Designing Our Next GPU Test Suite

240 Comments

View All Comments

• Locrian - Sunday, March 14, 2010 - link

  I'd like to see a review or section of a review dedicated to testing mainstream gamer cards with professional art applications. Comparing Geforce 4XX and Radeon 58XX with some of the Fire and Quadro pro cards. Testing things like 3dsmax render times as well as general real time view window performance/responsiveness. Same for Maya. Test real time shader plugins for said programs to see if one brand handles that better than another. I dunno, maybe this is beyond Anandtech and more for a 3D-specific website. But its something I am interested in seeing. Then there are the sculpting programs also. I believe Mudbox is the one that currently uses the graphics card while I think Zbrush still does not. Photoshop, Aftereffects, etc. See if they offer worthwhile value for the pro visual applications vs the ridiculously expensive pro cards.

• yyrkoon - Sunday, March 14, 2010 - link

  Refreshing your blacklist, so the rest of us can actually read the comments without a crap load of annoying spam.
  
  Maybe not important to some, but I personally like reading at least some of the comments to read about things that I may have/have not thought about myself.

• LyCannon - Sunday, March 14, 2010 - link

  How about using applications that use the GPU for processing, like Photoshop, Badaboom, Boinc, etc?

• charmedmeat - Sunday, March 14, 2010 - link

  Well, I know they're few & far between, but some Open CL benchmarks of some type (compression & cryptography come to mind) would be nice. Also, when they come out, some DirectX 11 tesselation heavy games would be good.

• Ramon Zarat - Sunday, March 14, 2010 - link

  Hi.
  
  Anandtech already focus on the essential with a very good strategy. Only 3 resolutions that reflect the power of the card and only 2 set of resolution: High and lower end. AA is only used when necessary. It's simple, efficient and cover over 90% of real life scenario.
  
  I want to attest why Anandtech strategy is so spot on, so lets refer to STEAM data, which is probably the best data reference source you can get:
  http://store.steampowered.com/hwsurvey/">http://store.steampowered.com/hwsurvey/
  
  lets start with resolution. The following list represent 70.45% of all used resolution:
  
  1024 X 768 @ 12.93% (Very surprising!)
  1280 x 1024 @ 19.31% (THE most used resolution)
  1440 x 900 @ 10.75%
  1680 x 1050 @ 18.43% (Second most used)
  1920 x 1080 @ 9.03% (The highest resolution used by a SIGNIFICANT amount of users)
  
  Resolution below 10%, but higher than 5%:
  
  1280 x 800 @ 6.61%
  1920 x 1200 @ 5.86%
  
  Progressing and receding:
  
  All resolution at and below 1280 X 1024 are receding. The biggest drop is 1024 X 768. All resolution over 1680 X 1050 are progressing. The highest gain is 1680 x 1050.
  
  Now we can see why Anandtech has chosen those 2 sets of resolution
  
  Low end set:
  
  1024 X 768 NO AA
  1280 x 1024 With 4X AA
  1680 X 1050 With 4X AA
  
  High end set:
  
  1680 X 1050 with 4X AA
  1920 X 1200 with 4X AA
  2560 X 1600 with 4X AA
  
  I wouldn't change a thing. This 2 sets strategy is perfect. 1680 X 1050 is on the 2 sets and will be the most used resolution of 2010-2011, taking over the receding 1280 X 1024.
  
  Now moving to the test platform. Here, I have some comments to make. Let see what STEAM have to say first:
  
  69.06% of the player use Intel. So using Intel for the platform is right.
  
  Intel speed:
  
  1.7 to 2.29Ghz = 16.53% (Combined 1.7 to 1.99 and 2.0 to 2.29)Median speed is 2.0Ghz
  2.3 to 2.69Ghz = 26.72% Median speed is 2.5Ghz
  2.7 to 3.29Ghz = 19.51% (Combined 2.7 to 2.99 and 3.0 to 3.29)Median speed is 3.0Ghz
  
  Intel core:
  
  1 core = 17.97%
  2 core = 55.37%
  4 core = 25.62%
  
  HyperThreading:
  
  11.27%
  
  CPU/platform conclusion:
  
  In the light of this data, I think Anandtech should use 2 platforms, 1 for each resolution set. A dual core running at 2.5Ghz with HT Off for the low end resolution set and keep the current quad core 3.33Ghz for the high resolution set.
  
