Potential New Anandtech Video Card Benchmarking Suite?

[Termie]

Just noticed that Jarred Walton greatly expanded the benchmarking suite for his test of the HD7970M vs. GTX680M:

http://www.anandtech.com/show/6369/avadirect-clevo-p170em-part-2-gtx-680m-grudge-match/4

I'm not posting this to discuss the outcome of this particular test, but rather to start a discussion of the games he's added to the suite and whether we as forum members would recommend that these games be used for upcoming video card tests. Here's what he had to say:

Our 2012 gaming suite currently ranges in age from over two years old (Civ5) to as recent as 11 months old (Batman), with an average age of around 16 months. Hardware and games have both changed during that time, so we dug through our games folder and added a few other titles to the mix. Okay, truth be told, we actually have quite a few gaming codes from AMD and NVIDIA (as well as Guild Wars 2 from ArenaNet), and we figured a selection of games from both sides should be more or less “fair”. To that end, we’ve benchmarked eight additional games: Borderlands 2 (NVIDIA), Diablo 3 (“agnostic&#8221, DiRT Showdown (AMD), Guild Wars 2 (“agnostic&#8221, Max Payne 3 (NVIDIA), Sleeping Dogs (AMD), Sniper Elite V2 (AMD), and The Witcher 2 (NVIDIA). That’s three NVIDIA “TWIMTBP” (The Way It’s Meant To Be Played) games and three AMD “Gaming Evolved” titles, so overall things should be relatively even. Here’s how the two mobile GPUs stack up using the latest available drivers (NVIDIA 306.97 and an as-yet-unreleased 12.9 Hotfix from AMD), at all three of our target settings.

I know a lot of folks around here think that the more games that are tested, the better the test, and therefore link to other sites more often than Anandtech. There's also a lot of talk about Nvidia-bias vs. AMD-bias, and it seems Jarred is trying to account for that. So what do we think of the following lineup?

(1) Batman: Arkham City (old)
(2) Battlefield 3 (old)
(3) Civilization V (old)
(4) Dirt 3 (old)
(5) Elder Scrolls: Skyrim (old)
(6) Portal 2 (old)
(7) Total War: Shogun 2 (old)
(8) Metro: 2033 (old)
(9) Starcraft II (old)
(10) Crysis: Warhead (old)
(8) Borderlands 2 (new)
(9) Diablo 3 (new)
(10) Dirt Showdown (new)
(11) Guild Wars 2 (new)
(12) Max Payne 3 (New)
(13) Sleeping Dogs (new)
(14) Sniper Elite V2 (new)
(15) The Witcher 2 (new)


First, assuming Anandtech can actually test 15 games in every test, any games you'd use in place of the ones above?

Second, assuming Anandtech can only test 10 games as it has in the past, which 10 games would you choose? Keep in mind that Jarred already had to drop Starcraft II, Crysis: Warhead, and Metro 2033 to keep the list down to 15.

UPDATE - 10/18/12
Based on input from the community, as well as Anandtech editors Jarred Walton and Ryan Smith, here's a suggested list for Anandtech to consider for the 2013 gaming suite:

(1) Battlefield 3 (Single-Player for repeatibility)
(2) The Witcher 2 Enhanced Edition
(3) Total War: Shogun 2 (Ultra Settings)
(4) Crysis: Warhead - to be replaced with OR joined by Crysis 3 depending on the product tested
(5) Metro: 2033 - to be replaced with Metro: Last Light
(6) Dirt Showdown
(7) Sleeping Dogs
(8) Batman: Arkham City (for Unreal Engine testing)

Note: Ryan is in need of new strategy and sim games - one suggestion is Anno 2070, and Project CARS on the racing sim side.

 

[RussianSensation]

1) Dirt 3 - I would drop it. Why have Dirt 3 and Dirt Showdown? Since Crysis Warhead is dropped, why not use Crysis 2 in place of Dirt 3?

2) I would either revise Skyrim testing or drop it in its current form since AT does not use ENB mods and the benchmark continues to be primarily CPU limited in the test suite at 1080P. Without ENB mods, it's more suitable to test low and mid-range GPUs, not higher-end ones. Imo, if Skyrim is used, AT's team has to incorporate ENB mods as only then does this game truly become GPU demanding for high-end GPUs and actually stresses VRAM well.

