Unigine is a real game engine, but Heaven is a benchmark. A couple of portions of the benchmark (the dragon is one, and I forget the other) tessellate to an unreasonable degree. Surely you can see the problem with tessellating to the degree that most of the triangles you generate are simply discarded and never used for a single pixel--and not because they're behind some other object, either. A game developer who intentionally codes that into a real game and won't fix it should be fired for incompetence. You can do things like that to artificially hurt performance in a synthetic benchmark, but you don't do it in a real game.

You can have synthetic benchmarks built on real game engines. Of Cryptic wanted to, they could release a benchmark for Champions Online that has 200 mobs scrambling around and fighting each other. Performance in that wouldn't be indicative of real gameplay, as Cryptic isn't dumb enough to put something that gratuitously overtaxing on hardware in the game.

Actually, if you really want to, you could try typing "/renderscale 4" (without quotes) to slow your frame rate to a slide show. One could benchmark various cards while moving around in Champions Online in the live version of the game like that, and the results might be interesting in an esoteric sense. But it wouldn't be representative of real gameplay. That would really just be a synthetic test of how much video memory you have, and how well your card can cope with needing vastly more than is available.

The excessively tessellated parts of the Heaven benchmark are like that. Yes, they're built on a real game engine. But they're not relevant to performance in real games.

-----

I realize that there's a "consider the source" rejoinder to the "AMD says that..." bit. But if AMD (or Intel or Nvidia) comes out and says that 16x SSAA isn't a good idea in modern games, that doesn't mean it's wrong. It's transparently obvious that it's true. One could perhaps quibble about the 16 pixels per triangle figure. Maybe one person would say 10 is better, and another would say 20 is more reasonable. But surely less than 1 is just plain stupid.

The GeForce GTX 570 has now launched, so I've added it to the list at its MSRP of $350. I've added the GTX 580 to the list as well, but I'm still not buying it as a good value at $500. The GTX 570 offers about 85% of the performance of the GTX 580 at 70% of the price.

Furthermore, with the Radeon HD 6950 and 6970 coming next week, any argument of "but it's the fastest single GPU card on the market!" is likely to be very short-lived. I expect the GTX 570 to still be a decent value at $350 after Cayman launches, but the GTX 580 will simply be laughable at $500.

Some GTX 570 reviews did say in the conclusion that Cayman was going to release next week. They didn't put it that explicitly, as they are under NDA, but they pretty strongly hinted that there was important competition for the GTX 570 and 580 coming next week, and there's only one thing that could possibly be. And it's not Larrabee.

Sorry if this is the wrong thread to ask this in, but here goes:

All things being equal, if I went from my current GTX 260 to a GTX 460, will I see a worthwhile increase in performance? Right now, I can run CO on everything *except shadows* up to high at about 30-50FPS. Will I be able to turn shadows up to med or high, and still see an overall increase in performance, or am I better off holding off?

Some general info for my PC:
Athlon II X3 445 (3.1Ghz per core)
4Gb DDR3 RAM (DDR3 1333 IIRC)

Thanks!

A GeForce GTX 460 1 GB is faster than a GeForce GTX 260. But it's not a lot faster. Do you really want to pay $200 for something that is only 30% faster or so?

The natural modern successor to the GTX 260 is the GeForce GTX 570 that launched yesterday. AMD's competitor for it, the Radeon HD 6950, is coming next week. The GTX 570 is more than twice as fast as a GTX 260, and the Radeon HD 6950 likely will be, too. That's enough for an upgrade to be worthwhile--but it might also be enough to be out of your budget.

Well, that's certainly some food for thought. Do you know anything about the GTX 560 which is due out soon? Anyway, I'll probably hold off for a bit, then...

Thanks!

I haven't seen any credible rumors on the GeForce GTX 560. Instead, there is speculation which makes sense and is likely to be true. There is likely to be a GeForce GTX 560 early next year based on a "GF114" GPU chip that is a base layer respin of GF104, the way that GF110 was a respin of GF100. It will basically be what Nvidia meant to deliver as the GeForce GTX 460, but couldn't because their yields were awful. I'd expect a fully functional GPU chip clocked somewhere around 750-800 MHz, with memory clocked around 1 GHz. New features are unlikely; at best, there could be something that was supposed to be in GF104 but didn't work right and had to be disabled in drivers.

The GeForce GTX 560 will probably be comparable in both performance and price to the Radeon HD 6870 that is already on the market. It will have significantly higher power consumption than the 6870 (e.g., a TDP around 180 W), but not high enough to be a problem--and markedly lower than the GeForce GTX 470 that currently holds that price and performance position in Nvidia's lineup. After the GTX 560 launches, someone looking for a nice $240-ish card will easily be able to go either AMD or Nvidia and get a good deal from either side, unlike now, where AMD clearly wins that market segment because of the GTX 470's problems with temperature, noise, and power draw that arguably violates the PCI Express specification.

