Geez, I want 8800GTX support soon... That's gonna make things much more interesting with the possibilities CUDA can bring.

Nvidia, you're not fooling anyone.. The architectural differences between the G80 and its die shrink G92 are about nil. You're just not supporting them because you'd rather have people think the 9 series is an actual improvement, and maybe sell some of them to unwitting fools.

One more nail in the 'buying another Nvidia card' for me.. Being treated like an idiot for the purpose of marketing features for newer cards.

I want G92 support like you promised..

ShadowFoldI want G92 support like you promised..

I'm having no problems running this on my 8800 GT :)

so 177.44 = no Physx??

wolfso 177.44 = no Physx??

Yup. That's a stable release though. Looks like NV didn't want to publish the 177.39 as stable.

I wonder if this is TRUE 100% Physx? Or just a command subset. Does the PhysX code require a special compile? (ie. GPU Physx code) Or does it work with original PhysX code? This is going to be such a pain for the user if you need to select, on game install, WHICH code version you want. If you then plug in, or unplug, a real Physx board, you need to reinstall the game code.

I'd be very interested in somehow benchmarking a Physx enabled routine (or graphics demo), with:

1./ CPU+GPU only (normal drivers)
2./ CPU+GPU only (Physx emulator drivers)
3./ CPU+real Physx+GPU

Even better if option 1 could include Physx emulated on the CPU. Therefore under all 3 tests we would have the same data to be rendered.

My thoughts:
1./ The Physx emulator in the GPU will probably be able to do *some* calculations much faster than an actual Physx board. But I think only *some*. Why? Because of the incredible performance of integer calculations on a stream processor, but much worse performance on FP calculations.
2./ To make the GPU Physx emulator work at speed, it will have to implement a lot of precalced math tables. I would imagine that these tables would be competiting with other data in the GPU cache.
3./ The GPU Physx emulator is going to use up quite a lot of resources in the GPU, whether stream processors, cache, memory (up to 128MB), etc. Consequently, there will be fewer stream processors available for rendering = low graphics performance even before rendering the *new* objects associated with Physx candy.

With the Physx now priced at EUR 83 new (approx GBP 60, $120), we are getting to a price point that if you WANT Physx, you can buy it as a discrete processor and get the benefit of zero resource hit. If these prices come down further, to say EUR 50 (and on PCIe), then I think the Physx hardware could become interesting again.

btarunrYup. That's a stable release though. Looks like NV didn't want to publish the 177.39 as stable.

yeah it seems they gave teh hardcore nerds (us) a taste of them and WHQL phyx is on its way :D hopefully with full G8x and full G9x support

Here are some benchmarks:

www.techarp.com/showarticle.aspx?artno=560

some light reading for your amusement: www.hothardware.com/News/NVIDIA_Responds_To_GPU_PhysX_Cheating_Allegation/ also both 177.39 & PhysX 8.06 work just fine and stable on my 8800GTS 512mb sli config. Just mod the inf. file. That's what I did.:rockout:

Download Links seem to be broken, at least they don't work for me. Here are working ones:
WinXP x86
WinXP x64
WinVista x86
WinVista x64

You can find the modded inf's on your own.

bowmanNvidia, you're not fooling anyone.. The architectural differences between the G80 and its die shrink G92 are about nil. You're just not supporting them because you'd rather have people think the 9 series is an actual improvement, and maybe sell some of them to unwitting fools.

One more nail in the 'buying another Nvidia card' for me.. Being treated like an idiot for the purpose of marketing features for newer cards.

1.) There are some differences between G92 and G80, it isn't just a simple die shrink. The G92 has about 78 Million more Transistors on it, more texture units, and a lower memory bus. NVidia had CUDA more in mind when they were creating the G92, and hence it is easier to implement CUDA based technologies on it than it is with the other older cores.
2.) nVidia just aquired Ageia in Febuary, so they have only had 4 months to get this technology up and running on their cards, I say they have gotten pretty far in that time frame.
3.) PhysX isn't even working on all the G92 cards yet, give it some time. It isn't like nVidia is making a big deal out of PhysX support, most people don't even know it is there, and if they really wanted to market it for just their new cards, why is the 9800GTX supported and not some other logical choices? I would think they would support it on the 9800GX2, since that is their next high end card after the GT200s, or if they really only wanted to support new cards, they would only support the GT200 cards, the fact that they are supporting other G92 cards disproves your theory. The technology is new, give nVidia some time to get it up and working on other cards before you start bashing them and claiming conspiracy.

Fixed that bad link. They're all fine now.

They're all still broken for me. Though it might just be me.

Odd. I've opened and checked all links, they're opening. Quick fix: Open each link, click on the address in the address bar/location bar and press enter? The URL's are all valid.

Even hitting enter doesn't do it. They all have a %20 (meaning there is an extra space in the url) at the end of the url, when I remove that they then work. Some browsers will automatically remove the %20, some won't, which is why it probably isn't working for me.

All links are fixed now.

newtekie1Even hitting enter doesn't do it. They all have a %20 (meaning there is an extra space in the url) at the end of the url, when I remove that they then work. Some browsers will automatically remove the %20, some won't, which is why it probably isn't working for me.

Where? :confused:


Alright Newtekie, I changed them all to the links in your post. However, there's a problem. While the old links took users to their respective NVidia servers closest to them geographically, yours was a direct link to the files on NV's American servers. (Example, my using that old link would have taken me to NVIDIA's Indian servers (that are located just 30 miles from where I live :D). If TPU mirrors these files, please download them from the Front-page under "Today's Files" list (whenever it comes up) for better speed (if you happen to live outside of the US).

just tried it on my 9800gtx and scores went up from p6750 (gpu 5453 / cpu 23526) to p7748 (gpu 6068 / cpu 45740)!!! nice... :cool:

bowmanNvidia, you're not fooling anyone.. The architectural differences between the G80 and its die shrink G92 are about nil. You're just not supporting them because you'd rather have people think the 9 series is an actual improvement, and maybe sell some of them to unwitting fools.

One more nail in the 'buying another Nvidia card' for me.. Being treated like an idiot for the purpose of marketing features for newer cards.

Ever take that attitude when purchasing your automobiles?

Manufacturers have been screwing consumers for decades now.