Introduction
You can't fail to notice that it's fall refresh time for the graphics card industry. With new products seemingly arriving at relentless pace, it can be hard to sort everything into performance order. Luckily that's what sites like HEXUS are for, to keep you on top of things, keeping that performance order list in check.With NVIDIA recently releasing its own new range topping card, GeForceFX 5950 Ultra, it makes sense that ATI have also done the same. Indeed, a few weeks prior to 5950U, ATI already had their new baby sampled and in the hands of people like me. With cards just about to hit retail in big numbers here in the UK, and cards available elsewhere worldwide and in OEM form for a little while, it's about time we looked at Radeon 9800XT.
I sneakily dropped performance numbers from 9800XT into my 5950U article, but we hadn't discretely reviewed it prior to that article going live. Naughty us, but time to put that straight today.
So with 9800XT sat atop ATI's performance pile, their most expensive graphics card by a margin, it makes sense that 9800 Pro, the previous champ, gets pushed down the order a little. The easiest way to see if that's true is to glance at the specs. Here's 9800XT sandwiched in between its NVIDIA nemesis and 9800 Pro 256MB in some 3-way table action.
| GeForce FX 5950 Ultra | Radeon 9800XT | Radeon 9800 Pro 256MB | |
| GPU Name | NV38 | R360 | R350 |
| Transistor count | Unknown | Unknown | Unknown |
| Manufacturing process | 130 nanometre | 150 nanometre | 150 nanometre |
| Pixel pipelines | 8 (**4) | 8 | 8 |
| Pixel shader units | 1 per pipe | 1 per pipe | 1 per pipe |
| Memory bus width | 256-bit/32-byte | 256-bit/32-byte | 256-bit/32-byte |
| Texturing units | 1 (**2) per pipe | 1 per pipe | 1 per pipe |
| Core clock | 475MHz | 412MHz | 380MHz |
| Memory clock | 950MHz DDR | 730MHz DDR | 700MHz DDR |
| Pixel fillrate | 3800 Mpixels/sec (**1900Mpixels/sec) | 3296 Mpixels/sec | 3040 Mpixels/sec |
| Texture fillrate | 3800 Mtexels/sec | 3296 Mtexels/sec | 3040 Mtexels/sec |
| Memory bandwidth | ~30.40GB/sec | ~23.36GB/sec | ~22.40GB/sec |
In terms of performance increase over 9800 Pro, we can see a rough 8% increase in core clock with all that entails (pixel/texel/shader performance increases) and a less than enthusiastic ~7% increase in memory clock. With 9800 Pro 256MB able to sample 22.4GB/sec of data from the memory bus, per second, a huge increase wasn't exactly demanded by 9800XT. It might seem that NVIDIA have bought all the fast DDR on the planet for their cards, leaving ATI with none, but with 9800 Pro 256MB managing just fine with what it has, there's no realistic need to slap 9800 with ~1GHz DDR. Indeed, it would only serve to embarass NVIDIA in memory bandwidth limited scenarios and in marketing terms ATI only seek to keep a slim performance margin, leaving headroom for future products instead.
So while we know that the top Radeon has less impressive numbers compared to the to GeForceFX, we know that all GPU's aren't created equal and that more MHz doesn't automatically mean more performance. 9800XT claims to be the fastest graphics card on the planet, ATI don't say things like that without a good foundation.
We've covered the implementation details of ATI's high end Radeon's before, with things like SMARTSHADER, the pixel and vertex shader engine that gives 9800 the bulk of its class leading DX9 support and performance, and SMOOTHVISION, the anti aliasing and texture filtering unit that makes things look good. If you want more talk on stuff like that, Tarinder's look at the Connect 3D 9800 Pro, here, is as good a place as any to look.
So where does the new speed come from? We know that 9800XT isn't just a faster clocked R350 that powers 9800 Pro, it's a new revision called R360. What do we get with R360 that wasn't there before?
