It looked to me like the memory clock was stock at 3505(7210 effective) which is where my 1070 fell in P2 state without NVI manipulation.

Ah right, no that's me forcing the clock with NVI. default 980 P2 state drops it to somewhere around 3GHz and change (just over 6 GHz effective). 10 series has memory controller improvements in raw latency and bandwidth with same memory chips, then improved memory compression as well.

Ohhh! Sorry I thought you had a Pascal. I've been running my 970's with a 7200Mhz effective memory clock with a NVI batch command file. Boosting memory speeds helps a lot with Einstein tasks.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

It looked to me like the memory clock was stock at 3505(7210 effective) which is where my 1070 fell in P2 state without NVI manipulation.

Ah right, no that's me forcing the clock with NVI. default 980 P2 state drops it to somewhere around 3GHz and change (just over 6 GHz effective). 10 series has memory controller improvements in raw latency and bandwidth with same memory chips, then improved memory compression as well.

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.

Hi,
I've got 4 GTX1080 reference design cards. I run the fan at 95% and the temps are 57-65 C. The nvidia-smi reports the cards drawing each 130W-140W most of the time and 170W under high load. I run P2 state, 2020Mhz for GPU and 10126MHz for memory.

How much can the PCIE + 8 pin connector supply together?

Petri

Supposedly, 75w through the PCIe motherboard connector and 100W 150W through the 8pin PCIe power supply cable. That is only 175W total and looks like 5 watts shy of maximum TDP. I see where the nvsmi document states the maximum power for the 1080 card is 225W. Not sure how that works, must be 225W instantaneous load and more likely 180 watts for sustained load.

[Edit] corrected power delivery

Yeah that makes sense. Peak 225 but continuous 175. Explains alot...

Hi,
I've got 4 GTX1080 reference design cards. I run the fan at 95% and the temps are 57-65 C. The nvidia-smi reports the cards drawing each 130W-140W most of the time and 170W under high load. I run P2 state, 2020Mhz for GPU and 10126MHz for memory.

How much can the PCIE + 8 pin connector supply together?

Petri

Supposedly, 75w through the PCIe motherboard connector and 100W 150W through the 8pin PCIe power supply cable. That is only 175W total and looks like 5 watts shy of maximum TDP. I see where the nvsmi document states the maximum power for the 1080 card is 225W. Not sure how that works, must be 225W instantaneous load and more likely 180 watts for sustained load.

[Edit] corrected power delivery

Thank you. So I'm getting near max but there is still something to improve.

---

To overcome Heisenbergs:
"You can't always get what you want / but if you try sometimes you just might find / you get what you need." -- Rolling Stones

Hi,
I've got 4 GTX1080 reference design cards. I run the fan at 95% and the temps are 57-65 C. The nvidia-smi reports the cards drawing each 130W-140W most of the time and 170W under high load. I run P2 state, 2020Mhz for GPU and 10126MHz for memory.

How much can the PCIE + 8 pin connector supply together?

Petri

Supposedly, 75w through the PCIe motherboard connector and 100W 150W through the 8pin PCIe power supply cable. That is only 175W total and looks like 5 watts shy of maximum TDP. I see where the nvsmi document states the maximum power for the 1080 card is 225W. Not sure how that works, must be 225W instantaneous load and more likely 180 watts for sustained load.

[Edit] corrected power delivery

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

Keith and Grant

Here is the 980Ti installation guide.

They have not posted one for the 1080 on their website yet
http://www.evga.com/support/manuals/files/980_Ti_HYBRID_Installation_Guide.pdf

Zalster, thanks for the link. It explains the process to me. I always wondered how much surgery you had to do with the hybrid cooler kit.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

Hi,
I've got 4 GTX1080 reference design cards. I run the fan at 95% and the temps are 57-65 C. The nvidia-smi reports the cards drawing each 130W-140W most of the time and 170W under high load. I run P2 state, 2020Mhz for GPU and 10126MHz for memory.

How much can the PCIE + 8 pin connector supply together?

Petri

---

To overcome Heisenbergs:
"You can't always get what you want / but if you try sometimes you just might find / you get what you need." -- Rolling Stones

Hey Keith,

I'm always aggressive with my fan curves, 80C is the 100% of fan power.

The way the EVGA hybrid kits work, they connect the pump mother and radiator fan to a pass thru connector that you connect to the blower fan connector on the GPU.

So all 3 are powered from the GPU blower fan connector.

