Phsyx a performance indepth look, with what I have (c:

[Kayden]

The world of graphics have always been always been a troubled sea to navigate, especially if you sail under one particular flag. No matter which flag you sail under however not many have said nVidia’s Phsyx acquisition hasn’t been without its troubles with performance, nVidia specific hardware to run it and then developers reluctant to jump on board because they believed it would limit the reach of their games to the PC market. These obstacles have not slowed the Phsyx monster from pushing forward, even getting a few hacker’s to force nVidia GPU and drivers to work alongside ATI GPU’s so they could run Phsyx, this was short lived but still proved it could be done if nVidia would loosen their reigns.

Whether you agree with them keeping this proprietary or not, it is now thus I decided to take a stab at running some benchmarks with my machine and with a 9600 GT I have laying around. This way I can try to give a contrast to see whether or not it would be best to pair a newer card with an older for Phsyx or just let the new one handle it. The games I chose for this are Mafia II and Batman AA, the stand alone benchmark I used is new it’s called Mars GPU Phsyx Benchmark it came out only last week. To quickly go over my spec they are Intel i7 950 (OC to 4 GHZ), Corsair Dominator 6 GB (OC to 1754), 2x EVGA 580 Water Cooled GTX580 FTW in SLI configuration (OC to 850 Mhz Core & 4196 Mhz Mem) and EVGA 9600 GT.

To start were going with Mafia II. An odd note everyone should be aware of when using Mafia II as a benchmark, when you do your first benchmark it will give your results and then do another benchmark right after it the difference will be about 10 to 15fps higher, this was consistent for several benchmarks so it is not a fluke. The only explanation I can give to this is caching, something is being precached after it has been ran the first time and it’s cleared when you exit the game, I have only included number from first run to prevent bias on the games part. All setting were Maxed at 1920x1080, AA was turned on but Vsync was not enabled to get the best possible amount of FPS. Here are the numbers

580 SLI Mode & 9600 GT Dedicated to Phsyx - High Phsyx: 42.3 FPS Rank C
580 SLI Mode & 9600 GT Dedicated to Phsyx - Med Phsyx: 77.0 FPS Rank B
580 Single GPU & 9600 GT Dedicated to Phsyx - High Phsyx: 40.0 FPS Rank C
580 Single GPU & 9600 GT Dedicated to Phsyx - Med Phsyx: 64.7 FPS Rank B
580 SLI Mode & 580 (2) set to Phsyx Not Dedicated - High Phsyx: 52.4 Rank C
580 SLI Mode & 580 (2) set to Phsyx Not Dedicated - Med Phsyx: 67.6 Rank B
580 Single GPU - High Phsyx: 41.3 Rank C
580 Single GPU - Med Phsyx: 58.4 Rank C
580 Single GPU - No Phsyx: 78.3 Rank B
580 SLI Mode - No Phsyx: 124.8 rank A

That game defiantly takes Phsyx seriously but it isn’t the only we have to test, here comes Batman AA 16xQ AA, detail level Very HIGH, Vsync NO and everything else Enabled. It should be noted when going to Medium Phsyx the game drops the Cloth Flags, about 40 to 60% of the smoke disappears, about 50% of the cob webs disappear as well and I was unable to judge the amount paper but it didn't seem like much, if any.

580 SLI Mode & 9600 GT Dedicated to Phsyx - High Phsyx: Min 43, Max 126, Avg 102
580 SLI Mode & 9600 GT Dedicated to Phsyx - Med Phsyx: Min 50, Max 182, Avg 115
580 Single GPU & 9600 GT Dedicated to Phsyx - High Phsyx: Min 46, Max 128, Avg 89
580 Single GPU & 9600 GT Dedicated to Phsyx - Med Phsyx: Min 49, Max 131, Avg 98
580 SLI Mode & 580 (2) set to Phsyx Not Dedicated - High Phsyx: Min 79, Max 214, Avg 137
580 SLI Mode & 580 (2) set to Phsyx Not Dedicated - Med Phsyx: Min 88, Max 231, Avg 152
580 Single GPU - High Phsyx: Min 64, Max 179, Avg 115
580 Single GPU - Med Phsyx: Min 65, Max 192, Avg 120
580 Single GPU - No Phsyx: Min 97, Max 227, Avg 156
580 SLI Mode - No Phsyx: Min 137, Max 379, Avg 262

The benchmarks are defiantly proving interesting aren’t they? Let’s move on to Mars really quick before we start to draw any conclusions or speculation. Since the Mars benchmark is so new and based on the Unreal tech there is no SLI support for it yet, thus all my testing was done with a single GPU running the GFX and either the other 580 is running the Phsyx or the 9600. There was also no way to change the amount of Phsyx being used in this benchmark

580 Single GPU & 9600 9600 GT Dedicated to Phsyx: Min 21, Avg 34
580 Single GPU & 580 (2) set to Phsyx Not Dedicated: Min 23, Avg 43
580 Single GPU - Phsyx: Min 22, Avg 52
580 Single GPU - No Phsyx: Min 48, Avg 75

That benchmark wasn’t as in-depth as I would have liked it to have been but it served its purpose for looking at a situation for people who only have 1 GPU and want to run another one for Phsyx.

