My experience with overvolting nVidia cards has been...

[ArchAngel777]

My experience with overvolting nVidia cards has been underwhelming. I have personally found that the extra voltage does very little to increase my overclocking head room. To give an exampe:

My MSI HAWK 460 can hit 822/1644/2000 on stock voltage. The temps are very low, 60's under full load. If I max out voltage in MSI afterburner, I can't even get it to hit 850 stable. This is so very unlike CPUs. You feed a CPU more voltage and see some return on your investment. Sure, there is diminishing returns, but it is at least predictable.

If this were the only nVidia card I tried to overclock via overvoltage, I wouldn't have started this thread. However, noticed this exact same scenario with my other 460 GTX (difference brand) and my older 280 GTX. None of them have ever increased their head room much, if any, by increasing the voltage and my 280 GTX (being a high end card) didn't really overclock much at all, not that I expected it too.

With that said, after my testing I would always just put voltage back to default and leave it at whatever it was able to handle at stock voltage and let it be, which I am totally fine with. But, I guess, I wanted to see if others have had the same experience or if perhaps they have had a totally different experience and found that voltage had a large impact on their clocks speeds.

My theory: I think many of these cards are actually underclocked (especially in the case of the 460) while retaining their near maximal voltage. It would be like a Sandy Bridge 2500K shipping with 1.5v as stock, yet still leaving clock speeds at 3.4Ghz (underclocked). So you can overclock to 5Ghz on stock (1.5v - in this scenario), then you think "Ahh, but then I will add more voltage" but find you don't get much further because the diminshing returns already took place long before. You found that this added voltage just doesn't do much, so you get a meager 5.1Ghz or 5.2Ghz out of 1.6v.

Edit ** The reason I specifically asking about nVidia cards is because I have no experience with overvolting and overclocking AMD cards. So, there is no hate on ATI. I loved my 6770M in my laptop.

 

[Ben90]

*deleted* because it was significantly less fanboyish at the end of OP than I had initially thought

 

[Rifter]

Overclocking has always been hit or miss. With anything, CPU's or GPU's.

 

[ArchAngel777]

My experience with overvolting ATi cards has been underwhelming. I have personally found that the extra voltage does very little to increase my overclocking head room. To give an exampe:

My XFX 4890 can hit 870/995 on stock voltage. The temps are decentish, 70's under full load. If I max out voltage in rivatuner, I can't even get it to hit 900 stable. This is so very unlike CPUs. You feed a CPU more voltage and see some return on your investment. Sure, there is diminishing returns, but it is at least predictable.

If there were the only ATi card I tried to overclock via overvoltage, I wouldn't have started this thread. However, noticed this exact same scenario with my other 9800 XT (difference brand). None of them have ever increased their head room much, if any, by increasing the voltage and my 9800 XT (being a high end card) didn't really overclock much at all, not that I expected it too.

With that said, after my testing I would always just put voltage back to default and leave it at whatever it was able to handle at stock voltage and let it be, which I am totally fine with. But, I guess, I wanted to see if others have had the same experience or if perhaps they have had a totally difference experienced and found that voltage had a large impact on their clocks speeds.

My theory: I think many of these cards are actually underclocked (especially in the case of the 460) while retaining their near maximal voltage. It would be like a Sandy Bridge 2500K shipping with 1.5v as stock, yet still leaving clock speeds at 3.4Ghz (underclocked). So you can overclock to 5Ghz on stock (1.5v - in this scenario), then you think "Ahh, but then I will add more voltage" but find you don't get much further because the diminshing returns already took place long before. You found that this added voltage just doesn't do much, so you get a meager 5.1Ghz or 5.2Ghz out of 1.6v.

Ben, check out my post history. I gave the reason while you were posting as to why I referenced nVidia. I have no experience with ATI, so it would be unfair to include ATI on my personal observations. You, on the other hand, are trolling this thread.

 

[Grooveriding]

I think GPUs are designed and intended to push the envelope more than CPUs are in terms of maximizing available headroom. Nvidia and AMD want to put out a product as fast as possible and due to this put out products operating near their limits for the most part.

Also, CPUs are a necessity in all computers, many are work-place related and/or reliability and longevity are critical and much more of a concern, so they are put out with conservative clocks and voltages. GPUs not so much, consumer video cards are much more prone to upgrades, and many computers have no use for a discrete card.

Intel puts a lot of weight in longevity and reliability. Since Core2 it's been a breeze to get a 50% overclock on any CPU they've put out. I haven't seen a GPU that can do that without some insane cooling in a long while. The Sapphire 7970 discussed in another thread on this forum right now with the planned 1325core clock is the nearest I have seen to a 50% overclock on a GPU without exotic cooling in a long while.

It's so rare to see a CPU ever get RMAed, whereas video cards are probably one of the components with the highest failure rate and most often RMAed. I think this has a lot to do with them operating at their limits as well as there being more areas of failure due to it being more than a chip, but also a PCB full of components. A lot of us have seen cards where it has been the core failing though, how often have we seen someone posting about having to RMA their CPU ? Even with all the huge 50% overclocks we've been pushing for years now.

TLDR. GPUs are generally already near their limits because of tight competition and the greater emphasis put on penultimate speeds among the market for discrete cards. CPUs do not fall so much into this area, having a lot of headroom to spare, and I think Intel is just that good at making stuff. :ninja:

 

[3DVagabond]

I agree a lot with Grooveriding. I think voltage is run higher with GPU's because they are running close to their max clocks and the need for as many stably operating chips as possible at those settings.

 

[kami]

It's hit or miss. While I wouldn't call a 22% overclock on your 460 bad, I can see why you wouldn't be happy since I have seen many people take the Hawk to 1GHz. In my experience GPUs seem to be much more of lottery than CPUs for extreme overclocks and yes CPUs respond much better to extra voltage.

 

[ArchAngel777]

It's hit or miss. While I wouldn't call a 22% overclock on your 460 bad, I can see why you wouldn't be happy since I have seen many people take the Hawk to 1GHz. In my experience GPUs seem to be much more of lottery than CPUs for extreme overclocks and yes CPUs respond much better to extra voltage.

I may not have been clear enough in my post. I am very thrilled with my 460 overclock. I have no complaints. These are some of the best cards I have ever purchased. I just made an observation that voltage didn't really do anything for increasing clocks further. These 460's I got right now in SLI and rock stable at 822/1644/2000 (560 reference speeds).

And yes, I agree with Groover as well. But still waiting to hear from people who have had success with voltage increases.

 

[notty22]

Have you tried isolating the core when you overclock. I have 2 Cyclones myself.
Max voltage is 1.087, I can run SLI at 890, but I have to lower the memory o/c quite a bit, compared to running a single card. Memory, Stock is 900, with core o/c of 890 , I run the memory @ 940.
I know with one card, with a similar or lower core clock, I was able to get a higher memory o/c, but it also did not return as much performance. Compared to core o/c.
On these cards , as most gtx 460's there is no active cooling on the memory chips.
Overclocking the core raises the operating temperatures at least 10c, and that's measured at the core. Who know's what hot pockets of air might be around one of the mem chips.
That's my theory on this behavior anyways.
Stock SLI clocks of 725/900 return 78.7 fps in the AVP benchmark.
With 890/940, returns 91.6 fps.

 

[BallaTheFeared]

With max voltage you could be tripping the cards overvoltage protection (ovp), assuming you haven't already, I'd suggest toning back the voltage and start out with small increases in voltage for 830 then 840 and on up.

Some cards just tap out sooner than others, there just isn't much more you can say about it. CPU's are the same way, some people can do 5.3GHz+ with their SB chips, others can't push much past 4.5GHz.

Another option, if temps are under control is to do a bios flash allowing more voltage (up to 1.21v).