tachyon_john
Newbie
Offline
Activity: 28
Merit: 0
|
|
March 11, 2014, 04:52:20 AM |
|
the basic idea is to enable an emergency shutdown and/or a throttling based on temperature so you don't risk your GPU's life because a fan has failed (nVidia drivers are throttling hard at 95 deg C but I've seen my MARS cards get to 100 deg C during testing). Also these values are an important data points to send through a remote monitoring API - without these the monitoring would be next to useless.
I would love to see a possibility to throttle the card(s) to reach a temperature/utilization target. nvidia-smi shows the fan-speed and temp but not the "GPU Utilization" as nvidia-settings does... nvidia-settings never showed the gpu utilization of my 780 under ubuntu 12.04 I dug into this a bit tonight. The 'nvidia-settings' app uses a different mechanism to talk to the NVIDIA driver than nvidia-smi does, and so the availability and range of information it provides is different in several cases. NVIDIA has open-sourced the code for nvidia-settings, and as a result the API for communicating with the driver via an X server extension is also available as open source. In comparing what I can query from that API vs. what I can query through the more CUDA-oriented NVML interfaces, the lineage of the two interfaces becomes obvious. There are a few nice bits of information that currently don't appear to be available via NVML such as the utilization percentage for the hardware video encoder/decoder on the Kepler chips, additional temperature sensors, fan RPMs, and other PCIe and utilization information that NVML currently reports only for Tesla GPUs (or at least so it would seem). The one thing I don't like about using the nvidia-settings methods for querying data from the X server is that there are plenty of cases where you'd rather not run an X server if you don't need it, and the NVML library and nvidia-smi work great in such a case while nvidia-settings and it's own NV-CONTROL X extension are not usable in that scenario. It is possible that NVIDIA might be convinced to add some of the key missing data from nvidia-settings into nvidia-smi, but we'll have to see. In the short-term I may see if I can make my code additionally talk to the X server when one is running. We'll see how it goes.
|
|
|
|
liomojo1
|
|
March 11, 2014, 05:09:30 AM |
|
His hashrate sucks for that high of an overclock. Getting stable 330kh with a +125 +600 overclock. I assume you are using -H 1 for this result?
|
|
|
|
jsamul
|
|
March 11, 2014, 06:40:48 AM |
|
I did some testing tonight on my 2 750ti , and noticed that only the x64 version needs the chrome trick , x86 have better results about 4-5 kkhs than the x64 -with or without chrome. So x64 needs chrome and x86 no . Sadly my max overclock stable for minig is 1355 core and 3196 mem. Max hashrate with this settings 320. May be when the rizers arrive i will try more cause now heat is an issue too cause both card are stack together and make 77-80 C. oh my... i use the EVGA FTW cards, and even with factory OC those cards do not go over 55°C while mining
|
|
|
|
jsamul
|
|
March 11, 2014, 06:43:16 AM |
|
i do not see any noticable improvement for keccak with these drivers and 750Ti
|
|
|
|
Finnminer
Member
Offline
Activity: 74
Merit: 10
|
|
March 11, 2014, 07:12:20 AM |
|
How do the non-Ti 750s perform compared to the Ti's? The reason I'm asking is that Amazon managed to send me regular 750s instead of the Ti's I had ordered and I'm considering my options. I'm not too keen on the idea of sending the cards back and waiting forever to get correct cards instead..
|
|
|
|
liomojo1
|
|
March 11, 2014, 07:12:36 AM |
|
I use EVGA GeForce GTX 750 Ti which is small and have only one fan.
|
|
|
|
cbuchner1 (OP)
|
|
March 11, 2014, 07:40:44 AM |
|
How do the non-Ti 750s perform compared to the Ti's? The reason I'm asking is that Amazon managed to send me regular 750s instead of the Ti's I had ordered and I'm considering my options. I'm not too keen on the idea of sending the cards back and waiting forever to get correct cards instead..
250 kHash/s with OC is what I've heard... Christian
|
|
|
|
cbuchner1 (OP)
|
|
March 11, 2014, 07:42:17 AM Last edit: March 14, 2014, 04:27:54 PM by cbuchner1 |
|
I dug into this a bit tonight. The 'nvidia-settings' app uses a different mechanism to talk to the NVIDIA driver than nvidia-smi does, and so the availability and range of information it provides is different in several cases. NVIDIA has open-sourced the code for nvidia-settings, and as a result the API for communicating with the driver via an X server extension is also available as open source. In comparing what I can query from that API vs. what I can query through the more CUDA-oriented NVML interfaces, the lineage of the two interfaces becomes obvious. There are a few nice bits of information that currently don't appear to be available via NVML such as the utilization percentage for the hardware video encoder/decoder on the Kepler chips, additional temperature sensors, fan RPMs, and other PCIe and utilization information that NVML currently reports only for Tesla GPUs (or at least so it would seem). The one thing I don't like about using the nvidia-settings methods for querying data from the X server is that there are plenty of cases where you'd rather not run an X server if you don't need it, and the NVML library and nvidia-smi work great in such a case while nvidia-settings and it's own NV-CONTROL X extension are not usable in that scenario. It is possible that NVIDIA might be convinced to add some of the key missing data from nvidia-settings into nvidia-smi, but we'll have to see. In the short-term I may see if I can make my code additionally talk to the X server when one is running. We'll see how it goes.
