AMD stream processors are more flexible.
Nvidia are not really better at gaming when you take price into consideration.
AMD usually undercut Nvidia by a wide margin but they do this by having less ROPs on the board.
This has the effect of making MSAA (pixel smoothing) less efficient on AMD boards.
AMD has recently negated this Nvidia gaming advantage by pushing game developers to use FXAA instead.
Topic: nvidia vs amd (Read 1795 times)
AMD is much better for mining.
It is debatable that NVidia is better at gaming. Even if one card is superior there are lots of cards at lots of price points. Also both companies have traded top spot in benchmarks for years. Competition is a good thing it means better products at lower prices for all consumers regardless of which brand they prefer.
However in OpenCL NVidia has higher floating point power (good for scientific simulations) while AMD has superior integer performance. Hashing (mining uses hashing not encryption) is all integer math so an AMD chip is going to be superior. However what is the "kill blow" for relative performance is AMDs chips have an instruction which reduces a complex rotation which normally takes 3 operations plus a delay into a single operation. That single instruction adds about a 15% to 20% "bonus" to AMD GPU performance.
There is another difference but it is growing smaller. NVidia has generally designed chips around a "fewer but more complex/powerful shaders" concept while AMD has designed chips around a "massive number of simple shaders per core" concept. SHA-2 hashing is very simple and thus fits more efficiently into "a lot of simple cores" model. However overtime AMDs shaders have gotten more and more complex while NVidia has added more and more shaders so this difference is closing. It is also one reason why the relative performance of a 7000 series isn't double that of a 5000 series despite the 30% higher clock and 50% more transistors . The 7000 series is "faster" but not as fast as it would be if the 7000 series was just a die shrink of the 5000 series.
So the combination of better integer performance, instructions which provide a significant speed improvement, and a higher shader count all make AMD a better fit for mining (and a worse fit for other OpenCL tasks). There is no reason NVidia couldn't design their chips differently they just have chosen not to. In most industrial and scientific tasks floating point performance is more important. If anything AMD will likely need to improve floating point performance (possibly at the expense of integer performance) in order to get most OpenCL design wins.
However in OpenCL NVidia has higher floating point power (good for scientific simulations) while AMD has superior integer performance. Hashing (mining uses hashing not encryption) is all integer math so an AMD chip is going to be superior. However what is the "kill blow" for relative performance is AMDs chips have an instruction which reduces a complex rotation which normally takes 3 operations plus a delay into a single operation. That single instruction adds about a 15% to 20% "bonus" to AMD GPU performance.
There is another difference but it is growing smaller. NVidia has generally designed chips around a "fewer but more complex/powerful shaders" concept while AMD has designed chips around a "massive number of simple shaders per core" concept. SHA-2 hashing is very simple and thus fits more efficiently into "a lot of simple cores" model. However overtime AMDs shaders have gotten more and more complex while NVidia has added more and more shaders so this difference is closing. It is also one reason why the relative performance of a 7000 series isn't double that of a 5000 series despite the 30% higher clock and 50% more transistors . The 7000 series is "faster" but not as fast as it would be if the 7000 series was just a die shrink of the 5000 series.
So the combination of better integer performance, instructions which provide a significant speed improvement, and a higher shader count all make AMD a better fit for mining (and a worse fit for other OpenCL tasks). There is no reason NVidia couldn't design their chips differently they just have chosen not to. In most industrial and scientific tasks floating point performance is more important. If anything AMD will likely need to improve floating point performance (possibly at the expense of integer performance) in order to get most OpenCL design wins.
my best guess amd use way more engines allowing it to process more processes per secound where nvidia focus on shaders and game important things. i atually think overall amd do the better gpu its level for most thing where nvidia are more game focused and less process focused
why is amd better at encryption/decryption?
why is nvidia better at gaming?
why is nvidia better at gaming?
Jump to: