It's all fun and games until someone loses a GPU.
All we can do is try to educate/warn.. others will skip the reading/research and/or not care...
MOST Pascal cards are simply POWER LIMITED. This is where this all started. If you have a single 8-pin connector, you will ALWAYS be power limited. Flashing another BIOS isn't going to add an additional 6 or 8-pin PCI-e. TDP/MAX POWER increases are possible by flashing another BIOS. There are some vendors that offer a "GTX 1080" and will also offer a "O/C GTX 1080" where the identical PCB is used but the BIOS is slightly tweaked to offer a factory overclock. There's NO GUARANTEE but it might be possible that the "O/C BIOS" has a higher TDP/MAX POWER limit. Some have reported this and have been successful but it's really not being tracked/logged to share. It's a matter of fishing through 60 pages of posts.Known safe flashing
= reference to reference. IDENTICAL custom PCB to custom PCB. Both assuming the memory type is identical also.
Any cross-flashing using a BIOS from a different PCB design (reference or custom, if it's different than yours) is taking a risk
(up to and including bricking) and if you don't brick it you may find oddities like someone mentioned they lost 1,000 max RPM on the fans. Instability because the voltage controller is different and voltage is unstable. Artifacts because the memory is running at the wrong voltage and/or timings. It's REALLY NO DIFFERENT than flashing a BIOS from a different model motherboard with the same chipset as yours.. in theory... although the MB flashing would be much more risky you get the point. It's a BIOS intended for a DIFFERENT GPU DESIGN (even if it has the same GP104 chip soldered).
Right now we have the luxury of all the cards being new and there's LIMITED brands of memory in use (some reduced risk now). Over time
, just like every other Maxwell GPU, there's going to be a mix of memory... Micron, Samsung, Eplida, Hynix (SHYNIX) will be out there and the risks when this happens will INCREASE. People are already noting that different memory brands are started to be seen. You really want the BIOS for your specific memory type because I know for a fact that vendors use a different BIOS when they change memory types (at least with Maxwell cards).
A couple FPS isn't worth it... it really isn't..
What we REALLY need is the BIOS editor (and much patience)
. This should allow us to manipulate the behavior to get the most out of our GPUs. Even these "O/C BIOS'" are limited and would be further modded if we had the editor. This means puzzling throttle points could potentially be removed (voltage, power and thermal). Custom fan curve could be added in BIOS to compensate for pushing the GPU harder. Boost clocks could be modified to not be variable (WYSIWYG). The maximum power for each PCI-e connector could be added. These are all changes (and there's more) that have NOTHING TO DO (directly) with trying to overclock the GPU higher. These changes would take the STOCK BIOS and manipulate the behavior at those STOCK parameters and basically help maintain the same maximum overclock you currently have without throttling down like everyone is reporting.
You want to overclock higher? It's probably MORE of a silicon lottery than ever before.
It seems to me that ASIC% and voltage is almost completely irrelevant with Pascal.
We saw this trend (with a MOD
= ASIC% meant something. Additional Voltage mattered and usually meant higher overclocks (almost guaranteed). A low ASIC% may put you over the threshold of not being able to supply enough voltage to reach the particular chip's maximum overclock therefore a high ASIC% was desired. Too much voltage meant too much heat.Maxwell
= ASIC% was almost irrelevant. Additional voltage helped some but not all. Mixed bag. All Maxwell chips could reach maximum overclock potential of any given chip with the voltage range that was realistically available (between 1.2v and 1.29v) which meant that having a low ASIC% didn't mean you couldn't reach your max overclock it just meant you might have slightly higher heat output when reaching it.Pascal
= ASIC% may be completely irrelevant. Voltage doesn't seem to matter much for most. The fabrication process is so tiny now (remember they skipped 22nm and went from 32nm to 16nm). We haven't been able to thoroughly test this yet. So far, indications point to additional voltage leading to instability not higher performance. As the trend shows, ASIC% and voltage is even LESS important overall now.