Printable View
Cool, just can't seem to dl the file. Is there an alternate place to get the file?Quote:
Originally Posted by Krizby87
Do you know if they changed the mem timings to make them more stable?
I'm trying to be fair to companies and not RMA cards I've modded. That's why I want a 100% stable card before I start in with the soldering iron. Even if they can't detect the mod once I remove it.
Ram timing is the same 31/39/22/9/10/6/8.Quote:
Originally Posted by Lerun
I will try it and see what happens.
"Only" 1600 for an unmodded Vmem is very, very good so I do not see anything abnormal there. When you get outside the maximum limit, everything can happen.Quote:
Originally Posted by SF3D
I simply do not understand why so many of you are pointing in the direction of the memory.
There are so much more arguments why it concerns the GPU, and I cannot find one single logical reason why it should be the memory.
A memory chip that malfuntions in 3D should also malfunctions in 2D, which is not the case here.
The 7900GTX has no problems with the same chips, etc, etc.
Flytek and Humeboy suggested not to install Coolbits.Quote:
Originally Posted by flytek
I could hardly believe that this was going to get me any further, but for the sake of trying, I installed forceware again without coolbits and used rivatuner instead , and guess what ???
Everything works again just as troublefree as before the problems started.
I have no explaination why a systems works properly with coolbits, then starts desintegrating and when finally replaced by rivatuner everything works fine again.
I am still very suspicious if this is going to stay, or if it is just a short delay in the dying process.
The cards when they die exhibit the SAME symptoms as they do when the memories oveclocked too far. The vMem regs get MUCH hotter than the vGPU.Quote:
Originally Posted by t024484
The GTX may use the "same" chips, BUT the regulation on the GTX is MUCH better than that of the GT. IF Nvidia used the same regulation on the GT as they did on the GTX then the GT's would simply be downclocked GTX's. They'd loose money.
The vMem regulation IS the issue on these boards. Not sure why a driver reload/reboot fixes the issue, sure theres someone with 1000* more knowledge on the subject out there.
The memory in 90% of the cases IS the issue with these cards.
<edit> As for the 2d/3d issue, the cards NOT loaded in 2d whereas it is in 3d (no s&*t sherlock...). Belive the issue is overloaded regs which overheat and crap out. Hence not occuring in 2d mode.... </edit>
Hmm.. would be worth a try, have the same problem as you, having Coolbits installed... :)
Best Regards :toast:
Hey Guys,
Although there seems to be lots of threads in other forums this was one of the first ones I've read over here at extreme systems (I thought all you guys knew something none of us did), lol! Like many I have also had these problems, in fact I'm on my forth eVGA 7900GT RMA replacement. With the first 3 I began having the problems within 4 days of installing the card.
Then about 3 weeks ago eVGA announced that there was indeed a hardware based "Issue" that they had fixed by "Slightly Modifying" the replacment cards they were sending out? Of course they have been very vague about it the fix, where it is and what it is? So 1.5 weeks ago I filled out a special form they had on their message board for people who have been persistantly having the same issues with their 7900GT's, they said once the form was submited they would call/e-mail the persons back to take care of it. Well the next day I got a message from a Jacob Freeman who sent me a special "Cross-Shipment Authorization" form so that I could get my "Slightly Modified" replacments card first, and then ship out my third defective card within 30-days.
Then about 3 days ago I got my replacment which was the Signiture Series card with a modified BIOS so that the clockspeeds would match my 7900GT CO card (500/1500Mhz). Fingers crossed so far it was been working great even though I've been doing some serious gaming/benchmarking to push the card. This card does indeed run at 1.45v in 3D mode and 1.2v in 2d mode (checked with multi-meter). I'm not exaclty sure why they would do this except maybe that these cards have had a better track record so far with RMA's?
I realize that the SS card is fairly new in eVGA's line up but does anyone know if they actually do have a better track record than the other cards? When I put both my regular GT CO card and my new SS card side by side to compare the components they are certainly different in many ways. I'm no expert so I'm not sure what componets do what but I do know they are differnt in many ways. Well thanks for any help you can give guys and good luck to anyone who is currenlty having issues with their 7900GT's!
Anyone with a signature series based card able to get a hi res photo of the back of it? Might be worth seeing if theres any difference between them and a normal EVGA. May well be that its *just* a voltage boost to the core which has fixed the problem (although this is unlikely since people running over stock vGPU are still having issues) or that the regulation has been uprated to cope with the increased voltage....?
The changes to the SS board where the non SS boards have to be pencil modded or soldered to change the Vgpu voltage are already well communicated. Everything was already there on the PCB, but no components where installed.Quote:
Originally Posted by Roger_D25
With the newly installed mosfet switches, it is now possible to have a dynamic setting of te Vgpu by the GPU itself. Theoretically it is even possible to mount to components on boards not equiped with these parts and have your own SS fabrication.
But in all sense, the skills required to do this are beyond the average amateur.
My guess for the why is, most of the time your PC is in 2D, and the 1.2 Volt, applied to the GPU, let the GPU live much longer. Why should EVGA add extra components if 1.45 Volt all the time is just as good ?
