Designing Our Next GPU Test Suite

240 Comments

View All Comments

• xeopherith - Monday, March 15, 2010 - link

  I don't exactly think the games matter at all. I think the focus needs to be on running multiple GPU's for comparison, multiple resolutions likely including 1280x1024 and then higher widescreen resolutions, and eliminate the CPU as a bottleneck as much as possible by using games that really push the GPU rather than CPU.
  
  It would be nice to have a comparison between the GPU in both an AMD and Intel based system. Yes an Intel CPU will get a higher performance in most games than AMD but disregarding the CPU performance is the card reaching it's potential in each situation?

• Targon - Monday, March 15, 2010 - link

  One thing that you almost never see is a comparison of the same video card on different motherboards. For Intel processors, you have Intel and NVIDIA chipsets, for AMD processors, you have AMD and NVIDIA chipsets available.
  
  Now, it is very possible that a NVIDIA video card will work better on a NVIDIA chipset for example, or that Intel may tweak things to intentionally or unintentionally hurt the performance of AMD/ATI Radeon cards. It would be nice to see more comparisons that try to expose where a given chipset is the best choice for a given video card.
  
  So, for Radeon cards, use a Phenom 2 X4 965 processor based system and see if the performance is better with an AMD chipset compared to NVIDIA.
  

• Lockson - Monday, March 15, 2010 - link

  Two games that would be very relevant to me to be benchmarked are:
  
  -Supreme Commander 2
  -Dawn of War 2: Chaos Rising
  
  I'd prefer these, other strategy games are also welcome of course! I just post this because I sometimes feel strategy games are somewhat under represented.

• gentlearc - Monday, March 15, 2010 - link

  I think benchmarks are often times misinterpreted by those new to AnandTech. I believe many people look at a review with one game in mind. Often times it simply isn't possible to accommodate everyone. However, I believe it would be beneficial to investigate a standard framerate/experience you are trying to achieve. This can be a bit subjective, but I do believe on the more popular older titles most people aren't asking if it will be able to run the game, rather, they are asking how well it will run.
  
  Take the hardly ignorable WoW or Crysis. Benchmarks are just graphs without the knowledge of AnandTech there to interpret them for your new viewers. I believe you would gain a larger viewership if you dedicated a bit more narrative and explanation for these games. Contrary to the popular phrase, "but will it run Crysis," people want to more than just that. They want to know what tangible benefits they may see from this new card.
  
  I think it would be beneficial to show a few major games with different options enabled to show how the playing experience is improved. Options such as viewing distance, AA, shadows, and resolution.
  
  On that note, reviews linked to articles describing how to optimize certain games would be helpful. Many gamers have little knowledge as to why you get a certain FPS and they don't. Just as you reference many of your articles with an article, that will really help to spread the education on how and why you are conducting your benchmarks as such.
  
  Which leads to "optimising games." I don't believe you have to actually review a game or provide an in depth technical overview of each game. Instead, I see this as an opportunity for AnandTech to educate viewers about the many options within most games. Perhaps this is a place to encourage your forum to tackle the task of showing for each game what are the different tiers of gaming options users should select when they want to improve their gaming experience.
  
  

• slaveII - Monday, March 15, 2010 - link

  
  I would suggest that benchmarking popular games with built in benchmarking tools - such as Resident Evil 5 - would be useful since it is then relatively easy for a reasonable proportion of those reading the article to judge the relative benefit of the reviewed GPU to the GPU they are currently using. Repeatability at home is key to this suggestion.
  
  Also, I agree that a spread of game types is desirable to cover the interests of the maximum number of readers in a given review.

• MrJim - Monday, March 15, 2010 - link

  I would love to see a new simulator, perhaps Black Shark or Rise of Flight. Maybe not perhaps DX11 but interesting for us who like sims.

• Scali - Monday, March 15, 2010 - link

  I think that the next battle of GPUs will partly be fought on the GPGPU-front. Games are starting to use DirectCompute or OpenCL for post-processing effects, and nVidia has been using GPGPU-accelerated physics for a while...
  
  So I would like to see some GPGPU-related benchmarks in there. SiSoft Sandra has some, GPU Caps Viewer has some... You could probably also use the samples from the DX SDK, nVidia Cuda SDK and ATi Stream SDK (they contain OpenCL and DirectCompute samples aswell, to keep it on a level playing field).
  
  If an OpenCL- or DirectCompute-based physics solution arrives (Bullet?), it will be very interesting to do some benchmarking on that aswell.
  
  Aside from that, it will be very interesting to have some tessellation-heavy tests in there, as it seems to be the biggest difference between ATi's and nVidia's upcoming DX11 architecture, graphics-wise.

• LordSojar - Monday, March 15, 2010 - link

  I agree with prior suggestions of adding Aliens VS Predator (2010) as one of the DX11 test beds. It has excellent use of tessellation, and certainly can stress modern cards enough to get a good gauge of performance.
  
  STALKER: Call of Pripyat is another great DX11 to test, as it can really bring cards to their knees at full settings/high resolution.
  
  Synthetic benchmarks are useless to most users, and I disagree with adding them, as it will only distort some readers' views of actual performance. They are so dependent on system settings, tweaking, and overclocking to push figures up, they don't represent realistic performance in any way to most end users. They simply will make reviews take longer, and skew the data in some cases for most readers.

• Overmind - Monday, March 15, 2010 - link

  AvP3 because it has Tesselation.
  S.T.A.L.K.E.R. CoP - the updated Xray engine (with DX11).
  UT 3 - old but still relevant.
  CoD MW2, BFBC2 - two currently very played games.

• Nathelion - Monday, March 15, 2010 - link

  How about Shattered Horizon? Pretty demanding if you turn the eye candy up, and built on DX10 from the ground up.