vacancies advertise contact news tip The Vault
facebook rss twitter

Review: ATI Radeon 9800 Pro 256MB

by Ryszard Sommefeldt on 7 June 2003, 00:00 4.0

Tags: ATi Technologies (NYSE:AMD)

Quick Link:

Add to My Vault: x


We've been waiting for this time since consumer graphics finally took off with the Voodoo. Since then, as PC's finally had the pixel pushing horsepower to render 3D at speed, games have been gaining in visual quality and complexity up to a point where larger framebuffers haven't just been a gimmick. They've been a necessity. And while a larger than 128MB framebuffer has been common on high end 'professional' cards for a long time, in the consumer market it has never been the case.

Until now that is. Recently we've seen the launch of two expensive consumer graphics behemoths, one each from ATI and NVIDIA, sporting the magical quarter GB of onboard card memory and it's the ATI entrant I'll be looking at today.

We've talked about 9800 Pro and R350 in previous articles so I wont cover old ground, other than to say that it's a 'traditional' 8 pipeline, 1 TMU and shader unit per pipeline architecture.

So while we've seen the Radeon 9800 Pro in 128MB guise before, and we know what performance is like, is there a real need for a 256MB version? Also, is there any difference between the two, other than the extra memory?

For starters, the 256MB version debuts at slightly higher core clock, 400MHz to be precise, compared to the 380MHz on a 128MB 9800 Pro. Memory clock is slightly up too, at 700MHz DDR rather than 680MHz DDR on the smaller board.

Second of all, ATI have chosen to debut DDR-II memory modules for the first time on one of their solutions. Functionally, as far as memory goes, it's no different to DDR-I in terms of what we care about on a graphics accelerator.

So in terms of raw numbers, what do we get.

Radeon 9800 Pro 256MBRadeon 9800 Pro 128MB
R350 R350
Pixel pipelines 8 8
Pixel shader units 8 8
Memory bus width 256-bit/32-byte 256-bit/32-byte
Texturing units 1 1
Core clock 400MHz 380MHz
Memory clock 700MHz DDR 680MHz DDR
Pixel fillrate 3200 Mpixels/sec 3040 Mpixels/sec
Texture fillrate 3200 Mtexels/sec 3040 Mtexels/sec
Memory bandwidth ~22.40GB/sec ~21.76GB/sec

So, a whopping 5% increase in fillrate and texturing performance over the 128MB version and a monsterous 3% increase in memory bandwidth.

The only real difference then is the 256MB framebuffer. Even though the R350 was designed with the ability to address this amount of memory from day one, I can still invisage a latency penalty on the memory controller of some sort from the increased size, therefore a clock increase to smooth that out wouldn't hurt.

So we're left with a double sized beast of a card to contend with and try and exploit.

From the outset after recieving the test sample, it became apparent that finding a win case for the larger framebuffer was going to be hard. We may have to rely a little on synthetic benchmarks to show you where you'd need it, but hopefully the performance drop graphs will help.

Let's take a look at the test setup, before proceeding to our benchmarks.

I apologise for the lack of pictures, my new digicam arrived a couple of days after it left my hands. Despite the extra memory size, memory density didn't make the board overly large. Molex power was of course required, and it used the same heatsink as on a Built By ATI 9800 Pro 128MB. It was a fairly noisy fan, but nothing too bad.

Check out PC Pro's pictures in the next issue of their print mag if you are desperate, I sent the board to Gareth when I was done with it.