I will raise this issue with nVidia engineers. Thank you for bringing it to my attention. I am also not running an X server on my mining rig equipped with three GTX 780 Ti. The mainboard has onboard graphics (Intel) and I prefer to use it as it induces no lag while mining. So the NV-CONTROL X extension can't work in this setting. Christian
|
|
|
|
cbuchner1 (OP)
|
|
March 11, 2014, 08:01:59 AM Last edit: March 11, 2014, 08:15:58 AM by cbuchner1 |
|
Stratum for blake256 was fixed. Happy hashing with the github version! Strangely my hashing speed won't currently exceed 570 kHash/s on my GTX 780, and changing the setting from 80% TDP to 106% TDP doesn't help either. HUH?
The Google Chrome trick seems essential here: Boosts hashes on my desktop PC from 770 kHash to 1150 kHash/s (this is for four GPUs combined)
I will now merge in some GPU temp / fan monitoring code, but I will keep this extra output optional.
Note that you can go to T1000x32 with the blake256 method (other hashing algos limit you to x24 with the T kernels)
Christian
|
|
|
|
DemosMirak
|
|
March 11, 2014, 08:11:09 AM Last edit: March 11, 2014, 08:26:50 AM by DemosMirak |
|
Hello, I just installed the new 335.23 driver on my windforce 3X GTX 770, and now Cudaminer is giving the following message: '[2014-03-11 08:58:15] Unable to query CUDA driver version! Is an nVidia driver installed?' I use the newest version of Cudaminer given, in combination with CUDA manager (version 1.2.1). If I try to run the miner outside of manager, it starts and then rapidly disappears. My set-up is as following:
Corsair Vengeance 16 GB : 2 x 8 GB Toshiba DT01ACA100 1 TB Kingston SSDNow V300 120 GB Thermalright Macho Rev.A Intel Core i7 4770K / 3.5 GHz Asus B85-PLUS Gigabyte GTX 770 OC Windforce 3X Windows 7 (completely up to date)
Edit: When I use the rollback function to get back to my previous driver, the error keeps persisting.
Edit 2: Performing a clean install as we speak.
Edit 3: The clean install did not resolve the issue, I did it twice to be sure, but the problems still persists.
|
BTC: 13enECLM3M3gjQDoBKouXuYFG4zXaDdDPx LTC: LRTbQNQcRjZV51PivQdhK7zpMtJYPouqR9
|
|
|
cbuchner1 (OP)
|
|
March 11, 2014, 08:12:29 AM |
|
Hello, I just installed the new 335.23 driver on my windforce 3X GTX 770, and now Cudaminer is giving the following message: '[2014-03-11 08:58:15] Unable to query CUDA driver version! Is an nVidia driver installed?' I use the newest version of Cudaminer given, in combination with CUDA manager (version 1.2.1). If I try to run the miner outside of manager, it starts and then rapidly disappears. My set-up is as following:
I would try reinstalling the driver once more with the "clean" install method (that's an optional checkbox to check during driver installation) Christian
|
|
|
|
ManiacMiner
|
|
March 11, 2014, 08:32:34 AM |
|
Hello, I just installed the new 335.23 driver on my windforce 3X GTX 770, and now Cudaminer is giving the following message: '[2014-03-11 08:58:15] Unable to query CUDA driver version! Is an nVidia driver installed?' I use the newest version of Cudaminer given, in combination with CUDA manager (version 1.2.1). If I try to run the miner outside of manager, it starts and then rapidly disappears. My set-up is as following:
I would try reinstalling the driver once more with the "clean" install method (that's an optional checkbox to check during driver installation) Christian Or try delete old driver with this utility in safe mode http://www.guru3d.com/files_details/display_driver_uninstaller_download.html
|
(つ ͡๏ ͜১ ͡๏ )つ[̲̅$̲̅(̲̅5̲̅)̲̅$̲̅]ε=ʕ ͡๏ ͜১ ͡๏ʔ=з
|
|
|
DemosMirak
|
|
March 11, 2014, 08:51:16 AM |
|