  It's either that or test both set with an overclocked quad running at 4.2Ghz (highest speed on air) and RAM at 2600Mhz to eliminate all platform bottleneck and isolate GPU performance.
  
  Which approach is best? With Crossfire / SLI / Dual GPU card solution, it's clear that they are CPU limited in some scenario. On the other hand, only a minority overclock their quad to 4.2Ghz. Maybe then provide only 1 result @ 4.2Ghz @ 2560 X 1600 and only with multiple GPU?
  
  Now lets talk about the games. There is 2 approaches. The best selling or the best rated. I would consider both and make a top 10 out of that.
  
  Top rated:
  
  http://www.gamespot.com/games.html?type=top_rated&...">http://www.gamespot.com/games.html?type...ge_type=...
  
  AND
  
  http://www.ign.com/index/top-reviewed.html?&so...">http://www.ign.com/index/top-reviewed.h...onths&am...
  
  
  Top sale:
  
  http://kotaku.com/5486293/pc-sales-charts-bringin-...">http://kotaku.com/5486293/pc-sales-charts-bringin-...
  
  AND
  
  http://www.amazon.com/gp/bestsellers/videogames/22...">http://www.amazon.com/gp/bestsellers/videogames/22...
  
  
  What I would like to be added is Directcompute and more importantly, OpernCL, GPGPU benchmark. SiSoftware for synthetic OpenCl benchmark would be a nice start. I've also found a little Directcompute benchmark HERE: http://www.ngohq.com/attachments/graphic-cards/250...">http://www.ngohq.com/attachments/graphi...mark-dir...
  
  If you insist on non-synthetic GPGPU benchmark, then try the folging@home GPU application that runs on both ATI and Nvidia : http://folding.stanford.edu/English/DownloadWinOth...">http://folding.stanford.edu/English/DownloadWinOth...
  
  
  That's it! :)
  
  Ramon

• Nimiz99 - Monday, March 15, 2010 - link

  I have to agree with the OP on this, 3 resolutions and no more. I'd even argue only 2 resolutions, because two things happen:
  1) Running a game that relies on RAM as it fills the GPU's ram
  2) Running a game that pushes the limits of what the chip can do
  
  Higher resolutions generally focus on (1) and lower resolutions focus on (2)...so why bother having any more than that (granted the cutoff point will vary, but having 1280x1024 and a higher one like 1920x1080p or 2400x1600 could accomplish that.
  
  Earlier in the thread there was also the argument for more RTS games...most of the gaming I do is RTS/RTT and FPS. From that perspective I'd like to see Starcraft 2 upon release (and I know it doesn't push Graphics as hard as it could) and then any FPS that pushes graphical boundaries.
  
  thanks and keep up the good work
  
  Finally, I still like to see synthetics...I enjoy benchmarking and although I won't buy a card based on a 3Dmark score, I still like to see it. So please keep at least 3Dmakr in y'alls set-up.

• Jamahl - Monday, March 15, 2010 - link

  69% of steam gamers use intel, but most of them are dual core pentiums. Should those be used for benchmarking too?
  
  i7's should not be getting used for gaming, that much is clear.

• primagen - Sunday, March 14, 2010 - link

  I would like to see the following: Shattered horizon, [A very good looking pc exclusive game] Stalker: COP, AvP [DX 11], Battlefield [very popular- dx 11 3d/eyefinity], the newest total war game [Napolean?]. That about sums up my desire for new games. I would however, like to see older card/CPU comparions as well. I do not have an I7 so I would like to know what I can expect out of my cpu coupled with my gpu.

• deathbycomputer - Sunday, March 14, 2010 - link

  I had a lot of trouble finding CoD MW2 performance reviews. It didn't make sense to me how a game so popular like that one did not get the attention of hardware testers; seems to me like only games that are demanding of high-end hardware are used on reviews, which makes sense of course. However, other popular/mainstream games that virtually everyone plays are not included in the Gaming Performance section of a CPU and/or GPU review.

• RoseRendeR - Sunday, March 14, 2010 - link

  I dunno why ya just don't test da games in da "Top Rated PC Games"-list compiled by Gamespot?
  Since they're the top rated PC games, OBVIOUSLY!
  I mean, that's da way "I" would work!
  Go figure!
  
  Then again, I'm just lil' ol' me! YOU're da pro's babe!
  
  RoseRendeR