Designing Our Next GPU Test Suite

240 Comments

View All Comments

• Quidam67 - Monday, March 22, 2010 - link

  I'm with you.
  
  The aversion to synthetic benchmarking seems almost religious. I'm pretty sure most people already get that synthetic benchmarks don't directly correlate to ingame performance. Most sights that include these tests normally say as much when they include the results.
  
  If the objective is to make synthetic benchmarks disappear off the market, then Anandtech is being naive, or has an inflated view of it's punching power in the industry.
  
  Synthetics exist because they offer detailed performance anaylsis that isn't possible within a regular application not designed specifically to provide performance metrics.
  
  Yes, it's a flawed system, which is why no relies only on these values, but to banash them entirely from a test-suite is a mistake in my opinion. And yes, Anandtech is still a great site -keep up the good work.
  

• JonnyDough - Wednesday, March 17, 2010 - link

  The thing about synethetic benchmarks is that other sites use it, and you can compare cards online on the developer's website. So it really isn't needed. Older taxing games however, aren't usually used which is why it would really be great to see Anandtech use them!

• SlyNine - Sunday, March 14, 2010 - link

  I know this is about GPU testing. But I would like to add when testing CPU's with games like supreme commander. Check the simulation time ( how long it takes to play through a replay) not just the FPS.
  
  Take Empire TW, It's important to see just how fast the CPU can simulate it. Simulation time in RTS's can be very important.

• Jediron - Monday, March 15, 2010 - link

  Afcource this is about GPU testing. However, i never seen written in stone that GPU's must be tested with the latest and greatest of CPU's, highly overclocked.
  
  No, that's a factor "we" made up ourselfs. Wrongly so, if you ask me. It's perfectly fair to test GPU's on more moderate, populair types of CPU's; nothing wrong with that. In fact; most people would
  appriciate those results more then "another i7 Extreme @4Ghz results page".
  
  Give us the results with dual's, triple's, quad cores. With cheaper cpu's and the best a man can get. That's what we are waiting for! That will be the day we can see immidiatly how much the cpu affects the performance of the GPU, without having to visit multiple sites and reviews.
  
  Anantech, you have a honorable task lying ahead of you; grab it!
  

• rocky12345 - Sunday, March 14, 2010 - link

  Just a few notes
  I think there should be a base set of games older ones say 2 years old
  maybe at least 2 of them that were known to work good on most hardware.
  This way when you test mid range & lower end cards they will at least get more than 3 to 4 fps. The other thing I noticed is when testing mid & lower end cards the testers always seem to think a $50 card should be able to run at the same resolutions as a $300 to $700 & tests them at the top of the pile resolutions like they would with the top end cards lets face it a $50 card is not meant to run at 2048x1536 with max details & it should not be tested at that extreme. So maybe having 2 sets of bench marks one for the high end cards & one for the the budget cards that way everyone gets to see how their potential card performs. Integrated graphics would fall into the lower end card test suite. Lets face it the mid & lower end cards out number the high end cards probably 500 to 1 out in the wilds of peoples systems.
  
  The other thing is to test on different CPU's not everyone has a Hex core i7@4Ghz. I think using different CPU's from both Intel & AMD is a good thing it will show potential buyers how a graphics card performs on a host of different CPU's not just the ultra high end Intel platform.
  
  The last point is have a set base of games for each of the two benchmark setups the low end & the high end but always have a selection of current games where you could rotate in to shake things up a bit.
  
  I won't even get into The way it's meant to be played games to much but having to many type's of games like this that are heavily catered towards one company skews the whole test. I am not sure what AMD calls their vender supported titles but same goes for them as well. Maybe there could be a whole extra test suite setup just for those type of games lol.
  
  My point is if test sites stop testing the heavily Nvidia & AMD sponsored games maybe both companies would focus more on a OpenCL & DX 11 platform where everyone is on an even playing field.

• Jamahl - Sunday, March 14, 2010 - link

  I'd love to see a multiscreen eyefinity/whatever nVidia call theirs benchmark although perhaps that is best for individual titles. 5040x1050 would be good as a lot of people will be looking at three cheap 22" screens.

• Ryan Smith - Monday, March 15, 2010 - link

  There will be Eyefinity/NVSurround testing. It'll be limited to the high-end cards that can already drive a single monitor with power to spare, but we'll have it.

• Roland00 - Sunday, March 14, 2010 - link

  If they people are buying new monitors, then they will be hard pressed to find any 16:10 screens now a days. 5760x1080 or 4800x900 would be more helpful.

• demomanca - Sunday, March 14, 2010 - link

  I don't know that more games will help, but to show how the GPU's scale on difference CPU's I think is important. No point showing a card will run a game at 50fps, but only when it's got a 6ghz i7 behind it.

• PrinceGaz - Sunday, March 14, 2010 - link

  I disagree. If I want to know how fast a game will run on a specific CPU, I'll read a CPU review article.
  
  In a graphics-card review, I want to know what the graphics-card is capable of when it is not bottlenecked, even if that means pairing it up with a far faster CPU than I currently have.
  
  If there's time for CPU scaling on a given card, by all means add it to the review, but a graphics-card reviews should always focus first and foremost on how the card performs with the fastest available system.