GeForce PhysX for all GF8/9 with FW175.16

This assumption is wrong.
Many games are CPU-limited on the fastest GPUs available today, sometimes even if you set the resolution and AA to maximum.
People also seem to think that Crysis is NOT heavy on the CPU just because it is heavy on the GPU.
This is also wrong. As I already said, run the CPU benchmarks for Crysis, where there's a lot of physics going on, and you're CPU-limited. So Crysis does fully saturate the CPU, just not all the time, because it doesn't always have to process a lot of physics. When it does, it struggles though.

 

This is a bit silly.
Instead of using a second videocard, you can just buy a faster primary card, and run the physics on that. Probably more cost-effective.
The only reason why you'd want a separate card is if you want the absolute best performance, and things like money, power consumption or number of cards in the system aren't an issue.
For everyone else, a single card running both physics and graphics is the most elegant and most cost-effective solution. And a lot of people already have these cards, so they don't have to do anything to their hardware, just install the new driver.

 

There are official drivers out now, 177.83.
They include the whole PhysX software suite aswell, should work right away.
You can download the drivers a lot of Cuda/PhysX related stuff in the PowerPack: http://www.nvidia.com/content/forcewithin/us/download.asp

 

Yep, thanks. found the link from EliteBastards as amended from my post above.

US

 

This is where the problem may be; what is the perfect solution? If you're looking for perfect, well, you may take negatives and make mountains out of them right now. One may look for improvements and nice steps in the evolution of Physics for dynamic gaming for the end-user, imho!

 

Thanks for the link!

 

In case you haven't noticed the first official GPU PhysX driver landed just a few weeks ago. You're comparing it to a game that was in development years ago? Come on, don't twist things to try to make your point. You're also conveniently ignoring what has been accomplished with UT3 and GRAW thus far. The fact is that physics on CPUs have been around forever and nobody has been able to generate the kind of effects shown so far on GPUs. How about you give GPU physics a chance ?

People buy unnecessary $100 sound cards. And they buy unnecessary $1000 SLI rigs. I don't see this being a problem if there is a pronounced benefit. What's your alternative? You haven't provided evidence that current CPUs are capable. So should we all just wait until CPUs are fast enough?

 

Excellent, I've been waiting for this

Cheers!

 

But what exactly *is* the perfect solution?
By the looks of it, both the CPU and GPU have their uses.
The CPU is good in unpredictable, serial code.
The GPU is good in predictable, parallel code.

CPUs are good at doing a lot of things, such as running an OS, taking care of hardware I/O and all that stuff... and running a generic application.

The GPU has stream processors that can be used for accelerating graphics, physics, audio and video processing and general solving of linear system (mostly scientific applications).

I'd say the perfect system consists of a good blend of both types of processors.
What the perfect blend is, depends on what software you run on it, and what choices the developers made. A bit of a chicken-and-egg problem. Will you go for a faster CPU and a slower GPU, or for a slower CPU and faster GPU? Does the choice of developers decide what people will buy, or will developers decide based on what people have bought?

 

Installing new drivers

How do you guys install new NV drivers? i.e. do you just install over the existing drivers or do you unistall the existing first and then install the new ones?

Just curious!

 

I always uninstall the previous ones and reboot before I install the new ones (and then I reboot again).
I'm not sure if it's required anymore, but in the past (way back in the Windows 98/2000 era I guess) I sometimes had issues with not all files being updated properly, so the driver I ended up running was a mix of old and new libraries, with some nice unexpected results.

 

I uninstall, use Guru3d's Driver Cleaner(Just about the only good thing they've ever done), reboot, install new drivers, then reboot for the final time.

 

Cheers guys, i'm taking that route as well now. I tried the 'install over the top' option and was having issues with UT3 freezing in the physx levels.

I'm now in the middle of deleting the drivers and starting over!

 

I delete the drivers, reboot, run CCleaner and then install the drivers. I have never used that DriverCleaner software and have never had any driver related buggery following that process at least with ATI and Nvidia's stuff.

 

Yea, I don't quite trust things like drivercleaner, to be honest.
They might delete more than I bargained for.
I'd only use such a tool as a last resort, when you're having trouble uninstalling or reinstalling the normal way.
It is certainly not a required procedure for a normal driver update. The uninstaller should already clean up everything that needs to be cleaned (and in most cases the installer will first trigger the uninstaller of the previously installed driver).

 

These articles seem relevant to this discussion:

http://www.fudzilla.com/index.php?option=com_content&task=view&id=8859&Itemid=1

http://www.fudzilla.com/index.php?option=com_content&task=view&id=8862&Itemid=1

 

Wow, these guys are really looking for anything they could possibly throw at nVidia/PhysX, don't they?
I don't find it at all surprising that the performance is affected by CPU load. Obviously the CPU still has plenty to do. Their tests show that both get the same framerate under heavy load. If anything, that would be an indication that both have the same CPU overhead per frame. If the GPU dropped below the PPU performance, then they might have a bit of a case (although not much, because by offloading physics to the GPU or PPU you actually reduce the overall CPU load).

 

Well, in a real game you can't exactly dedicate 80% of the available CPU time to feeding GPU physics. If doing Physics on the GPU requires that level of CPU utilisation, then it may not be of any practical use except in synthetic benchmarks.

 

As I said, their results show that there's no more overhead for the GPU than for the PPU, they both get around the same framerate.
The whole article is nonsense. There's already plenty of proof of games using PhysX, this is a non-issue. FUD-zilla indeed...

 

Stop reading into something so far. Use of "perfect" wasn't meant literally as the perfect solution but more accurately as "the right" solution.

Throwing things around that don't change how you play the game... big accomplishments. Again those are nothing more than tech demos included in the game. That's nothing at all. The time line situation applies to both GPU Physics and CPU based physics with the availability of more free cores. I'm not the one stating the time lines out of order.

The vast majority of computer gamers use onboard audio. A even lesser amount use SLI. Stop trying to paint this picture of $2000 computers being normal for gamers, it's not. Games have already shown CPUs are capable of improvement and continued improvement along those lines. What more evidence do I need beyond telling you to go play a game. The most obvious game is of course Crysis as its physics are certainly the most impressive I've ever experienced or seen that had major gameplay ramifications.

In this time I'll be patiently awaiting a game with PhysX used properly and see the results. Like I said already I'm not calling for GPU physics advancements to stop entirely, even if someone people keep trying to say I am, and I'll personally be installing this driver package and checking out some of the content in the "power pack."