Hello gain G3D peeps, got another issue with folding i have a potent rig basting a quad core phenom II processor, a radeon 4850 and a 6950, hacked to a 6970. the two major problems are : 1) the 6970 wont fold properly 2)the core processor seem to take significantly longer than the 4850 My 6k gpu is recognized by the 6.32 F@H client thanks to a forcegpu flag, bu when i try to run F@H, it stops at 300/10000 every time my processor is OC to 4.2 Ghz but it takes 18hours to work a single WU with a ppd averge of like 300+ Any ideas? i'll provide more info if necessary
Well, since you hacked the 6950 into 6970: 1) It could be possible that the unlocked shaders were defective so they are giving you errors. Your games and benchmarks may work perfectly fine but not for F@H. A simple test is to do this: 2) Since the GPU has only 2 6-pin connectors, when you unlocked the shaders and run F@H, it may be overloading the GPU thus giving you errors (Details). Regarding your CPU, was it idle when you checked its PPD? Do note that running ATI/AMD GPUs will use significant amount of CPU cycles so if those CPU cycles are "stolen" from FahCore_a3, it will cause a non-linear slowdown thus will have a negative impact on your PPD. I suggest that you try: 1) Only running the SMP Client (Do not fold on the ATI/AMD GPUs) 2) Run the SMP Client with -smp 3 while folding on 4850 3) Run the SMP Client with -smp 3 while folding on 6970 4) Run the SMP Client with -smp 3 while folding on 4850 & 6970 5) Run the SMP Client with -smp 2 while folding on 4850 & 6970 Compare the PPD between the above setup and stick with the one that gives the highest PPD. Do note that the hours spend to fold isn't really useful since different WUs take different times. However, you should consider the points and for a quad, getting 300+ PPD is extremely low. I am assuming that you have already qualified for the Bonus Points (Details).
That's the same sort of PPD I was getting untill I qualified for bonus points/PPD now it's upto 2738PPD and that's on an P6074 WU which will take me some 17+ hours to crunch As PantherX Has stated before in numerous other help me posts any cycles taken away from the CPU buy gpus folding will slow the SMP folding down in an non linear way but even so my points for 24hrs has gone from 5022 to over 8K and you might think about trying the newer GPU3 v6.41 client aswell
Thanks for the very thorough replies! unfortunately i've found nothing that suggest the card is faulty. also if the cards can have as much as 20% surplus power sent to them, then the hundred or so shaders granted from the unlock should not overload the card. lets assume that the 6970 is working fine... what else could i do? any flags i forgot or try the gpu3 version?
I got my card to work folding using GPU3 and the R800 via FAH GPU Tracker V2. I'm only getting ~4300 PPD though, so I don't think I'm going to fold with this. Temps stay under 80C with 50% fan
shoot 80c for only 4300PPD is a bit ridicules I get 2900 on my HD5770's and don't go over 65c and thats OC'd to 940/1275 I would have assumed those 6970's would be a wee bit cooler due to die shrink and better shader design
My cards both rock Zalman coolers, the 69(70) carrying the massive 3000a on its back ildes around 30 with loads in upper 60's but im not gettign anywhere near 4k ppd! i'd get around 2500 and its oc nearly to max! also i've been using HFM.net to manage them , should i try the fah tracker?
Regards of what you use to set-up, it will always use the FahCore to process the WU so it doesn't increase the points or anything (For AMD/ATI GPUs, it will use the same FahCore_11 and it won't get any updates since all efforts are going towards FahCore_16). The only difference is how easily the setup of your F@H Clients become. Since you have an ATI/AMD GPU, have you considered using Environment Variables: (Guide 1, Guide 2) which can significantly effect the efficiency while folding?
I agree with PantherX environment variables do make a considerable difference in performance... Without them I get around 2400ppd on each HD5770 with them I'm getting upwards of 2900ppd and that's only because fahcore_11 isn't exactly well optimized for HD5xxx or HD6xxx cards hopefully when F@H v7 arrives and openCL is being used by our GPU's they'll increase output dramatically
