The head of graphics product group of Advanced Micro Devices, the former ATI Technologies, said in an interview that the recently introduced Nvidia Corp.’s GeForce GTX 200 graphics processing unit (GPU) was the last high-end monolithic “megachip”. The comment follows vows by high-ranking Nvidia execs to continue making high-end graphics chips that are massive in size and costly in price.
“I predict our competitor will go down the same path for its next GPU once they see this. They have made their last monolithic GPU,” said Rick Bergman, the vice president of AMD who heads graphics product group, reports EETimes web-site.
ATI has been utilizing multi-chip graphics setups to address the high-end market of graphics cards that are used by gamers and enthusiasts for about a year now, whereas Nvidia Corp. has been using the approach for over two years, but has been claiming for about a quarter now that single-chip high-end graphics cards are better than dual-chip approach.
According to some estimates, the die size of Nvidia code-named G200 GPU that powers GeForce GTX 200-series products is about 600mm², which means that only about 97 of such chips may fit onto a single 300mm wafer. Provided that the yield of such chips is hardly high, whereas a single wafer costs several thousands of dollars, such large GPUs do cost a lot to manufacture, not talking about the amount of money and time to develop and test such large products.
Usually developers of graphics chips have to disable certain defective parts of their large graphics processing units and install them on less expensive graphics cards to be able to sell even imperfect chips.
Earlier this year Nvidia’s chief scientist said that AMD is unable to develop a large monolithic graphics processor due to lack of resources. However, Mr. Bergman said that smaller chips allow easier adoption of them for mobile computers. Even though Nvidia’s currently provides faster graphics processors for mobile market at the moment, it would hardly be able to offer G200-class mobile hardware in the near future.
“There’s no way this new Nvidia core will be in notebooks this fall,” Mr. Bergman claimed.
Comments currently: 11
Discussion started: 06/17/08 10:18:57 PM
Latest comment: 07/02/08 03:51:28 AM
Expand all threads |
Collapse all threads
[1-8]
1.
Two words: die shrink.
[Posted by: boner | Date: 06/17/08 10:18:57 PM]
2.
die shrink i dont think nvidia would do that, because they have no competition, and die shrink will just eat there profit more, they just sell it at higher price.
nvidia need to do that in order to maintain they status quo of being the best in graphics even if its illogical sometimes.
and in my own opinion AMD is just making excuses for not competing with nvidia in this generation.
[Posted by: clueless | Date: 06/18/08 12:10:06 AM]
3.
Multicores are Just Better.... More Yield, Run Cool...Cheap.... :D
http://free-and-useful.blogspot.com
[Posted by: Geek | Date: 06/18/08 01:41:40 AM]
4.
The problem with multicore is that developers are still not ready for it. There are some things that are just not suited to multicores. Every data set has a latency (from input data to output data). Paralelization only hides the latency by doing more data sets in the same time. But if you don't have independent data sets, you cannot benefit of paralelization.
GPU graphics is massively paralel, but it still does a lot of housekeeping. Since 2 (or more) GPUs cannot (yet) access the same data/memory (vertex coorinates, textures), the CPU has to copy them to both GPUs memories. This takes time, sometimes even hiding the gain of the 2nd GPU.
Also, in CPU area, the multiprocessing is very old (I remember seeing the SMP protocol in the '90) and there is a specific protocol for 2 or more CPUs to access the same memory. This was developed because of needed server performance, so a lot of money was spent. But this was not worth trying on GPU (until now ?!?!?!).
[Posted by: mathew7 | Date: 06/18/08 03:26:56 AM]
5.
Multi GPU cards have been on the market ever since the days of 3dfx, and they have ALWAYS had problems, even still today over 15 years later.
When are people going to wake up to this?
[Posted by: EndPCNoise | Date: 06/18/08 07:55:05 PM]
6.
I cant see this GPU ever being used for mobile unless there is a die shrink
[Posted by: alpha0ne | Date: 06/20/08 12:07:08 AM]
7.
Fuck nvidia, its to expensive
[Posted by: Wet Fevogina | Date: 06/22/08 02:34:35 AM]
+ expand thread (3 answers)
- collapse thread
to do what?you gotta be retarded buying computers for gaming anyway......
[Posted by: muzzel Mo | Date: 06/24/08 05:24:23 PM]
i can share his frustration NVIDIA thinks it can just overprice is gfx cards 50% for only 10% performance increase... NOT JUSTIFIED!!!
[Posted by: nick | Date: 06/27/08 08:51:55 AM]
The only one retarded here is you, I'm afraid.
Seek some professional help... really... you need it.
[Posted by: advisor | Date: 06/28/08 10:36:47 AM]
8.
why not make a gpu that you can put on a motherboard like a normal cpu. add all the good stuff on the die and add like 2 memory dimms like gddr4. in place of useing a addin card all these times
[Posted by: Benny | Date: 07/02/08 03:51:28 AM]
[1-8]