WOW! That is really hot at 100% fan speed. I know that all the 1080 reviews document gaming temperatures, but I thought that even with GPGPU work at 100% utilization, the cards would be able to keep from thermal throttling at 100% fan speed. After all you are not hardly even using most of the card as the memory controller, video engine and bus interface loads are almost nil most of the time.

I was surprised to find that my new Noctua NF-F12 iPPC 2000 fans come with 16" fan length leads. I actually had to bundle most of the length up for tidiness. Those leads would be able to reach anywhere even on E-ATX or server motherboards.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

Keith and Grant

Here is the 980Ti installation guide.

They have not posted one for the 1080 on their website yet
http://www.evga.com/support/manuals/files/980_Ti_HYBRID_Installation_Guide.pdf

---

Hey Keith,

I'm always aggressive with my fan curves, 80C is the 100% of fan power.

The way the EVGA hybrid kits work, they connect the pump mother and radiator fan to a pass thru connector that you connect to the blower fan connector on the GPU.

So all 3 are powered from the GPU blower fan connector.

I first disconnected the radiator fan from that set up when I noticed the temps not coming down and replaced it with a aftermarket fan with a 4 pin that I could connect to the motherboard.

That worked for a while with much better temps but it was in the last 3 weeks when I noticed the speed of the GPU wasn't even getting to stock, much less boost speeds.

I had to manually overclock the GPU just to get t stock and even more to get it to boost speed

I'm convinced that the power out of the blower fan connection isn't sufficient to power the stock blower and the hybrid pump motor.

In previous GPUs they had 8+6 pins that provided sufficient power to cover this.

But in the 1080 reference design, I think the card is underpowered by that single 8 pin connector.

Grant asked if I could connect the hybrid pump somewhere on the Mobo? I would need to disassemble the card and disconnect the pass thru and see if I could be an adaptor to power it. There probably is. Thought my guess is it's going to require a molex as there are no fan pins on the motherboard near by so it would probably require a very long extension or better just a molex adaptor.

I may just revert this card back to it's reference design after the WOW and test. I bet the speed will go back to normal. If it does, then I know for sure.

If that is the case, it means no more adapting base model 10x0 GPU to hybrids as they are deliberately underpowering those models to prevent hybrids. :(

If I was to ever go for a hybrid again, I'd go for the EVGA 1080 hybrid built on a FTW design with 8+6 pin config.

Zalster

Edit...

Ok, I may have been overly po when I said they deliberately underpowered it...

Things always look better when you take a break. Maybe it's an unforeseen issue that they haven't seen yet.

Will have to wait and see

Edit 2...

I've on the EVGA website and looking at the 1080 hybrid kit. They say it will work with the reference design but there is no PDF installation guide on their site so I can look at how this thing is powered. It appears to look like previous generation hybrid kits. A PDF would be nice. I'll keep looking but I still think the card is underpowered.

---

I'd need to look at the power connection that it's plugged into and see if I could find some adaptor.

If not, then the only other option is to return the card to it's original condition.

After the WOW, I'll look into that, lol...


Edit..

At the end of the video by Jay, he also replaced the fan on the radiator and connected it to the motherboard so he could also control it's speed to lower the temperature more. So I may not be so far off, lol

The other option, (expensive) would be to use a motherboard that has additional power input for the video cards supplementing the standard 75W through the PCIe connector. At least that is my understanding of what the additional 8pin PCIe connector that plugs into the motherboard near PCIe slot 0 on my ASUS ROG motherboard. The manual says it helps supply sufficient power if the board is fully populated with video cards. Not sure exactly how the extra power gets to the card through the slots though. Does it put power on pins of the slot that don't normally get used? I've never seen the pinout diagram of the PCIe slot connector so just guessing.

Hey Keith,

I always have those plugged in but in this case it doesn't help the GPU as it's already maxed out as that single 8 pin isn't providing enough power to the 1080. That explains why other venders are adding the additonal pins to the cards. There is a problem providing sufficient power via that single pin to the power requirements of that card


Zalster

I'm not exactly sure how everything gets powered up and from where with the hybrid cooler on the card. Is there some way to just have the stock blower fan powered from the card and move the cooler pump and radiator fan over to a motherboard fan header? I think the smaller radiator AIO pumps and fans can be powered from the CPU fan header. Even my 240mm AIO gets its power from the CPU header. I do know that some of the AIO systems use a SATA power connector for power though. For controlling the externally powered AIO pump and fan, you could use Speedfan for example or get a separate discrete fan controller which also gives you many more fan ports. Extra cost of course. If the situation of the AIO cooler fan cable is possibly not long enough, you could get an extension cable to extend its length to whatever available fan header is on the motherboard. An extension cable comes standard with my Noctua fans.