But if that's the level of performance you want (maybe 60% faster than what you have), then there's no real reason to wait unless you're an Nvidia fan or don't have the money yet. You can get a Radeon HD 6870 now.

If you're hoping for price drops, it's going to be a while. The GTX 560 will perform comparably to a Radeon HD 6870, but will probably cost more to build than the much more expensive Radeon HD 5870. Nvidia isn't likely to start a price war from that position, as they'd quickly price themselves into losing money on every card they sell. AMD already brought down the cost of a given level of performance in that market segment with the 6850 and 6870, so I don't expect them to cut prices too much more, either.

Next year, both AMD and Nvidia will move to 28 nm HKMG process nodes, and that should bring down costs considerably for a given level of performance. When those new cards come out, then you'll see the price drops. AMD's cards, probably the Radeon HD 7000 series, of the "Southern Islands" generation will probably be nearly the same architecture as their "Cayman" cards, the Radeon HD 6950 and 6970, that launch next week, so you'll kind of be able to extrapolate and know what to expect--well, if you can guess specs like the shader count, that is. Nvidia's cards will be a new Kepler architecture. Those might be a GeForce 600 series, but Nvidia increments the first digit whenever they feel like it, so it doesn't necessarily correspond to a generation of parts. Performance characteristics won't be known until those launch. Hopefully, Nvidia will finally have an architecture that can compete with AMD's on performance per mm^2 (as they haven't since RV770 launched) and performance per watt (as they haven't since the switch to 40 nm).

That's largely waiting on TSMC and/or Global Foundries to get the process nodes working, and then AMD and Nvidia to figure out the quirks of the process nodes well enough to make GPUs on them with mass production-worthy yields. If you want to be optimistic, such cards could come next summer. If you're more of a pessimist, they could be delayed all the way into 2012. Well, Nvidia might take a while to release Kepler, as they could run into architecture glitches once they've got the process node down. Trying to move to a new process node and a new architecture at the same time is very hard--which is why Intel adamantly won't even try it in CPUs, and AMD tries to avoid it, too. That's probably also a major factor in why Fermi was so late.

I've now added the Radeon HD 6950 and 6970 to the list, at their respective MSRPs. Their TDPs are made up numbers from me, as typical gaming power figures that AMD offers are a lower type of estimate than the traditional TDP, while maximum board power as enforced by PowerTune is a higher type of estimate.

It's worth noting that in order to get something faster than a Radeon HD 6870, you end up getting significantly less performance per dollar.

The problem is that the program in question, well that's the game you're playing, has to actually use all your four cores, but most games are not programmed that way. I'm less sure about that, but I believe that the OS also needs to be able to put all cores to work for a specific task. Correct me, if I'm wrong on that point. That would be the reason, why only 2 of your cores actually show up working when gaming. Under Linux, if you have a program that can be split into several tasks, you could adress all available core via xargs. Example: converting several audio/video files or writing random numbers on a partition (you would have each core generating random numbers for a specific part of the partition and write the data to it). Well that doesn't work with games. At least I'm not aware of a way to do this, but maybe someone else knows more.

As has already been mentioned, the GPU is way more important than the processor for the game to run smoothly, you could use a pretty crappy CPU and still run some nice games, if your graphics card is up to the task.

For a game to run smoothly, the processor is very important. A bad processor can't run the game smoothly at all.

For a game to run smoothly at high graphical settings, the video card is very important. A slow video card can probably still run the game smoothly at low graphical settings, but not at high ones. A bad enough video card (say, really old integrated graphics) won't be able to run the game smoothly at any settings.

Champions Online certainly benefits some from more than two cores. Performance isn't just restricted by how many cores the program can use, but also by how well it can distribute the load between the cores. Suppose that a game divides the processing work into 6 threads such that one does 50% of the computations and the others others each do 10%. If you have a quad core processor, Windows 7 will spread out the threads so that it uses all four cores. But you'll be constrained by the one core that has to do half of the work, and the quad core processor won't perform much better than a dual core that put the five small threads all on a single core--if Windows is smart enough to do that, which it might not be.

I'd suspect that the person claiming two of his cores are maxed out by Champions Online while the other two go unused doesn't really have a quad core processor. A dual core processor with hyperthreading will show four cores in Task Manager, but there are only two real cores there. Champions Online doesn't take advantage of hyperthreading, which is why the two virtual cores go unused.

I've added the Radeon HD 6950 1 GB and GeForce GTX 560 Ti to the list. I've also reduced prices on the Radeon HD 6870 and Radeon HD 6950 2 GB to reflect AMD's price cuts. I've likewise reduced the recommended prices on the Radeon HD 5850 and 5870 to keep them in line with the other new prices, though those cards seem to be disappearing.