Related articles

• Now let’s turn to a typical mid-range PC. Here, we use a (simulated) Intel Core Duo with clock rates ranging from 1.8 GHz to 3.0 GHz. This encompasses a range of processors from the dual-core E2000 all the way through the E6850. Likewise, the Core Duo Series E8000, E7000, and E5000 also produce similar results at similar clock rates, so that we can summarize results from a huge number of target CPUs with our tests. In addition, we don’t expect the smaller caches in lower-end models to make much difference, nor to see much impact from the differences between 65 and 45 nm processes. AMD CPUs also scale similarly at increasing clock rates. The AMD model we tested in the previous round at 2.6 GHz matches up in the middle of this range, somewhere between the 1.8 GHz and 2.0 GHz Intel dual-core models. Our heavily-overclocked graphics card falls somewhere between a Radeon HD 4850 and a Radeon HD 4870, and is generally in the same performance league as a GeForce 9800 GTX+. This is pretty close to an average gaming PC nowadays (as represented here in the Gaming PC 1 configuration). We also tested on Vista in this case, so we could compare DirectX 9 and DirectX 10 results. Test 1: Average Frame Rates for DirectX 9 vs. DirectX 10 and GPU vs. CPU PhysX Average frame rates were our next area of interest, along with outliers on the low side because these represent frame-rate hiccups that are most likely to interfere with game play. Once again, we present cumulative values for our three game scenes and animated sequences and produce some interesting results. Overall, graphics settings were upped to middle values, because we couldn’t really play the games when settings were increased any further than that. We found it very interesting that when running Shader Model 4.0 under DirectX 10 at lower clock rates (that is, on less powerful CPUs), we observed higher frame rates than we did with DirectX 9. At 2.6 GHZ, both sets of values were nearly identical, whereas DirectX 9 appeared to benefit more from higher clock rates than did DirectX 10. At first, we were inclined to question these results, but further testing with other scenes only confirmed our initial findings. The obvious difference between GPU-assisted PhysX versus CPU calculations isn’t as noticeable in game play as it is on the graph, because fast-motion sequences usually occur at higher frame rates. After a while during game play, one learns to sense which scenes are most likely to act like speed bumps. Test 2: Maximum Frame Rates: DirectX 9 vs. DirectX 10, and GPU vs. CPU PhysX Next, we evaluate maximum frame rates. Here, the measured values corresponded perfectly to our subjective impressions. The same picture is painted here: weaker CPUs benefit most from DirectX 10 as compared to more powerful ones with higher clock rates. These results either demonstrate the benefits of offloading physics calculations onto the GPU or indicate that the differences really aren’t as great. The resolution chosen also plays a role when it comes to frame rates, but we never found a situation where scenes that were fluid and smooth in 1280x1024 became unplayable at 1680x1050. Frame rates only decrease somewhat, so that you can play reasonably fluidly on a 20” to 22” monitor. ATI cards lack PhysX hardware support, so they don’t do quite as well in their overall evaluations. We tested a Radeon HD 4870 with 1 GB of graphics RAM, using CPU physics calculations and Shader Model 3.0, and it performed nearly the same as the Nvidia card. Also, it did not slow down with AA enabled. The Radeon HD 4870 wasn’t exactly trouble-free though, and required a small trick to fix—more on this later. Playability With an Intel dual-core processor at 2.4 GHz or higher or an Athlon 64 X2 at 2.8 GHz or higher, and a mid-range $100 graphic card or better, Cryostasis offers fluid, pleasing play as long as you don’t boost the graphics settings too high. Here, the advantage goes to the Nvidia cards, given higher frame rates that make the game more fluid and visually pleasing. We saw no big differences between turning GPU-assisted PhysX on or off during our tests. The shipping game is quite different in this respect compared to the demo.  Visuals As is usually the case at minimum settings, mid-range graphics settings similarly fail to impress us with graphical delights. Modest increases in hardware capability don’t do this game real justice. They can’t come close to doing what high-end systems can.

