The main reason most of us are here is because we like eeking ostensibly "free" performance out of our hardware and there are very few things better at it that cut-down variants of powerful hardware. Lower clocked CPUs, stream-processor reduced graphics cards and RAM from an alternative parts bin are all things that can be as good as their bigger brother for a fraction of the price.With the GTX465 it should have been easy to get it up to around the level of the big daddy GTX480, but by utilising this very standard reference design they've taken a card that was too hot with a huge cooler, and reduced the heat it puts out AND the heat it can disperse.
Somehow we've ended up with a card that has all the faults of the GX480, but with none of the eye-watering performance that made living with it more bearable.
We've always been very aware than nVidia don't consider their cards to be solely useful as something to crunch polygons for the latest game, but rather an alternative processing solution. It's a very viable business model on paper. Unfortunately it just doesn't seem to bear fruit in the real world.
Starting with the good, Physx is, partially by virtue of its maturity, a really excellent bit of technology. The difference it makes is both visible and stark. Crysis Warhead's minimum frame-rates attest to how much the dumping of some load off of the CPU can affect your experience. Anyone who's watched the slide-show that is the second CPU test in Vantage will be astounded by how smooth it is when the GPU is helping crunch the numbers. Sales on nVidia cards would go up if only they'd allow people to run them as a Physx card alongside a ATI card. However I can appreciate why they jealously guard their technologies.
Much less successful are the CUDA and Tessellation spearheads. In reverse order, the Tessellation doesn't provide the huge visual improvement we were hoping. Of course this is still a very early technological leap and so perhaps in time it will be as amazing as Direct3D was all those years ago. However we don't review on theoretical possibilities. Even if we did the results can't be denied and the HD5850, which by no means has placed a good deal of its eggs in that basket, performed just as admirably.
CUDA should give massive boosts to anything that can hand off some of the calculations to the GPU. However we just didn't see that. Maybe if you've got a very slow processor you will see the benefits, but the realities are that not many people are going to attempt huge video or rendering tasks on a 2GHz single core. Considering the small file size we tested with, 30 seconds slower is a hell of a lot.
The GTX480 shows immense ability that is only some heat-taming away from greatness. By bringing the price down to the middle of the market, everything else has been brought down too. If this came with the GTX480 heatsink then some hefty overclocking with the excellent ASUS Voltage Tweak BIOS could make the GTX465 a serious contender. As it is the heatsink we have barely copes with the card at stock, and the performance at stock is absolutely nothing to write home about.
Pricing isn't helping. Currently at retail for £240 this is the equivalent of a HD5850, hence todays comparitive tests. Clearly it's just not up to that level of performance. If you want a budget card your money is much better spent on a HD5770, or even two for this money. If you want to overclock, go HD5850. If you need the extra features such as CUDA, you really would be better off just investing in a better CPU.
With decent cooling and some hardware maturity the Fermi breed of cards definitely has a hell of a lot going for it. But right now heat is the major issue. As a card that's all heat and no performance the GTX465 is impossible to recommend. If you absolutely have to have nVidia in your machine we'd recommend at least holding off until the GTX470 has come through our labs.
Thanks to ASUS for providing our GTX465 for review. Discuss in our forums