Tom's Hardware Tests DX11 across multiple cores

[Mira Yurizaki]

Note to Mods: I was considering putting this in the news section, but I figured it was better here since it offers data on everyone's favorite question: Will this bottleneck my graphics card?

 

This is pleasantly surprising to find this morning. Tom's Hardware this morning posted an article showing various DX11 games and how they run across multiple cores. Or rather, they ran it with 2, 4, 6, 8, then 10 cores. These were actual cores and not threads. They gathered the frames per second and frame time as well. They also had some interviews with some developers about how they program for a multithreaded environment. Since it's an 8 page article with a bunch of stuff per page, I'll "briefly" summarize it.

 

Their test setup: The processors were: Core i3-6320, i7-6700K, i7-6850K, i7-6900K, i7-6950X with HyperThreading turned off and clocked to 3.9GHz. The GPU used was a GTX 1080 FE.

 

The games:

Note: They tested at 1080p, 1440p, and 2160p. For the most part, these comments reflect 1080p and 1440p. 2160p is the great equalizer.

• Ashes of the Singularity: You definitely need at least four cores, but there were incremental improvements at 8 and 10.
• Battlefield 4: Two cores offer enough performance, but frame time variance suffers a lot. Adding more than 4 cores doesn't help performance
• GTA V: Four cores is necessary, as two cores offers half the performance. Adding more cores does improve things but not appreiably so. Also frame frate variance suffers a lot with two cores.
• Hitman (2016): Four cores is necessary as two cores gets half the performance of four.. Going to six cores from four offers incremental improvements and a bit less frame time variance.
• Metro: Last Light Redux: Two cores can get up to 80FPS at 1080p, but four or more gets up to 140-160FPS. At 1440p, two cores gets the same 80FPS while the other cores drop down to about 100FPS. Frame time variance however is consistent (though 2 cores does have more spikes)
• Middle Earth: Shadow of Mordor: Two cores is fine. However Tom's Hardware noticed 10-cores causes frame rate variance to be all over the place.
• Rise of the Tomb Raider: For some reason they didn't get two cores at 1080p, but two cores at 1440p does do respectably well (70FPS vs 90FPS for the rest of the tests). Frame time variance is pretty bad though for two cores.
• Project Cars: Two cores offers playable frame rates at least, but horrible frame time variance. You should have four cores for this game.
• The Division: Two cores can get up to 70FPS at 1080p, vs 90FPS on the other cores. Frame time variance is bad with two cores though.
• The Witcher 3: Two cores performs about the same at 1440p with larger variances in frame time. If you're playing at 1080p, you should get four cores.

So overall summary: A lot of modern games benefit from having four cores, but getting 60FPS with two cores is possible, just with frame time variance all over the place.

 

Note, they added a "Part 1", so either they'll test more games, or go maybe go DX12 :3

[AresKrieger]

6 minutes ago, M.Yurizaki said:

or go maybe go DX12

To me that is the logical next step

[pagani123]

well looks like i am set for now with my 4820k

[Herman Mcpootis]

4 cores... MUZ GET DUAL E5 2670 

#totallynotoverkill

 

seriously though, greatly considering dual e5 for my next build(assuming china.com doesn't screw up). hope them extra cores don't screw up the performance.