Secret Weapon of AMD Forza motorsport apex + hd7870 Do you get same behaviour without extra voltage (and without latest patch which I will try later) on 16.5.2 crimson? Just tried with patch and same settings(all vram options lowest, others highest) and 2 laps finished without any artifact on +75 MHz on top of dx11 absolute minimum oc. 1213mV 1225 MHz gpu 1200 MHz vram msi-ab says %99 gpu usage
Ok I didn't get it. You mean that the card can OC better for Forza, or that you get some kind of weird boost in clocks?
GPUPI uses opencl api which already has 100 MHz more overclocking headroom than dx11. Motorsports uses direct-x 12 api for everything. But I will try gpupi anyways and put result here. Here is result: OpenCL 2.0 AMD-APP is ready. Timer: HPET (14.32 MHz) OpenCL GPU: AMD Radeon HD 7870 GHz Edition (20 CUs, 1100 MHz(actually is 1225)) Compiling OpenCL kernels ... done. Calculating 1.000.000.000th digit of PI. 20 iterations. Allocated device memory : 335546368 Bytes Batch Size : 20M Reduction Size : 64 00h 00m 01.849s Batch 1 finished. 00h 00m 03.711s Batch 2 finished. 00h 00m 06.419s Batch 3 finished. 00h 00m 12.208s Batch 4 finished. 00h 00m 17.743s Batch 5 finished. 00h 00m 19.582s Batch 6 finished. 00h 00m 21.450s Batch 7 finished. 00h 00m 24.120s Batch 8 finished. 00h 00m 29.702s Batch 9 finished. 00h 00m 35.037s Batch 10 finished. 00h 00m 36.874s Batch 11 finished. 00h 00m 38.737s Batch 12 finished. 00h 00m 41.440s Batch 13 finished. 00h 00m 47.231s Batch 14 finished. 00h 00m 52.768s Batch 15 finished. 00h 00m 54.608s Batch 16 finished. 00h 00m 56.470s Batch 17 finished. 00h 00m 59.138s Batch 18 finished. 00h 01m 04.721s Batch 19 finished. 00h 01m 10.057s PI value output -> 5895585A0 Statistics Calculation + Reduction time: 68.594s + 1.429s and this is with a default r7 240: Code: OpenCL 2.0 AMD-APP is ready. Timer: HPET (14.32 MHz) OpenCL GPU: AMD Radeon R7 240 (5 CUs, 1100 MHz) Compiling OpenCL kernels ... done. OpenCL GPU: AMD Radeon HD 7870 GHz Edition (20 CUs, 1100 MHz) Compiling OpenCL kernels ... done. Calculating 1.000.000.000th digit of PI. 20 iterations. Allocated device memory : 671092736 Bytes Batch Size : 20M Reduction Size : 64 00h 00m 01.476s Batch 1 finished. 00h 00m 03.154s Batch 2 finished. 00h 00m 05.205s Batch 3 finished. 00h 00m 10.413s Batch 4 finished. 00h 00m 14.847s Batch 5 finished. 00h 00m 16.687s Batch 6 finished. 00h 00m 18.365s Batch 7 finished. 00h 00m 20.481s Batch 8 finished. 00h 00m 25.065s Batch 9 finished. 00h 00m 29.746s Batch 10 finished. 00h 00m 31.579s Batch 11 finished. 00h 00m 33.073s Batch 12 finished. 00h 00m 35.065s Batch 13 finished. 00h 00m 40.271s Batch 14 finished. 00h 00m 45.012s Batch 15 finished. 00h 00m 46.703s Batch 16 finished. 00h 00m 48.323s Batch 17 finished. 00h 00m 50.490s Batch 18 finished. 00h 00m 55.017s Batch 19 finished. 00h 01m 01.531s PI value output -> 5895585A0 Statistics Radeon R7 240 calculated 14.0% (60.972s + 0.549s) Radeon HD 7870 GHz Edition calculated 86.0% (58.492s + 1.236s) and only r7 240 (at default): Code: OpenCL 2.0 AMD-APP is ready. Timer: HPET (14.32 MHz) OpenCL GPU: AMD Radeon R7 240 (5 CUs, 1100 MHz) Compiling OpenCL kernels ... done. Calculating 1.000.000.000th digit of PI. 20 iterations. Allocated device memory : 335546368 Bytes Batch Size : 20M Reduction Size : 64 00h 00m 10.871s Batch 1 finished. 00h 00m 21.888s Batch 2 finished. 00h 00m 38.043s Batch 3 finished. 00h 01m 12.720s Batch 4 finished. 00h 01m 45.755s Batch 5 finished. 00h 01m 56.626s Batch 6 finished. 00h 02m 07.646s Batch 7 finished. 00h 02m 23.563s Batch 8 finished. 00h 02m 56.979s Batch 9 finished. 00h 03m 28.813s Batch 10 finished. 00h 03m 39.682s Batch 11 finished. 00h 03m 50.701s Batch 12 finished. 00h 04m 06.855s Batch 13 finished. 00h 04m 41.532s Batch 14 finished. 00h 05m 14.568s Batch 15 finished. 00h 05m 25.440s Batch 16 finished. 00h 05m 36.458s Batch 17 finished. 00h 05m 52.375s Batch 18 finished. 00h 06m 25.791s Batch 19 finished. 00h 06m 57.628s PI value output -> 5895585A0 Statistics Calculation + Reduction time: 413.710s + 3.882s All have same result and performances related to frequencies(1225,780) multiplied by cores(1280,320) So, anyone else tried forza motorsports with extra overclock with all vram consuming settings set to bottom and all others at top?
I do all my OC based on the lowest possible denominator, and then reduce it a bit on top. I never OC per game/program.
But 3dmark firestrike and cloudsomething and catzilla crashes even at chip default(not even factory default) frequencies. I cannot back from that. All other dx11 games work fine up to %15 oc. I am just thinking about this may be %22 for all dx12 games. Vithout extra voltage. Just tried 1240 MHz (dx11 limit + 90 MHz) and it played well for 2 laps then I clicked exit game, it crashed at menu when exiting. No artifacts, just locked. Maybe because of -6 mV from default voltage?
It is stable at 1150/1375 for other games and benchmarks but not stable at 1000/1200 for 3dmark_2013 and catzilla. CPU and RAM are stable in prime95 which is ridiculously power consuming.
Finished last map with 3 laps using 1230 MHz on default voltage (max oc for dx11 + 80 MHz). No lock, no artifact. I cannot try all combinations of settings but most gpu exhausting combination passed a map with rainy weather and soily terrain. Next map(rio) failed. It freezed in last lap but hitting esc saved it from infinite freeze maybe. Then I exited game. Now I will try +75 MHz again in this map. Edit: +75 MHz successful in last map. Maybe +70 MHz is needed for all maps? Graphics to compute ratio is not same between maps(and games) so it could be very hard to find a minimum oc level for dx12. But dx11 lets you find max oc for pure graphics while opencl finds pure compute limits(which is 100 MHz more than dx11 for my card).
look into raising your gpu voltage or seeing if a different bios revision uses different voltage....my card would crash every time brand new until i flashed a different bios or raised the voltage with afterburner. Your issue COULD be the same
But changing bios changes memory timings and many other things that idk so this changes performance. But I wouldn't say no to a memory tightener software so I have better performance at default vram frequency.
Usually for intense games I increased the power in AB to 25%+. I think from factory my 280x was a little power starved at the configured clocks. Similarly I could configure 5-10% OC on the GPU and drop the vram to 1300mhz and no power modification.