With the launch of the Radeon HD 4870, AMD showed us that they are not only still in the game, but they are back. Their hardware either out performed or was on par with NVIDIA hardware priced significantly higher. As a result, NVIDIA was forced to significantly adjust their prices downward to remain competitive. And, currently, they are competitive in terms of pricing and performance.

But in this business competitive isn't always good enough. AMD came out of the gate swinging for a knock out. And they did a good job of winning a significant amount of mindshare. They built a good product and priced it very aggressively at launch. While our concern is the current state of things, and our recommendation will be for the part that gives our reader the best value, that's not how every graphics card enthusiast sees it. The things companies do (like initially selling their hardware at way too high a price) can significantly affect the position of some enthusiasts.

That might be one reason NVIDIA went down the core 216 path with the GTX 260. A name more like GTX 265 would have been nice, and we already talked about how much we don't need dozens of parts all with slight tweaks and price differences cluttering up the market. But, the core 216 did help make NVIDIA's hardware more competitive (even if it didn't put it over the top). And more competitive is a good thing. Better competition does nothing but benefit the consumer, and we love to see it. If NVIDIA took that step because they want to win back some mindshare then that's fine with us. All we care about is what performs best in a price class, because that is what benefits our readers.

Of course, the core 216 might also have been anticipating the eventual availability of the Radeon 4870 1GB. In which case, the core 216 falls short.

Well, it isn't just that the GTX 260 falls a little short. The fact is that the extra RAM really does make a significant difference in many high quality high resolution situations when playing current games. We didn't expect the gains we see here, and combined with the original stellar performance of the Radeon HD 4870, we have to say that we are impressed.

The Radeon HD 4870 1GB has the same number of GDDR5 chips on board, but the devices on the 1GB model are double the density of the 512MB part.


Yes, these are different parts. The top one is the 512MB version and the bottom one is the 1GB.

The 512MB card we have uses Qimonda GDDR5, while the 1GB model we tested has Hynix devices. As GDDR5 is still pretty new, it is likely that the delay in getting out the 1GB model of the 4870 had to do with delays in getting a high enough supply of high density RAM.

Anyway, we mentioned that the GTX 260 core 216 doesn't quite keep up now that the 4870 has twice as much RAM. Let's take a look ath exactly how short it falls, first in terms of how much performance we gain over the original model and then in absolute terms.

The Test and Performance Imrpovement
Comments Locked

78 Comments

View All Comments

  • carmaster22 - Thursday, September 25, 2008 - link

    How come you don't include the NVIDIA 9800 series cards anymore?

    They were proven to perform just as well and better than the GTX series and there's many people that have them.
  • SiliconDoc - Friday, October 3, 2008 - link

    It's to make it as confusing as possible to we the consumer. No matter WHAT review site I go to - they are absolutely CERTAIN to leave out a couple of KEY cards in the reviews - so that it makes it absolutely near IMPOSSIBLE to make a reasonable decision without endless HOURS of finding, comparing, checking the system stats, of various reviews....
    It's like corporate code - but what really happens is the goobers are thinking, thinking, thinking - and they think so much and so !bleepin! hard, that they come up with some cool points, and interesting facts - but alas - you still don't know what you want to know.
    Whatever, it's so frustrating - I'm sending $100 paypal to the first author that actually satisfies a good lineup in review.
  • Goty - Thursday, September 25, 2008 - link

    The only 9800 series card that could outperform the 4870 was the 9800GX2, and even that fell behind the 4870 and the GTX200s when you started cranking up the resolution and IQ.
  • daniyarm - Thursday, September 25, 2008 - link

    8800gt SLI beat 4750 in several tests even at hi-res. People that own single or sli 8800 or 9800 want to know how the cards compare. What's the point of a review that compares only new gen cards and give absolutely no information for people that want to know if they need to upgrade or not.
  • daniyarm - Thursday, September 25, 2008 - link

    I meant 4870.
  • Patrick Wolf - Thursday, September 25, 2008 - link

    Where do you get your info?

