NVIDIA GeForce FX Benchmarks
During their launch event, NVIDIA showed the first game benchmark numbers for the GeForce FX compared with the GeForce 4 Ti. Just remember that this is NVIDIA's data, and should be treated as much as a marketing pitch as a guide to future performance. It makes you wonder, though, what possibilities the FX holds.
| NVIDIA GeForce4 Ti4600 | NVIDIA GeForce FX | |
|---|---|---|
| Doom III
1280x1024-32 High Quality |
20,9 FPS | 49,8 FPS |
| 3D Mark 2001
Nature 1280x1024-32 4xFSAA + 8x Aniso |
16,2 FPS | 40,6 FPS |
| UT 2003
Asbestos HQ 1280x1024-32 4xFSAA + 8x Aniso |
39,3 FPS | 108,4 FPS |
The benchmarks were made on a 3 GHz Pentium 4 with 512MB RAM running Windows XP. The following benchmarks show the OpenGL Workstation performance tested with ViewPerf5 in 1280x1024-32 in CDRS03.
| Onyx
Infinite Reality |
NVIDIA
Quaddro 900XGL |
NVIDIA
GeForce FX |
|
|---|---|---|---|
| ViewPerf 5
1280x1024-32 CDRS03 |
168 | 690 | 1830 |
These benchmarks were run under Windows 2000 SP3 on a P4 system with 2GB RAM.
NVIDIA GeForce FX Memory Bandwidth
There was a lot of confusion about the memory bandwidth of the DDR2 Modules NVIDIA is using on their new GeForce FX card and how to calculate it. Most information available on DDR2 explains that DDR2 has twice the data-bandwidth than DDR - made possible by a four-bit prefetch instead of two-bit, used with DDR.

Let's begin with the cards we already know:
GeForce 4 Ti:
16 Bytes * 325 MHz * 2 = 10.4 GB/s
The buswidth of the memory is 128Bit, which means that 16Bytes of data can be transfered by clock. Since the DDR memory is using a prefetch of two, the data rate is doubled. Let's take a look on the Radeon 9700 PRO, also using DDR memory:
32Bytes * 310 * 2 = 19.8 GB/s
You have to take 32Bytes here, since the memory bus on the card is 256 bit wide. Now let's face the GeForce FX. The card is using DDR2 memory, which means it's using a prefetch of four and doubles the amount of data transferred again - in theory. If a card is running with 1GHz DDR2 datarate, the modules can be run at a quarter of that: moderate 250MHz. That's what people mean when they say that DDR2 is a cheap solution with a lot headroom. You can also read that in this Jedec whitepaper on page 6.
But NVIDIA is using Samsung DDR2 modules with a dram cell frequency of 500MHz - only half the data frequency. This means that the DDR2 memory on GeForce FX behaves just like DDR memory with just higher clock frequencies.
So here we go:
16 Bytes * 500 MHz * 2 = 16 GB/s
This goes along with a Samsung whitepaper on the DDR2 modules NVIDIA is using for the GeForce FX. It says that one module (32Bit) has a single Bandwith of 4GB/s. This means 16GB/s for 128Bit. GeForce FX is using two banks with four modules each - if you wondered after counting the number of chips on the card.