they both used reference cards only in their reviews. non of them includes AIB versions, just like all of the american review sites. But the french one does. ex: in the 980 ti review. it has reference 980, 980 ti, AIB versions of both and the best of 290x in their reviews. which was exactly what I was looking for. it isn't perfect, but it is as good as it gets for what I want. perfect would be if it included 780 ti AIB n 290x reference numbers in their charts too.
AnandTech is probably one of the least biased of all the top North American sites when it comes to GPU reviews. I do think they ripped R9 290 series too much and didn't do any testing on many after-market 7950/7970/R9 290 cards, but they claim AMD never sent them samples. Hard to blame them for that I guess.
TechSpot almost always exclusively switched to after-market models for AMD cards. They are from New Zealand I believe. Also, the owner is a straight shooter - he recommends i5 over i7 for most gaming rigs and I've never seen bias from him. He showed HD5800 series tanking in DX11 games with tessellation and the same for Kepler in modern games. He also values price/performance (unless some sites that threw this metric under the bus in favour of perf/watt) but recognizes the need for flagship cards and perf/watt cards for HTPC.
I also do like Sweclockers. Even though they test reference AMD cards, they run them in Uber mode and from time to time they do have reviews of specific after-market cards.
Hardware.fr and Computerbase.de often have some of the best testing for SKU specific AIB cards when it comes to noise levels and temperatures. Guru3D and TPU is also a great source for that but instead of doing round-ups, they compare SKUs individually.
For all of these sites, I look a the data but I don't let the editor's preferences sway me. For example, TechReport keeps hyping up Borderlands games and Project CARS, both that happen to be NV-sponsored games but the
same site removed Dirt Showdown when it favoured AMD's cards, sighting how the advantage of Compute for global illumination is an unfair advantage for AMD, but ignoring how ProjectCARS was specifically developed ONLY on NV cards - basically ProjectNvidia or NvidiaCARS is what that game is. Sometimes obvious bias is obvious.
But yes, TechReport and PCPer were the first to push FCAT usage, PCPer even went as far as hiding the fact NV gave them the tool & told them to examine frame latency & runt frames.
^ Dirt Showdown removed from TR's testing. ProjectCARS is in it and
promoted as the best racing game since sliced bread. How come that site is never "in the moment" with any AMD GE titles? Now expect to see Project CARS in every GPU review for 12-24 months from that site....