I like my PC gaming and even though decent titles with decent Physx are rare, I do appreciate what Physx effects can add to a game. But, with almost nothing concrete - in regards to simple information - from Nvidia, trying to buy the right card seemed to be a gamble, literally. My GTX 680 can struggle to maintain decent frame rates @ 2560x1440 with high levels of Physx stuff, so I was hoping a dedicated card would smooth things out. I decided to go with the 650Ti SSC, on paper it looked more than capable and it wasn't silly money. A few facts and thoughts..... Mafia 2 Benchtest. 320.14 BETA Drivers, max settings, AA ON, Physx OFF, VSync Off, average FPS 2560x1440 GTX 680: 56.2FPS GTX 680 / GTX 650 (dedicated to Physx): 56.1 FPS GTX 680 / GTX 650 (dedicated to Physx) AA OFF: 93.1 FPS CPU Physx: 16.8 FPS You may think a 'Physx OFF' test is pointless, but I've heard people say that enabling a dedicated card in the NVCP can have a negative impact on performance. In the case of Mafia 2, it doesnt. At 1440p, the mediocre AA in Mafia 2 removes a large chunk out of the performance, almost halving it. And I know, my CPU isn't exactly cutting edge, but nor is it a slouch. Take what you like from the CPU result. I take this from it, CPU = Bad Physx. Let's say no more on the matter. Mafia 2 Benchtest. Max settings, AA on, Physx HIGH, VSync Off, average FPS 1920x1080 GTX 680: 46.7 FPS GTX 680 / GTX 650 (dedicated to Physx): 82.9 FPS I don't play at 1920x1080, but for those that do, and think that a GTX 680 is more than enough for a game approaching it's 3rd birthday, think again. Yes, 46.7 FPS is perfectly playable but just look at what happens when you throw a GTX 650Ti SSC in the mix. Virtually an 80% performance increase! I've seen many, many people state in forums that a dedicated Physx card is a waste of money, especially when running a GTX 680 or similar. This test states otherwise. 2560x1440 GTX 680: 37.5 FPS GTX 680 / GTX 650 (dedicated to Physx): 52.3 FPS GTX 680 / GTX 650 (dedicated to Physx) AA OFF: 78 FPS At 2560x1440 there is still a massive increase, basically 40%. But, with the added pixel count, turning AA off isn't such a big deal and 78 FPS average is very nice. With VSync and TB forced expect a solid 60 FPS. What's interesting here is that running high Physx with a dedicated card is almost free; with only a few FPS dropped when compared to Physx off, the extra card is doing it's job. Whereas the GTX 680 alone drops almost 20 FPS. An average of 37.5 FPS proves that the GTX 680 simply can't do Physx at the same time. The GTX 650 hit a maximum of 48% usage in Mafia 2. So definitely earning it's money. Anyway, that little benchtest session took longer than I wanted it to, time to talk simple. Alice Madness Returns utilises the dedicated card, but only maxing it out @ 14% usage, so hardly taking advantage of the extra cash spent. Removing the frame rate cap makes it nice to play, but the Pepper Grinder smoke effect, causing the game to slow down, isn't helped by the GT 650. What's strange here is that the GTX 680 is still maxing out while the GTX 650 sits about doing very little. What's that all about? Is the game not offloading the Physx properly? Borderlands 2 looks bloody amazing maxed out with Physx set to high. And thankfully the GTX 650 does make a significant difference. I play at a fixed 60 FPS and I mostly get it, so higher frame rates don't interest me. What I do like is the Physx intensive scenarios playing out far more smoothly. The GTX 650 showed a maximum of 42% usage whilst playing the 'Boom and Bewm' mission. With the GTX 680 not having to worry about Physx, the game runs smooth, the GTX 650Ti SSC showing that it can keep up with the rendering speed of the GTX 680. I took a look at the Batman AC benchtest: Running @ 2560x1440, with DX11 enabled, yielded a 25% increase in performance with a dedicated Physx card, the GTX 650 helping the game to almost sit at a solid 60 FPS. Without it, the average was 42 FPS, with a minimum of 15 FPS. Running in DX9 left the GTX 680 with enough headroom to sort the Physx out, too. So the GTX 650 only makes a difference in DX11, showing figures of usage in the high 30s. I was going to say that's a shame considering DX11 in Batman AC is not ideal, but since updating my drivers to the 320.14 BETA I don't see anymore flashing textures. Yes, the odd jerk and stumble is still present - albeit not so bad - but DX11 plays almost acceptably now, at least for those that care. So, there you go. I apologise to those who think I've gone on and on, but hey, you didn't have to read it all! The thing is this: Physx is almost taboo on the web; Nvidia doesn't say much about it, even though it's clearly still serious about it. Trying to find out if it's worth the money is one thing, but then having to choose the correct card for the job is another. Is it the CUDA count that matters? The clock speed? The architecture? Nobody seems to know for certain. Does Nvidia even know? If it does, it's keeping it a secret for whatever reason. Nvidia pushes SLI and 3D Vision in a big way - which many consider not worthwhile - but Physx gets barely any mention. For these reasons I wanted to share my experience and I hope this helps someone and maybe even gets a good discussion going on this niche. Simply put, if you like Physx then a dedicated Physx card is not a waste of money. One thing that has surprised me: the amount 1080p can benefit from a dedicated Physx card, even when a GTX 680 is the main card, is incredible. Is the 650Ti overkill for Physx? I don't know. Ask Nvidia.