NVIDIA vs Hydra - Take 1

[LydianKnight]

NVIDIA's PhysX shortsightedness angers developers (http://www.brightsideofnews.com/news/2009/9/28/nvidias-physx-shortsightedness-angers-developers.aspx)

Ill explain why this function was disabled.

Physx is an open software standard any company can freely develop hardware or software that supports it. Nvidia supports GPU accelerated Physx on NVIDIA GPUs while using NVIDIA GPUs for graphics. NVIDIA performs extensive Engineering, Development, and QA work that makes Physx a great experience for customers. For a variety of reasons - some development expense some quality assurance and some business reasons NVIDIA will not support GPU accelerated Physx with NVIDIA GPUs while GPU rendering is happening on non- NVIDIA GPUs. I'm sorry for any inconvenience caused but I hope you can understand.

Best Regards,
Troy
NVIDIA Customer Care

That's just pure bull****... isn't it easier to say they're not allowing any system with two different graphics cards powered by Lucid's Hydra chip to have PhysX because they will be losing some cash due to motherboard manufacturers relying on the Hydra chip instead of having to pay a fee for a SLI certification?

Hah...

[CaptNKILL]

I'll just post my response from another thread...
http://www.nvnews.net/vbulletin/showpost.php?p=2092189&postcount=7

As I said in another thread, if everyone could use their current geforce cards for Physx for years to come with acceptable performance, physx would no longer be a selling point for their high end cards.

I agree that they've done the end user a major disservice by disabling ATI+Nvidia Physx in drivers, but the move makes sense to me.

Why bother spending all that money on physx to sell high end graphics cards if it just gives people less of a reason to buy your high end cards and more of a reason to buy cheaper (but similarly performing) ATI cards and keep their existing GT200 series card for Physx for the next 3-4 years?

If nvidia allowed mixed GPU physx support they would most likely stop promoting it and move on to something else that actually gave people a reason not to buy competing graphics cards.

It makes absolutely no sense from a business perspective to allow people to use one of your primary exclusive features with a competing product.

[Muppet]

Personally ATI should just pay Nvidia some sort of royalty so they can use PhysX. I know I want it in my future games.

[K007]

yep as sad as it is...they can't have things like this available as a option

while i understand what nvidia does with the whole ati/nvidia phyx thing it is still lame ><

wouldn't it be easier to just buy a actual ageia phyx card?

[LydianKnight]

It makes absolutely no sense from a business perspective to allow people to use one of your primary exclusive features with a competing product.

Sorry if it looked like ranting, but it wasn't. I agree with your response, I would do the same, competition is competition after all and you have to try (at least) to stop your enemies dead on their tracks, it's just I think this kind of comment is just silly...

They could just tell 'we don't want PhysX to run in a platform where a product from competing companies are present, if you want that, go somewhere else...' or something similar.

We all know NVIDIA and ATI are rivals for life, they don't need to politely cover that fact with a silly statement, at least I think it's more than obvious and they don't need excuses.

[LordJuanlo]

I understand the business decision, but as a end user I'm pissed. Let's say I buy a nVidia 9800 card because I want PhysX. I play games and have PhysX in this card, all good. After some months new games arrive and I notice that performance is bad, so I decide to keep my nVidia 9800 card as a dedicated PhysX card and buy another card for graphics.

A friend lets me try his ATI card on my rig, I launch a game and I see PhysX is working nicely, the ATI and nVidia card are working together, so I go and buy an ATI card for graphics and keep the GeForce 9800 for PhysX, I'm happy. Because IT WORKS.

And after some months, nVidia decides that it won't work anymore. They release a new driver that disables PhysX on my system.

They should have done it from day 1!!!

It's too late now, some users could have bought nVidia cards for their ATI systems just for PhysX because it worked. Or some nVidia users could have bought ATI cards as primary graphics keeping their nVidia cards for PhysX.

When PhysX was released, nVidia did not tell anyone that all GPUs on your rig should be nVidia. Look at PhysX FAQ (http://www.nvidia.com/object/physx_faq.html) on nVidia page, they have added this line a few days ago:

Can I use an NVIDIA GPU as a PhysX processor and a non-NVIDIA GPU for regular display graphics?
No. There are multiple technical connections between PhysX processing and graphics that require tight collaboration between the two technologies. To deliver a good experience for users, NVIDIA PhysX technology has been fully verified and enabled using only NVIDIA GPUs for graphics.

Appart from this being a LIE to its customers, this line has not been added (yet) to the spanish (http://www.nvidia.es/object/physx_faq_es.html) FAQ, so any Spanish user that can't read English doesn't know that nVidia is blocking PhysX on systems with a non-nVidia GPU installed. He could go and buy an nVidia that would refuse to work as PhysX card.

Too bad, and too late

[LydianKnight]

When PhysX was released, nVidia did not tell anyone that all GPUs on your rig should be nVidia.

Why should they? PhysX is a propietary technology created in the beginning by Ageia Technologies, and now it's property of NVIDIA... they can do whatever they want with it...

It's quite logical to think if you're using PhysX (by NVIDIA) you should use an adequate graphics card (by NVIDIA). Unfair in terms of interoperability but makes sense...

[LydianKnight]

wouldn't it be easier to just buy a actual ageia phyx card?

supposedly the main advantage of integrating the PhysX middleware into graphics cards through the use of some specific codepaths and hardware bits is the bandwidth required for adequate deployment of physics calculation data, the original PhysX processor by Ageia was tied to a PCI slot with the possibility of having a PCI-E x1 slot card? that leaves you with 250MB/s of raw data, and that's without the bus overhead due to the 8b/10b data encoding, so you have a raw data rate of 80% of that speed.

A graphics card like the upcoming GT300 has something like 286GB/s if I remember correctly... so it's like 1000 times faster (with the same 20% penalty for data encoding, I guess) so makes sense on having the graphics card do the job.

About a dedicated PhysX card using a full PCI-E x16 slot? Isn't that what NVIDIA is selling with their Tesla cards? (although I haven't been able to see a single card out on the internet for sale).

[LordJuanlo]

Why should they? PhysX is a propietary technology created in the beginning by Ageia Technologies, and now it's property of NVIDIA... they can do whatever they want with it...

Because they are clearly saying it NOW. Of course they can do whatever they want, they bought AGEIA. But if they didn't inform about the limitations before, and they do it now, it's because it was a mistake.

They said 'you need an nVidia GPU for PhysX', ok I install one... together with an ATI card for graphics.

Today they say 'your system can't have non-nVidia GPUs installed for PhysX to work'... well they should have said it before.


No big deal today, because older 185 drivers work pretty well, but future PhysX drivers will probably require a 190 driver...

[walterman]

With some luck, we could see a new physics engine based on DirectX 11 Compute Shaders that will run on any GPU. Also, i hope that DX11 CS will finally unify all this CUDA/OpenCL/... mess, with just one standard.

[jolle]

NVIDIA's PhysX shortsightedness angers developers (http://www.brightsideofnews.com/news/2009/9/28/nvidias-physx-shortsightedness-angers-developers.aspx)



That's just pure bull****... isn't it easier to say they're not allowing any system with two different graphics cards powered by Lucid's Hydra chip to have PhysX because they will be losing some cash due to motherboard manufacturers relying on the Hydra chip instead of having to pay a fee for a SLI certification?

Hah...
What did Hydra have to do with this? wasn´t this the old disable PhysX when a Non NV card is rendering?

Couldnt you still use 2x NV cards with Hydra instead of SLI?
Or 2x ATi cards.
Or 1x ATi card + 1x NV card, just not with physX running on the other board.
Using a dedicated NV board for PhysX, wouldn´t be using Hydra anyway.

With some luck, we could see a new physics engine based on DirectX 11 Compute Shaders that will run on any GPU. Also, i hope that DX11 CS will finally unify all this CUDA/OpenCL/... mess, with just one standard.
OpenCL runs on anything, DC is windows only.
I´d imagine OpenCL would be the better standard, specially for crossplatform stuff. (wii, PS3 etc)

[Q]

As much as I like Nvidia cards, the company has a real knack at telling half-truths and omitting information. This situation takes me back to the "sure the 6800 Ultra can accelerate HD video" disaster.

[JH24]

I can understand Nvidia's position on this, but as stated above, they should have done that from the beginning and now they've created unncessary confusion. Not a good move for your image, Nvidia.


Of course, with the HD 5800 series, Nvidia was probably worried many people would buy an AMD card and use their previous Nvidia cards for Physx. Again, I can understand that, as I was actually planning to use my current Nvidia card as Physx card with a HD 5850 in a new computer. Oh, well. I can throw that idea out of the window now, I guess.

[hell_of_doom227]

Since i am coming from Nvidia world, i had chance to compare single HD5870 vs SLI 280GTX setup and in real gaming regarding to Physx. My conclusion is that Physx is not much of a deal, you can live without it. It doesn't really add to the gaming experience. It would be nice to have it, but it's nothing essential.

[jolle]

For a variety of reasons - some development expense some quality assurance and some business reasons
Translated:
We payed for it, there could be tech issues and we´d rather not expand testing configurations, and we really just dont want to support it.

[hell_of_doom227]

I think this smells like another Court trial. Since Nvidia sells and advertises card as physx, somebody will sue Nvidia for disabling it just because competitor card is detected.

[LydianKnight]

What did Hydra have to do with this?

Didn't know there already was a disabler for non-NVIDIA cards in regards of PhysX, but... it's a bit suspicious NVIDIA is making this silly statement now the Hydra chip is so close to market, sounds like a way to say 'don't even think you're gonna run one of our cards with the 'others''

[LydianKnight]

Couldnt you still use 2x NV cards with Hydra instead of SLI?

And you think the people at NVIDIA will just sit on their sofas looking how motherboard makers just opt to skip the SLI certification licenses in favor of Hydra? Bet not, or at least it wouldn't be that strange if they try to counter-measure that with some sort of software probe or method to disable PhysX if something 'strange' is detected...

[Vorgus]

Physx is an open software standard any company can freely develop hardware or software that supports it.

Personally ATI should just pay Nvidia some sort of royalty so they can use PhysX. I know I want it in my future games.

