GeForce FX 5900 Ultra Preview

The GeForce FX 5900 Ultra 
A Preview And Performance Analysis With NVIDIA's New Killer

By - Dave Altavilla
May 12, 2003

HotHardware.com Test Setup Pentium 4, Canterwood and 4 killer Graphics Cards

Before we jump into the benchmarks, please take a look at our system test setup and specs below.  It is important to note that we have only tested an ATi Radeon 9800 Pro 128MB card in the following benchmarks.  ATi's latest 256MB DDR2 based R9800 board did not make it to our lab in time for testing in this article.  As such, we will be coming back with a side by side benchmark comparison of the 256MB Radeon 9800 Pro card, later this week, for a proper apples to apples comparison.  With that said, let's get it on!

Pentium 4 Processors at 3GHz  - 800MHz System Bus
Motherboard and RAM Config
Abit IC7-G "Canterwood" Motherboard
512MB of Kingston HyperX PC3500 CAS 2 RAM
CAS Timings were 2-2-2-5
NVIDIA GeForce FX 5900 Ultra  256MB
NVIDIA GeForce FX 5800 Ultra  128MB
ATi Radeon 9800 Pro 128MB
ATi Radeon 9700 Pro 128MB

Other Hardware and Software:
Seagate Barracuda V SATA 120GB HD
Windows XP Professional w/ SP1
NVIDIA Detonator FX Drivers Version 44.03
ATi Catalyst 3.2 Drivers
Intel Release Chipset Driver  v5.00.1012
Intel Applications Accelerator RAID Edition v3.0.0.229

3DMark 2001 SE and 3DMark 2003 The Angry Onion And The FutureMark  Tests

Interestingly enough, 3DMark 2001 at 1024X768 res, gives a slight edge to the Radeon 9800 Pro, until your enable AA.  Then the NV35 clearly takes the lead.  This could point to a bit of driver immaturity for the GFFX 5900 Ultra but that is to be expected, since we're about a month away from official release of the card.

At high res and with 4X AA enabled, the GeForce FX 5900 Ultra really begins to distance itself from the Radeon 9800 Pro, by almost 1000 points.  In terms of AA performance, the 5800 Ultra is clearly bringing up the rear here.

3DMark 2003 gives a clean sweep to the NVIDIA cards here and the GeForce FX 5900 Ultra takes the lead handily by over 400 marks.  If NVIDIA was unhappy with the fact that they had to optimize drivers for a benchmark that, in their opinion, was not representative of real-world game rendering, then they certainly re-prioritized their efforts.  What is interesting is that the NV30 and NV35 are within a couple hundred points of each other.

Quake 3 Arena and Serious Sam SE