what i'd really like to see are some full cover AIO video cards. Why do we have to settle for hybrid or go full custom to get the entire card water cooled? I know the cost would be higher, but i think there's a market for something like this, at least on the higher end video cards.
There is a market, but unlike traditional CPU watercoolers, there is additional necessity to cool VRM components, which would ordinarily be cooled by passing air from a fan. So while the placement of the GPU compared to the mounting holes for the GPU remains consistent, the VRM and power delivery chips vary from board to board, even amongst the same family of GPUs (GTX 970, for example).
The best we see is hybrid GPU mounting brackets that accept common CPU AIO kits, such as the NZXT G10 Kraken. It has a mounting bracket to hold modern style NZXT/Corsair AIO watercooling kits, and has a fan to actively cool VRM and memory chips, which would normally have been cooled by the air cooling kit.
In essence, what I'm saying is that until there's some sort of standardization for how/where VRM and memory is located onboard for GPUs, I don't expect an AIO watercooling kit to be available for GPUs (so that it's universal), without having to have a hybrid solution with fan.
HBM memory being built onto the GPU die will go a long way towards moving memory chips surrounding the GPU in a central and common location, which would benefit from an AIO kit, but that still leaves VRM chips being located in random places from board to board.
That's exactly the reason why an AIB partner could make a full cover AIO cooler, PCB and cooling system are designed (sort of) in the same house. So there are no variations in VRM and memory positions to account for, so long as the cooler remains compatible with Inno3D's own PCB. I don't really see why Inno3D has to settle for a hybrid cooling solution like this one.
This makes especially no sense since for GTX 1060's, thermals are rarely the limiting factor, it just depends on the silicon lottery. Any dualslot open air cooler has enough thermal headroom to reach a GPU's maximum clockspeed. This hybrid cooler will not allow any higher clockspeeds than any other cooler, it just takes up more space and keeps the card a little cooler. At least a single slot full cover liquid cooler would enable some different form factors (you require some space for the radiator, but only one PCIe slot).
I didn't see it scanning the article but does this have at least an 8-pin power connector? I thought with all Pascal cards the main limiting factor for OC was power delivery, not heat.
The main limiting factor of pascal is it's inability to go past 2050mhz. All the power delivery in the world wont fix that, as the dual 8 pin, 14 phase VRM 1080s have shown. The stock 1060, as ananadtech showed, can already hit it's max clock ceiling.
Absolutely, the TDP is just not there to warranty a AIO card - sure this is a premium product if the pump and the fan are trully silent. But then again you can already get air cooled 1060 with a large air cooler that is dead silent.
As an owner for a r9 295x2 with a sigle 120m rad cooling solution that is perfectly adequate - these just seem really funny to me.
maybe.. but theinsanegamer.. insinuated that all pascal doesnt have the inability to go passed 2050 on the core ... seems, when i searched to see how fast the 1060 could go. so i could get an idea how far i can push it before i would need to be cautious, mine isnt the only one that can go passed 2050 ....
Anybody who has read the reviews of pascal knows I'm insinuating that pascal has difficulty overclocking much past 2050. going over 2100 is entirely up to silicon lottery.
I mean, if we want to be pedantic, YOU insinuated pascal was TDP limited, when that is definitely not the case. You were not even in the same ballpark as the actual issue of Pascal's architecture not clocking past 2050-2100 easily.
Well, no, this is objectively really a bit silly :) Maybe some point if they'd made it a short card, but honestly even the short 1060's do OK noise wise.
Maybe for machine learning, where cards could be maxed out for days at a time. I bought a 1060 for that exact purpose. Would I pay the additional $100 or whatever this is going to cost? No. I'd get a 1070 and underclock it.
Not only is this card quite ridiculous in terms of cooling a 120W GPU with water, but (I don't know if this article states it, I spotted it somewhere else) the truly atrocious thing is that it requires a 4-pin Molex connector for the pump. Yes, you read that correctly. Molex. On a brand new GPU in 2016. What the...?
It could have a floppy disk power connector instead... I still can't believe opening a new modern power supply box and finding a modular floppy disk power connector.