HardOCP's testing today showed that Skyrim can be maxed out completely on GTX680/HD7970 with 12x Edge Detect on the Radeons and 8xMSAA + 2x TRSSAA on GTX680 at 2560x1600. In other words this game is going to be useless to test future GPUs for another 12 months without ENB mods, or it should be dropped, or please force Super-Sampling for it.

3) For BF3, I'd try to focus on BF3 Armored Kill testing rather than the single-player BF3. I think it would better reflect the current landscape for these players.

4) Diablo 3? No way. This game is a total waste of time for testing high-end GPUs that already go well over 100 fps at 1600P. It would only take up space in the test suite that frankly should be left for Metro Last Light and Crysis 3 instead that can be added early in the year. This game is perfectly playable even on a GTX550Ti / HD7770 card at 1080P and gets > 100 fps on a GTX660. Seriously, don't do it and leave that slot for a more demanding game in 2013.

5) Use Witcher 2 Enhanced Edition, not the original one as some graphics effects have been upgraded. Great game to test Uber-Sampling.

6) We've already seen Portal 2 SSAA performance and it seems NV has a huge advantage here that's driver related. If Portal 2 continues to be used, we are going to see the same results over and over, NV dominating in 4xSSAA. This would be no different if we used Trine 2 or Mass Effect 3 as AMD performs better due to drivers in those. Let's see another game with SSAA. Actually, if AT won't use ENB mods for Skyrim, they could include Skyrim SSAA in its place. This game is very popular and people are likely to use SSAA in this game due to so much foliage and open/bright environments. Skyrim would be perfect replacement for SSAA over Portal 2.

7) Batman AC - no longer GPU demanding for high-end cards. When GTX660Ti and 7950 are getting 80 fps at 1080P in a 3rd person action adventure game, there isn't any point testing it for another 12 months. Either force 8xMSAA or drop it. 8xMSAA in this game is a great test to find out ROP and memory bandwidth bottlenecks. Otherwise, it's a pretty outdated benchmark.

8) Shogun 2 - enable Ultra settings at 1080P with MSAA or this benchmark continues to be useless at 1080P for modern GPUs. The vast performance delta in AT's test data between 1080P and 1600P in Shogun 2 shows that at 1080P max settings must not be enabled. If Ultra settings in Shogun 2 cripple GPUs with less than 1.5GB of VRAM, great, we want to see that! This actually helps us know that some games are starting to need more than 1.28GB of VRAM.

Right now in AT's Shogun 2 testing GTX660Ti gets 112 fps at 1080P. In reviews that test Shogun 2 at Ultra, this card gets less than half of that. Let's enable those Ultra settings for a truer reflection of real world testing.

9) Civilization V - I would also consider adding 8xMSAA testing. This actually helps us see the impact of additional memory bandwidth and extra ROPs that high-end cards such as GTX680 have over say GTX660Ti. This game is slowly becoming outdated for testing modern GPUs and Crysis 3 or Metro LL or Medal of Honor Warfighter would probably be better in its place.

10) I would maybe consider swapping some of those older games with a more GPU demanding game such asArma II / Day-Z mod / Reinforcements or wait for Arma III and add that.

My other thoughts:

Note 1: I think AT should hold off until December to revise the test suite for next year. Medal of Honor Warfighter and Far Cry 3 could prove to be great GPU benchmarks and could easily replace some of the old/non-GPU demanding games in the suite (Dirt 3, Batman AC, Portal 2, Civilization V, Diablo 3, Skyrim), especially Medal of Honor Warfighter serving as another data point for future Frostbite 2.0 games. Since this engine will be used extensively in the future, this game could give us another glimpse outside of BF3 of how modern and future GPUs can handle Frostbite 2.0 with perhaps more repeatable testing results than BF3 multiplayer testing. Since this game launches this month, why not wait another 2 weeks and see if it's GPU demanding? If it is, let's throw it in over older/non-GPu demanding games.

Note 2: One of the main issues of AT GPU reviews this year was that no new games were added throughout the year. I realize this was done to make comparison in the Bench from year to year easier but if this same approach is used, next year this website will miss out on Metro Last Light and Crysis 3. I think AT should make an exception and drop 2 slots so that later next year these 2 games are added. For example, Diablo 3, Dirt 3, Batman AC, Portal 2, Civ V, Skyrim. A lot more people will care about Metro Last Light and Crysis 3 than these 6 games. Missing out on what are going to be arguably the most GPU demanding games of 2013 but keeping these 6 older games makes little sense as a lot of people are going to be looking how modern GPUs perform in Metro LL and Crysis 3. Maybe AT can make an exception and add these 2 games throughout the year and not wait until 2013 to do it.