    The 9800 GX2 can be had for under $300 and is also very comparable to the 8800gt SLI. If you have an SLI mainboard and an 8800gt, a very cheap and viable upgrade would be to throw in another 8800gt. If you lack the SLI mainboard, bite on the GX2 and sell your current card.

    Same goes for the 9800 GTX and GTX+ if you're going SLI with them.

    The following graphs speak for themselves. All the above solutions are still a great contender, worthy of inclusion.

    Crysis: http://www.anandtech.com/GalleryImage.aspx?id=3192">http://www.anandtech.com/GalleryImage.aspx?id=3192
    Oblivion: http://www.anandtech.com/GalleryImage.aspx?id=3192">http://www.anandtech.com/GalleryImage.aspx?id=3192
  • Patrick Wolf - Thursday, September 25, 2008 - link

    Oops.
    Oblivion: http://www.anandtech.com/GalleryImage.aspx?id=3189">http://www.anandtech.com/GalleryImage.aspx?id=3189
  • JarredWalton - Thursday, September 25, 2008 - link

    Considering you just linked a recent review that has all of the pertinent information, then throw in the cluttered nature of those charts, and I'm of the opinion that dropping most of the cards and just keeping the more recent stuff makes a lot of sense.

    FYI, outside of a few games (The Witcher 2xAA, ETQW 4xAA, COD4 4xAA) the 9800 GTX+ is very close to the performance of the GTX 260. That's understandable, since they have similar architectures. Here's the theoretical performance overview:

    GTX 260:
    192 SPs at 1242MHz = 715.4 GFLOPS
    Core clock of 576 MHz = 36.9 GT/s texture fillrate
    28 ROPs at 576 MHz = 16.1 GP/s pixel fillrate
    448-bit RAM at 999MHz = 111.9 GB/s

    9800 GTX+:
    128 SPs at 1836MHz = 940 GFLOPS
    Core clock of 738 MHz = 47.2 GT/s texture fillrate
    16 ROPs at 738 MHz = 11.8 GP/s pixel fillrate
    256-bit RAM at 1100MHz = 70.4 GB/s

    So the GTX 260 has substantially more bandwidth (59%) and pixel fillrate (36%), while the 9800 GTX+ has more theoretical GFLOPS (31%) and texture processing power (28%). The GTX 260 ends up faster overall - I'm not sure it ever trails - but there are many games where the difference between the two is only about 10%. The 8800 GT, for the record, is usually 65 to 75% of the performance of GTX 260.
  • Spoelie - Thursday, September 25, 2008 - link

    The 9800GT is a rebadged 8800GT, which sits below any ATi 4 series .
    The 9800GTX+ is competitive with the 4850, but that's a lower price/market segment (target resolution 1600x1200/1680x1050 ?).

    This is an article about cards that run newer games on 1920x1200 and higher comfortably, and well, the 9 series just don't make that cut anymore. Anandtech included all possible contenders except maybe for some SLI configurations.
  • Jovec - Thursday, September 25, 2008 - link

    AT is wrong in what many, if not most, of us want in these reviews. Simply, we have our specific system, and want to know if a single upgraded part or parts (when it comes to new platforms) justifies the purchase price. Obvioulsy impractical, but there is no reason AT can't build and keep 1-2 systems per year, for a 2-3 year moving window, of the most common mid-range builds and include those benchmarks as a reference in all reviews.

    The best example of this is the 8800GTS which I assume many of us still own. We don't care how the 4870 runs on the Intel Core 2 Extreme QX9770 @ 3.20GHz used in the review, we care if the card will boost our FPS enough to warrant purchasing for our computer. It's a different type of comparison than the GPU-limited and CPU-limited tests they currently run, but very useful to the majority of us.

    We can say "My system is very close to the Fall '07 system, so if I buy this new card I'll get a similar performance increase."

Log in

Don't have an account? Sign up now