Originally Posted by Alatar
Why do these guys almost always benchmark their games at unplayable settings?
Do they actually need to test 4xMSAA at 1600p on singe card solutions?
Also crappy OCs are crappy. The 7970 is barely OCd and I bet the Titan and the 780 (even though less so) are throttling due to stock bioses.
Yeah, they do. MSAA is a lot more important at lower resolutions. I like these foreign reviews like hardware.fr because they don't really play by the rules as much. Nvidia has always had very poor IMCs and memory bandwidth intensive situations have nearly always been their problem. It makes sense, AMD basically writes the spec for GDDR.
What I'm getting at is that American reviews are very cushy and they are always trying to do things to NOT make Nvidia, AMD, or Intel angry so they can keep getting free review samples. Nvidia knows that they have bandwidth problems (Tom's proved it with their bandwidth limitations testing when GK104 fell behind Pitcairn in memory bandwidth when GK104 theoretically has more, it's simply not coming close to theoretical performance), so they don't want people reviewing their cards in bandwidth intensive situations. Yeah, it'll happen, but they sure don't like it. Ask TT and some other sites what happens when you don't review by Nvidia's rules, they just cut you off completely.
Originally Posted by almighty15
If a 7970 GE can get that close to a stock 780 with that poor overclock ( Beats it in some cases ) then the 8970 could very well end up being faster with those extra shaders to help push it along.
I recall reading on B3D forums that Tahiti's problem is that it is ROP limited and Hawaii is going to significantly increase ROPs. If that is the bottle neck, performance will increase by more than the amount of shaders which were added.
Originally Posted by HeadlessKnight
Originally Posted by geoxile
HWbot shows the average OC on air for the 680 is 1219. For the 7970 it's 1209. If you're going to say the 680 was OC'd badly you have to say the same for the 7970.
I personally know people who can't get their 7970s above 1150 MHz gaming stable 24/7. So yes this 7970 is a fair represent of what an average 7970 can do. But the GTX 680 sample is garbage, even MSI GTX 680 Lightning is faster than that at stock. Now I am not blaming them because I know this is the luck of the draw
Also it appears some people forgot about this benchmark that it is not only applied to hardcore overclockers, but also is relevant to average people or people who overclock on air cooling. So getting a 7970 that can do 1400 MHz on water or a Titan that can do 1250 MHz and putting them in that benchmark is out of the question.
My card is really, really finicky. I can't see to overclock memory and GPU core at the same time. Memory does 1600mhz just fine with the core at stock, core can do 1300mhz fine with the memory at stock. Set core to 1275 and memory to 1500 and driver crashes all day. I think part of it may be power delivery or powertune or something. I just OC the core now and then mess with memory a little bit, it seems for me that core has a greater effect.
I do think it's fair though, and I like reviews like this where they have time to collect data showing what average overclocks are and they can test them there. It really reduces the risk of seeing a good review with good clocks and getting a poor chip and being disappointed. However it gets difficult to keep things fair this way, as was mentioned earlier, Nvidia boost frequency can be reported as base or turbo. I just find it odd that people will complain about AMD touting turbo speed as high clock with FX 9000 series and then they turn around and report their highest turbo GPU speed as their overclock speed. Seems a bit contradictory to me.