4 ms·

I believe Nvidia is purposely discouraging OEMs from altering their stock clocks. At least that was true for the 1070 Ti. The big thing I usually look for is

by ivl 8y ago

I believe Nvidia is purposely discouraging OEMs from altering their stock clocks. At least that was true for the 1070 Ti. The big thing I usually look for is if the PCB layout matches the reference layout. Else liquid cooling (mostly for how much quieter it is under load) becomes much more of a pain.

saganus 8y ago
> The big thing I usually look for is if the PCB layout matches the reference layout. How do you do that? I mean, where do you look for the PCB layout of a potential card you will be buying? Do you physically open the box and take a picture? Also, why would the PCB layout make any difference? I mean, I've been buying 3D cards since the original Voodoo but I just never heard of this metric. Sounds interesting but would be nice to have some more details :)
- ivl 8y ago
  I was lucky to very early on have one of those Voodoo cards, so it's kind of amazing how far we've come. Unfortunately, it's a true pain to determine whether or not a card matches reference. The reason is simple: if the layout changes, there's a good chance the water cooling block won't correctly align with the high points on the card. Having that happen means two things: you will have points where the block won't be making contact, or you will have to bend the PCB, or you will have a component not being cooled. All really bad things. This can even be impacted by the thermal pads used by the OEM. I mostly check to see if they're using the reference cooler design. I haven't bothered with AMD/ATI cards in ages, so this only really applies to Nvidia products. Cards with the blower style coolers with a single fan towards the rear of the card, with an output near the video outputs are almost always reference. Cards with coolers the OEM designs often are, but sometimes are not. That's mostly where things are annoying. Thankfully, water cooling OEMs often have specs from the start as to which cards will match if you're planning on buying day on. An example: With reference layouts the 980 and 980 Ti ought to be compatible. On some EVGA cards, you have memory mounted on the rear of the card, that makes backplate installation impossible depending on which backplate you use, as it'll teeter: either a 980 backplate, or one meant to be compatible with both the 980 and the 980 Ti. Worse, that can mean having to find a different set of screws to mount the water block you wish to use.
- cc439 8y ago
  There's no definitive rule for determining whether a card is using rhe reference PCB design or not, but generally, any card using the reference cooler (1) and most card SKUs released on a new GPU family's launch day will use the reference PCB design. As for why you's want the reference PCB for water cooling, well that is easily explained by looking at water cooled GPU heatsinks (2). A Graphics Card is composed of much more than the GPH itself. They basically stuff all the components of a mini computer (power management, memory, processor) onto a single board and each one of those subcomponents needs cooling in order to achieve optimal performance. A GPU heatsink that neglects the VRMs or memory will not be able to be overclocked as far as a GPU where those parts are actively cooled and in a liquid cooled system, there is typically very little airflow within the case. Slap an all-in-one water cooling system intended for a CPU onto a GPU and you'll often get memory issues right off the bat even with stock clocks and overheated VRMs will lead to throttling after a relatively short time in heavy use. The fans on a GPU provide enough direcf airflow to keep those components within spec, but remove the fans and replace them with nothing and you lose that necessary airflow. Since each card manufacturer uses their own PCB layout (once they figure out what to improve or lower costs on versus the reference design), there is no way to guarantee a heatsink will work with anything but the reference design. Card manufacturers place their VRMs all over the place and sometimes even stray from spec in memory placement so heatsink manufacturers can't standardize on anything but the one design guaranteed to have a market presence which is the OEM reference PCB. 1* Here's an article covering an old Nvidia reference cooler design: https://techbuyersguru.com/founders-edition-vs-open-air-video-card-coolers-definitive-analysis https://techbuyersguru.com/founders-edition-vs-open-air-vide... 2* Here's a GPU water heatsink: http://www.phanteks.com/PH-GB1080.html http://www.phanteks.com/PH-GB1080.html

by ivl 8y ago

saganus 8y ago

> The big thing I usually look for is if the PCB layout matches the reference layout. How do you do that? I mean, where do you look for the PCB layout of a potential card you will be buying? Do you physically open the box and take a picture? Also, why would the PCB layout make any difference? I mean, I've been buying 3D cards since the original Voodoo but I just never heard of this metric. Sounds interesting but would be nice to have some more details :)

ivl 8y ago
I was lucky to very early on have one of those Voodoo cards, so it's kind of amazing how far we've come. Unfortunately, it's a true pain to determine whether or not a card matches reference. The reason is simple: if the layout changes, there's a good chance the water cooling block won't correctly align with the high points on the card. Having that happen means two things: you will have points where the block won't be making contact, or you will have to bend the PCB, or you will have a component not being cooled. All really bad things. This can even be impacted by the thermal pads used by the OEM. I mostly check to see if they're using the reference cooler design. I haven't bothered with AMD/ATI cards in ages, so this only really applies to Nvidia products. Cards with the blower style coolers with a single fan towards the rear of the card, with an output near the video outputs are almost always reference. Cards with coolers the OEM designs often are, but sometimes are not. That's mostly where things are annoying. Thankfully, water cooling OEMs often have specs from the start as to which cards will match if you're planning on buying day on. An example: With reference layouts the 980 and 980 Ti ought to be compatible. On some EVGA cards, you have memory mounted on the rear of the card, that makes backplate installation impossible depending on which backplate you use, as it'll teeter: either a 980 backplate, or one meant to be compatible with both the 980 and the 980 Ti. Worse, that can mean having to find a different set of screws to mount the water block you wish to use.
cc439 8y ago
There's no definitive rule for determining whether a card is using rhe reference PCB design or not, but generally, any card using the reference cooler (1) and most card SKUs released on a new GPU family's launch day will use the reference PCB design. As for why you's want the reference PCB for water cooling, well that is easily explained by looking at water cooled GPU heatsinks (2). A Graphics Card is composed of much more than the GPH itself. They basically stuff all the components of a mini computer (power management, memory, processor) onto a single board and each one of those subcomponents needs cooling in order to achieve optimal performance. A GPU heatsink that neglects the VRMs or memory will not be able to be overclocked as far as a GPU where those parts are actively cooled and in a liquid cooled system, there is typically very little airflow within the case. Slap an all-in-one water cooling system intended for a CPU onto a GPU and you'll often get memory issues right off the bat even with stock clocks and overheated VRMs will lead to throttling after a relatively short time in heavy use. The fans on a GPU provide enough direcf airflow to keep those components within spec, but remove the fans and replace them with nothing and you lose that necessary airflow. Since each card manufacturer uses their own PCB layout (once they figure out what to improve or lower costs on versus the reference design), there is no way to guarantee a heatsink will work with anything but the reference design. Card manufacturers place their VRMs all over the place and sometimes even stray from spec in memory placement so heatsink manufacturers can't standardize on anything but the one design guaranteed to have a market presence which is the OEM reference PCB. 1* Here's an article covering an old Nvidia reference cooler design: https://techbuyersguru.com/founders-edition-vs-open-air-video-card-coolers-definitive-analysis https://techbuyersguru.com/founders-edition-vs-open-air-vide... 2* Here's a GPU water heatsink: http://www.phanteks.com/PH-GB1080.html http://www.phanteks.com/PH-GB1080.html