The benchmarks show us a few confusing things, but let me explain a few things first. The term Phsyx doesn’t correlate to anything really specific to something happening in the game, like Mafia II’s benchmark doesn’t have smoke or cobwebs, just like Batman AA doesn’t have exploding cars and windows. This is just something specifically interacting with other objects nothing more, so it could be paper being kicked around or rocks from a stone wall being kicked around, this is why the numbers looks so different.

The benchmark for Mafia II shows that having an older midrange video card for a Phsyx card is a benefit because it mainly uses particles. The 9600 GT is dominant in the Med benchmarks but not High most likely because of it being an older card. If I had more confidence in the card, I would over clock it and see what I could get out of it but I have other uses for it so I am not going to push it.

The benchmark for Batman AA is where the 9600 GT gets steam rolled in all areas, I even noticed when I went to the 580 for Phsyx (single GPU or in SLI) that the FPS was much smoother. The launcher tells me that for the game to have High Phsyx it is recommended for you to have a 260 video card and a 9600 GT dedicated to Phsyx, I met the requirements for the Phsyx but it just wasn’t holding its own against the 580. This is because Batman AA uses much more complex Phsyx in the game then Mafia II. I had a 260 GTX from PNY and I still had a 8800 GTX WC, while waiting to find out if I would get my other 8800 back from BFG because of a seal leak, I played Batman AA with my 260 running Phsyx and the 8800 ran the GFX because I would get 50+ FPS with the 8800 as the GFX card but if it was the Phsyx the 260 could barely put out 30 FPS. The game benefits from having a very power GPU running Phsyx and it still shows here today.

The Mars benchmark did many of the same things that Mafia II, it did have some complex water but it was very brief. The FPS numbers also reflected just having a Phsyx GPU of any kind helped but the overall performance gain was not as significant going to the 580, I suspect this maybe a driver issue rather than a hardware one.

To sum this up, is it better to have a dedicated GPU if you’re running a mid-range card and have older model laying around? Yes, the performance gain may not be whole lot maybe only 20% but anything is better than nothing, just remember you may not be able to run Phsyx at high, just like you might not be able to run BF3 on High with a 560. Those running a single top end 580 GTX or 570 it’s probably in your best interest to just let your card do the Phsyx work and forgo a dedicated Phsyx GPU, the Batman AA showed the greatest improvement but Mafia II did have a slight gain without using one. For all the SLI users out there just let the 2nd or 3rd card be auto detected to run as a Phsyx slave do not dedicate it, you would just lose out on the benefit of having the SLI so it would be a lose/win situation.

 

[Rob Williams]

I can't believe that there hasn't been a single comment on this! Sorry man, I ended up forgetting all about it and didn't spot it when perusing the forums :-(

An odd note everyone should be aware of when using Mafia II as a benchmark, when you do your first benchmark it will give your results and then do another benchmark right after it the difference will be about 10 to 15fps higher, this was consistent for several benchmarks so it is not a fluke.

Unigine's Heaven benchmark has the same 'problem'. The second run might have a similar average FPS, but the minimum FPS is almost guaranteed to be much higher (an SSD might prevent this from happening, I am not sure).

Those running a single top end 580 GTX or 570 it’s probably in your best interest to just let your card do the Phsyx work and forgo a dedicated Phsyx GPU

I have to admit, I find it interesting that in Batman, SLI was a detriment more than a benefit. With a single GTX 580 and High PhysX, the game averaged 115 FPS, while with the same configuration in SLI mode, it was 102 FPS. Those sorts of differences are also seen at Medium PhysX also. This surprises me given that this is a game that NVIDIA pushed hard. You'd imagine a showcase game like this should be able to benefit greatly from SLI. That said, the performance all-around is still excellent so it's not even going to be noticed.

The same goes for the results where the 9600 GT wasn't part of the equation. Straight-up SLI performed far better than it did with the 9600 GT GPU being dedicated. This is sort of expected, I guess, as NVIDIA doesn't consider the 9600 GT to be that appropriate for PhysX computation nowadays. The company quicker recommends something on par with a GTX 260 (GTS 450 might be suitable also).

I guess the real lesson is that no two games are alike, and experimentation is the only real option if you want to make sure you are getting the best performance possible. It's too bad not all games were alike in this regard to make things simpler ;-)

 

[Kayden]

Unigine's Heaven benchmark has the same 'problem'. The second run might have a similar average FPS, but the minimum FPS is almost guaranteed to be much higher (an SSD might prevent this from happening, I am not sure).