You said originally that you weren't happy with the 80° C. GPU temps with the stock blower configuration and that is why you moved to the hybrid solution. Did you try to bump the fan speed curves of the card with some other software fan control solution or did you just accept the stock Nvidia fan curves. I think JayzTwocents had a video where he showed that the stock fan settings were not aggressive enough to prevent the card from thermal throttling and were made for noise control concerns and not performance.

I have been controlling the fan speed curves with SIV now for quite a while. Back with the 670's I was using the manual temperature set point method and now with the 970's and 1070's I use the custom 6 point fan control curves that are in the later releases. Has worked quite well for me so far.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

I'd need to look at the power connection that it's plugged into and see if I could find some adaptor.

If not, then the only other option is to return the card to it's original condition.

After the WOW, I'll look into that, lol...


Edit..

At the end of the video by Jay, he also replaced the fan on the radiator and connected it to the motherboard so he could also control it's speed to lower the temperature more. So I may not be so far off, lol

The other option, (expensive) would be to use a motherboard that has additional power input for the video cards supplementing the standard 75W through the PCIe connector. At least that is my understanding of what the additional 8pin PCIe connector that plugs into the motherboard near PCIe slot 0 on my ASUS ROG motherboard. The manual says it helps supply sufficient power if the board is fully populated with video cards. Not sure exactly how the extra power gets to the card through the slots though. Does it put power on pins of the slot that don't normally get used? I've never seen the pinout diagram of the PCIe slot connector so just guessing.

Hey Keith,

I always have those plugged in but in this case it doesn't help the GPU as it's already maxed out as that single 8 pin isn't providing enough power to the 1080. That explains why other venders are adding the additonal pins to the cards. There is a problem providing sufficient power via that single pin to the power requirements of that card


Zalster

---

I've never seen the pinout diagram of the PCIe slot connector so just guessing.

They exist on t'interweb.

http://pinouts.ru/Slots/pci_express_pinout.shtml

Looks like the number of power lanes is fixed, the same for everything between 1x and 16x.

I doubt the additional PCIe connector on the motherboard would increase the rating of any single PCIe slot, but would probably be intended to take the load off the general-purpose motherboard power connector and the interconnection layers of the motherboard if multiple PCIe cards were present, each drawing the permitted 75W.

Thanks for the pinout link, Richard. As you state there are only 5 pins total that supply card power between 1x and 16x. Three +12v pins and two +3.3v pins. Most benefit from an additional slot power connector would seem to be reduction in the current load of the motherboard interconnection layer traces as you state. Or maybe some signal noise reduction benefit.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

Jason, that nvml snippet, is that after you manipulated the power target and temp limits of the card with Nvidia Inspector? Or is that the stock settings without NVI adjustment?

Precision X running, but nothing attempted as agressive (so as to keep power down and higher stability while in development). NVinspector has the memory clock override on p2 active, and precision is driving a custom fan curve (I'm sensitive to whine from working in the audiophile & recording industries when younger, so cannot really leave things completely on auto)

It looked to me like the memory clock was stock at 3505(7210 effective) which is where my 1070 fell in P2 state without NVI manipulation.

One of the greatest benefits of the new 1070's is the complete removal of the whine of the cards compared to my 970's. I too am sensitive to the whine. I just have had to live it for the last couple of years. I already want to replace the other pair of 970's in the other cruncher to get rid of that whine. Finances need to recover first.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

I've never seen the pinout diagram of the PCIe slot connector so just guessing.

They exist on t'interweb.

http://pinouts.ru/Slots/pci_express_pinout.shtml

Looks like the number of power lanes is fixed, the same for everything between 1x and 16x.

I doubt the additional PCIe connector on the motherboard would increase the rating of any single PCIe slot, but would probably be intended to take the load off the general-purpose motherboard power connector and the interconnection layers of the motherboard if multiple PCIe cards were present, each drawing the permitted 75W.

Jason, that nvml snippet, is that after you manipulated the power target and temp limits of the card with Nvidia Inspector? Or is that the stock settings without NVI adjustment?

Precision X running, but nothing attempted as agressive (so as to keep power down and higher stability while in development). NVinspector has the memory clock override on p2 active, and precision is driving a custom fan curve (I'm sensitive to whine from working in the audiophile & recording industries when younger, so cannot really leave things completely on auto)

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.