I've got a factory shrink-wrapped 5 1/4 " floppy drive I'm dying to put into my Skylake box, if only I could find a motherboard with a floppy drive connector on it, you know, to see oodles of treasured data I haven't had access to in years. :-)
I actually use the 3.5 inch floppy power connector for a Cyber Acoustics CA-1000 speaker box in my gaming desktop. It's basically a pair of cute little speakers crammed inside a 5.25 inch bay-compatiable box that comes with wiring to run audio from your stereo out on the back of the motherboard, back inside the case via an empty expansion slot using a plate that has the wire punched through it and into the speaker box. It needs floppy disk power to drive the speakers so, on the off chance I ever plug a monitor back into my gaming PC, I'll be able to hear things without cluttering up the world with big, ugly speakers that waste space on my desk. So yeah, floppy power connectors are still relevant in certain situations.
Yes, absolutely it's not great quality. I've owned three CA-1000 in-bay speaker sets over the years and all of them have been consistently poor. Stereo seperation is basically nonexistent for the obvious reason that both speakers are crammed into the same casing and neither is in an independent chamber. Having opened one of them in the past, I found a couple of cheap paper cones in plastic molds that have them firing upward at the top of the box's interior rather than forward out the front grills. I think this was done so that larger diameter speakers could be placed inside to get a bit more bass. Power was rated at something like 1.5 - 2 watts. Anyway, they're probably best dropped into the "barely adequate" category, though they do get loud enough to fill a small room and don't distort until the very top end of their volume dial.
Since my gaming box is a headless streaming machine only, audio is wrapped up in with video and pushed over the network and I hear it through the headphones plugged into my laptop on the other end of the wire. I only have them becasue my Lian Li's wiring for the front audio ports doesn't want to play nicely on my current motherboard. In the event I eventually stop streaming, I'm going to want working front audio and the CA-1000 has a headphone connection. It's also nice enough to move that connection from the very bottom of the case to closer to the top so my wire run doesn't need to be as long and I get more headphone slack.
In the long ago past when some of the guys I worked with invited me to LAN parties, I used older model CA-1000s in MicroATX cases to get sound without extra fuss. The lack of stereo seperation and poor quality didn't hurt me too much when I was rocking Battlefield 1942 or UT99. Anyway, I don't think Cyber Acoustics is even producing the CA-1000 anymore. If I ever do need desktop speakers again, I'll probably pick up something that's powered off a USB port, but still has a stereo connector for analog audio. Dell shipped a nice set of them with some of their Optiplex systems that also fit in the "yup, it makes noise" category and they don't waste a lot of desk or room space with stupidly huge boxes packed full of subwoofers.
I bet this is going to cost as much as a 1070. Which would make it a terrible deal and a pointless product. Although seeing as even 1080s can be more than adequately cooled on air it's guaranteed to be a pointless product.
If you want to make a low TDP card like GTX 1060 watercooled then at least make it attractive . That is a single slot card. or even better , Low Profile single slot card .
There is no point in watercooling for a low TDP card outside this...
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
33 Comments
Back to Article
FalcomPSX - Monday, December 5, 2016 - link
what i'd really like to see are some full cover AIO video cards. Why do we have to settle for hybrid or go full custom to get the entire card water cooled? I know the cost would be higher, but i think there's a market for something like this, at least on the higher end video cards.JoeyJoJo123 - Tuesday, December 6, 2016 - link
There is a market, but unlike traditional CPU watercoolers, there is additional necessity to cool VRM components, which would ordinarily be cooled by passing air from a fan. So while the placement of the GPU compared to the mounting holes for the GPU remains consistent, the VRM and power delivery chips vary from board to board, even amongst the same family of GPUs (GTX 970, for example).The best we see is hybrid GPU mounting brackets that accept common CPU AIO kits, such as the NZXT G10 Kraken. It has a mounting bracket to hold modern style NZXT/Corsair AIO watercooling kits, and has a fan to actively cool VRM and memory chips, which would normally have been cooled by the air cooling kit.
JoeyJoJo123 - Tuesday, December 6, 2016 - link
In essence, what I'm saying is that until there's some sort of standardization for how/where VRM and memory is located onboard for GPUs, I don't expect an AIO watercooling kit to be available for GPUs (so that it's universal), without having to have a hybrid solution with fan.HBM memory being built onto the GPU die will go a long way towards moving memory chips surrounding the GPU in a central and common location, which would benefit from an AIO kit, but that still leaves VRM chips being located in random places from board to board.
