OVER THE PAST MONTH, we've seen a zillion reviews, big and small, of both Nvidia and ATI's new GPU chippery.

The Geforce GTX280 and 260, as well as Radeon HD4870 and 4850, do provide a major performance jump compared to their predecessors for the first time in nearly two years, and also get closer to each other, diminishing the green-vs-red performance delta.

What about that darn physics thing, especially how the Ageia Physx support in the new Nvidia GTX cards affects the CPU portion of the 3DMark Vantage benchmark, letting the marketing droids repeat the "affected" scores to their benefit? And, the ATI fanboys complaining (cough!) how Nvidia rigs all and sundry?

Since, usually, physics calculations depend heavily on the CPU and are well threaded, we built a platform where there's a minimum skew possible due to Nvidia's on-chip Physx support: the eight core, 4GHz Intel Skulltrail, liquid cooled with Asetek LCLC set.

We also pulled out the old Asus Ageia Physx PCI card, and ran all graphics card combinations both with and without it. What were the cards used? Asus GTX280 TOP 1GB, with a GPU running at a very decent 650MHz, was the highest end entry, while Leadtek GTX260 896MB card played the less expensive slower cousin here. Two reference ATI cards, the HD4870 and HD4850, both with 512MB memory and reference clocks, filled in the picture.

While Leadtek's entry is a decent package with Neverwinter Nights 2 game and accessories, Asus' box focuses on raw performance. It is one of the fastest GTX280 entries around, with high factory overclocked settings of around eight per cent average, and it hasn't failed on any single benchmark we ran.

Windows Vista Ultimate 64 was used to run the 3DMark06 and 3DMark Vantage sets, with the newest drivers we could get hold of, including the Nvidia Physx support and the patched ATI one.

3DMark06 was thrown in more as a reference, as you can see how its scaling compares vs its newer brother. All tests were run at default settings, without SLI or Crossfire parallel GPU attempts this time.

Here are the results:

Now with that extra Ageia card:

Looks interesting? As you can see, once the CPU is powerful enough, it makes little difference if the GPU does physics - even in the 3D Vantage CPU 2 benchmark where there's no 3-D graphics to speak of, it's all "physical".

In this case, both Nvidia and ATI cards still benefit from the addition of the good old and rarely-used Ageia Physx card anyway. Even the percentage gain is similar as you can see.

In the ATI case, even the GPU benchmark portions seem to have gone up a bit - keep in mind there's a perception among some that 3DMark Vantage is quite Nvidia friendly, yet this ATI stuff gains just as much.

Yeah, no one would plonk five thousand quid into a Skulltrail and then fit in an HD4850, but you get the idea: the balanced CPU/GPU configuration still matters most, even in highly synthetic benchmarks like this.

The point is that, even if you have "only" a typical quad-core PC, these ATI cards will still be able to depend on the CPU-bound physics for a while. GPU-bound physics has a future nevertheless, and there's nothing wrong with Nvidia's move to offload it via CUDA.

The point stands on what 3D resources would be sacrificed within that GPU for simultaneous physics processing, though. Or maybe, plug in a spare (future) Geforce GTX240 or such into a spare PCI-E slot just for this? µ