The Inquirer-Home

ATI's RV670 can beat R600

When 256-bit GPU is more powerful than 512-bit one
Mon Oct 01 2007, 22:12

WHAT HAPPENS IF your previous generation product is nine-month late overheating monster? In all cases, the solution to the problem is to make something radically new and hope for the best.

This will not be the case with next-generation, alleged gap-filler refresh from AMD. The engineering team has gone on a limb to produce a succession of products starting with RV670 (followed by R680 and R700) that will really change the landscape of 3D as we know it.

You might feel that previous statement is a bit overhyped, but we're finally starting to see the change in 3D space. After seeing abysmal DX10 performance of Radeon HD 2600XT and GeForce 8600GTS, we were in danger of pronouncing GeForce 8800GTS as the entry-level part for gaming in applications that utilize DirectX 10 API. With Crysis, Hellgate of tomorrow, and Call of Juarez, Lost Planet and World in Conflict of today, looking at 10-15 fps is really not an option if PC gaming wants to survive.

As you can see, RV670 sports quite sporty image compared to bulky R600 of yesteryear (picture courtesy of NV6800, member of Coolaler Forum)

We have known about RV670 developments for quite some time now, and seeing that this 55nm GPU will feature single-slot cooling at standard and dual-slot cooling at insane frequencies, you can start thinking what will happen in the world of water-cooling board designs. We only hope that AMD can nurture the fragile, low-volume system of water-cool manufacturers.

AMD put R600 on a 55nm diet that shred more than 100W TDP - now the RV670 in its fastest form will eat around 120W less than fastest R600-based boards. As you might have guessed it, RV670 should feature all 320 superscalar shader units (64 fatties, 256 fast ones) present on R600, only two real changes being an updated API support - Shader Model 4.0 is now 4.1 (checkbox feature, don’t get swayed by this number), and putting a more mainstream friendly 256-bit memory controller (cost of PCB for R600 was just horrendous). This deal enabled ATI to put the RV670 on a PCB very similar to X1950, changing the whole cost game. RV670 is expected with a retail price between 229 and 279 dollars, but we would not be surprised if a low-clocked part would retail for 199 Dollars/Euro.

You can expect that partners will spot DirectX 10.1 stickers all around the package, but this GPU has a different purpose altogether. Real purpose is to offer more than decent performance in playing the latest titles with all the details plugged to the max. If one card is not enough, AMD will support two, three and four graphics cards in a single system. The load of four RV670 boards will be less than two 2900XT boards, so the difference is striking.

When it comes to the name, it is still too early to tell, but Radeon HD 2950 GT/Pro/XT monikers are a definite possibility. This decision will be done by AMD marketers, but not all is decided, as some sources from Far East are inclined to say.

This will probably be the first time in history that the part with narrower memory controller (256-bit) will beat wider part (512-bit), at least in some tests. We were told to pay special attention to titles that are heavy on longer shader code. Expect that Radeon HD 2900XT will speed pass through RV670 in all tests that require heavy bandwidth load, such as FullHD (1920x1200) or XHD (2560x1600) variants. Bear in mind that AMD will push for multi-GPU configurations for these users, and if you opt for a single-slot board, it should be no problem to put 2-3-4 boards in the same system.

Pages of magazines and web-sites dedicated to ICT industry will sport three huge wars in next sixn months: G96 versus RV670 and G98 versus RV635 (55nm die-shrink of 2600XT) in November, followed by real G92 versus R680 in Q1 2008. Life can't be more fun than in 3D world. Compared to fight between AMD and Nvidia, McLaren vs. Ferrari looks like child's game. µ


Share this:

blog comments powered by Disqus
Subscribe to INQ newsletters

Sign up for INQbot – a weekly roundup of the best from the INQ

INQ Poll

Heartbleed bug discovered in OpenSSL

Have you reacted to Heartbleed?