Printable View
I just had another idea why the memories won't clock well: the copper plate is really really hot UNDER 3D (under 2D it's okay!). Perhaps the copper cooling plate produces that much heat, that the memories won't clock any further (kind of hot spot / heat accumulation). To confirm this idea, somebody would have to buy another cooler like that one:
http://www.maplin.co.uk/images/Full/47129i0.jpg
However then one has to find a solution for cooling the HSI Bridge...
There is another 7900GT 512MB card that uses the reference PCB shown here
http://www.itestate.com.au/products_...ode=PD00003810
Its made by Sparke and is called the Calibre, the web page is down ATM but it uses the 1.4 ram, same as the 256MB version so I would imagine its a lot better than the Gainward/Palit/XpertVision version of the card that OC's like a dog
The PCB looks identical so maybe the Vmods used on the 256MB version are the same
Page is up again http://www.calibrestyle.com.tw/main.asp
So U are saying that the mem isn't undervolted? Why the heck does it clock so poorly? I'd really like to get one of those but there's just too many question marks...Quote:
Originally Posted by persivore
Keep up the good works guys!
PS. Maybe they're planning to go with GLH version and that's why they messed up the mem here
hmmm. i like the look of the sparkle :D
anything 512mb on a new pcb would be worth testing
i'll try hunt one down. thanks alpha0ne
edit: @czuk - i actively cooled the ram and there was sufficient paste.
The memory is overvolted if anything. Its possible that the overvolting is reducing the overclock, but I wouldn't think it would decrease it to the extent thats being seen on these cards.Quote:
Originally Posted by toolboy
I wonder how it works with this one:
Gainward 7900GT 512mb Golden Sample Goes Like Hell edition 1.2ns DDR3 PCI-E
it says 580/1650 :confused: - mistake?
I've also seen one 7900GT 512 GLH on ebay.de and as far as I remember it was only 550/1400...
they've iether got a secret trick to hold the ram back or there are massive binning margins on these pcb's.
stock on the legends was 550/1400
Where the hell did my post above go ?
@flytek - have you tried underclocking the memory?
From what I've seen on other forums Gainward 7900 512 (at least quite a few of them) seems to have better 3DMarks results running 660-670 rather than the default 700MHz lol
The cooler has nothing to do with it, the PCI-E cards dont have a Artic Cooling Nv5 rev3 Silencer.
Cooling the ram doesnt help much , most of the heat is transferred via the PCB.
I have a Gainward 7800 GT 512 MB non-Glh with 1.4 ns Infineon Rams and in the beginning only 1200 Mhz were possible with 1.92 Volts .(the GLH uses 2.02 volts) . But with a new bios with reduced timings 1500 Mhz were possible und also a quite healthy gain in 3D Mark. So no speed binning , just a little less voltage and timings too tight.
I just received some hot information about the Gainward Bliss 7800GS+ card:
At the moment it is not possible to change fan speed in the card BIOS. Neither the defaults of 2D=20% and 3D=50% do work, nor the user made changes via Nibitor!!!
It's simple to validate that; change 3D speeds to 100% by using nibitor - you won't recognize any difference in noise level or airflow compared to 2D=20%.
Gainward says that Nvidia has prohibited external access (biossided.) on fan control.
All Gainward could do was changing the card ID, voltages and clocks. In the opinion of Gainward that step of Nvidia was necessary in order to ensure the reliability of the GPU.
Furthermore Gainward said that this is a common problem for many double-slot cooled graphics cards, that had a single-slot cooling solution in the beginning.
As a technical note I was able to obtain the information, that the fan isn't DC-controlled anymore but pulse width controlled.
Gainward already has adressed this issue to their design facillity in Taiwan. They hope that they will get a modified BIOS soon.
So, to clarifiy, has anyone compared the Goes Like Hell to the Golden Sample in any detail?
Do the PCBs differ? Are the bioses and timings any different? Are the voltages Different?
I assume it has higher core voltage as a start but why is it capable of higher memory on worse chips?
Is it likely to be deliberate or are there physical differences between the cards and this is just a side affect? Would flashing a golden sample with a glh bios mess it up?
I have a Golden sample coming soon as it was on a great offer, im keen to help crack this memory problem but i have no real experience with these things.
The person you spoke to at Gainward is, um, .... mistaken. Proof comes in the new ntune 5.0, which allows me to change the fan speed on my 7900GT to whatever I like.Quote:
Originally Posted by czuk
Note that this is controlled through a simple applet and no BIOS chage is required.
This 1.24V could be Vref, which is used as a comparison level for Data Input, CMD, and address signals. It is defined in the data sheet as as .7*Vddq, so if you are running your memory at 1.8V, then your Vref would be 1.26V. If this board is running it's memory at 2.0 Volts and the voltage you're looking at is indeed Vref, then there may be a problem.Quote:
Originally Posted by persivore
For a 2.0V Vddq, Vref should be at 1.4V. Usually a reference voltage like this is set through a simple voltage divider off of Vdd, so check some resistors around there and see if you can find one with 2.0V on one side and this 1.24V (or 1.4V) on the other.
Sorry to be a bit out of the loop in terms of electronics but if this is the Vref and its lower than it should be. What affect would this actually have? Does it cause instability at higher clocks? Thanks
My card should be coming today, i dont really have a motherboard to try it in at the moment but i might be able to use my dads.
Vref is critical. When the receiving circuits in the memory chip have to decide if the signal at their input is a 1 or a 0, it compares the signal to Vref. Any voltage above Vref is a 1, anything below Vref is a 0.Quote:
Originally Posted by Scarlet Infidel
If Vref is off by more than 2% of Vdd, then you are out of spec. If it's off by 5% ... who knows. If it's WAY off then your memory won't know 1 from 0 and you are utterly hosed. At what point it begins to affect overclocks is anyone's guess, but I think it's probably best to keep it within the 2% spec.
Yes i read up more on it, thanks for that info too. Basically if the vref turned out to be low is it because its been set too low deliberately, a design mistake (seems unlikely given its importance) or somethign else.
Would it be a case of simply increasing it or would this put a strain on other components. Basically is it likely to be set low or just not up to the job anyway.
Ive got my card now, i can test resistances till the cows come home but i have nothing to test it in for the time being for voltages and affects.
Honestly, I'd also have a hard time believing that it's a design mistake. It's too important and too easy to do correctly (so that Vref tracks Vdd automatically). On the other hand, this is the first time (to my knowlege) that OEM's are overvolting parts. I wouldn't be shocked to learn that something unexpected fell out as a result.Quote:
Originally Posted by Scarlet Infidel
If you trust the Samsung data sheet, Vref should be at 1.4 Volts. I don't know if setting it there is possible or if it is even the right thing to do. Samsung may have told nVidia that it's safe to overvolt Vdd but to leave Vref alone. I just don't know. All I can say is that it looks like a violation of the spec. Other than that it's all speculation.
I've sent a note about this to evga. I'll post here if I get a response.
Well as always, what they consider safe and what we do at our own risk are very different things. Im looking forward to their reply.
Its a shame theres not a little more activity on this thread, xtremesystems is always great at getting to the bottom of these things.
Would make sense to me that either vddq or vref (or both?) are a little off the specs...even the 512MB cards equipped with 1.2ns ram seem to hit a wall at around 1550Mhz although they should be quite capable of doing that speed.
So imho there's a general memory voltage problem with these cards...
Maybe i'm wrong...but shouldn't vddq always be half of vmem? :confused:
http://flytek.sa3doc.co.za/screens/7...persivore1.jpg
So if that 1.24V reading should in fact be vddq it's about 0.2v too high.
If anybody has an idea how to change that 1.24V (either up -for vref- or down -for vddq-) i'm willing to play the guinea pig as soon as my card returns from rma (got an already used one so i sent it back).
ok guys the solution to this little puzzle is to buy a sparkle calibre 7900gt 512mb.
the vreg design is the same as the 256mb 7900gt's and they should clock similarly.
i'm vmodding one within 48hours and will post results.
the core vmod is the same digital vmod as the 256mb cards and the 1.4ns memory seems to clock better than the legend/gainward style pcb's.
there will also be the same issue of risking burning the memory vreg if pushing the memory too hard.
I posted how I think the 1.24v voltage can be increased in post #86. Read a few of the posts above that one to see what I'm refering to in post 86.Quote:
Originally Posted by OneProdigy
Thanx, Persivore! I'll give it a try as soon as i have my card back.
Any idea on how to decrease that 1.24V?
Decreasing the resistance of the resistor above the 667ohm resistor should decrease the voltageQuote:
Originally Posted by OneProdigy