Sorry Captain, the CPU can’t push any more!

The full release of StarCraft II is nearly here and we are getting you prepared by detailing GPU performance across current and previous generation offerings. Will the new game require the likes of the GeForce GTX 480 to run or will your Radeon HD 4850 make it through? UPDATE: New ATI driver enables antialiasing options – fully tested inside!
UPDATE 8/2/10: Over the weekend ATI went ahead and released a new Catalyst 10.7 beta driver that allows for AA to be enabled via the control panel.  Check out the updated information for benchmarks and image quality comparisons!

Introduction and Overview

If you aren’t excited about StarCraft II, then you aren’t a real PC gamer.  Argue your points all you want about this being the exact same game with 3D models rather than 2D sprites (and you are mostly correct); the fact of the matter is no PC game has garnered as much attention in recent months as SC2.  The original StarCraft, and the Broodwars expansion, were easily my biggest time-sink in gaming through my high school years so I am fairly attached.  I am ready to jump back into the world that Blizzard has created and I think millions of gamers across the world are as well.

StarCraft II Performance Review - Even your mom can play (UPDATED) - Graphics Cards 34

Today I am going to show some of our performance results from the StarCraft II beta in order to prepare you for the release of the full retail game on July 27th.  I know that many of you that either didn’t make it into the beta or didn’t have time to try it out and might be curious how your system and GPU will stand up to the challenge.  I’ll take you through my observations on CPU limitedness, GPU scaling and even the question of anti-aliasing with StarCraft II.

The CPU May Be Holding us Back

For testing purposes we are using the SC2 replay system that allows you to save and playback an entire multiplayer match.  I played through several dozen matches, slaving away for our readers, to find instances of performance lag, low frame rates and what I would call your “typical” scenarios.  I narrowed my replays down to a pair that seemed pretty interesting, but each provided very different frame rates. 

One consisted of a modest battle between lower grade forces early on in a match and frame rates ranging in the 60-100 mark.  The second was a more mass-onslaught of Zerg and Protoss that brought our system further down in frame rate.  My first reaction was to use this second replay as our testing scenario for benchmarking but I found some interesting results when looking at the CPU utilization. 

StarCraft II Performance Review - Even your mom can play (UPDATED) - Graphics Cards 35

What you see here is a graph of the CPU utilization for the above mentioned replays.  You can see that even though REPLAY02 (the ‘heavier’ of the two) had noticeably more units on the screen (and 8 players total rather than 2 players) and a lot more battle data to deal it didn’t spike the processor any more than our more casual replay (REPLAY01).

Initial performance data showed that a GTX 460 wasn’t much slower than a GTX 480 on the REPLAY02 results however and that was something that definitely isn’t indicative of the relative GPU performance of the two cards.  Results on the more casual REPLAY01 scaled much better though with noticeably higher frame rates.  It seemed then that even though overall CPU utilization looked similar there was something keeping the scores in our heavier replaying from scaling the way we wanted. 

I decided to break it down further and monitor the utilization of each of the 8 threads our Core i7-965 was capable of:

StarCraft II Performance Review - Even your mom can play (UPDATED) - Graphics Cards 36

Here you can see there is one thread that is basically maxed out with just a few others hovering in the 20-30% utilization range in our REPLAY01 test.

StarCraft II Performance Review - Even your mom can play (UPDATED) - Graphics Cards 37

Replay02 doesn’t differ much – a single thread is pegged on our system with another thread in the 45% range and a couple of others around 10%. 

To me this is obviously indicative of a game that was built with single and dual-threaded processors in mind which is disappointing, but not all that surprising considering the varied audience that the game will draw in.  While you might think Blizzard could have utilized more of the available processing power for AI, battle stats, etc, they obviously chose not to focus on that and instead work on other areas to get the game out on the schedule they desired. 

For our use then the first, smaller battle-based replay fit the bill for what we wanted to test: GPUs.  With that replay the performance scaled quite nicely from GPU generation to generation and allows us to have results of interest for PC gamers.  I do intend to follow up after the full release of the game with some CPU scaling results as well, but with the beta being disabled today I don’t think we’ll have time before. 

« PreviousNext »