Jump to content
  • entries
    63
  • comments
    51
  • views
    20,321

Benchmarking with 3D mark over the years

Mira Yurizaki

538 views

After getting an upgrade to a GTX 1080, the honorary benchmarking run went into effect as soon as I got the system up and running. My choice for benchmarking is 3DMark mostly because I have my results tracked on Futuremark's website. Still, the results do tell me enough about my systems across the years to see what improved or what didn't.

 

3DMark 03

I remember when I first ran this benchmark on a Athlon XP 2500+ and a GeForce FX 5600. And that machine chugged along, barely getting 30FPS in most of the tests. I don't have it recorded, but I'm pretty sure it only had a 4-digit score. And I remembered the glorious day I could finally run this benchmark suite at 60FPS.

 

The main systems I have listed are:

  • Core 2 Duo E8400 w/ GeForce 8800 GT 512MB: 39210
  • Core i5-2500 w/ GeForce GTX 670: 51774
  • Core i5-4670K w/ GeForce GTX 670: 83417
  • Core i5-4670K w/ GeForce GTX 980: 148169
  • Core i7-6700 w/ GeForce GTX 980: 135828
  • Core i7-6700 w/ GeForce GTX 1080: 141541

There's definitely one takeaway from this test now, you can use it as a CPU benchmark for your games! Well, that depends on how actual multi-core rendering comes along. The interesting points are:

  • Between the i5-2500 and i5-4670K results. Just changing up the CPU added 30,000 points. I have no real explanation for this unless the i5-2500 ran at 3.3GHz the entire time and the i5-4670K ran at 4.0GHz (what I overclocked it to), but even taking architectural improvements into account, it shouldn't increase the score by nearly 40%.
  • Upgrading the GTX 670 to the GTX 980 nearly doubled the score, but that was expected.
  • Going to the i5-4670K to the i7-6700 seemed to be expected, since the i7-6700 (note there's no K here) runs at a lower clock speed. Even if I was bottlenecking the graphics card, the GTX 1080 still gets a measurable lead. The tests it excelled at were Battle For Proxycon and the feature tests.

3DMark 06

I skipped 3DMark 05 since it's mostly the same as 06. But since I mention 3DM05, I did run the Athlon machine on it. Needless to say it didn't like it very much (seconds per frame anyone?).

  • Core 2 Duo E8400 w/ GeForce 8800 GT 512MB: 12161
  • Core i5-2500 w/ GeForce GTX 670: 19740
  • Core i5-4670K w/ GeForce GTX 670: 33004
  • Core i5-4670K w/ GeForce GTX 980: 34498
  • Core i7-6700 w/ GeForce GTX 980: 32636
  • Core i7-6700 w/ GeForce GTX 1080: 32702

Mostly the same relative results as 3DMark 03. Between the i5-2500 and the i5-4670K with the GTX 670, the FPS was boosted in the latter machine with the CPU test scores barely nudging. I do have a run with a laptop running a Core i7-4700HQ and the CPU test was appreciably better than either the i5 machines, suggesting 3DMark 06 had a pretty good multithreaded CPU test.

 

3DMark Vantage

This is where the Core 2 Duo  machine stops being recorded. But for some reason I'm also missing the i5-2500 machine results. The preset for this run was the Extreme preset.

  • Core i5-4670K w/ GeForce GTX 670: 17359
  • Core i7-6700 w/ GeForce GTX 980: 28705
  • Core i7-6700 w/ GeForce GTX 1080: 40843

So the thing to note here is surprisingly, 3DMark Vantage is not low end enough to expose bottlenecking. Granted this is the Extreme preset, but the i7's did score appreciably better in the CPU test.

 

3DMark 11

For this one I seem to be missing the i5-4670K's results! Oh well, it's still nice to see how the old machine did. Actually for good measure, let's throw in the laptop run I have in here too. This time I used the Performance preset.

  • i5-2500 w/ GeForce GTX 670: 7716
  • i7-4700HQ w/ GeForce GTX 765M: 4237
  • i7-6700 w/ GeForce GTX 980: 15205
  • i7-6700 w/ GeForce GTX 1080: 20448

The funny thing here is the laptop beat the i5-2500 in 3DMark 06, but here it's getting it's rear smacked down by it. It's also nice to see that relatively speaking for DX11, the i7-6700/GTX 980 combo is twice as powerful as the i5-2500/GTX 670 combo. But seeing an appreciable amount (like up to almost 70%) upgrading to the GTX 1080 certainly instills confidence that my processor isn't bottlenecking.

 

3DMark

And here we come to the latest and greatest. But there's now five tests... and yes I did run the tests I could for my machines... Here are the results:

 

Ice Storm Extreme:

  • Core i5-2500 w/ GeForce GTX 670: 120828
  • Core i5-4670K w/ GeForce GTX 670: 119531
  • Core i5-4670K w/ GeForce GTX 980: 156462
  • Core i7-6700 w/ GeForce GTX 980: 146869
  • Core i7-6700 w/ GeForce GTX 1080: 147316

It's interesting to see where the graphics cards stop giving improvements to a low-end enough game and where raw single-threaded CPU power is needed to carry it.

 

Cloud Gate:

  • Core i5-2500 w/ GeForce GTX 670: 14411
  • Core i5-4670K w/ GeForce GTX 670: 15934/17527
  • Core i5-4670K w/ GeForce GTX 980: 20131
  • Core i7-6700 w/ GeForce GTX 980: 28115
  • Core i7-6700 w/ GeForce GTX 1080: 29920

The i5-4670K w/ GTX 670 is with and without an overclock. This benchmark seems to favor multiple threads by the jump between the i5-4670K and the i7-6700. The GeForce 900 and Radeon R9 300/Fury series seems to be the last GPU the results will greatly improve upon though.

 

Sky Diver:

  • Core i5-4670K w/ GeForce GTX 670: 16006
  • Core i5-4670K w/ GeForce GTX 980: 24353
  • Core i7-6700 w/ GeForce GTX 980: 28799
  • Core i7-6700 w/ GeForce GTX 1080: 35357

Sky Diver likes more threads it seems, and the technical brief explains that it runs the physics parts in multiple threads (up to 96). Otherwise between the two GTX 980 runs, the graphics score was the same. Though it's nice to see that with a GTX 1080, the score jumped up nicely.

 

Fire Strike:

  • Core i5-2500 w/ GeForce GTX 670: 5476
  • Core i5-4670K w/ GeForce GTX 670: 5693
  • Core i5-4670K w/ GeForce GTX 980: 10720/5624
  • Core i7-6700 w/ GeForce GTX 980: (DNR)/6058
  • Core i7-6700 w/ GeForce GTX 1080: 16548/9542

The first/only scores are for the vanilla Fire Strike. The second scores are for Fire Strike Extreme. Similar to Sky Diver, the physics test wanted more logical cores, otherwise the graphics performance remained the same.

 

Time Spy:

  • Core i7-6700 w/ GeForce GTX 980: 4175
  • Core i7-6700 w/ GeForce GTX 1080: 6527

Going off on this alone, it seems to me that obviously the older the game or the technology (I'm guessing Ice Storm uses a DX9 render path), the more sensitive to single threaded performance it is. DX10 games may not be as CPU sensitive, as I did see a performance improvement expected between the GTX 980 and the GTX 670 despite the i5-4670K getting a slight edge over the i7-6700 if the test was pegging the CPU more. DX11 games? Well anything reasonably high end on the CPU side won't be bottlenecked as far as I know.

0 Comments

There are no comments to display.

×