[Tyrian]

I've been looking at getting a lower end graphics card for my system to serve as a dedicated PhysX card.  I know they're not in widespread use yet, but it's something that I've been wondering about.  I've done some Googling on the topic, but I can't really find an answer that solves my question. 

My question is this:  How critical is the GPU's onboard memory size to the performace of the card?  (I.E.:  If I had two EVGA GTS 250s running as PhysX cards, and the only difference between them was a 512MB memory bank vs. a 1GB bank, would the card with the larger memory space attain better performance, all other things being equal?)

[Nuke]

supposidly you can run physx off of a card that you also run graphics on. but i cant really name any games that use physx.

[BloodEagle]

I was under the impression that all modern NVidia cards have (and automatically use) built-in PhysX chips. And I've never even heard of using older cards that way. Is it actually possible to do that, and would it be effective?

Regardless, to my knowledge (see: outdated), physics processing is mainly speed over size, so there shouldn't be any real difference between the two cards if they're used specifically for that purpose.

------

but i cant really name any games that use physx.

Unreal Tournament 4(?) "uses" it.  And I've seen a number of recent game demos that install (without telling you) the PhysX software. :/

[Thaeris]

UT4?

Last I checked there was UT3...

[CP5670]

There are several games that use it now, although I can't think of any game where it's actually desirable. Either it does nothing noticeable or it causes a big performance hit.

I don't know how much a dedicated card helps with the performance, but the memory size will hardly make any difference at all. It uses shaders on the GPU. Don't bother buying a card for it unless you already have an extra one lying around.

[Ghostavo]

I was under the impression that all modern NVidia cards have (and automatically use) built-in PhysX chips. And I've never even heard of using older cards that way.

IIRC, Nvidia cards use CUDA to do PhysX processing, so any graphics card that can run CUDA, can run PhysX. There is no special dedicated hardware needed to enable PhysX. Theoretically, modern ATI cards could also run PhysX through Stream, and I remember some talk about it a year or so ago, but it seemed to have never materialized as far as I know.

[Thaeris]

Given everything I've heard about ATI cards, why the heck would you want to get one of those anyway?

[BloodEagle]

Given everything I've heard about ATI cards, why the heck would you want to get one of those anyway?

loltroll!

[Thaeris]

Troll?

Please do elaborate... I'm quite serious. I've not heard one good thing about ATI cards on this forum to date!

[Herra Tohtori]

For one thing, Radeon HD5 series is already on the market and doing pretty well, GeForce 300 series is not.

Basically, in computer world different generations of hardware can alternate in which is better. Sometimes NVidia makes bad decisions that affect the customer poorly, sometimes it's been ATI. Sometimes it's Intel in processors that is better, sometimes AMD. Currently, Intel CPU's and AMD GPU's are the better option than AMD CPU's and NVidia GPU's respectively.

Looking at NVidia's GPU's, the 8800 GT was one of their latest cards that offered better performance to price ratio than concurrent Radeon cards, unless I'm much mistaken. The 9 series and even 200 series hasn't been up to the performances of concurrent AMD cards (your mileage may vary, this is based on some reviews I've been reading, and I can't really remember all the sources to cite here). Additionally like I said, NVidia's new Fermi or G300 series of GPU's isn't even ready yet, much less in production, so AMD cards are gaining headway on the market of latest generation GPU's and they are apparently doing rather well in reviews. NVidia is going to enter late on this generation of GPU market and it is likely going to do it with cards with lower performance/price ratio than AMD's equivalent versions. But like said, it's a cyclic thing, some generations of ATI cards are not so good (HD1000 series for example with it's infamous GLSL problems with FS2_Open comes to mind) and some generations of NVidia cards are not so good (GeForce 5000 series, GeForce FX...). Since they are already late, it also likely will mean that the first GeForce 300 cards on the market will be rushed and buggy, but that's just my conjecture on the matter. Time will tell how they will succeed. Maybe the delays will mean the next NVidia cards will be awesome and win, but I doubt it.

ATI/AMD cards are also doing better and better on Linux side as the drivers are developing quite nicely, and from a pure open source perspective, AMD cards are better choice than NVidia cards which only have the restricted, proprietary drivers available as opposed to open-source AND proprietary drivers for ATI/AMD cards.

[CP5670]

Not only is Fermi very late, but Nvidia has reportedly also stopped production of most of their existing higher end cards and the ones being sold are all overpriced as a result. With the 5800s now becoming widely available, there isn't any reason to get an Nvidia card over $150 now.

I think the one advantage Nvidia has is the superior third party utilities. Both companies' own driver control panels are crappy, but nHancer and Rivatuner are more powerful and more frequently updated than ATI Tray Tools.

some generations of ATI cards are not so good (HD1000 series for example with it's infamous GLSL problems with FS2_Open comes to mind)

The 1900 line was actually very good for its time, and was easily the performance leader back then. ATI's biggest failure in recent years was probably the 2900 line.