Jason, that nvml snippet, is that after you manipulated the power target and temp limits of the card with Nvidia Inspector? Or is that the stock settings without NVI adjustment?

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

I'd need to look at the power connection that it's plugged into and see if I could find some adaptor.

If not, then the only other option is to return the card to it's original condition.

After the WOW, I'll look into that, lol...


Edit..

At the end of the video by Jay, he also replaced the fan on the radiator and connected it to the motherboard so he could also control it's speed to lower the temperature more. So I may not be so far off, lol

The other option, (expensive) would be to use a motherboard that has additional power input for the video cards supplementing the standard 75W through the PCIe connector. At least that is my understanding of what the additional 8pin PCIe connector that plugs into the motherboard near PCIe slot 0 on my ASUS ROG motherboard. The manual says it helps supply sufficient power if the board is fully populated with video cards. Not sure exactly how the extra power gets to the card through the slots though. Does it put power on pins of the slot that don't normally get used? I've never seen the pinout diagram of the PCIe slot connector so just guessing.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

Some digging with the python bindings for nvml (which nvidia-smi is based on), reveals there is a lot of interesting info hiding in there, that might be able to tell us more about the power limits, throttling reasons etc. Seems a bit easier than deciphering the command line utility options. Example portions from my 980 (my p2 state mem clock overrides are engaged via nvidia inspector):

 ...
<attached_gpus>1</attached_gpus>
  <gpu id="b'0000:01:00.0'">
    <product_name>b'GeForce GTX 980'</product_name>
    <product_brand>GeForce</product_brand>
    <display_mode>Enabled</display_mode>
    <display_active>Enabled</display_active>
...
    <fan_speed>43 %</fan_speed>
    <performance_state>P2</performance_state>
...
    <fb_memory_usage>
      <total>4096.0 MiB</total>
      <used>824.47265625 MiB</used>
      <free>3271.52734375 MiB</free>
    </fb_memory_usage>
...
    <utilization>
      <gpu_util>68 %</gpu_util>
      <memory_util>50 %</memory_util>
      <encoder_util>0 %</encoder_util>
      <decoder_util>0 %</decoder_util>
    </utilization>
...
    <temperature>
      <gpu_temp>75 C</gpu_temp>
      <gpu_temp_max_threshold>96 C</gpu_temp_max_threshold>
      <gpu_temp_slow_threshold>91 C</gpu_temp_slow_threshold>
    </temperature>
    <power_readings>
      <power_state>P2</power_state>
      <power_management>Supported</power_management>
      <power_draw>61.74 W</power_draw>
      <power_limit>225.00 W</power_limit>
      <default_power_limit>180.00 W</default_power_limit>
      <enforced_power_limit>225.00 W</enforced_power_limit>
      <min_power_limit>120.00 W</min_power_limit>
      <max_power_limit>225.00 W</max_power_limit>
    </power_readings>
    <clocks>
      <graphics_clock>1379 MHz</graphics_clock>
      <sm_clock>1379 MHz</sm_clock>
      <mem_clock>3505 MHz</mem_clock>
    </clocks>
    <applications_clocks>
      <graphics_clock>1240 MHz</graphics_clock>
      <mem_clock>3505 MHz</mem_clock>
    </applications_clocks>
    <default_applications_clocks>
      <graphics_clock>1240 MHz</graphics_clock>
      <mem_clock>3505 MHz</mem_clock>
    </default_applications_clocks>
...
    <process_info>
      <pid>6604</pid>
      <process_name>b'C:\\BOINC_Data\\projects\\setiathome.berkeley.edu\\Lunatics_x41zj_win32_cuda50.exe'</process_name>
      <used_memory>N\A</used_memory>
    </process_info>
    <process_info>
      <pid>3236</pid>
      <process_name>b'C:\\BOINC_Data\\projects\\setiathome.berkeley.edu\\Lunatics_x41zj_win32_cuda50.exe'</process_name>
      <used_memory>N\A</used_memory>
    </process_info>

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.

I'd need to look at the power connection that it's plugged into and see if I could find some adaptor.

If not, then the only other option is to return the card to it's original condition.

After the WOW, I'll look into that, lol...


Edit..

At the end of the video by Jay, he also replaced the fan on the radiator and connected it to the motherboard so he could also control it's speed to lower the temperature more. So I may not be so far off, lol

---