facebook rss twitter

Review: GAINWARD FX 5200 POWERPACK PRO/660 TV/DVI PCI

by Tarinder Sandhu on 1 October 2003, 00:00

Tags: Gainward FX 5200 Powerpack PRO/660 Tv/dvi PCI, Gainward

Quick Link: HEXUS.net/qatx

Add to My Vault: x

Setup and notes

Basic setup

  • Intel Pentium 4 3.2GHz 800FSB CPU
  • EPoX 4PDA2+ i865PE Springdale motherboard
  • 2 x 256MB Corsair XMS3500 memory run at 2-6-2-2 @ DDR400
  • Intel reference cooler
  • IBM 41.5GB Hard drive
  • Pioneer 105 DVD/RW
  • 420w Samcheer PSU
  • Samsung 181T 18.1" TFT

Video Cards used

  • Gainward GeForce FX 5200 PowerPack Pro/660 TV/DVI 64MB
  • Tyan Tachyon G9600 Pro 128MB (400/600)
  • NVIDIA reference GeForce FX 5600 Ultra Rev. 2 128MB

Software

  • Windows XP Professional Build 2600.xpclient.010817-1148
  • Intel 5.00.1012 chipset drivers
  • ATi CATALYST 3.5 drivers (6360) for the 9600 Pro
  • NVIDIA Detonator 45.23 for the Gainward FX 5200 and FX 5600 Ultra Rev. 2
  • 3DMark 2001SE v330
  • Quake III v1.30
  • UT2003 Demo build 2206
  • Comanche 4 benchmark

Notes

The first question that required an answer was whether the Gainward FX 5200 pixel-pushing muscle could outstrip the meagre 133MB/s bandwidth constraint imposed on the PCI bus it would run off. Or, putting it another way, does the PCI interface limit the card's potential performance. If a card is powerful enough, it will continue to scale in benchmarks that don't totally swamp its GPU. A case in point is, say, a Radeon 9800 Pro. Strap that card into a 1.6GHz P4, running with DDR200 memory and run a Quake III HQ benchmark at 1024x768x32 (without imposing a hit on the GPU via AA and AF). Try that test again with a 3.2GHz P4 and you'll find a linear increase in performance. The card is just being fed enough geometry to keep it ticking over. We'd hazard that it continues to scale up to 4GHz. Only at these sky-high speeds would one expect the card to become the bottleneck.

In the case of the Gainward FX 5200, A 3.2GHz P4 was underclocked to 1.6GHz and run with dual-channel DDR200 memory; effectively half its normal speed. The aforementioned Quake III timedemo was run and reported a benchmark figure of 78.1 FPS (1024x768x32). The CPU was then set back to its normal running speed and memory was at the standard DDR400 speed. The benchmark then returned a result of 78.3FPS - which is a statistically insignificant rise in performance, given that the base system now had twice the geometry-pushing ability. Could the lack of improvement be blamed on the impoverished PCI bus bandwidth, or was it case of card limitations on a 1.6GHz P4 ?. Overclocking the card, though, produced statistically significant increases in performance, even at 1.6GHz. That infers that the limiting factor was not the bus bandwidth, but the stunted architecture of the card. In other words, an AGP version of the card would fare too much better, except in the management of textures, perhaps. Here we have a modern 3D accelerator that isn't limited by the archaic 133MB/s PCI bus. A perfect candidate for PCI running or an extremely limited design .... you decide.

2D quality was, subjectively speaking, pretty good, and that's at 1280x1024x32 and on a Samsung 181T TFT run through the HD15 connection. DVI usage, too, was clean, crisp and more than palatable. This FX 5200 features a 400MHz RAMDAC. What that means to you and me is a maximum resolution of 2048x1536x32 @ 75Hz, which is more than good enough for most home users. The card's passive heatsink is appreciated, but we would have also appreciated a half-height card, such that it would fit into many a smaller system, systems which often do without an AGP port. The card was benchmarked with 64MB of system memory allotted for texture usage.

To give the readers an idea of where the Gainward FX 5200 PCI-based card stands in relation to cards a notch or two up in ATI and NVIDIA's ranges, a standard 9600 Pro (Full DX9 compatibility, 8x AGP, 0.13-micron, 128-bit bus, 400/650 clocks, 4/1 rendering system, Z + colour compression) and NVIDIA's very own FX 5600 Ultra (Full DX9, 8x AGP, 0.13-micron, 128-bit bus, 400/800 clocks, 4/1 rendering, Z + colour compression). Faster cards, sure, but they'll provide a decent yardstick to measure this card by.

Simple overclocking tests showed that it was stable at 295MHz core and 422MHz memory - a reasonable rise from stock.