... unless you get a screaming deal and also don't mind similar levels of "screaming" noise.
Ha! I like what you did there.
Tom, also keep in mind that although more "powerful" than the PCIe versions, they also consume more power than their PCIe equivalents (which also means more heat). There might be some exceptions out there, but none that I am aware of.
& on eBAY, you can gat V100 from $500+ with PCIe: eBAY . So 4x V100 PCIe would cost you the same as board & 4x V100 SMX2. ???? & you can sell more easily PCIe cards, once you are done with them. ????
BTW, those V100 cards come without fans & when you put fans on them - they are quite noisy in a case (server or otherwise). ????
Yes it uses 12VHPWR connector. But since the 3090 Ti does not have anywhere near the power consumption of the 4090 that has been burning up, there have been very few, if any reports of melted connectors on the 3090 Ti.
And the 12VHPWR connector is cross-compatible with the new 12V-2x6 connector that is on the newest supplies.
I wouldn't worry about it as long as you obey the rules about bending the cable too close to the connector and fully insert it into the card socket.
Someplace on the Website Ian&Steve C posted some comments about setting up a Titan V for boinc.
Ian,
Could you post what your using for Graphics/MTM OC's? And if you are using Power restrictions to what wattage?
Thank you.
I have already demonstrated that I appear to have misremembered one MTM OC and got a lovely "artifacted screen" in various locations as well as apparently making the Keyboard/Mouse unresponsive. :)
Tom M
==edit==
And Googled up some OC posts where they were doing +200 Graphics and +200 MTM.
A Proud member of the O.F.A. (Old Farts Association).
in most cases you dont need to overclock the titan v. i dont set any overclocks on them. i set persistency mode and fan speeds to 80% and that's it.
under compute loads, the core will only reach a maximum clock of 1335MHz. overclocking will not change this. the posts you found were likely from people running gaming benchmarks. when the card detects a 3D application, it allows higher boost clocks, but you will be limited to 1335MHz. there is a command to lift this restriction (sudo nvidia-smi --cuda-clocks=OVERRIDE) which allows boost clocks to rise to your set power limit. I havent found that this is very useful for me. you get maybe 5% more resultant performance at twice the power draw. the stable/efficient 1335MHz default clocks are doing well for me and i would recommend staying with that for cool/efficient/quieter operation.
i have not found mem overclocking to be stable or effective. even quite small mem OCs can crash the system. don't bother
for Einstein BRP7 you probably wont see more than about 135W of power. for Einstein O3AS you will be around 100W or less. so power limiting probably wont do anything, you'll be under the limit already.
I do set power limits around 160W, just in case some new project shows up that hits the card harder and i don't want the card to run hot or exceed system limits.
Ian&Steve C. wrote: ...
)
Ha! I like what you did there.
Tom, also keep in mind that although more "powerful" than the PCIe versions, they also consume more power than their PCIe equivalents (which also means more heat). There might be some exceptions out there, but none that I am aware of.
Tom M wrote: I stumbled
)
You need to make more research, as you are not gaining any advantage with SMX2 interface: https://www.gpuzoo.com/Compare/NVIDIA_Tesla_V100_PCI-E__vs__NVIDIA_Tesla_V100_SMX2/
& on eBAY, you can gat V100 from $500+ with PCIe: eBAY . So 4x V100 PCIe would cost you the same as board & 4x V100 SMX2. ???? & you can sell more easily PCIe cards, once you are done with them. ????
BTW, those V100 cards come without fans & when you put fans on them - they are quite noisy in a case (server or otherwise). ????
non-profit org. Play4Life in Zagreb, Croatia, EU
Thank you for the additional
)
Thank you for the additional details on how badly I would screwup myself :)
The $199 per v100 gpu sxm2 card was a real attention grabber. But that is the ONLY low priced part in the whole system.
PCIe here I come.
Tom M
A Proud member of the O.F.A. (Old Farts Association).
I was poking around at rtx
)
I was poking around at rtx 3090 ti's and discovered something else I had missed.
It looks like it uses the exact same connector as the rtx 4090 (has it melted yet?) connector.
Tom M
A Proud member of the O.F.A. (Old Farts Association).
Yes it uses 12VHPWR
)
Yes it uses 12VHPWR connector. But since the 3090 Ti does not have anywhere near the power consumption of the 4090 that has been burning up, there have been very few, if any reports of melted connectors on the 3090 Ti.
And the 12VHPWR connector is cross-compatible with the new 12V-2x6 connector that is on the newest supplies.
I wouldn't worry about it as long as you obey the rules about bending the cable too close to the connector and fully insert it into the card socket.
https://einsteinathome.org/ho
)
https://einsteinathome.org/host/12830750
Ian&SteveC,
This Titan V 4 GPU system can run under a 1600 watt PSU?
Thank you.
Tom M
A Proud member of the O.F.A. (Old Farts Association).
yes. it only uses like 500W
)
yes. it only uses like 500W total running Einstein (not using the CPU to run any CPU projects)
_________________________________________________________________________
Using "buy it now"
)
Using "buy it now" comparisons on eBay it looks like rtx 3090's are running maybe $100 or less higher than rtx 3080 ti's.
Unfortunately, the rtx 3090 ti's seem to still be a bunch higher.
A Proud member of the O.F.A. (Old Farts Association).
Someplace on the Website
)
Someplace on the Website Ian&Steve C posted some comments about setting up a Titan V for boinc.
Ian,
Could you post what your using for Graphics/MTM OC's? And if you are using Power restrictions to what wattage?
Thank you.
I have already demonstrated that I appear to have misremembered one MTM OC and got a lovely "artifacted screen" in various locations as well as apparently making the Keyboard/Mouse unresponsive. :)
Tom M
==edit==
And Googled up some OC posts where they were doing +200 Graphics and +200 MTM.
A Proud member of the O.F.A. (Old Farts Association).
in most cases you dont need
)
in most cases you dont need to overclock the titan v. i dont set any overclocks on them. i set persistency mode and fan speeds to 80% and that's it.
under compute loads, the core will only reach a maximum clock of 1335MHz. overclocking will not change this. the posts you found were likely from people running gaming benchmarks. when the card detects a 3D application, it allows higher boost clocks, but you will be limited to 1335MHz. there is a command to lift this restriction (sudo nvidia-smi --cuda-clocks=OVERRIDE) which allows boost clocks to rise to your set power limit. I havent found that this is very useful for me. you get maybe 5% more resultant performance at twice the power draw. the stable/efficient 1335MHz default clocks are doing well for me and i would recommend staying with that for cool/efficient/quieter operation.
i have not found mem overclocking to be stable or effective. even quite small mem OCs can crash the system. don't bother
for Einstein BRP7 you probably wont see more than about 135W of power. for Einstein O3AS you will be around 100W or less. so power limiting probably wont do anything, you'll be under the limit already.
I do set power limits around 160W, just in case some new project shows up that hits the card harder and i don't want the card to run hot or exceed system limits.
_________________________________________________________________________