I have 11 matching Titan V water blocks. it took a lot of searching and luck to get them.
but for 30-series and 40-series cards, it's less an issue of finding blocks as it is finding blocks that are 1-slot. most blocks for these generations are 1.5 slot thickness or so, making it not possible to stack them all next to each other.
I have 11 matching Titan V water blocks. it took a lot of searching and luck to get them.
but for 30-series and 40-series cards, it's less an issue of finding blocks as it is finding blocks that are 1-slot. most blocks for these generations are 1.5 slot thickness or so, making it not possible to stack them all next to each other.
But if you water cool them can't you push them a bit harder than you are right now? Also you could get a cheap I5 or AMD cpu and motherboard and put the extra gpu's in there and water cool them as well. With a cheap non cpu crunching secondary setup your biggest electrical cost would still be the gpu's and the power supply, and you re already paying for the gpu's electricity right now.
Even water cooled, pushing graphics clocks normally only gets you one extra clock +13Mhz bin over defaults. The way that the Nvidia GPUBoost algorithm works, is that the card will naturally overclock itself up to the thermal and power limits set. And go no more.
So setting outrageous graphics overclocks is pointless, the card will just ignore them. I normally give my watercooled cards a 40-60Mhz clock boost and the card will probably hold that extra clock bin for 80% of the computing time. No card is ever running past about mid 50's even in the hottest room with 83° F. temps so nowhere near stock thermal limits. Only power limits control the GPUBoost.
You can overclock the memory quite a bit but really should only use the overclock to get the downclocked memory clocks, because of the Nvidia detected compute penalty, only back to the spec graphics clocks of P0 state. I use 400Mhz more than the stock P0 clocks and don't have any memory failures because of too high boost normally.
Depends on the project, but in general most BOINC gpu apps don't really respond emphatically to high memory clocks. On some projects, the improvement in memory bandwidth does matter, but they are few.
+1
)
+1
A Proud member of the O.F.A. (Old Farts Association).
I am curious. Considering rtx
)
I am curious. Considering rtx 3080 ti's.
Is there brand/model that could be liquid cooled to fit 7 onto an AsRock Epycd8 motherboard?
My inspiration is Ian&Steve C's doing it with 7 Titan V's.
Tom M
A Proud member of the O.F.A. (Old Farts Association).
Tom M wrote:I am curious.
)
For this purposes, I use Tech Power-Up site, which has a great GPU database.
Information you are seeking is here: https://www.techpowerup.com/gpu-specs/geforce-rtx-3080-ti.c3735
Go down & you will see all 3080Ti versions & recognize the models with "hybrid or water cooling".
non-profit org. Play4Life in Zagreb, Croatia, EU
Tom M wrote: ... My
)
You might/will have troubles getting WBs ....
Not only for Titan V or Titan Xp ...
Stay cooled !
S-F-V
San-Fernando-Valley
)
That would be the trick. Finding after market water blocks that feed into each other while making the rtx 3080 ti GPU 's pure single slot GPU's.
I would also have to find a 2500 watt PSU unless I left the cover of the case off :)
Unless I power limit the GPUs to fit into my 1600 watt PSU.
A Proud member of the O.F.A. (Old Farts Association).
I have 11 matching Titan V
)
I have 11 matching Titan V water blocks. it took a lot of searching and luck to get them.
but for 30-series and 40-series cards, it's less an issue of finding blocks as it is finding blocks that are 1-slot. most blocks for these generations are 1.5 slot thickness or so, making it not possible to stack them all next to each other.
_________________________________________________________________________
Ian&Steve C. wrote: I have
)
But if you water cool them can't you push them a bit harder than you are right now? Also you could get a cheap I5 or AMD cpu and motherboard and put the extra gpu's in there and water cool them as well. With a cheap non cpu crunching secondary setup your biggest electrical cost would still be the gpu's and the power supply, and you re already paying for the gpu's electricity right now.
Even water cooled, pushing
)
Even water cooled, pushing graphics clocks normally only gets you one extra clock +13Mhz bin over defaults. The way that the Nvidia GPUBoost algorithm works, is that the card will naturally overclock itself up to the thermal and power limits set. And go no more.
So setting outrageous graphics overclocks is pointless, the card will just ignore them. I normally give my watercooled cards a 40-60Mhz clock boost and the card will probably hold that extra clock bin for 80% of the computing time. No card is ever running past about mid 50's even in the hottest room with 83° F. temps so nowhere near stock thermal limits. Only power limits control the GPUBoost.
You can overclock the memory quite a bit but really should only use the overclock to get the downclocked memory clocks, because of the Nvidia detected compute penalty, only back to the spec graphics clocks of P0 state. I use 400Mhz more than the stock P0 clocks and don't have any memory failures because of too high boost normally.
Depends on the project, but in general most BOINC gpu apps don't really respond emphatically to high memory clocks. On some projects, the improvement in memory bandwidth does matter, but they are few.
yeah i don't watercool
)
yeah i don't watercool anything for overclocking ability. most of my cards are power limited anyway.
I water cool some of them for density and/or noise reasons.
_________________________________________________________________________
Ian&Steve C. wrote: yeah i
)
Woof you could move up a bit them from the bottom of the top 50 if you free them up then.