I did not know that about Heaven benchmark, of course I only run it once and then close the App so I haven't seen it, I am going to check it out but that is good to know.

I have to admit, I find it interesting that in Batman, SLI was a detriment more than a benefit. With a single GTX 580 and High PhysX, the game averaged 115 FPS, while with the same configuration in SLI mode, it was 102 FPS. Those sorts of differences are also seen at Medium PhysX also. This surprises me given that this is a game that NVIDIA pushed hard. You'd imagine a showcase game like this should be able to benefit greatly from SLI. That said, the performance all-around is still excellent so it's not even going to be noticed.

The same goes for the results where the 9600 GT wasn't part of the equation. Straight-up SLI performed far better than it did with the 9600 GT GPU being dedicated. This is sort of expected, I guess, as NVIDIA doesn't consider the 9600 GT to be that appropriate for PhysX computation nowadays. The company quicker recommends something on par with a GTX 260 (GTS 450 might be suitable also).

I guess the real lesson is that no two games are alike, and experimentation is the only real option if you want to make sure you are getting the best performance possible. It's too bad not all games were alike in this regard to make things simpler ;-)

Another thing to keep in mind about the Batman AA is that the game recommends a 9800 GTX as a dedicated Phsyx card when at High. This makes me think that they have really worked on the efficiency of Phsyx on the higher end cards, which is fine if you can afford it but when you go to midrange even a few gen behind it's just going to hurt you. It is nice to see that if you put the money into high end cards your money is going further then just GFX but this can only be said in my exp with 580. I don't know how a 560 or anything will hold up, would be interesting to see. (c;

 

[marfig]

I'm terribly sorry I didn't reply to this before, Kayden. Inexcusable since you did a fantastic job and your work alone should probably have deserved centerpiece as full TG article. To my defense, I sort of bookmarked your post on the back of my head for later reading. I understood its importance and my own curiosity about it. But the time it showed up I was being busy and it eventually went to that place were all things on the back of my head go -- a big black hole with two orbiting stars; void and null.

These benchmarks, they sort of reinforce my uneasy feeling about Phsyx. The fact it is a proprietary nvidia-only technology is sure one of them. Trying to enforce such a thing can only make sense if the technology is indeed radical enough to make people purchase nvidia just to get the coolest effects on the block. Because it isn't that of an explosive technology... it becomes detrimental. The only thing that is currently keeping it going is deals between nvidia and publishers or developers. The market simply isn't reacting to it and... to put it mildly... a lot of money was spent developing a technology no one really cares much about.

But those benchmarks sort of confirm my suspicious that Phsyx is a technological dead end. You will excuse me I disagree with you on a couple of points, so here it goes:

1. All I see is in fact no important benefit to running a Phsyx dedicated card. The gain is insubstantial considering that the game will otherwise be running at a very acceptable FPS rate already. Anything that is put on top of that to take full benefit from Phsys is simply not much relevant, when one considers they just doubled their power consumption and may have had to make modifications to their cooling methodology.

Arguably as a game makes more and more use of Phsys the gains will be larger and some barrier could eventually be broken that could justify the adoption of another card. But what game would do that, when this is a proprietary technology we are talking about? Who on their right mind would want to sell their game to nvidia owners only?

2. Single cards seem to me the best approach. And again we need to be practical about it. A GTX 560 will run BF3 on high just fine if I can accept a minimum of 30 FPS. If I can't, I can move to another smaller monitor. I understand the rage is on 1920x1080, but neither I like 16:9 (16:10 is my thing), neither I actually need such a high resolution it that means getting better FPS and avoid being so close to the 30 fps danger zone. There's no SLI, no $500 card, no Phsys dedicated card, that can beat a well-placed resolution downsizing. For anyone with the money, shoot for it! But in terms of efficiency, it's pretty much wasteful except for the hardcore gamer. It's also not economically sound.

...

Phsys was sort of a disappointment to me. It was one of those technologies with a vast potential for exploration, but that ended up delivering so little. Not because there's something wrong with it (I don't think there's isn't, but I could be wrong), but because of how it was applied. A good idea on the wrong head, that's what it was. As if it hadn't been Newton the one thinking about gravity, but Jack the Juggler, who didn't have a clue on what to do about it other than add a couple more balls to his juggling act.

An open technology would have raised adoption and stimulate further developments (we could be having an entire different discussion around here, had that happen). But also, this technology could have had a very big impact on certain scientific fields, but instead was focused on games, gaming and gamers. Not the worst bunch, we are; But we couldn't give a flying arse either. As we actually didn't. So it's all one big waste of technology.