 

[Rvenger]

Wow, well done RS.

 

[OILFIELDTRASH]

RS should be doing video card reviews, benches, etc. for some website.

 

[wand3r3r]

Well said, I agree with Russian.

Basically drop the games that are well over 100fps. Keep the demanding games with a balance of them and not just the ones which highlight a particular brand.

Also the critical point: update the drivers especially when they increase performance dramatically.

 

[raghu78]

anandtech should have a game test suite update every 3 months. Newer games need to take the place of older games on a continuing basis. Till recently there was not a single game released in 2012 in that test suite. only with the HD 7970M vs GTX 680M comparison newer games have been added. Alan Wake, Max Payne 3, Dirt Showdown, Sleeping Dogs, Guild Wars 2, Diablo 3, Borderlands 2 are few of the good games out this year. My game list would be.

1. BF3
2. Crysis 2
3. Batman AC
4. Witcher 2 Enhanced edition
5. Max Payne 3
6. Borderlands 2
7. Guild Wars 2
8. Total War Shogun 2
9. Dirt Showdown
10. Sleeping Dogs

In December Shogun 2 can be replaced by MOH Warfighter or Farcry 3. anandtech is retaining games for such a long period so as to form a comparative annual GPU bench. but it really does not let users know of the performance in the latest titles and thats not a good thing.

 

[KompuKare]

Ok after the first bunch of patches and everything, I guess Skyrim is not no longer stressing anything much. (It was never such a GPU hog since it runs okay(ish) on my old 5770).

But instead of dropping it I would love to see Anandtech using it as benchmark of insane amounts of user mods. As in try to go for something with even more mods than Computerbase are using and really try and see how a heavily user-moded game performs.

I know some people might on about how user texture mods are unrealistic because users texture are not optimised like textures written by professional developers etc. I would acknowledge those critiques but counter that Skyrim is hardly the only game to feature user textures and having a Skyrim+Insane being benched in GPU reviews would give an indication of what a card is capable. After all, without modding we might as well be playing on a console...

 

[DaveSimmons]

I prefer to keep at least one or two "old" games like Skyrim on the list so you can compare generations of cards against each other.

The other approach would be to be sure to include cards from one and two generations back in some reviews. That might be better since then they'd be using the same versions of games and graphics drivers.

There have been many posts here asking "what's the current equivalent to my <x-years-old card>" or "how does new-card compare to my old-card" so this is something people do care about. Real games work much better for this (IMHO) than synthetic benchmarks.

 

[VulgarDisplay]

I understand why they would want to keep the bench suite consistent for comparison's with old generations of cards. I think however that once AMD and Nvidia have their 28nm refreshes up against each other would be a good time to update everything.

Which ever company releases first I would use the old bench suite to compare to the old cards. Once both companies release their new hardware is when you should bring out the new games because that's what your readers want to see. If they don't get it from you they will go elsewhere to find it.

Maybe after using the bench suite for a more direct comparison you could bring in a few newer games at the end of the review.

 

[DominionSeraph]

anandtech is retaining games for such a long period so as to form a comparative annual GPU bench. but it really does not let users know of the performance in the latest titles and thats not a good thing.

Anandtech is a hardware site, not a software one. With games being launched half finished with dozens of patches dropping in the first month, you're looking at a ton of continuing work for data that is going to quickly become useless artifacts.

I like that there is consistency between reviews because it makes research so much easier.

 

[moonbogg]

Looks good, but I would ditch dirt showdown and sniper elite because those games suck and noone likes them.

 

[tviceman]

I think Metro2033 should stay in the testing suite until Metro Last Light comes along. Warhead is still a good benchmark for the texturing performance. There are other games that can offer more demanding benchmarks too - like Anno 2070. The new Hitman game is coming soon, as is Assassin's Creed III.

I don't think mods should ever be used for benchmarking video cards in video card reviews, though. It becomes a much, much more complicated process and throws the apples-to-apples out of the box comparisons out of the window. Mods can have glitches or anomolies that normal Q&A departments could find, mods might not work with driver updates, mods might not work with game updates, mods themselves are updated in and out of unpredictable cycles, and are often left unsupported when unfinished. Mods are great and I used them in Fallout 3, Fallout Vegas, Oblvion, Skryim, and other games but using mods and the performance impact of such mods should be their own articles.

