an advantage.
Conclusion
As expected, AGP 8x can safely be considered a marketing feature for now. Although it sounds impressive, it offers no real benefit for the current generation of games. A GPU like the GeForce4 MX440 would be unable to handle the amount of data the 8x interface is capable of supplying anyway, and it would be much slower than the on-board memory as well. The most telling sign that the switch to AGP 8x is more of a calculated marketing move than a push for desperately needed hardware resources is the fact that the only two NVIDIA cards that are being “upgraded” to the new interface are also the company’s mainstream cards, the best sellers MX440 and Ti4200, while the heavyweights Ti4400 and Ti4600 remain unchanged.
At least the switch brings with it the additional and more tangible upgrade of a clockspeed increase for the MX440-8x, bringing it into the performance regions of the MX460. This also does a lot to close the performance gap to the Radeon 9000 Pro, even allowing the 8x MX to overtake it in Jedi Knight II. Nonetheless the MX chip is still inferior to the Radeon 9000 Pro where 3D features are concerned, since it is still a DirectX 7 chip. The Radeon 9000 Pro, on the other hand, comes equipped with both vertex and pixel shaders, thanks to its close relation to the Radeon 8500. Strangely, this will have less impact on OEM sales than a nice big number like “8x.” Crazy world? My thoughts exactly!
New cards with NVIDIA GeForce4 Ti4200-8x and MX440-8x
Some pictures of MX440-8x and Ti4200-8x boards from several companies that have already announced products based on these chips:
Leadtek WinFast A180 (GeForce4 MX440-8x).
Leadtek WinFast A180 AGP8x LED.
Leadtek WinFast A280 (GeForce4 Ti4200-8x).
New cards with NVIDIA GeForce4 Ti4200-8x and MX440-8x, Continued
Gainward GeForce4 PowerPack! Pro/600-8X XP “Golden Sample” (GeForce4 MX440-8x).
Sparkle SP7300M4 GeForce4 MX440-8x.
Triplex TX-680 ” (GeForce4 MX440-8x) – Features: 2,8ns Memory, Clockspeed 300 Core / 600 Mem.
Triplex Ti4200-8x.
ASUS V9180 Magic (GeForce4 MX440-8x).
ASUS V9180 TD (GeForce4 MX440-8x).
ASUS V9180 Video Suite (VS) – (GeForce4 MX440-8x). Clockspeed 305 Core / 570 Mem.