LinkBack URL
About LinkBacks
I've recently been working on an article that discusses the recent problems, since the SKUs launched, with texture filtering in their drivers. I won't give the game away since we'll likely publish very soon, but I do want to talk about a recent failure of ours to bring the issue up sooner, and I think .blogz is a good place to do so.
My remit, as Group Technical Editor, especially since I'm the one responsible for our analysis of new GPU products, is to evaluate not only the performance claims of a new graphics product, but also its other facets.
Of those, image quality is paramount these days, especially with the advent of hardware with so much basic texturing ability that today's best graphics accelerator has, literally, 200 times the texture sampling ability of the first Voodoo Graphics boards.
And it's image quality that I've not focussed enough on recently. I wrongly assumed that since G70 shares the same texture filter hardware that the NV4-series has, that its texture filtering IQ would be the same. I'm happy (enough) with NV40's texture filtering IQ, so I passed on that assumption to the new hardware.
The truth is it's not the same, and I wasn't prudent enough in checking, on launch, that that was the case. 7800 GTX launched at the end of June, over 2 months ago, and it's taken me that long to get round to putting down on paper [*] what's going on, mulling it over and presenting a theory to the right people, and then subsequently writing it all down for the web.
It's not often I feel like I've not done enough research into what I'm writing about, but it feels so here. Subsequently it almost feels like I've let you guys down, but I still stand by my overall conclusions about the two 7800 SKUs in terms of recommendations to buy. And that's key for us here, really.
Anyway, the article should be up soon, and thankfully it's an overall positive piece for NVIDIA, since they've been active in coming up with a fix and honest with me about what was going on, confirming my testing and analysis, so that I could write it all down. That's to be commended.
Of course, they're culpable in letting it through Q&A, or indeed consciously engineering the 'bug' into the driver, but at least they've offered a solution that seems to work.
Hopefully the article and NVIDIA's driver that I tested will be available some time soon, and sorry for missing out the proper investigation into IQ at the time of launch.
[*] I'm not sure I've ever talked about this publically, but I'm a big fan of actually using the paper and pens approach to technical writing. Often while I'm writing, when I get to the end of a significant section, or concept I'm trying to convey, I'll print out what I've typed up so far, make myself a brew and sit down away from the computer with the printout and a pen and sit and ponder it.
In the filtering article's case, it happened that I sat down with a notebook and a pen and scribbled out some notes and diagrams regarding texture mapping, MIP chains and anisotropic filtering, since they're all very visual concepts (and my skills with mspaint.exe are fairly ****e). It really helped while working through it all, to get it down on paper and sit away from the whinge of my PC and ponder it in peace and quiet.
I'd urge the other folks banging out copy for us to do the same from time to time, especially when stuck in a rut with writer's block.
It's also very advantageous, and this is something Bob and I have chatted about in the past and he's spot on, in spotting spelling errors and other naughty typos, when they're on a piece of paper, rather than on the screen.
PD and I often joke that I don't spell check my copy, and in the "using software to do it" sense it's correct; I honestly don't. But I definitely print my stuff out and try and spot it there whenever possible. It's not always doable, due to time pressures, but every little helps, eh?