 

[RussianSensation]

Anandtech is a hardware site, not a software one. With games being launched half finished with dozens of patches dropping in the first month, you're looking at a ton of continuing work for data that is going to quickly become useless artifacts.

I like that there is consistency between reviews because it makes research so much easier.

How do other hardware websites do it then? If you fix the test bench to 2012 games, the review becomes outdated as more demanding games come out and people start looking to other websites for their upgrade questions. Also, consider the consequences of this since we are in a period of console port stagnation. If we finalize the test bench now, AT could end up missing 3-5 GPU demanding games in 2013. You cannot be a world class GPU hardware website and not update the game selection at least once in 12 months. I think they should especially make an exception this time since Metro LL and Crysis 3 will launch next year and these 2 games are way more important for GPU upgrading than games like Dirt 3, Batman AC, Civilization V, Skyrim, Diablo 3, and Portal 2. Why buy a $500 GTX680 to play Diablo 3 or Dirt 3 when modern GPUs get > 100 fps? Do you really need to see how GTX780 will play in those old games? If a triple-A FPS or racing games launches in 2013, are you suggesting it's ok to not test it until 2014? Think about it, these 6 games I listed can already be maxed out on modern GTX680/7970 cards without pushing SSAA. What difference does it make how fast they run on an HD8970/GTX780? Continuing to keep them is not going to leave any room to fill in the 15 spots for future games that actually will need GPU upgrades and we'll be anxious to read about. This is especially true since Metro 2033 and Crysis Warhead will be dropped as 2 very GPU demanding games will be dropped by 6 non-GPU demanding games will remain.

As a general point, why does AT not have the Asus HD7970 Matrix reviewed when so many other hardware websites did?

AT should make a decision if it wants to continue investing $ into PC hardware reviews or start competing with GSMArena and Phone Area on smartphone reviews. The lack of proper GPU Roundups, Motherboard roundups in the last 2-3 years has somewhat undermined AT as the premium hardware website to go to for reviews. If AT really wants to improve their membership / readership base for GPU reviews, it's not a very good practice to not update the GPU test suite for 12 months when so many competitors are doing it.

Maybe some compromise can be reached. Here is my other alternative. For example, fix the Test Bench with games in 2012, but if we get MSI Lightning and Asus Matrix style of enthusiast high-end GPUs in 2013, do a separate GPU round-up in 2013 and in that round-up add the more advanced GPU demanding games that will push cards like 8970 and 780 to their limits. Otherwise, the reviews risk becoming outdated for >$300 cards by mid-2013. This would be more work but it would still allow the "Main" GPU reviews for reference cards to keep the same game selection and be useful for the Test Bench, while for people who are upgrading to the latest and greatest, they could see how $500-600 flagships (For example, have an MSI Lightning HD8970 vs. Asus DirectCUII GTX780 roundup) duke it out in the most GPU demanding games of 2013. I think this way you end up catering to both the mainstream and budget gamers who might want to see how GTX750Ti does in Diablo 3, while also not neglecting the enthusiasts.

In contrast, the current approach is well suited for sub-$300 GPUs but is ill-suited for testing >$300 GPUs because AT can't really push Shogun 2 to Ultra for example without crippling everything with less than 1.5GB of VRAM. Then we end up seeing 100+ fps for $500 GPUs which serves little purpose.

I really feel that could be a very simple solution if adding games dynamically throughout 2013 is not viable. Pick the fastest single and dual-GPU AMD and NV cards by summer-2013 (as by then both 8970 and 780 should be launched) and do a separate review mid-year with updated test suite just for those 3-4 cards including Metro LL, Crysis 3 and whatever else GPU demanding games may be out to stress performance in more modern games and at the same time see how Cross-fire vs. SLI scaling compares.

 

[djnsmith7]

I'd vote for BF3, as that's currently the only relevant game for me. However, you can't leave Crysis 3 (when available in the future) out of the mix, as it's always great to see where the bar is set.

 

[bryanW1995]

4) Diablo 3? No way. This game is a total waste of time for testing high-end GPUs that already go well over 100 fps at 1600P. It would only take up space in the test suite that frankly should be left for Metro Last Light and Crysis 3 instead that can be added early in the year. This game is perfectly playable even on a GTX550Ti / HD7770 card at 1080P and gets > 100 fps on a GTX660. Seriously, don't do it and leave that slot for a more demanding game in 2013.