Macpoedel - Wednesday, December 7, 2016 - link
That's exactly the reason why an AIB partner could make a full cover AIO cooler, PCB and cooling system are designed (sort of) in the same house. So there are no variations in VRM and memory positions to account for, so long as the cooler remains compatible with Inno3D's own PCB. I don't really see why Inno3D has to settle for a hybrid cooling solution like this one.This makes especially no sense since for GTX 1060's, thermals are rarely the limiting factor, it just depends on the silicon lottery. Any dualslot open air cooler has enough thermal headroom to reach a GPU's maximum clockspeed. This hybrid cooler will not allow any higher clockspeeds than any other cooler, it just takes up more space and keeps the card a little cooler. At least a single slot full cover liquid cooler would enable some different form factors (you require some space for the radiator, but only one PCIe slot).
Ej24 - Monday, December 5, 2016 - link
I didn't see it scanning the article but does this have at least an 8-pin power connector? I thought with all Pascal cards the main limiting factor for OC was power delivery, not heat.TheinsanegamerN - Monday, December 5, 2016 - link
The main limiting factor of pascal is it's inability to go past 2050mhz. All the power delivery in the world wont fix that, as the dual 8 pin, 14 phase VRM 1080s have shown. The stock 1060, as ananadtech showed, can already hit it's max clock ceiling.Cellar Door - Monday, December 5, 2016 - link
Absolutely, the TDP is just not there to warranty a AIO card - sure this is a premium product if the pump and the fan are trully silent. But then again you can already get air cooled 1060 with a large air cooler that is dead silent.As an owner for a r9 295x2 with a sigle 120m rad cooling solution that is perfectly adequate - these just seem really funny to me.
Qasar - Monday, December 5, 2016 - link
tell that to my asus strix GTX 1060.. currently running at 2114 core and 1958 ramMeaker10 - Tuesday, December 6, 2016 - link
That's silicon lottery rather than the PCB.Qasar - Tuesday, December 6, 2016 - link
maybe.. but theinsanegamer.. insinuated that all pascal doesnt have the inability to go passed 2050 on the core ... seems, when i searched to see how fast the 1060 could go. so i could get an idea how far i can push it before i would need to be cautious, mine isnt the only one that can go passed 2050 ....TheinsanegamerN - Tuesday, December 6, 2016 - link
Anybody who has read the reviews of pascal knows I'm insinuating that pascal has difficulty overclocking much past 2050. going over 2100 is entirely up to silicon lottery.I mean, if we want to be pedantic, YOU insinuated pascal was TDP limited, when that is definitely not the case. You were not even in the same ballpark as the actual issue of Pascal's architecture not clocking past 2050-2100 easily.
TheinsanegamerN - Tuesday, December 6, 2016 - link
ej24, not you. When will anandtech get an edit feature?TheinsanegamerN - Monday, December 5, 2016 - link
Is there really a need for a watercooler on a GPU that only pulls 120 watt?Qwertilot - Monday, December 5, 2016 - link
Well, no, this is objectively really a bit silly :) Maybe some point if they'd made it a short card, but honestly even the short 1060's do OK noise wise.lazarpandar - Monday, December 5, 2016 - link
Maybe for machine learning, where cards could be maxed out for days at a time. I bought a 1060 for that exact purpose. Would I pay the additional $100 or whatever this is going to cost? No. I'd get a 1070 and underclock it.I dunno this does just seem silly overall.