Hello, I just installed the new 335.23 driver on my windforce 3X GTX 770, and now Cudaminer is giving the following message: '[2014-03-11 08:58:15] Unable to query CUDA driver version! Is an nVidia driver installed?' I use the newest version of Cudaminer given, in combination with CUDA manager (version 1.2.1). If I try to run the miner outside of manager, it starts and then rapidly disappears. My set-up is as following: Corsair Vengeance 16 GB : 2 x 8 GB Toshiba DT01ACA100 1 TB Kingston SSDNow V300 120 GB Thermalright Macho Rev.A Intel Core i7 4770K / 3.5 GHz Asus B85-PLUS Gigabyte GTX 770 OC Windforce 3X Windows 7 (completely up to date) Edit: When I use the rollback function to get back to my previous driver, the error keeps persisting. Edit 2: Performing a clean install as we speak. Edit 3: The clean install did not resolve the issue, I did it twice to be sure, but the problems still persists. Hello, I just installed the new 335.23 driver on my windforce 3X GTX 770, and now Cudaminer is giving the following message: '[2014-03-11 08:58:15] Unable to query CUDA driver version! Is an nVidia driver installed?' I use the newest version of Cudaminer given, in combination with CUDA manager (version 1.2.1). If I try to run the miner outside of manager, it starts and then rapidly disappears. My set-up is as following:
I would try reinstalling the driver once more with the "clean" install method (that's an optional checkbox to check during driver installation) Christian Or try delete old driver with this utility in safe mode http://www.guru3d.com/files_details/display_driver_uninstaller_download.html The program you recommended trying did not work, as I am apparently missing the language file. What apparently did work was restarting the computer in safe mode, as it does work now, oddly enough.
|
BTC: 13enECLM3M3gjQDoBKouXuYFG4zXaDdDPx LTC: LRTbQNQcRjZV51PivQdhK7zpMtJYPouqR9
|
|
|
|
ivanlabrie
|
|
March 11, 2014, 09:59:04 AM |
|
That works just fine but using DDU (display driver uninstaller, google it) is simpler and more convenient. :p Christian, demand Nvidia to make a 4gb equipped GTX 750 Ti please, thank you
|
|
|
|
bigjme
|
|
March 11, 2014, 10:00:13 AM |
|
And get them to try and fix the need for chrome lmao
|
Owner of: cudamining.co.uk
|
|
|
cbuchner1 (OP)
|
|
March 11, 2014, 10:01:38 AM |
|
And get them to try and fix the need for chrome lmao
so far I do not know how to achieve the same effect. Should I open a DirectX window and render a little spinning cube or what Ah, now I know: I will have to integrate flash based advertising!
|
|
|
|
restless
Legendary
Offline
Activity: 1151
Merit: 1001
|
|
March 11, 2014, 10:02:18 AM |
|
The Google Chrome trick seems essential here: Boosts hashes on my desktop PC from 770 kHash to 1150 kHash/s (this is for four GPUs combined)
What's the reason for needing this Chrome trick? Card thinking it should enter 2d state?
|
|
|
|
bigjme
|
|
March 11, 2014, 10:31:01 AM |
|
And get them to try and fix the need for chrome lmao
so far I do not know how to achieve the same effect. Should I open a DirectX window and render a little spinning cube or what Ah, now I know: I will have to integrate flash based advertising! Hey if nvidia want to send you adverts I'm fine to have them ;-) support Christian through nvidia ads!
|
Owner of: cudamining.co.uk
|
|
|
djm34
Legendary
Offline
Activity: 1400
Merit: 1050
|
|
March 11, 2014, 11:11:06 AM |
|
ok I ran a bit with the blake algo. Now it works. The 780ti is able to reach at T2880x32 975~980Mhash/s overclocked around 950Mhash/s on stock setting. The strange thing is the TDP which is well below 100% while running and the gpu usage is around 97%
How do you turn-on the temp monitoring ? I didn't see it.
On an other matter, I bought a new USB riser. With the previous I had lots of crash and freeze and with the new I haven't got one since I installed it. Apparently the problem was due to some readout problem. I saw before a crash that MSI Afterburner was reporting a temperature of 184°C (causing I suppose a panic crash...) while the card was running at 70°C strangely this was not seen in the HWinfo log file... (VRM temps was ok too...) So basically over 10 risers you should expect to have a few bad ones...
|
djm34 facebook pageBTC: 1NENYmxwZGHsKFmyjTc5WferTn5VTFb7Ze Pledge for neoscrypt ccminer to that address: 16UoC4DmTz2pvhFvcfTQrzkPTrXkWijzXw
|
|
|
|