So this could be one of the "Issues" that EVGA reported to extend the life of the board.
What I am really interessed in, can you find any differences around the Vmem regulator, the ISL6549, between the two boards that you have, and does the regulator plus one of the two mosfets next to the regulator becomes just as hot on your SS board??
Highlander,Quote:
Originally Posted by Highland3r
What I dislike in your mail, is that you are just posing wild assumptions as being the thruth.
1) "When you overclock the mem, the symtoms are the same as when the card dies" ?? Where is the prove for that .
2) "The GTX has a MUCH better regulator for Vmem" ?? Is the regulator on the GT worse because it gets hotter? That would not be my definition of good and bad. The regulator on the GT is designed for 1320 Mhz. Overclocking the mem means more current demand, so the poor guy gets hotter. But is it not good because of that ? No way, it is probably an excelent regulator until the prove for the opposite is there. With proper cooling everything should be O.K.
Without proper cooling your GPU would also die, but is it a bad GPU because of that ??
3) "The memory is the cause of trouble in 90% of the cases" ?? Where is the prove?
By replacing coolbits with rivatuner, my board seems to function again.
Does Samsung memory not like Coolbits?
It is not my intention to piss you or anybody else off, but before pretending to have some black and white knowledge, you should be damned sure.
1. personal Experience... I've pushed GT's to far and they artifact. You get the same flashes/refresh issues on desktop when the rams pushed too far and FPS and scores drop. My XFX recently died. Guess what happened? It artifacted, refresh problems when switching 3d -> 2d and shock scores/fps dropped too..
2. If the regs on the GT's were as good as the GTX then they WOULDN'T overheat would they..? Yes the regulation is designed for stock running, but that ISN'T what I said.. The GTX is designed to run the ram at GTX speeds, and the GT at GT speeds...
3. See my above comment for #1 mate.
The reason IMHO Evga are using different 2d and 3d voltages is probably to keep heat down. I've not seen the superclocked cards personally, but I'd imagine they run the same coolers as the "normal" cards. 1.45v is a LOT for the stock cooler to cope with. Again personal experience has shown this :)
I'm not trying to make wild assumptions. I only say things like I see them.. If someone tell's me I'm wrong, and presents valid proof to backup their comments trust me I'll be the first to agree :)
Does replacing Coolbits with Rivatuner make it stable in games as well, or only benchable??
Best Regards :toast:
Dude, you sound like your dead set against it being the memory on these cards that's causing the problem. I'm on my second card and it's acting up also. If I under clock my memory all runs good no artifacts so that is proof enough for me as when I try to run the card at stock or with oc on the memory it artifacts or crashes the system.:fact:Quote:
Originally Posted by t024484
I'm not using coolbits either so no issue there. The card will do 580 on the core all stock but the memory won't even run it's rated speed any more without artifacting. X1900XT is looking better and better.:toast:
Everything I have tried including FEAR is running without any problems (so far), but the last thing I would like to suggest is that Riva Tuner is always better than Coolbits. It happens to be so in my case, but the reason why is written in the stars.Quote:
Originally Posted by DTU_XaVier
I am absolutely not against any cause.Quote:
Originally Posted by MisterFix
All I say is that this and similair Samsung gddr3 is used all over the world without any problems. Why should it be the cause of trouble in the 7900GT. Because of that it seems unlikely.
On the other hand, there is such a thing as a power supply that could malfunction, so far no prove of that, but it is already more likely than the Memory chips.
Next to that is the Clock generator. How well is this circuit designed, and how is the clock instructed to change the frequency, probably by the GPU, but we don't know.
A fact is that the memory is not producing its own clock, in fact the memory knows nothing about clocks although it should be clocked in the proper way. So when in an overclocking situation, like you describe, the memory gives up, is it the memory, the power supply or the clock generator. It can also be the GPU that is not able to read and write to the memory in the proper way or being not able to instruct the clock generator properly.
There are so many reasons, so without any further knowledge or hard facts, it is just guessing.
I think time will learn
That is the logical conclusion, so I do agree. Especially when the same volt controll cuircuts are used in 2D and 3D.Quote:
Originally Posted by t024484
I was just reaching because for my card it seems like memory problems.
I have also this strange ting, for HL2 based games I can overclock the mem, no problems. Running 3Dmark DeepFreeze og F.E.A.R the card crash every time. Need to go down to nVidia stock speeds (1320) to make it stable :confused:
I'm also agreeing that it's a heat issue, if they speced the IC for 1320, then this might be why OC'ed cards are having this issue? One would think that that there would be some headroom. I also think I remember reading something about some bad components comming out of china, and to stick with the japanese made ones. But this is just speculation on my part.
t024484 you seem the person with the firmest grasp of the situation and done som good investigations of the problem. What is you'r take?
Don't (personally) believe the Ram chips themselves are "going bad" its more a case of the regulation thats failing.
Hi guys,Quote:
Originally Posted by t024484
I agree with t024484 mostly, though I think the other guys meant including the Vreg as well when they were stating that memory is the issue.