jordanclock - Monday, December 5, 2016 - link
Not that it will make a huge difference but this model is rated at 150W.shabby - Monday, December 5, 2016 - link
It seems they want you to think this is a high end card, big heatsink = big powahdoggface - Tuesday, December 6, 2016 - link
Is there a need for an AIO water cooler on a 65-95w TDP CPU? Probably not, but it is way quieter...Valantar - Monday, December 5, 2016 - link
Not only is this card quite ridiculous in terms of cooling a 120W GPU with water, but (I don't know if this article states it, I spotted it somewhere else) the truly atrocious thing is that it requires a 4-pin Molex connector for the pump. Yes, you read that correctly. Molex. On a brand new GPU in 2016. What the...?HomeworldFound - Monday, December 5, 2016 - link
It could have a floppy disk power connector instead... I still can't believe opening a new modern power supply box and finding a modular floppy disk power connector.TheinsanegamerN - Tuesday, December 6, 2016 - link
You know there is that ONE person with a skylake box and a 3.5" floppy drive.Ktracho - Tuesday, December 6, 2016 - link
I've got a factory shrink-wrapped 5 1/4 " floppy drive I'm dying to put into my Skylake box, if only I could find a motherboard with a floppy drive connector on it, you know, to see oodles of treasured data I haven't had access to in years. :-)BrokenCrayons - Tuesday, December 6, 2016 - link
I actually use the 3.5 inch floppy power connector for a Cyber Acoustics CA-1000 speaker box in my gaming desktop. It's basically a pair of cute little speakers crammed inside a 5.25 inch bay-compatiable box that comes with wiring to run audio from your stereo out on the back of the motherboard, back inside the case via an empty expansion slot using a plate that has the wire punched through it and into the speaker box. It needs floppy disk power to drive the speakers so, on the off chance I ever plug a monitor back into my gaming PC, I'll be able to hear things without cluttering up the world with big, ugly speakers that waste space on my desk. So yeah, floppy power connectors are still relevant in certain situations.HollyDOL - Wednesday, December 7, 2016 - link
That has to provide terrible sound quality...BrokenCrayons - Wednesday, December 7, 2016 - link
Yes, absolutely it's not great quality. I've owned three CA-1000 in-bay speaker sets over the years and all of them have been consistently poor. Stereo seperation is basically nonexistent for the obvious reason that both speakers are crammed into the same casing and neither is in an independent chamber. Having opened one of them in the past, I found a couple of cheap paper cones in plastic molds that have them firing upward at the top of the box's interior rather than forward out the front grills. I think this was done so that larger diameter speakers could be placed inside to get a bit more bass. Power was rated at something like 1.5 - 2 watts. Anyway, they're probably best dropped into the "barely adequate" category, though they do get loud enough to fill a small room and don't distort until the very top end of their volume dial.Since my gaming box is a headless streaming machine only, audio is wrapped up in with video and pushed over the network and I hear it through the headphones plugged into my laptop on the other end of the wire. I only have them becasue my Lian Li's wiring for the front audio ports doesn't want to play nicely on my current motherboard. In the event I eventually stop streaming, I'm going to want working front audio and the CA-1000 has a headphone connection. It's also nice enough to move that connection from the very bottom of the case to closer to the top so my wire run doesn't need to be as long and I get more headphone slack.
In the long ago past when some of the guys I worked with invited me to LAN parties, I used older model CA-1000s in MicroATX cases to get sound without extra fuss. The lack of stereo seperation and poor quality didn't hurt me too much when I was rocking Battlefield 1942 or UT99. Anyway, I don't think Cyber Acoustics is even producing the CA-1000 anymore. If I ever do need desktop speakers again, I'll probably pick up something that's powered off a USB port, but still has a stereo connector for analog audio. Dell shipped a nice set of them with some of their Optiplex systems that also fit in the "yup, it makes noise" category and they don't waste a lot of desk or room space with stupidly huge boxes packed full of subwoofers.
Assimilator87 - Monday, December 5, 2016 - link
**Article Correction**In the chart, I believe the reference 1060's memory clock should be 8.0 Gbps.
Assimilator87 - Tuesday, December 6, 2016 - link
and the Boost Clock is 1708 Mhz, not 1683 Mhz.Flunk - Monday, December 5, 2016 - link
I bet this is going to cost as much as a 1070. Which would make it a terrible deal and a pointless product. Although seeing as even 1080s can be more than adequately cooled on air it's guaranteed to be a pointless product.Death666Angel - Tuesday, December 6, 2016 - link
It's 365€ for this 1060/6GB vs 395€ for the cheapest 1070. The poor fools who choose this abomination.HollyDOL - Tuesday, December 6, 2016 - link
Kind of reminds me pple drilling holes in their car exhausts so their 900ccm 34kW engines sounded 'cool' :-)BrokenCrayons - Tuesday, December 6, 2016 - link
Wouldn't buying a 1070 make more sense than adding a complex and costly cooling solution to a 1060 in order to overclock it?sna1970 - Tuesday, December 6, 2016 - link
If you want to make a low TDP card like GTX 1060 watercooled then at least make it attractive . That is a single slot card. or even better , Low Profile single slot card .There is no point in watercooling for a low TDP card outside this...
zerothxv - Sunday, November 5, 2017 - link
Where can I buy this? Does it even exist? Searched all major pc part retailers and it doesn't show up at all.