Hey everyone, i read quite a lot about hardware and i found this interesting article on the french website (Hardware.fr). I translated it for you guys because i think it's interesting.
Check this out and sorry for the mistakes, my primary language is french. Post your comments and make this thread stay!
" As you already know if you read our articles on a regular basis, Quad Core CPUs are not yet well used in actual games. If in fact the use of Dual Core CPUs is more spread out, not many games can use full benefit from additional cores.
With some tests to observe the dependence of Quad Cores, we arrived to the conclusion that one parameter has more impact than expected : the graphic card, or rather, it’s driver. In our protocol test the CPUs, we use to games, Crysis V1.2 (using it’s integrated benchmark_cpu2) and World in Conflict V220.127.116.11 (also using the integrated benchmark). In both cases, the tests are done using 800*600 to limit the impact on the graphic card and concentrate on the limitations of the CPUs. Details are set to the Maximum so the level of stress on the CPUs is as high as possible.
Here are the results we obtained with an E8400 and an Q9650, both clocked at 3 GHZ. We put in both a Radeon 4870 (Catalyst 8.11 hotfix) and a GTX 280 (Forceware 180.48). DX9 and DX10 under vista 32bits :
The results speaks for themselves! With DX10 and an Nvidia card, the gain is about 21 to 24% when you pass from 2 cores to 4 cores. With the ATI card, the gain is minimal. On the other hand, with DX9, the gains are less important : only 10 to 11% gain with Nvidia, and this time, but only with Crysis, we note an almost similar amelioration with the Radeon. We have to say that in DX9 and DX10 mode under Vista, the multi-threading is not managed the same way.
The obtained results clearly show that the DX10 Nvidia drivers are better suited for multi-thread than the DX10 AMD ones, which are almost not or not multi-thread friendly. We can see that the Nvidia card is more suited to substantial gain from a Quad Core CPU, which can’t be taken lightly when you are in a game in which the CPU can limit the performance ( Note: sorry for the last sentence, my English is not A1..)
From Intel’s saying, 25 to 40% of the CPU load in a game is linked to Direct3d and to the graphic card driver. The threading from this load is not negligent, and it would be more than useful that AMD get is nose in this issue to rapidly give at least something comparable to Nvidia in their drivers."
And since this is a cpu bottlenecked test, howd they do at gpu stress levels? Or, is this bench implying that, if the results change and even up more at higher res, that tho nVidias drivers may be better at low res, their cards are inferior at higher res? Not enough info, incomplete overall, and as dagger said.
probably not since it talks about PERCENTAGE increase not absolute frame rates
Isn't that the point though? A more powerful gpu, in theory, would be bottlenecked more severely than a less powerful one. Once that bottleneck is removed, the more powerful gpu would revert to its original level, which means a bigger relative performance increase.
How controlled were the tests? Same rig? Did you only change the card and CPU? Did you make sure that there were no background programs running? Were the processors running at the same frequency? (probably the biggest effect on the outcome, more so than the rig changing)
Since the question is about THREADING efficiency, I'd have like to see a test that MAXIMIZES CPU bottleneck on a dual core compared to a quad core at the same frequency and see what, if any, are the performance gains and if there is a difference in this gain between the two driver sets.
Thats like saying a single core at the same clocks should do as well as a dual or a quad. Adding cores does more than just help a game, anything else thats running will be helped by those added cores, even on a bench setup, as the OS usues certain amounts of clock cycles all by itself.
I tend to agree with dagger. Think of it this way. The faster card will start off with higher fps, even at low res/cpu bound scenarios, but will also gain the most as the clocks or cores are increased, proven in many benches. If you go doen low enough on the gpu scale, a so so card wont see any improvements, just those that adding cores to the overall load breings, which is minute, but there . Itd been a better test to see a G260 vs a 4870 used. Also, certain things, so many parameters at these small res effects different cards in different ways
They should've compared it to an HD4KX2 both because it's more of a match, and also because of the higher CPU overhead for the X2 versus a standard 4870. As dagger points out it's not the best of tests for what they should be trying to show.
It's interesting, but it also doesn't mean what the OP think it atcually means the opposite, as in if you don't have a quad core then your efficiency in some soltuions may be down. Few reviewers use dual core nowadays, so most results already show dual core efficiency, so the thing to not would be that your dual core might be a bottleneck from time to time.