AMD to Nvidia: Two chips are better than one
Advanced Micro Devices announced on Monday its most powerful graphics technology to date, going after Nvidia in the rarified--and closely watched--enthusiast game segment.
This also marks the current performance pinnacle of AMD's strategy to beat Nvidia at the high end by building comparatively smaller chips and then ganging them together for better performance.
The ATI Radeon HD 4870 X2 graphics board houses two 4870 graphics processing units (GPUs) and competes with Nvidia's fastest board, based on the GTX 280. In chip-to-chip competition, Nvidia's GTX 280 generally beats a single 4870 in performance because it's bigger and faster: the Nvidia chip packs 1.4 billion transistors onto one chip, while ATI has about 950 million.
But because AMD puts two chips on one board and has improved chip-to-chip communication, the 4870 X2 is is expected to equal or exceed the Nvidia chip in many cases.
AMD has introduced a more advanced cross-GPU connection technology based on the PCIe Generation 2 standard. And the 4807 X2 can use two gigabytes of memory, compared to most high-end boards that use a maximum of one gigabyte. It also uses memory based on the new GDDR5 standard.
AMD says the 4870 X2 delivers over 3X the bandwidth of the its previous dual-GPU board, the 3870 X2
(Credit: AMD)One of the central challenges for AMD is to make sure the performance scales up efficiently when more chips are added. This is the crux of AMD's strategy: instead of building a large monolithic--albeit fast--chip as Nvidia typically does, AMD takes smaller chips and gangs them together for better performance.
To date, the results for multi-GPU performance have been problematic, typically another board will deliver only 1.5 times better performance. AMD is targeting 1.8 the performance with two chips running games in high resolution, and with four of them, about 2.5, according to earlier comments from Jon Peddie of Jon Peddie Research.
Game PC vendors expect good things. "(The 4870 X2 is) more than a match for a single Nvidia GTX 280, and depending on the title sometimes a match for two GTX 280s," said Kelt Reeves, CEO of game PC maker Falcon Northwest, responding to an email query. "Drivers are now ATI's only weak area, so the 4870 X2's performance and scaling with two 4870 X2s (QuadFire) often varies widely from title to title," he said.
In September, AMD is also expected to bring out the HD 4850 X2, a dual-chip board with slightly lower performance. The higher-end 4870 X2 is rated at 2.4 TFLOPs (or teraFLOPs a common yardstick for raw graphics chip compute power) and communicates with memory at 230GB per second, while the 4850 X2 is rated at 2.0 TFLOPs and has a memory bandwidth of 128GB/sec.
Both boards will integrate 1600 stream processors, which do parallel processing on streams of data.
The 4870 X2 is priced at $549. Nvidia preemptively responded to this by cutting the price on the GTX 280 to $499 in July.
The lower-end 4850 X2 will be available in September for $399.
Brooke Crothers has been an editor at large at CNET News, an analyst at IDC Japan, and an editor at The Asian Wall Street Journal Weekly, among other endeavors, including co-manager of an after-school math-and-reading center. He writes for the CNET Blog Network and is not a current employee of CNET. Disclosure. 
Consider what I'll call the 1.5 task problem if you will. Two CPUs have two separate tasks, but at some point these two separate tasks need to work with the same data. A very over simplified, hypothetical example follows.
CPU 1 is rendering objects with texture maps while CPU 2 is writing those texture maps to the video ram at the same time. Perhaps receiving them from the game. They're both working on two separate things very nicely. Now all of a sudden CPU 1 needs a texture map that CPU 2 has only written half of. What do you do? Well, CPU 1 can go ahead and draw with half a texture and it will look like crap, or it can wait for CPU 2 to finish up its work, but that's where it loses performance. For a millisecond or so CPU 1 doesn't have anything to do. You're back down to one CPU. When these wait conditions happen every few milliseconds or more your average performance takes a hit real quick.
That's a little rude. Maybe we should just stick to criticizing each other's opinions and spelling like we always do. If someone pops on in with a generally dumb opinion that's one thing, hammer away, but when someone stops by just trying to figure out why something works differently than the way they thought it would, you could be little more constructive don't you think? Gotta tell ya, I'll respect someone that admits something they don't know on a board looking for some insight long before I'll respect someone that just throws out random insults. Especially since I've been that guy on the boards looking for answers. Haven't you?
- by smokified August 13, 2008 5:18 AM PDT
- ATI/AMD has been a failing entity for a while now. Nothing they make comes close to the performance and efficiency as it's counterpart nVidia product. After AMD takes the time to "catch-up", nVidia has already had the time to move on to the next generation. AMD continues to play catch-up while nVidia continues to make groundbreaking developments.
- Reply to this comment
-
-
- by smokified August 13, 2008 4:11 PM PDT
- Lets also not forget that it is taking 2 chips and a newer memory standard for AMD to do what nVidia does with 1 chip and GDDR3 and GDDR4 standards.
-
-
(16 Comments)ATI is just a cheaper alternative to real gaming hardware.