well, i will try clean install later :)Quote:
Originally Posted by Ket
Printable View
well, i will try clean install later :)Quote:
Originally Posted by Ket
New vBIOSes uploaded. Check changelog for all details.
Very nice Ket. :) Thank you. :)
Do you have any knowledge on different bios providing better overclocks or are they all too similar to really make a difference?
The only thing that makes a difference is voltage on graphics cards. Its not like mainboard BIOSes where you can put different BIOS modules in to better match hardware configurations. There are exceptions to this rule though, particularly with ATI cards. My XFX HD4830s actually use a Asus vBIOS because the XFX vBIOS caused BSODs whenever I OC'd the cards in CF. The Sapphire HD5830 I have uses a Gigabyte vBIOS because it allows a bit more voltage, and hence a better OC. None of these are issues with nVidia cards though, so theres no need to use a different manufacturers vBIOS.
hey ket the link u upload says: THIS LINK IS TEMPORALY UNAVAIABLE..
edit: LINK IS NOW AVAIABLE.. don't know what happened.. anyway thank you 4 the excellent job KET!
Who wants to be a bit of a guinepig? Doesn't involve anything major just until I get my new GTX460 I can't do this myself. All someone has to do is use a modded vBIOS that I'll edit with lower 2D voltage for the new idle clocks. Looking to make it as power efficient as possible. 0.925v is probably enough voltage for 200MHz idle, but I need a tester.
Hello guys, I got another GTX 460 here. The brand is Twintech which I notice you don't have yet, and here's the bios (768mb version) :
http://www.mediafire.com/?x1ejhc6y8jh1c5l
I notice that the stock voltage is 1.012 which is weird, overvolted from factory? Anyway, I got a problem, that the card randomly throttles to 405 core. For example I might be stressing with furmark and then at some point it goes 405 core and all testing from there on is useless, and I need to restart in order to get it back to normal...
I read something about dual displays causing that, but even though I have a projector hooked up (via audio-video receiver first) besides my screen, the receiver is off and is not visible in nvidia control panel and this throttling stil happens....so I don't think it's related...
The twintech seems reference design, is it safe to try aby of those bioses in your list KET as long as it's 768mb? Is there any of them recommended for a reference card?
Actually 1.025v is the real nvidia stock reference voltage, however manufacturers have choosen to undervolt, and all is well since the gf104 is such a efficient GPU. So your 1.012v is more than most, but its still under the real stock.Quote:
Originally Posted by Konny
And about your throttle issue, are you overclocked, or stock when this occurs?
If your overclocked the driver will throttle the card down when it detects problems, i.e. your overclock is not stable. This is the norm.
If this is happening on stock, you may want to RMA, or I hear nvidia will have forceware 260.xx out today, try that if your still throttling on stock you have problems. FYI the card will throttle on less intensive 3d apps/games to save power, but absolutly not on furmark, unless there is a problem, i.e overclock too high, temps too high, or defective.
Of course it is when overclocked...Quote:
Originally Posted by CrazyNutz
OK great, now it makes sense.
One last thing, is there some way to figure out if the memory or the core has cause the problem when furmark fails or when you have this throttling?
Because I think my core is stable at 846 and my memory can pass Vantage at 2270, yet its not furmark stable there, so I was trying to lower it to find where I can pass furmark with 846/2xxx , and I keep getting errors after 2 hours for example and I have gone down to 2050 memory now which seems too far off from what vantage is stable at, so I wonder if maybe the core is not 100% stable but rather 98% and that brings the error after 2 hours and not the memory...
2h furmark is a real long time. I think if it passes for that duration, just about any game you play will be fine forever. Furmark puts an abnormally high load on the card, a completely unrealistic load to that of which a game would actually put on the card.
Even though you have a point, I disaggree.Quote:
Originally Posted by cantankerous
It's not about the stress, it's about the chance to crash, and yes furmark has a higher chance to crash during a given time than a game because it stresses more, but the fact that furmark crashes means the game can crash as well, it's just rarer.
It's all about the "chance to crash".
If you crash after 2h furmark, you can say this translates to a chance to crash during gaming of, let's say 3% per hour (totally random figure, just for argument's sake). This means that roughly every ~34 hours of gaming you will get 1 crash (you could of course get 3 in a row if really unlucky, or none for a few months), which might be totally fine for many people. 34 hours of gaming could be a week or a month for some people, so if they get 1 crash per this period, who cares?
But when I play starcraft online, this 1 time that it crashes and I get a loss, I will curse everything and everyone :P so I don't want even that to happen...
And generally I see OCing not as a way to get maximum performance, but as a way to get free performance without sacrifising anything. If I lose even 1% of stability then there is no point for me. I want to gain as much as I can while maintaining stability up to the stock level.
After all, let's face it, you get for example 40 fps with stock clocks, and then you OC and you get 45....big deal, in real-life do you notice those 5 more fps? Usually not...so getting something you won't really notice and in the same time getting a crash here and there that you WILL notice, is pointless...