If it's an open standard, they don't have to pay anyone any thing. Nvidia is only hirting them selves. I mush prefer to buy from companies who inbrace and follow open standard policies. But because corp. greed, that is hardly ever an option.

Hydra is much better then ether sli or xfire because it not only lets you mix brands, but vintages as well. If ether nvida or ati wants to continue pushing their propritary systems, they will at least need to make possible to combine cards of different core chips which Hydra lets you do. Get a board with crappy on board video, latter you get a good plugin card, you can still make that on bard chip help out even if it's just a little.

[hell_of_doom227]

Regarding to Lucid chip it will be really interesting. I wonder what happens if you have SLI and Lucid acting together. Perhaps SLI and Crossfire being sort of 'overwritten' by Lucid? It smells like another Court Case.

[jolle]

And you think the people at NVIDIA will just sit on their sofas looking how motherboard makers just opt to skip the SLI certification licenses in favor of Hydra? Bet not, or at least it wouldn't be that strange if they try to counter-measure that with some sort of software probe or method to disable PhysX if something 'strange' is detected...
Yeah I wouldnt be suprised if they did something.
Im just saying, as it is now, and with any relevance to the link, it should work fine.
Using ATi card for rendering and a NV for dedicated physX processing wouldnt use Hydra anyway.

But sure, they could prolly step in and try to rain on the parade, or buy out the company.
On the other hand, better multigpu could help sell more GPUs, some people shy away from SLI or Crossfire, but if its more seamless, more might go for it.
Depends on their math on which way would be more beneficial, sell more mobo chipsets, or sell more GPUs (which would be split between them and ATi ofcource).

Not being able to mix and match ATi and NV cards does NOT seem like a huge loss for Hydra setups though (even though you still CAN, for rendering), seems like asking for trouble to have both drivers installed, and you got discrepancies on how filtering is done, what AA support is offered etc.
Might even be so that you´ll notice which object in a scene has been drawn on which GPU, dunno.

Also the entire solution will have to prove itself aswell.
Working seamless with any title, API version and driver, they´re saying it should, but it´ll be interesting to see reviews and tests of this stuff.
If it works great, I wouldnt be very suprised if either AMD or NV bought them

[Muppet]

If it's an open standard, they don't have to pay anyone any thing. Nvidia is only hirting them selves. I mush prefer to buy from companies who inbrace and follow open standard policies. But because corp. greed, that is hardly ever an option.

Hydra is much better then ether sli or xfire because it not only lets you mix brands, but vintages as well. If ether nvida or ati wants to continue pushing their propritary systems, they will at least need to make possible to combine cards of different core chips which Hydra lets you do. Get a board with crappy on board video, latter you get a good plugin card, you can still make that on bard chip help out even if it's just a little.

Only if Hydra works. It's a great idea but i'm not sold on it until I can at least see it, or some benchmarks

[Atomizer]

supposedly the main advantage of integrating the PhysX middleware into graphics cards through the use of some specific codepaths and hardware bits is the bandwidth required for adequate deployment of physics calculation data, the original PhysX processor by Ageia was tied to a PCI slot with the possibility of having a PCI-E x1 slot card? that leaves you with 250MB/s of raw data, and that's without the bus overhead due to the 8b/10b data encoding, so you have a raw data rate of 80% of that speed.

A graphics card like the upcoming GT300 has something like 286GB/s if I remember correctly... so it's like 1000 times faster (with the same 20% penalty for data encoding, I guess) so makes sense on having the graphics card do the job.

About a dedicated PhysX card using a full PCI-E x16 slot? Isn't that what NVIDIA is selling with their Tesla cards? (although I haven't been able to see a single card out on the internet for sale).

Actually, thats not the main thing as I understand it, the problem with a dedicated physx card, is I am pretty sure things have to go from CPU->PhysX->CPU->GPU, where as with GPU physics, all it does is CPU->GPU, the GPU calculates physics, applied directly to the mesh, then renders it.
Of course, this is only for the fluff physics, any physics that that requires game feedback, like a projectile colliding with a monster/player that needs to do damage, or some other scripted effect, such as exploding on impact, will need to go back to the CPU, which is when the extra bandwidth will help.
Though, with CUDA, or now with DirectCompute, it should be possible to keep everything to do with physics on the GPU, completely bypassing any transfers between hardware.

And as mentioned, DirectCompute will help unify things now, so more games will take advantage of it.

Also, in terms of non-propietary physics...Havok? Easy to see something like that using OpenCL or DirectCompute now.

[LordJuanlo]

I hope nVidia will port PhysX to OpenCL or DirectCompute instead of CUDA. Otherwise it will be dead sooner or later, and this would be really sad, now that games like Batman are doing a nice use of it, it does not affect gameplay but it makes it look much, much better

[Muppet]

I hope nVidia will port PhysX to OpenCL or DirectCompute instead of CUDA. Otherwise it will be dead sooner or later, and this would be really sad, now that games like Batman are doing a nice use of it, it does not affect gameplay but it makes it look much, much better

As long as Nvidia still has the majority Graphics card sales. I don't think PhysX can die. That is unless Nvidia wants it to.