If the game doesn't support Crossfire, i.e. World of Tanks, then you'll always see a 100% GPU Load on the 2nd GPU. This is with CrossfireX. This happens especially with the newer R7 and R9 Series. Basically all games and benchmarks that only use single graphic card setups, GPU 2's load will always do 100%, and GPU 1 will be at 0%. This isn't an error or malfunction in the card. It is the nature of AMD Graphic Cards. Even complain to AMD about this on their forum, they will tell you that this is actually a natural function or state for their cards. So whether you use Driver Version 13.4 or 13.12, it won't really make a difference on GPU 2 at 100% load. Second GPU at 100% load is actually normal for AMD Cards. The only major difference between 13.4 and 13.12 is optimization, profiles changes, bug fixes, etc... Yes. You can take ULPS, set it to 0 in the registry, but that just means that both (if you have 2) GPUs will be running 100% loads on a game that only needs 100% loads, or throttles less loads on a single GPU. Both GPUs at 100% loads equate to more power consumption over time. You also need to take into account that GPU Loads are approached differently between NVidia and AMD Graphic Cards. GPU loads on NVidia cards try to mirror one another. If GPU 1 is doing 43% for half section of frames, then GPU 2 will do 43% load for the other half section of the same frames, in SLI. On the other hand, if GPU 2 does 88%, more than twice the first, then you know the SLI configuration is having scaling issue. AMD GPUs throttle 99 to 100% loads amongst the different Graphic Cards. The question then becomes for how long. If it has to do it for a long time, then the GPU load will always cap at 99 to 100%. When it's not required to do those loads for a small instant in time, the GPU Load could throttle down and up again. This occurs more often when ULPS is set to 1 because their is a change in power consumption requirements over time for different loads. While NVidia uses AFR 1 and 2, AMD only uses AFR 1, or whole frames from each Graphic Card. So AMD Graphic Cards always do 99 to 100% loads for whole frames, per GPU, when frames need to be produced.
Basically, when you can only utilize 1 AMD Graphic Card, and you're running a CrossfireX setup, you will always see a 99% to 100% load on GPU 2. What basically happens is that GPU 2, on Graphic Card 2, will produce the frames. Frames will be created by GPU 2, sent to Graphic Card 2's framebuffer, transferred to Graphic Card 1's framebuffer from the CrossfireX Bridge, and then it's sent off to the Display Adapter. We are talking about whole frames being transferred from one framebuffer to another framebuffer, in AMD's case. On the other hand, if you do play a game that does utilize both Graphic Cards, GPU 1 will produce frames on it's frame buffer while GPU 2 does the same. GPU 1 shoots a frame off to the display adapter. Once the Frame leaves the framebuffer from the first card, it will receive the frame from the 2nd Graphic Card's framebuffer, and push that frame off to the display adapter. From there, it's just a repetitive motion. Frame Pacing Software was designed and needed to regulate this process. Before the software, both GPUs would have a potential to produce partial frames instead of whole ones, and try to send it off to the display adapter. When the graphic card realized it didn't produce a whole frame, it would drop the frame. Nothing on a hardware level was regulating that to reduce runt frames from being sent off from the frame buffers. AMD implemented software to control the way those frames were being pushed out. AMD 7000 Series Graphic Cards had a potential to push even smaller Frame Times, but sacrificed the time needed to developed whole frames. That was the tradeoff. This is why AMD 7000 Series Graphic Cards had so much drop-age and bandwidth in their Frame Time Variance Graphs. With the PCIe Based CrossfireX, it's allowed AMD the ability to better control and regulate whole frames with their software, and how it's being transferred from one framebuffer to the other.