Anyway, I don't consider my OC worthwhile unless it passes more than 10 hours of furmark/linx/whatever...and when I reach a final point, I test my PC for a whole 24h.
OCCT GPU test with error checking on. If it shows errors your not stable.
You will know it's stable when your up to max temp, and error free for at least 20-30 mins
Does the flashing procedure in the OP flash both cards in an SLI setup, or just the top one?
No. GF104 GPUs have a hardwired VID, which varies from 0.937v to 1.0??v.Quote:
Originally Posted by CrazyNutz
Thanks a lot for the advice!Quote:
Originally Posted by CrazyNutz
And OMG, what a difference....
I had to go as low as 803 core to become error-free with stock voltage...
And I could pass 4 hours furmark with 846 core.... isn't that extreme?
Anyway, since I am not going for extreme fps, but want 100% stability, this error check ability of OCCT is great. I will reach a point where I am error-free for hours and leave it there....
Wrong! VID is changable via bios mod I do it, and so do others. It can vary, or you can set the range low/high to the same VID and lock it to anything up to 1.21v.Quote:
Originally Posted by Kelso
No prob:up: And here's a little more advice, set the mem to stock then try for more on your core/shader. when you get your core/shader clock at max stable, then start raising your mem clocks. you may not be able to get your mem clocks as high, but you may get your core/shaders up higher, which will benefit you more than the mem.Quote:
Originally Posted by Konny
One more thing, OCCT, and furmark put EXTREME loads on your GPU way more than anything else. Running these things for hours is putting an unnecessary beating on your GPU/MEM/voltage regulators. You could damage your hardware. 1 hour OCCT is more than enough.
Yeah I knew about that, it's been this way for decades, always the core gives you a much bigger gain.Quote:
Originally Posted by CrazyNutz
I tested the core and memory individually and to reach 1hour error-free I ended up with
core 803
memory 2034
Now I am testing both of those at the same time to see if this is stable, and if so, I guess it's teh max my card can do error-free.
It's stil amazing however how much higher it can bench/pass furmark....I could play with 846/2260 for hours in furmark, so I wonder when everybody says their max OC what exactly do they mean....cause if they mean error-free, then my 803 core seems pretty low, not to mention the stock vid of my card is 1.012...
Anyway, if it passes a couple of hours at 803/2036 I'll consider that stable and then tomorow I'll start playing a bit with extra voltage to see how high it can go. But I doubt it would be worthwhile as I expect 1.087 to get me around 850, so IMHO it won't be worth the risk (added temperature and wear on the core)...
Yeah most people post their max OC's as it can run some furmark, and 3d mark passes. Not saying thats bad or anything, that's what it's mostly about here ;)
For me Im like you, I want it to be as stable OC'ed as If it was stock.
Top card only. You must tell nvflash which card to flash otherwise it will just flash the primary card.Quote:
Originally Posted by bhavv
No, they don't. You can set GPU voltage anywhere between 0.82v and 1.2v via vBIOS mod.Quote:
Originally Posted by Kelso
Furmark/Kombustor/OCCT are absolutely useless. In the case of Furmark it just puts a lot of stress on GPU and VRMs, Kombustor is just crap, never seen it take the GPU over 15% utilisation. OCCT is just a torture test, not a stability test. If you want to truely test your graphics card properly for stability (GPU and mem) use Unigine Heaven 2.1. Set max AA/AF and resolution and just let it loop. GPU stays over 90% usage basically always and Heaven gives the memory a proper workout so you can test for artifacts at the same time. Kill two birds with one stone, if you will.Quote:
Originally Posted by Konny
I passed OCCT error test for 10 hours with 803/2034. Even 1mhz above those, gives errors pretty soon.
The Core real clock is 797.3 as reported by gpu-z sensor. I guess there are fixed frequences the gpu can take and you can't force exacly 800 for example. After 797.3 the next step is 810 (if you choose from 804 and above you get 810) in which it's not error free.
Kinda disappointing that I ended up lower than 800, but whatever...
I also tried adding voltage, I was expecting to hit around 850 with 1.087 and stil I wouldnt consider it worthwhile, but I was even more disappointed at that point. I couldn't even do 822 with 1.087v and I abandoned it completely, I don't really care to find if it can do 810/815....big deal...
Yeah I'm gonna blank that because I've seen real world utilisation. Heaven 2.1 puts extreme heavy load on GPU and memory and its based on a real game engine. To prove just how "good" OCCT is, I got lazy when OCing so used it to test CPU stability. Passed 2hrs, crashed in minutes in a real game.
I can't speak for GPUs, but on the cpu front I have to disaggree with you. If you use the Linpack mode, it's very picky, and as soon as I achieved 10 hours linpack stable on my i7 920 4ghz, it has never crashed for 8 months running 24/7...whatever I threw at it...Quote:
Originally Posted by Ket
I was using linpack. Theres just no substitute for the real world environment.
Inno3D 768mb/1GB bios dumps missing from that list ;)