9) Civilization V - I would also consider adding 8xMSAA testing. This actually helps us see the impact of additional memory bandwidth and extra ROPs that high-end cards such as GTX680 have over say GTX660Ti. This game is slowly becoming outdated for testing modern GPUs and Crysis 3 or Metro LL or Medal of Honor Warfighter would probably be better in its place.

Can't comment on the others as I've spent a few minutes on skyrim and BF3 (and 0 in all the others) in my life, but I've spent a buttload of time in D3 and CiV over the past couple of years. Your thoughts are spot-on about Diablo, and the statements on CiV are good as well.

 

[alexruiz]

A lot of good comments in here :up:
I'll add my thoughts:

Borderlands 2 in: Seriously? A GTX660 handles it perfectly. Besides being poorly optimized, it is not demanding. Once it gets decent optimizations, the faster hardware will just slice right through.If you want a frantic shooter in the suite, why not Serious Sam 3?

Metro 2033 out: Arguably the most or second most demanding game in the suite, and it is being dropped?

Blizzard games: Use them only for iGP testing.

Oh, and I totally concur on the suggestions to test games in the most played situations, meaning, BF3 (and AK) gets tested in MP. I understand It gets really hard for testing as repeatability goes down, but I cannot imagine that Anand and company couldn't set a dedicated server always in the same map to test the game and have forum members join to help with the testing

 

[RussianSensation]

That's a good point. BL2 is pretty much CPU limited at 1080P without PhysX. It's going to look like Skyrim all over again. I think Borderlands 2 would be a great candidate for future CPU testing.

This is the problem with the way the current generation of consoles affected us. While Crysis 1 / Warhead and Metro 2033 are old games, the irony is that they are way more GPU demanding compared to newer games like BL2 or Dishonored. The counter-argument is who plays Crysis Warhead or Metro 2033 still? How useful are they for testing if most people don't play them anymore? But games like Diablo 3 don't make much sense to test even though it's probably among the most popular PC games. Popularity vs. GPU demanding, what's more important?

The other potential issue is the idea that the test suite should be balanced between TWIMTPB vs. AMD Gaming Evolved vs. agnostic titles. NV barely spent $ this year working with developers on new titles which is why there are not many notable GPU-demanding titles on the TWIMTPB side besides PhysX in BL2.

Ironically, it is the AMD Gaming Evolved games that use complex lighting, HDAO, contact hardening shadows and more advanced anti-aliasing filters. Gamers don't really care if the title is AMD Gaming Evolved or TWIMTBP. They buy a game and want to know exactly if their GPU runs it well or not. If AMD GPUs run Alan Wake or Dirt Showdown faster, that's not biased, but actually the experience a gamer will have if he owns an NV card. Not picking a game because it's an AMD Gaming Evolved title, even if it happens to be GPU demanding, somewhat undermines the basis of testing GPUs.

Games like Hitman Absolution, Medal of Honor Warfighter, Far Cry 3, Tomb Raider will also be AMD Gaming Evolved titles. Start adding FC3, Metro LL, Medal of Honor Warfighter and even more of the test suite becomes AMD Gaming Evolved. But what can a gamer do if he wants to play those titles? It's not like an NV gamer who was interested in this games will not buy them only because those are AMD Gaming Evolved games. They should still be tested if they are GPU demanding.

It's hard to pick many GPU demanding titles and not 'favor' AMD Gaming Evolved accidentally simply because AMD spent more $ this year on winning AMD Gaming Evolved games and making sure they are not just console ports. The more of these GPU demanding games you use, the more 'AMD-biased' the test bench becomes.

How do deal with the idea that even more games coming out in the next 6 months are AMD Gaming Evolved titles? I think you stop worrying about it and start picking GPU demanding games, regardless if it's an AMD Gaming Evolved or TWIMTPB game. Therefore, I still think the older games should be replaced with Far Cry 3, Medal of Honor Warfighter if they happen to be GPU demanding, despite them both being AMD Gaming Evolved titles and offsetting the TWIMTBP vs. AMD Gaming Evolved game count equilibrium.

 

[cmdrdredd]

Keep Metro 2033. It still is relevant for Ambient Occlusion and MSAA performance with tessellation etc.

Blizzard games? hell no!

Borderlands? Can anyone with a video card made after 2009 NOT play this game at 1080p?

