PDA

View Full Version : nVidia has faster & cheaper GPU than HD 5870 already


Pages : 1 [2]

jlippo
09-22-09, 01:12 AM
if you were running the same physics with GPU acceleration you would be getting better overall performance. but like Lfctony said, games supporting Physx do more complex physics to really show off the feature. so at the end of the day overall performance is lower, but you get really slick physics effects in exchange.
Indeed, in sacred the leaves and particles hit everything in the scene with polygonal accuracy.
Cryostasis the water does the same, but particles have to take account every other water particle in vicinity as well.

In many cases those effects are also quite graphics, something like additional over 1000 moving objects is not a small thing.

If one want's to see what the performance difference is between CPU/GPU physics, the physx demos might be good test ground.
GPU even while rendering the graphics, simply trounces CPU in physics calculations.

warlockk
09-22-09, 02:59 PM
Apparently their marketing people found out that ati/amd is launching new generation before they do.
Remember the 'unified shders is bad idea' marketing a month or two before 8xxx series..
Happily we do not have to listen their BS for many months, soon we have new generation for both companies and we pretty much know the setting for the next 3 years. ;)

They use stream processors, not unified shaders like ATI. nVidia's SP's are much simpler to program, since they are all identical and do not need different patterns of code to crunch numbers. This makes it easier on programmers and games have better chances of being optimized to that architecture. Ati uses a "pattern" in its shaders, so each shader block of 6 uses 4 identical and 2 different ALU's. To take full advantage of Ati's hardware, programs would have to be coded in such a way. This is what complicates optimizing. So for each line of code that is not using that pattern, you only get 1/5 (+1 correction) shaders being utilized v 8/8 (+1 correction) for nvidia.

So essentially, the 800 (670 effective) unified shaders on the Ati card become 160* (134 effective), vs the nVidia's remain at 240 (216 effective). (** in a perfect world where only simple shaders are used)

So nVidia was right, ATi's architecture is not very efficient unless programmers spend lots more time optimizing their games for ATi.

That said, if games were optimized for ATi to take full advantage of their hardware, it would maybe give them the edge on performance - the only obstacle now is the fact that the core's clock on nVidia cards are almost double the speed...

-D

jlippo
09-23-09, 01:37 AM
Even if both companies use their ALUs differently, nvidia marketing said that separate ALUs for pixel shaders and vertex shaders is the way to go, just before 8xxx series launch.
Their reasoning was quite simple they had product in the market which didn't share ALUs for pixels and vertexes and Ati was touting about their way, so they simply downplayed the importance of the shared ALU approach.

As for which is easier to program to and which gets better performance with same amount of time used by programmers, it really isn't my area of knowledge and I will not go to that discussion with just my butt feeling. ;)

doom3ed_player
09-23-09, 07:09 AM
progress of technology is against short term shareholder buissness!
it need new build process -> new R&D , investement etc ...

if share were managed ALL investement , actually, i could play on 8086 ...
i'm sick of nvidia marketing fail . NO , CUDA will not supplent HPC cluster, neither physx will predominate gaming market .

don't listen to these coward.

jlippo
09-24-09, 12:59 AM
if share were managed ALL investement , actually, i could play on 8086 ...
i'm sick of nvidia marketing fail . NO , CUDA will not supplent HPC cluster, neither physx will predominate gaming market .

don't listen to these coward.
They have been on roll for couple of years, all those stupid naming changes of GPUs and cards.
I really liked the NVxx naming scheme, one could immediately see if new GPU was speed bump or proper new architecture.

trinibwoy
09-25-09, 11:15 PM
They use stream processors, not unified shaders like ATI

Sigh.

Lyme
09-25-09, 11:47 PM
nVidia ever since the PS3 has been using the Sony Marketing Playbook, and expecting people to believe anything.