Still, I don't think GPU can not stand a voltage of at least 1.4V. (If added more I am not sure.) The main reason is that GTX GPUs are at 1.4V, and I haven't heard of any technology that can distinguish voltage tolerances between chips that are from the same fabrication process without permanently damaging them. Therefore, I think 7900GT GPUs can stand a voltage so long as GTX ones can stand it. So the question here is that is there a 7900GTX died because of some 1.5+ vmod?
I don't think the Vreg being hot is a sign of the cards' dying. Mine is hot from the beginning, even before I vmodded it. It stays at 640/800 well at least for now-- it's about 2 months.
So let's hope your card stays well from now on.
I have planned for this weekend to do some thorough testing on the Vmem regulator, and try to find out why the upper mosfet and the regulator are getting so hot.Quote:
Originally Posted by Lerun
If I find a solution I will let it know.
So far Im happy and running 1.55 VGpu and no other mods, Dont dare to do the vmem yet :)
Tried removing Coolbits today, and started playing F.E.A.R...
550/800 for clocks... Went fine the first 15-20 mins, then the core reset to nominal, though the game kept running fine... When I went back to Windows, the problem was back... So, I suppose I just oc'ed the core too high, will keep testing :)
Best Regards :toast:
It now turns out it must have been my core causing the problem... have removed Coolbits properly (forgot to remove the registry entry) and have played around with ATI-tool... the mem went all the way to 1800 before artifacting, but the core only got to ~530-540...
So removing Coolbits has removed my problem :)
Best Regards :toast:
Today I spent quite some time in measuring the performance of the Vmem regulator.
The conclusion? Everything works as it should.
The regulator is pulling some 25mA, independent of the load or frequency of the memory, so with the 12Volt supply, it consumes 1/3 watt.
With a thermal resistance of 100 C/W, this mean a temp rise of 30 C.
Everything is within the specifications and nothig to worry about.
The Voltage regulator for the GPU has a thermal resistance of only 50 C/W, thats why it does not get as hot as the Vmem regulator.
The upper mosfet, connected to the 12Volt, has a thermal resistance of 62 C/W.
I have measured with my osciloscope a rise and fall time of 10 nsec.
Switching frequency is 630 Khz.
With these parameters the switching losses when this fet is turned on are:
12(I+1)/300 , where I represents the current that is supplied to the Ram.
When this is 6A, this dissipation is 0.28 watt
The switching losses when the Fet is turned off are: 12(I-1)/300.
With the same 6A, this gives 0.20 watt.
Then there are losses to discharge the lower Fet.
This results in roughly 0.12Watt.
As a last dissipation factor, whe have the current flowing through the fet given the on resistance and the duty cycle. This dissipation is some 0.08 watt
In total whe end up with 0.28+0.20+0.12+0.08 = 0.68 Watt in an ideal world.
Let's give it some extra margin and assume it is 0.8 watt.
With the 62 C/W, this give a temp rise of 50 C.
At a room temp of 20C it results in 70C, enough to burn your fingers.
There happens to be nothing strange to he fact that the Mosfet gets so hot.
But with proper cooling, just a small heatsink measuring 20*20*20 mm. temp will never get above 40 C, so well under control.
Now the performance of the regulator: With loads varying wildly, it stays as stable as a rock. There is no current limit coming into action with all the possible memory loads.
So my conclusion can only be: apart from the cooling aspect, this voltage regulator is as good as you can possibly wish,solid as a rock and not sensitive to OCP or OVP.
t024484,
First of all thank you very much for taking the time to fully test the regulator, this is great info to know! Especially with all the talk about this card's mosfet temperatures being so high, it's nice to know their within spec! Not only that but it also narrows down the possible problems area's by rulling out the regulator section.
Also thank you for giving my your thoughts on the SS card I recenlty got. After looking at my old GT CO card it's clear that all the additional components on the SS could be added the GT CO without much trouble if your know what your doing! I just got out of work about an hour ago and it's quite late, otherwise I'd fully compare both my cards and answer your eariler questions about the VMEM regulator area differences. I've got tomorrow off so I'll do it then. I'll also try to take some close up shots of the VMEM regulator area on both cards so that you can compare them yourself as you'll notice much more than I will.
As far as the temps go I'll use my non-contact infrared thermometer to accuratly measure the temps of any points you want (regulator and mosfets). Thanks again to24484 for your help, talk to you tomorrow.
Something came up, was reading alot and ppl talking about the load dist of the 12V rails.
Even if both my 12V rails are rated for 20Amps this seems not to be enough. Problem with my PSU is I think the two 12V rails are devided between CPU and rest. So it may not be that easy to test without tapping into the cpu plug.
I was going to destribute load by using the 2xmolex to PCI-E converter to see if I could get a more stable system.
Can also someone else with more 12V rails try this?
We might be underestimating the load on the PSU with overclocking and all. Also manny PSU does not perform well if loads are not balanced between the rails (3.3, 5 and 12)
Lerun, did you connect the 47Kohm resistor that I described here ?Quote:
Originally Posted by Lerun
http://www.xtremesystems.org/forums/...=94177&page=20
it is the last mail on this page.
In my case it removed the crashing completely.
Give it a try, and see what happens.