[S-99]

Don't buy a physx card, they're useless, and a stupid idea (it's dying off as well). There's no reason why physics capabilities can't be integrated into your normal graphics card. When they do it, who knows, who cares, don't invest in physx; it's dying pretty hard. If you want physx, it will eventually be integrated into gpu's.

[Herra Tohtori]

The 1900 line was actually very good for its time, and was easily the performance leader back then. ATI's biggest failure in recent years was probably the 2900 line.

For it's time, I'd have to agree - with drivers that made GLSL actually work. It's more problematic in long term view though, as the support for it has been removed from the drivers and Vista/7 users have consequently reduced useability.

I was just picking examples close to home, but yeah, I didn't hear good things about the HD2000 cards in general either, most likely because they were contemporary with late NVidia GeForce 7 and most of the GeForce 8 series, if I recall correctly, and those cards were pretty nice indeed...

[asyikarea51]

I always thought Nvidia had the better Linux driver. I guess I'm very wrong, or something else...

Or maybe there's just... something... I don't like about ATI drivers (sometimes their cards too), but since ATI cards are the performers at the moment it's hard to pass them up...

[Herra Tohtori]

I always thought Nvidia had the better Linux driver. I guess I'm very wrong, or something else...

I think the NVidia's proprietary driver is still somewhat more compatible with Linux, but AMD actually provides the specs for their cards which makes developement of open source drivers possible, and that alone is a detail worth considering.

Or maybe there's just... something... I don't like about ATI drivers (sometimes their cards too), but since ATI cards are the performers at the moment it's hard to pass them up...

I have had several bad experiences with ATI products in general, but like I said, it goes in cycles.

[Nuke]

i tend to go with ati as far as mobile gpus go, but other than that im usually a nvidia user.

[S-99]

Zombie dead walking named ned.

Resurrected is what i mean. I've used linux for a long time. I was a big fan of ati since 3dfx died and when ati came out with the radeon line of cards. Their best cards to date i think was the radeon 9000 series. They were fantastic performers for their time and actively made people not buy nvidia.

But, back on topic. The proprietary ati driver for linux has always sucked. The proprietary nvidia linux driver on the other hand is the exact opposite. It's more complete, has more active support, and is easy to live with. Normally with the nvidia driver you just install and you're good to go, as compared to a not so similar experience with the ati driver.

The nvidia driver is of much higher quality than the ati driver. Although i do want to see an opensource nvidia driver (the nouveau project)...being able to use the same driver for old and new nvidia cards is ****ing great as well as other things.

The nvidia driver for linux is so much better that for many linux users, when buying a new computer they favor nvidia. Meanwhile, the driver for ati is more like the ****ing company's after thought that they barely remember to do anything about.

Now you know more about why ati/amd made their driver opensource and are making available more documents about their cards for devs. "Why not off load the driver work to other people who actually want to do it?", that's what went through ati's head.

Even though nvidia's driver is still proprietary and the ati driver is now opensource, the nvidia driver is still loads better. The ati driver which hasn't been opensource for very long has yet to reach it's full potential since becoming opensource (something previously proprietary going opensource doesn't make it 2000% awesomer over night, but it does make the potential for it 2000% awesomer over night).

[Fury]

Drivers aside what makes the move from ATI's part really awesome, are the open specifications for their cards. It's a move towards more open hardware in area where inner workings of said hardware always had been closely guarded secret. There's no longer need to waste precious time trying to reverse-engineer to get something done that isn't supported by proprietary drivers. I'm sure OpenCL will also see a boost on linux platform thanks to the open specifications.

However, you're wrong about ATI's motives. They had to do this because of their upcoming Fusion product-line, which incorporates both CPU and GPU processing on same physical CPU slot/socket. As CPU's are nearly supported out of the box, they need similar support for GPU's as they roll the products out. It'd be much more difficult to achieve if specifications were closed as they used to be and that would hamper Fusion sales. As we know, GPGPU capabilities are very much sought out and even though OpenCL, Stream and CUDA helps, open specifications helps even more. In short, it was purely a business motivation to sell their future hardware for platform where GPGPU processing will be sought after with big money, aka workstations and servers.

[Bobboau]

ATI drivers suck, I had a cheap nVidia card that worked fine, now I have an ATI card that cost 3 times as much and I've had nothing but problems, I'm hoping a new driver release will addres these issues, but my experience with ATI drivers tells me they will only get worse.

also, doesn't nVidia now own physX?

[Nuke]

like i said the only time i buy ati is when im looking at laptops. when i do that i look to see if the gpu has dedicated memory or if it had shared memory. and go figure both times i was shopping for a laptop in whatever price range i was looking at that had dedicated graphics memory, it was an ati.