I hate to say it (and I don't mean to sound like a dick), but unless those numbers are averages, they don't make any sense. You can't have above a 100% increase in performance by only adding a second video card (of course, allowing for some sort of margin of error), which a lot of these numbers show. If these are averages, it might be useful to have a min/max FPS listed there as well because one of these sets of numbers goes from 16.58 -> 47.44 and that's an increase of almost 200% in performance (which can't happen).
Each video card can only do a maximum of X FPS at 100% load in whatever situation you're putting them in. So, if you had
perfect scaling (which rarely ever happens), X + X can only equal 2X; but again, that's theoretical maximum FPS in an absolute perfect world -- Both of which SLI and CF do not live in most of the time (although it can be close depending on the game).
Here's a link to show SLI and CF with damn good scaling in Far Cry 3 (although they don't seem to mention any specifics of the test setup beyond the GPUs):
http://www.hardocp.com/article/2012/12/ ... _preview/3Also, if you're testing only the GPUs for FPS, then you need to remove the CPU from the mix (not literally). If any of the CPU cores are pegged at 100%, then it could be easily skewing the results because both the CPU and GPU rely on each other to display frames on the screen. This means you need to find a good mix of settings so that the CPU load stays less than 100%, but can keep the GPU load high.
Finally, what exactly are you using as a benchmark for all of this? (Not that I would know either way since I don't play EVE). Is your test able to be constantly repeated with little-to-no variance in each pass? Honestly, when trying to benchmark an MMO (for multiboxing) I would probably focus more on average GPU loads as opposed to raw FPS numbers; because unless a game has a built-in benchmark that can be reliably ran over and over, then it's usually incredibly difficult to recreate the same exact situation with consistent results.
An example of this would be me sitting in any major city of any MMORPG at peak playing time and attempting to record data. The major problem with this is that I cannot guarantee that the same number of players/characters are going to be on the screen at the same time doing the same thing each time I go to record my numbers.
I also say this because raw FPS can be tricky to measure when using SLI or CF. In light of recent findings, both SLI and CF suffer (CF much more than SLI) from what is known as frame-pacing issues (AKA frame latency or frame rating); as well as, micro-stutter (which has been known for some time). The problem with a bad frame rating is that a program like FRAPS is not actually reading the correct FPS that is being displayed to you. This is something that would be impossible for me to explain in a single forum post, not only because it's lengthy, but because I don't fully grasp the concept myself; but if you are interested in learning about the subject, then this video is very informative:
http://youtu.be/2cH_ozvn0gAI guess in order to give players a more accurate idea of what your numbers mean (and this is just a list of suggestions):
1) In your case of looking for GPU data, make sure any outside variables such as CPU or disk access are not skewing the results -- This means they need to be removed from the loop (i.e. putting yourself in a situation (or using specific settings) that are least dependent on said variables).
2) Move away from looking at raw FPS numbers because they're most likely inaccurate since it's incredibly difficult to recreate a specific scenario while multiboxing in an MMO. I would say that seeing consistent GPU loads (at X video settings and Y number of clients (maybe even high video settings versus low video settings) and how they scale between CF/SLI and non-CF/SLI helps give a better picture of what players can expect with certain hardware.
3) Explain what you're doing when you're recording these numbers (or loads). If someone wants to see how their system or setup measures up against your setup, then it's likely they'll want to put themselves in the same situation/test to see where they stand.
And something I didn't mention in my wall of text...
Try to keep overclocked numbers out of your original data unless you're able to provide stock clocks alongside the overclocked data. A certain overclock is never guaranteed on any piece of hardware, so if you're showing numbers with overclocked hardware, and another user can't overclock their same hardware to what you've got (regardless of how mild of an overclock it is), then that only makes it more difficult to compare.
Overall, benchmarking can be quite tedious and that's why it takes hardware review sites, at the minimum in most cases, days to record data -- Even when they have specific systems set up dedicated to benchmarking. I'm not saying you need to take it to their level, but if you're trying to deliver accurate results for others to look at, then it definitely requires some work.
Fri Sep 13, 2013 12:47 pm 