Portal 2 and all source engine games are irrelevant. Even my old GTX280 got over 130FPS in that engine.

Skyrim should only be tested with ENB mods and 4k textures applied.

Calling Battlefield 3 old? No...just the way every hardware site tests it is utterly useless and we have been saying this on the forums since it released. You must test multiplayer on demanding maps with lots of players.

The other games really don't make a difference either way for me.

 

[MrK6]

Games in the suite need to be pertinent and demanding. I think we can retire Crysis and Portal 2. I'd rather see Crysis 2 than Crysis 1. However I don't think any game should be run beyond vanilla/official settings (Crysis 2 DX11 would therefore be fine). That eliminates Skyrim from my list, as it just isn't that demanding at vanilla settings.

I think something like the following covers a broad spectrum:
1. BF3
2. Guild Wars 2
3. Witcher 2 Enhanced edition
4. Crysis 2
5. Batman AC
6. Max Payne 3
7. Alan Wake
8. Dirt Showdown
9. Sleeping Dogs
10. Civilization V
11+. Wild card for new releases

 

[notty22]

F1 2012 was recently released, from Codemasters,

Anyone know of a site benching this game ?

F1 2012 sports an impressive 17 graphics options, and many sub-options, including Shadows, Particles, Vsync, and eight different levels of AA, divided into MSAA and EQAA types. For those who don't know, EQAA stands for enhanced quality anti-aliasing, and functions like a (sometimes) slightly more effective and slightly more expensive form of the traditional MSAA.

 

[Dankk]

Looks good, but I would ditch dirt showdown and sniper elite because those games suck and noone likes them.

That's a very arrogant thing to say. DiRT Showdown (while not very demanding) is a very fun game.

Sniper Elite V2's built-in supersampling makes it one of the most graphics-intensive titles I have in my library, so perhaps it would make a nice addition to the list. Rebellion has an extremely nice proprietary engine underneath their belts, regardless of what you think about their games. (I enjoyed Sniper Elite V2 as well)

Metro 2033 should stay on the list, but then again Metro: Last Light is also coming out fairly soon.

 

[maniacalpha1-1]

I know it would be a lot of work and everything, but how about devising a test suite of MMOs? For example, Planetside 2 and maybe one or two of the top MMORPGs.

Yes, I'm aware that for that to be relevant, it'd need to be updated every 3 months due to patch improvements.

But then, the same is true of a game like BF3, it too has patch improvements all throughout, which means the video cards coming out now may have slightly artificially higher FPS relative to cards reviewed when the game came out(or worse if something gets broken), unless they update their benchmarks...

 

[Majcric]

I don't care as long as they keep Metro 2033. (Favorite game of the past couple years.

 

[raghu78]

As suggested above I agree Metro 2033 should be on the list till Metro Last Light takes its place. I somehow completely screwed that up. Crysis 3 can take Crysis 2's place. MOH Warfighter or Farcry 3 can take Shogun 2's place.

1. BF3
2. Crysis 2
3. Batman AC
4. Witcher 2 Enhanced edition
5. Max Payne 3
6. Metro 2033
7. Guild Wars 2
8. Total War Shogun 2
9. Dirt Showdown
10. Sleeping Dogs

 

[RussianSensation]

For example, Planetside 2 and maybe one or two of the top MMORPGs....

Planetside 2 looks cool, but it's going to be very CPU limited on a modern GPU like GTX680 for real world gamers.

Even a Core i7 2600K holds back a GTX680 in that game.

I think like Borderlands 2, this game would make for a great CPU limited test.

F1 2012 was recently released, from Codemasters,

Anyone know of a site benching this game ?

Ya, I was thinking of suggesting this game to replace Dirt 3 but saw the benchmarks at TPU. It's a cake-walk for even a GTX660.

My list would be:

1. BF3 Armored Kill
2. Crysis 2 (replace it with Crysis 3 when it's released)
3. Skyrim with ENB mods + 4K textures or vanilla but force SSAA (if not Alan Wake)
4. Witcher 2 Enhanced Edition
5. Max Payne 3
6. Metro 2033 (replace this with Metro Last Light when it's released)
7. Guild Wars 2
8. Total War Shogun 2 but all settings on Ultra
9. Dirt Showdown
10. Sleeping Dogs
11. Arma II Reinforcements or Arma III

Depending on how GPU demanding these are:
12. Medal of Honor Warfighter
13. Far Cry 3