backtop


Print 77 comment(s) - last by Jkm3141.. on Oct 29 at 2:07 AM


AMD's RV670 processor runs on a single-slot cooler design, though working samples unveiled last week used a two-slot solution instead.  (Source: Coolaler.com)
Next-generation GPUs are the fastest things on the planet -- if they were released a year ago

Traditionally the Fall graphics refresh has been the battle of the titans -- ATI and NVIDIA both would debut behemoth video cards in an attempt to snag the headlines from one another.

Much of that changed when AMD acquired ATI last year.  Not only did ATI miss the Radeon HD 2900 launch window by almost six months, but NVIDIA's high-end GeForce 8800 became the undisputed ultra-high-end GPU as well.

This Fall, we will not get an ultra-high-end replacement from AMD or NVIDIA. Instead, November will be a clash of the sub-titans.  NVIDIA's mid-range G92 will go head-to-head with ATI's RV670.

ATI's RV670 has been called many things in the past. It was originally a 65 nanometer die-shrink of the R600 class GPU; then a 55 nanometer shrink. Taiwan Semiconductor Manufacturing Company, Asia's largest core-logic foundry, confirmed AMD would go with a 55nm R600 shrink in a memo forwarded to DailyTech earlier this year.

When TSMC debuted its 55nm process earlier this Spring, the company claimed "significant die cost savings from 65 nm, while offering the same speed and 10 to 20 percent lower power consumption."  Since R600 was manufactured originally on a 80nm node, thermal improvements should be fairly dramatic on RV670.

Last week at the World Cyber Games, Sapphire demonstrated a working RV670 using a dual-slot cooler.  Sapphire and ATI engineers alluded to DailyTech that this dual-slot configuration will likely be replaced with a single-slot solution by time of launch.

NVIDIA's G92 has also carried many names.  Originally slated as the 65nm "fill-in" GPU between GeForce 8600 and GeForce 8800, the company began changing documentation earlier this month as ATI's offerings began to firm up. 

NVIDIA confirmed the specifications of G92 with board partners earlier this week.  The GeForce 8800 GT will feature a 600 MHz core clock, a 900 MHz memory clock and a 256 bit memory interface.

Newest guidance from NVIDIA, released Monday, claims the 8800 GT will feature 112 stream processors and a shader clock locked at 1500 MHz. 

The one thing that didn't change on G92 is the process node.  NVIDIA's foundry partner, TSMC, forwarded a second memo to DailyTech confirming G92 is in mass production at the company's Fab 12 with samples available now on 65nm process node.  NVIDIA's GeForce 8500 and GeForce 8600 are manufactured on TSMC's 80nm node; GeForce 8800 GT will be the company's first 65nm graphics processor.

NVIDIA guidance suggests G92 will be here next month, followed by AMD's marketing blitz for RV670, RD790 and Phenom.  All three AMD offerings are expected to launch on the same day, which AMD distributors have penciled in for late November. Intel is expected to launch its 45nm Penryn processors on November 12, and any NVIDIA launch will likely coincide with that announcement.

Late last week, Maximum PC reported that NVIDIA senior vice president Dan Vivoli commented that NVIDIA would be releasing new hardware to go along with the upcoming title Crysis.  The confirmed launch date by Electronic Arts for Crysis is November 15, 2007

However, G92 fans might get a quick preview of the new GPU on October 29, 2007, when the company officially lifts the embargo on 8800 GT.

Neither AMD nor NVIDIA have released "firm" pricing for the products, though we can reasonably infer several key points regarding the price.  Since RV670 is effectively a smaller R600, performance will be very similar to existing R600-based cards on the market today.  However, since the card only utilizes a single-slot cooler and a considerably smaller die, the cost of these cards should be lower than existing R600s.

G92, which was originally called GeForce 8700 until just last week, has a soft suggested retail price of $250, according to NVIDIA board partners.  Since the GeForce 8800 GT will be launching first, it's fairly likely that AMD will adjust the suggested price of RV670 depending on the outcome of initial GeForce 8800 GT feedback.



Comments     Threshold


This article is over a month old, voting and posting comments is disabled

RE: Sub-titans...
By sxr7171 on 10/9/2007 11:36:20 PM , Rating: 2
Why not have cards that can all games at 1920x1200 with 4x AA minimum? They introduced up to 16x AA, and why not have a card that can run games with it enabled? I know it doesn't affect gameplay, but it looks very nice. Should we all stagnate and have 1680x1050 with no AA forever?

If anything we need to be moving to higher resolutions and higher PPI with our OSes supporting scaling. Now that's what would separate your Xbox 360 from a high end gaming PC.

Manufacturers should reach higher and not rest on their laurels.


RE: Sub-titans...
By hrah20 on 10/9/2007 11:58:49 PM , Rating: 3
(Why not have cards that can all games at 1920x1200 with 4x AA minimum?)

I was thinking the same.


RE: Sub-titans...
By Lakku on 10/10/2007 12:09:43 AM , Rating: 2
I didn't say rest forever, but considering the cards now can play any game at almost any resolution, most with AA, that video card companies have no incentive to release something new, especially when one company isn't offering much competition on the high end. Eventually, you may be able to get your wish, and that will come with time, but no company is in any rush to push there when the current situation is just fine for the most part.


RE: Sub-titans...
By murphyslabrat on 10/10/2007 1:03:12 PM , Rating: 2
The answer? because that will cost more. More power=higher quality design/silicon, and that equals more cost. The design itself takes longer to complete, with a higher chance of bugs; and costs more, as it takes up more die-space. With the advent of efficient 3-D layouts, this might change; or they just maintain the current level of performance, and you pay half the price for it. Then, of course, if you require higher clock-rates, you deal with higher-quality transistors; or being more selective about your chips, which ends up with more of them being tossed/budget-binned. Both of these solutions result in either higher cost-of-manufacture or lower yields, both of which also drive the price like a highway chase scene.

So, the reason why? Because most people prefer a physics/light/AI intensive game (aka, realistic) to beautifully rendered/smooth-edged/2560x1920 resolution-ed/etc. Quake 2 models. If resolution must be sacrificed to give me relatively cheap life-like scenes, I am all for it.


"There's no chance that the iPhone is going to get any significant market share. No chance." -- Microsoft CEO Steve Ballmer

Related Articles
Crysis System Requirements Revealed
October 9, 2007, 3:43 PM
Intel Sets Official "Penryn" Launch Date
September 18, 2007, 1:17 PM
AMD Unveils "Barcelona" Architecture
September 7, 2007, 3:03 PM













botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki