Can someone explain graphics cards to me?...

[Benjiman46]

Okay... I'm not as dumb as the title suggests, but I'm a bit confused.

 

Take the GTX 970, I'm lead to believe nvidia released the graphics card, then companies like EVGA, ASUS, Palit and MSI use that card as a 'reference' to make their own cards... Correct me if I'm wrong.

 

So, my question is, if I buy the cheapest nastiest GTX970 I can find, slap a water block on it, and overclock it higher than would be possible on any standard air cooler, would it be any different if I bought a more expensive card and did the same thing?

 

Takes a step back... lol

 

Cheers

[Tedster]

A more expensive non-reference card would be better because of improved power delivery systems...and other things that I don't know.

[Zelx]

That's why people buy reference cards to overclock in general.

Aftermarket ones just change their pcb and use a bettter cooling, allowing them to overclock it a bit better.

Also sometimes some useful gimmicks like dual bios can be included too.

[Mao_Zedong]

off topic: I would get a custom loop for your first build

 

On-topic: If it has the reference pcb, it will perform around the same as other cards with a waterblock.

[Vacsol]

Nvidia makes the chips and provides other companies with a licence to make cards.

 

All reference are almost exactly the same, however if you get a shitty card with a non-reference PCB, you won't be able to put most water blocks on it.

[Sauron]

Many brands go further than redesigning the cooler and actually design their own PCB. This (suposedly) allows for more overclocking, better power delivery etc.

USUALLY however waterblocks are only compatible with reference pcbs. Therefore if you buy any card that hs a reference pcb and slap a waterblock on it you should get the same experience regardless of brands. BUT, some brands take reference pcbs nd bin their gpus more, meaning you're more likely to get a good overclocker.

[minibois]

Nvidia makes the chips and a design others can base upon. These other companies makes other coolers (that sometimes are better) and sometimes they make the chip faster or give it more memory.

You can put a waterblock on all cards I think, but the reference cards are more known or whatever, so it is easier to find a waterblock for that (rather than for an Gigabyte card or whatever.. 

 

EDIT: some manufactures do have the same 'PCB' (basically the back, on which you can put a waterblock..)

[Benjiman46]

off topic: I would get a custom loop for your first build

 

On-topic: If it has the reference pcb, it will perform around the same as other cards with a waterblock.

 

I've got one if I were to buy a waterblock I'd just need more piping and fittings B) 

 

Thanks everyone!

[xBlizzDevious]

Okay... I'm not as dumb as the title suggests, but I'm a bit confused.

 

Take the GTX 970, I'm lead to believe nvidia released the graphics card, then companies like EVGA, ASUS, Palit and MSI use that card as a 'reference' to make their own cards... Correct me if I'm wrong.

 

So, my question is, if I buy the cheapest nastiest GTX970 I can find, slap a water block on it, and overclock it higher than would be possible on any standard air cooler, would it be any different if I bought a more expensive card and did the same thing?

 

Takes a step back... lol

 

Cheers

 

A reference design is normally the "low-end" model. Every manufacturer will make a card that is almost exactly the same though may use different components but the design and layout is the same.

A non-reference card may (or may not) have a different layout, different power delivery (VRM) and so on. Generally, the non-standard cards are of higher build quality and will reach higher clocks. In some cases (like the Asus Matrix cards) they even cherry-pick the GPU itself to make sure that it will overclock better than a standard card. The drawback of non-reference, is that not all of them will get custom made waterblocks so that idea goes out of the window. If you want to watercool a GPU, you're safest bet is with a reference card. Some high-end cards will get custom waterblocks built for them, however.

[DavidTheWin]

NVidia and AMD make a reference design, this covers chip/memory/power delivery/pcb layout etc. The vendors such as EVGA, MSI, Gigabyte and so on can then sell these reference designs and also make their own where they can improve the PCB layout with better power delivery, different memory capacity, different port combinations, larger PCBs, different/more/less power connectors and so on. Usually the biggest change a vendor will make is the cooling, better fans/better heatsink and so on. A more subtle change is that some will pay NVidia more to get better binned* chips so those ones are likely to overclock better. A good example is EVGA's classified (or ASUS's Matrix Platinum) series where they try to get the absolute best GPUs they can, then bin what they get themselves so that they can put the absolute best chips in their classified cards and put the more average chips into say their superclock models. Cards like these might have even more exotic features than the ones I mentioned above such as memory heating to stop the memory chips performing worse when being cooled using liquid nitrogen or dual bioses or voltage check points and so on.

 

The biggest advantage of a reference design is that they are usually cheaper (although a Palit Jetstream 980 is cheaper than an MSI reference 980, at this point you're paying for better warranty/support/brand name/possibly a better binned chip etc) and that you will always be able to find a waterblock to fit. Some waterblock manufacturers will produce a waterblock to fit more common aftermarket GPUs, I think I saw some EK blocks to fit ASUS DCUII GTX 780s but that's the exception.

 

*A quick explanation on "binning" (There are likely to be inaccuracies here). When a silicone wafer is made it comes out as a big disc with lots of chips on it. Chips closer to the centre of this disc will be of a higher quality than those to the outside. From this, the manufacturer will split those ones in the middle off to be their best models and the ones towards the outside to be the worse models. A good example here is Intel's 5960x, 5930k and 5820k chips. The absolute best of the best centre of the wafer chips become 8 core 5960x's. Chips just out of the centre that are slightly defective such as having a core that doesn't work or some of their cache doesn't work or whatever, the core/cache/etc is fused off and they get sold as 6 core 5930ks. The ones right out on the edge that don't perform quite as well as that become 5820ks. Since whether your chip was closer to the centre or not is a complete mystery to the customer this is what is known as the "silicone lottery". I buy a GTX980 and it maybe gets 50MHz boost before crashing I got a mediocre 980 chip that was maybe a few centimetres away from being a GTX970 instead. Maybe I get a 980 that under liquid nitrogen can hit 2100MHz, that would be a golden chip right from the centre of the wafer. Because of this, professional overclockers will likely go through multiple GPUs to find the best one to go chase a record. Similarly, a manufacturer such as EVGA or ASUS gets a certain number of chips from NVidia or AMD. I think they can probably pay more to get better binned chips but don't quote me on that. At this point they can do their own testing to determine which are the golden chips to go into classified/matrix platinum editions and which are the more average ones to go into their regular cards. To this end, you likely couldn't just buy the cheapest, nastiest GTX980 and overclock the balls off it because you will likely get a chip that just physically can't do that. Sure being able to keep the temperatures in check with watercooling would help you get the temperatures lower than you might otherwise be able to but in the end the chip likely just can't do as much as a chip that ended up in a classified card even if the classified card was left aircooled.