• The performance differences between ATI and Nvidia graphics cards are currently fairly tight and well-segmented—in other words, it’s easy for an enthusiast to hit a site like newegg.com with a budget in mind and, all else being equal, find the best card for the money. Case in point: you tell me you have $160, I’ll tell you to get a GeForce GTX 260 Core 216 and cash in on the rebates. You tell me you can spend $180, I could tell you to grab a Radeon HD 4870 with 1 GB. But all else is not equal, so recommendations aren’t quite that cut and dry. Nvidia preaches the gospels of PhysX and CUDA, with an occasional verse about GeForce 3D Vision. ATI sings the hymns of DirectX 10.1 and Stream. Depending on which company you put your faith in, one of those two messages is going to sound a little sweeter. Right now, the two are in full-scale conversion mode, trying to get everyone they can to pitch in their tithing for a little gaming salvation. Just as PhysX and CUDA are starting to take hold with high-profile titles enabling support, so too are software developers paying more attention to DirectX 10.1. As a superset of DirectX 10, DirectX 10.1 includes a handful of quality-enhancing features that, in some cases, will run on DirectX 10 hardware, but at a performance hit. For instance, the Gather4 function fetches four samples (2x2) where a DirectX 10 part would only be able to fetch one. The result should be more realistic shadow maps and better performance. The Stalker: Clear Sky demo lets us test that theory with a toggled check-box to enable or disable DirectX 10.1. Our first test pits all of the Radeon HD 4800-series cards in this story against each other, without MSAA for alpha-tested objects enabled. At 1920x1200, performance is fairly similar across the board, with all test cases except the Radeon HD 4870 X2 demonstrating small gains by moving to DirectX 10.1. We then turned on 4x MSAA for alpha-tested objects and re-ran the numbers, this time at 1680x1050, in an attempt to maintain somewhat reasonable frame rates. This time, a majority of the test cases show DirectX 10.1 incurring a small performance hit. Is the performance tradeoff worthwhile? For that, we’ll need to make an image quality comparison from within the game itself. Up top you’ll find the screen capture provided by ATI, right up against a wall, demonstrating that the DirectX 10.1 shadows are softer and arguably more realistic. I ran all over the opening area of the game trying to find a clear example of the difference made by DirectX 10.1 shadows and just couldn’t come up with an indisputable best-case scenario. Even by reloading saved points, it was impossible to generate the exact same scene twice. Nevertheless, given the option of enabling DirectX 10.1 and not seeing a significant performance hit, you might as well turn the feature on. With the Game Developer’s Conference recently past, ATI had a handful of DX 10.1 titles to discuss during its briefing, besides Stalker. Tom Clancy’s HAWX looks like a fun title with DX 10.1 screen space ambient occlusion and accelerated Gaussian shadows. There were two other lesser-known titles, plus the UNiGiNE game engine, on which several upcoming titles are purportedly based. As of right now, DirectX 10.1 isn’t making a huge impact, but because it will become a subset of DirectX 11, you can expect the extra features being enabled right now to work moving forward in Windows 7. The same couldn’t be said for the disruptive shift from DirectX 9/XP to DirectX 10/Vista.

• The Scientists' Opinions on Gaming Physics

  Physics has always been part of our games, as the term describes how objects relate to each other and their surroundings. But to render physics in a realistic way, a large number of extremely advanced calculations is required. This demands serious horsepower, which puts more and more strain on the CPU. And as the CPU already handles a lot of other tasks, the idea of a separate hardware unit to take care of physics - just as the GPU handles graphics - has been around for quite a while. Based on that thinking, Ageia created and released their PhysX card. Nvidia and ATI, though, have a different take on the problem, saying that physics can be done just as well, if not better, by adding another GPU to the equation rather than a separate physics processing unit (PPU). So it seems that opinions differ concerning which technique handles physics best. That's why we here at THG had a talk with a couple of scientists, in an attempt to try to shed some new light on this subject. Is physics really that hard to do? Why? And which technology works best? Join our discussion on this topic Speak out in the Toms's Hardware reader survey!