nVIDIA's New GeForce4 Ti and GeForce4 MX
The NV25 and NV17 Debut

By Dave Altavilla
2/6/02

 
A mere 4 months ago, we turned HotHardware readers on to NVIDIA's "fall refresh" product, the GeForce3 Titanium.  A new set of drivers, higher clocks speeds and all new branding of the GeForce3 product, was unleashed upon the casual and hardcore gaming markets, as well as the professional/desktop graphics space.  NVIDIA, known for blisteringly short product development cycles, rang the bell for round two of the GeForce3 punishment being dealt to their closes competitor, ATi.  However this time, ATi was ready and answered the bell with their Radeon 8500 product line that finally gave the high end enthusiast segment a real choice for high end 3D Graphics for the PC.  However, as we've seen so many times in the past, you can almost set your watch to NVIDIA's product development cycles and we knew true next generation technology would not be far around the corner for the company, now known affectionately as the "Chipzilla" of the PC Graphics Industry.

Like the proverbial 1-2 punch, NVIDIA comes out swinging again with two new graphics chips aimed at the value and performance segments of the market. Today we're taking the wraps off the GeForce4 Ti and GeForce4 MX GPUs from NVIDIA. 

Specifications and Features of the NVIDIA GeForce4 Ti and MX
New boards with two totally new GPUs
 


NVIDIA GeForce4 Ti 4600


Graphics Engine
NVIDIA GeForce4 Ti GPU @ 300MHz
 
Video Memory
128MB DDR Video Memory @ 650MHz DDR
 
Bus standard
AGP 4X / 2X / 1X
 
Connector
Dual VGA or DVI (with converter)
 
TV-out Connector (optional)
One S-VHS mini-DIN

NVIDIA Personal Cinema Ready (optional)

nView
Display technology provides the ultimate multiple display flexibility and control
 
Accuview AA
Delivers high resolution, high frame rate, full scene antialiasing
 
Lightspeed Memory Architecture TM (LMA) II
128- bit DDR interface, radically improves memory efficiency
"Quad Cache" - Pixel, Texture, Primitive and Vertex - On chip cache
Lossless 4:1 Compression of Z-Data
Second Generation Occlusion Culling

nfiniteFX II Engine
Drives complex geometry and animation w/ Dual Vertex Shaders and Faster Pixel Shaders

4.8Gsamples/sec


NVIDIA GeForce4 MX 460


Graphics Engine
NVIDIA GeForce4 MX GPU @ 300MHz
 
Video Memory
64MB DDR Video Memory @ 550MHz DDR
 
Bus standard
AGP 4X / 2X / 1X
 
Connector
One VGA and one DVI
 
TV-out Connector (optional)
One S-VHS mini-DIN

NVIDIA Personal Cinema Ready (optional)

nView
Display technology provides the ultimate multiple display flexibility and control
 
Accuview AA
Delivers high resolution, high frame rate, full scene antialiasing
 
Lightspeed Memory Architecture TM (LMA) II
128- bit DDR interface, radically improves memory efficiency
"Quad Cache" - Pixel, Texture, Primitive and Vertex - On chip cache
Lossless 4:1 Compression of Z-Data
Second Generation Occlusion Culling

Integrated TV encoder at 1024x768 resolution

Integrated full hardware MPEG- 2 decoder
Processes full frame rate, full screen MPEG- 2 video

Motion compensation and IDCT
Allows for DVD decoding with minimum CPU usage

HDTV ready


 

New technology enabling new features and performance levels:
Using TSMC's .15 micron wafer fab process, NVIDIA is able to drive core clock speeds on both chips to 300MHz.  In addition, next generation DDR SDRAM is also running at speed bins of 650MHz for the GeForce4 Ti and 550MHz for the GeForce4 MX.  Upon closer inspection of the board shots above, you'll notice that the DDR SDRAM components are now in tiny little BGA (ball grid array) packaging.  Not only are these higher density chips but these BGA packages typically have better noise, power and grounding characteristics, which allows higher clock speeds.  In addition, it seems the day has come that 128MB of Texture Memory on board is going to be a standard feature in high end product offerings.  At the new level of performance NVIDIA is claiming the GeForce4 is capable of, users will be able to run 1600X1200 resolutions at a full 4X Anti-Aliasing setting and at good frame rates.  With only 64MB of Texture Memory, 4X AA sampling wouldn't fit at 1600X1200 resolutions.  Now, with 128MB of DDR RAM on board, new games with larger more detailed textures should run smooth and clean with 4X AA, at 1600X1200.

An Electrical Engineering friend of mine at Lucent is always talking about "speeds and feeds".  That's what the semiconductor game is all about for the most part. In laymen's terms, what he is talking about is that sure, you can have a CPU or GPU that is clocked at break-neck core speeds.  That will always help overall performance.  However, if you can't fill the I/O pipes to and from that processor and fully utilize those extra clock cycles, you can think of your part as a Nitrous Oxide driven Drag Racer but stuck in Neutral.  The engineers at NVIDIA are also keenly aware of this and as a result, have incorporated their all new "Quad Cache" architecture on both the GF4 Ti and MX products.  This additional on chip cache will "feed" the high speed rendering pipelines of the GeForce4 Ti and MX, with faster on die memory, versus off chip frame buffer memory,  increasing overall throughput.  What's probably more impressive is the fact that NVIDIA was able to pull this new technology out of their collective hats, in 100 days from initial "tape out" to full run production volumes.

   

    

Lastly, NVIDIA is also incorporating new and sleeker looking GPU cooling techniques, with a new heat sink design that has a turbine type fan that blows cool air not only across the sink but the DDR SDRAM chips as well.  The current reference design does not have heat sinks on the RAM chips themselves but we are fairly certain there will be a few OEMs that bring boards to market with these installed.  Another fairly obvious variance is that NVIDIA is clearly pushing flat panel displays, offering Dual DVI Connectors (convertible to VGA with an included dongle) on the GF4 Ti and one of each VGA and DVI connector, on the GF4 MX.

 

 

A closer look inside the GeForce4 Ti and MX