Personalkly, if I had the money, I'd still probably go for the 4870 over the GTX 280. Preliminary benchmarks I see suggest that AMD has started to fix their problem with AA sapping performance... And I always use AA. Likewise, I just don't like the idea of a single one of my PC's components burning well over 200 watts of power. Likewise, I feel that as games get progressively more shader-heavy, AMD's cards will stand up a bit better to the test of time, as they hold a higher shader-to-texture performance ratio than nVidia's cards, with 20 stream processors for every TMU (800 vs. 40) compared to only 3 SPs per TMU for nVidia. (240 vs. 80) This seemed to show with previous generations of cards, where the Radeon 9700/800 series made the GeForce FX 5800/59xx series progressively more laughable as time went on, and similarly how the Radeon X8xxs holds a better edge over the GeForce 6800 now then when they first came out. Given that I don't buy new hardware but once every few years, that's something that matters to me.
😛
■DirectX 10.1 matters pretty much nothing, just as, in reality, DirectX 10 matters rather little; only a few games support it, and DX 10 modes look not a whole lot improved over the DX 9 modes avaiable for most games today. It'll likely be years before it's required; it was 4 years before DX 9 was required, for instance. (2002, with the release of DX9 itself and the Radeon 9700pro, to 2006, with the release of
Oblivion)
■The type of memory ITSELF is as irrelevant a number as the memory interface. In the end, it simply boils down to how they contribute to the memory bandwidth, as measured in GB/sec. In the end, both cards have well over 100 GB/sec, they just take two different routes to it; the GTX 280 opts to go with a pretty typical 2200MHz effective clock, (the high-end of what's possible with GDDR3) but multiplies it by a 512-bit interface. The 4870 stays with a 256-bit interface, but cranks the effective memory clock to 3600MHz, a speed previous impossible until the advent of GDDR5. Either way, both cards have vastly higher memory bandwidth than their predecessors like the 9800GTX and 3870, which relied on GDDR3 speeds on a 256-bit interface. And this increased bandwidth will show an effect as you crank up the resolution and enable the AA; those settings create an exponential increase in strain on the bandwith. Of course, as the math works out, the GTX 280 does hold a bit of an edge, (140.8 GB/sec to 115.2 GB/sec, a 22% increase) though both are still rather high.
■I'm highly doubtful that 1024MB of memory will really prove to be all that improtant in the near future like that. It's possible to make use of it through heavy modding (such as for
Oblivion) but in most games, 512MB is really all that they'd call for, even at 24" resolutions, it seems.
■The 4870 would OC better, hands-down. The GTX 280 represents clock speeds set at their absolute highest; GDDR3 wasn't really meant to surpass 2.0GHz, and its massive 24x24mm die runs stupidly hot and is strained for power. On the flip side, the 4870's GDDR5 is largely untested, and AMD took an extremely conservative route to clock it at the slowest speed it was designed for; testing by manufacturers have produced chips possible to hit or pass 4.8GHz, and apparently card owners have reported good results with the memory they have. Additionally, the GPU itself is also much smaller, cooler, and seemingly conservatively-clocked. As an additional plus, the card is rated at 150 watts or so, yet has two 6-pin plugs... Technically, that second plus is normally going to be almost entirely unused. (since the card's slot and each plug provide up to ~75 watts apiece)
■I'm not quite certain which is less bug-prone at the moment... Clearly, early on, both are going to have bugs relating to their drivers. Whatever record either company had with their graphics drivers can be thrown out the window at this point. However, as I hear, while some are having compatability issues with the GTX 280, most of the 4870's issues seem to deal with control of the fan, which is a less serious issue, so I MIGHT give this one to the 4870... Though I'd note that I don't really have all the information here.
■In this case, I'd LIKELY judge that the Radeon card is likely to see more improvement through driver updates than the GeForce, since the GTX 280, largely, seems to be an extension of the G80/G92, while AMD appears to have done some more serious overhauling for the RV770, so it's more liable that there are more kinks to iron out first.
Facebook
Twitter
Instagram