Problems with GoM and 3Dmark11 with AMD FX-6100 and XFX 6870

[stickg1]

I've been having trouble benchmarking my setup and was wondering if its just an error in the database? I have an AMD FX-6100 Six-Core processor OC'd to 4.5ghz and with turbo disabled futuremark sees that its a six core but in my results it always shows it as tri-core. And with my GPU it says my core clock is only running at 300mhz, it runs at 300mhz when in powersave mode but when I play games or do anything that needs graphics rendering it jumps up to 950mhz. I see it perform at 950mhz but Futuremark states that its running 300mhz. Its not too important because I know my computer performs well but I'd like to get an accurate 3DMark score to compare with friends and others.

[Jarnis]

I've been having trouble benchmarking my setup and was wondering if its just an error in the database? I have an AMD FX-6100 Six-Core processor OC'd to 4.5ghz and with turbo disabled futuremark sees that its a six core but in my results it always shows it as tri-core. And with my GPU it says my core clock is only running at 300mhz, it runs at 300mhz when in powersave mode but when I play games or do anything that needs graphics rendering it jumps up to 950mhz. I see it perform at 950mhz but Futuremark states that its running 300mhz. Its not too important because I know my computer performs well but I'd like to get an accurate 3DMark score to compare with friends and others.

The 3/6 core thing is due to the way Bulldozer is built. It has three modules that have support for two hardware threads. In other words, GoM and benchmarks currently see it as three core processor capable of running 6 threads. I've flagged this for consideration as to how to fix it.

As for GPU clock frequencies, SystemInfo scans the frequencies before benchmark runs and at that point the card is still in low power mode. Fixing this is far more complicated and at the moment I can't promise a solution anytime soon. I'm quite sure it runs at full power during the actual benchmark.

[Jarnis]

Seeing your post it gave me an idea you can pass along. They could add a physx number crunch like folding at home(just an example) to trigger the gpu to go to full power. It would not need to be anything that takes allot of time to do. just some equation that is fairly simple and only takes long enough for it to get through the hardware scan part of the sysinfo run. hidden of course..
Something like that should be relatively easy to code in..

That is the theory as to how to fix it but practical implementation, in a way that works on *all* GPUs out there (even ones that can't do compute tasks like folding), is not so easy.