vacancies advertise contact news tip The Vault
EPIC HEXUS COMPETITION: Win a Moto 360 or Omate X smartwatch.! [x]
facebook rss twitter

Review: ATI Mobility Radeon 9600 vs NVIDIA GeForce FX Go5700

by Ryszard Sommefeldt on 29 May 2004, 00:00

Tags: Alienware (NASDAQ:DELL)

Quick Link: HEXUS.net/qax2

Add to My Vault: x

Introduction

A recent review sample of Alienware's Area-51m Extreme laptop afforded me the chance to test out a couple of competing mobile graphics solutions. With its swappable graphics modules, the Area51-m Extreme is close to the perfect test platform for evaluating mobile graphics. With a very capable processor and plenty of memory, removing them as much as possible from the possible sphere of performance influence, it does a decent job of hosting today's shootout.

Alienware were kind enough to send two swappable cores with the sample: ATI's Mobility Radeon 9600 and NVIDIA's GeForce FX Go5700. Evaluating mobile graphics parts is slightly more involved than evaluating desktop hardware, due to imposed form factor limitations. For instance, taking into account power saving schemes available in each driver is prudent, but something never exposed or tested in a desktop accelerator.

Driver issues also conspire to make testing mobile graphics a frustrating experience. Finding up to date WHQL'd drivers that support mobile GPUs is a painful task. More often than not you're reduced to downloading desktop drivers, running a tool that adds PCI ID strings into the driver .inf, allowing your mobile GPU to be detected. But it's common, when using that method of driver creation, that the resulting driver drops its power saving module, leaving you unable to test its effects.

Finally, with certain driver versions come certain application specific optimisations for certain benchmarks, making benchmarking them objectively very difficult. Image quality is the recipient, driver optimisations often lowering quality on a given benchmark for faster scores.

So this article is a quick shootout on a few synthetic tests and benchmarks, along with some UT2003 real world testing, to get a handle on one facet of mobile performance. A proper real world test with our usual game suite will hopefully come at a later date. This article is in the same vein as my recent 6800 Ultra preview, a technical analysis of sorts.

I'll save talk on how the swappable modules on the Area51-m Extreme work, the main review covers everything, including nudey pics of the modules and the GPUs contained within. A quick basic spec table, then onwards to the testing.

Radeon Mobility 9600 GeForce FX Go5700
Core Clock 350MHz 350MHz
Memory Clock 400MHz DDR 400MHz DDR
Process 0.13µ TSMC(75M transistors) 0.13µ IBM (80M transistors)
Memory Bus Width 128-bit 128-bit
GPU M10 NV36
Pixel performance 1400Mpixels/sec 1400Mpixels/sec
Texturing performance 1400Mtexels/sec 1400Mtexels/sec
Memory bandwidth 6.4GB/sec 6.4GB/sec
Traditional render setup 4x1 4x1

It's worth noting that M10 takes its features from RV350, the core that powers the desktop Radeon 9600 Pro. That, crucially, means no Hierarchical-Z, ATI's mechanism for reducing overdraw penalties by discarding hidden pixels early in the pipeline. Since a Z-buffer depth test on a pixel uses memory bandwidth, without a mechanism for 'free' Z test M10/RV350 can suffer a bit when memory bandwidth is at a shortage. This usually comes when the resolution is high or anti-aliasing is on.

To be clear, early depth culling is still present in M10, it just chews up memory bandwidth rather than being free.

M10 has a pair of shader units, the same as NV36. Out of NV36's four pixel pipes, only two have vertex and pixel shader ability.