Show Posts
|
Pages: [1] 2 3 »
|
The only workaround to the Pascal TLB bug is to run at a higher clockspeed, but that means more power usage, as you discovered. Unfortunately I can't do this on my P100 16GB cards since all the clockspeeds are locked because it's a 'datacenter product' according to NGreedia.
its not really a bug... its a small cache size that was outgrown.. you should be able to change clock speed on dc product with nvidia-smi, at least I have been able to in the past unless they locked it in later drivers..
|
|
|
try to set mem -502, core +200/220 pl 75...
That got me to 40~41 Mhs but PL 75% = 225W. Lowest PL with nvidia-smi for a 1080Ti is 125W Settled for 34Mh/s @ 130W PL +145 CC +1600 MC Thanks everyone! I might play around with automated optimization.
|
|
|
Here's my current settings and results: CPU: Intel Xeon E5-2660v2 GPU Driver version: 390.144 CUDA version: 9.0, CUDA runtime: 8.0 sudo nvidia-smi -pl 158 DISPLAY=:0 nvidia-settings -a [gpu:0]/GPUGraphicsClockOffset[2]=+145 DISPLAY=:0 nvidia-settings -a [gpu:0]/GPUMemoryTransferRateOffset[2]=1850 => 48C P2 161W / 158Wethminer 0.18 or PhoenixMiner 5.4c both need ethlargement (-straps 2 does nothing) => 36.7Mh/sThanks for the info! System was installed in 2019 -- used nvidia-driver from debian repo and CUDA toolkit had to be installed from nvidia website at the time. Already watercooled  Tried settings from : https://miningchamber.com/gpu-mining/gtx-1080-ti-mining-settings/Core Voltage (VDD): NA Power Limit: 130w Core Clock (CClock): +40 Memory Clock (MClock): +1600 Result: 33 Mh/s As of today:  mind sharing your miner, settings and driver version?
|
|
|
https://bitcointalk.org/index.php?topic=5289066.0Low 40's is normal for the 1080ti and it will only get lower due to a known bug suffered by the 10x0 series excluding the 1070ti where it was fixed. 1080Ti has gone from 55 to 43mh in 5 years and 1070 has gone from about 33mh to 25mh. I thought the lack of update for the 10** series is why they lose hashrate performance easily? What is this bug update that you are talking about? I doubt the gtx1080ti can use 470 and 471 Nvidia drivers, you need to download the last old driver which I believe it's the 460 series, I don't know what Nvidia did this. Imagine the DAG data is a bunch of buildings in a city and we only have the phone number. We need an address book to map phone numbers to addresses -- this address book resides in the TLB. Unfortunately the size of the address book has exceeded the TLB size, so we no longer have fast look-ups for the entire address book - parts of it have to be recalculated (cache miss). NVIDIA figured a smaller TLB size was enough to last a while (512kb is enough for anyone). For gaming thats true -- not so much for ETH HASH. So theres no we to go back to full performance unless the DAG size is reduced. TLB is physical cache on the GPU (like CPU L1/L2 cache) so no way to get more with fabbing a new GPU chip. My hashrate is at 37 Mh/s, so there still room for improvement to low 40's Mh/s. Going to try upgrading driver version and cuda version and see if that gets me the last few Mhs..
|
|
|
Thanks for the info about the Translation Lookup Buffer (TLB) issue.
Driver version 390.144
Tried PhonenixMiner 5.4c from the other thread, tried -straps 2 but didnt work. Still needed ethlargement and got to 37 Mh/s
Using nvidia-settings version 390.144
Tried the latest driver a while ago and got really poor hashrate so stuck with what works..
maybe its time to try 4XX?
debian apt has 460.91.03 as latest
|
|
|
Hi all, whats the latest and greatest for eth hash in 2022?
Still using ethminer 0.18 and ethlargment. Used to get 55 Mh/s, now its dropped to 35 Mh/s -- this can't be pure from difficulty increase can it?
Found current 1080Ti benchmarks still getting 50+ Mh/s -- not sure if they are legit
Seeing threads on using t-rex miner, and phoenix miner being the current fastest.
Mining specs:
gpu: GTX 1080ti os: debian 11 driver: nvidia-legacy-390xx-driver
Whats the recommended miner and hashrate for ETH on a GTX 1080 Ti?
|
|
|
Update: Decided to release a Winblows supported version. I've tested it on 2 systems with Latest adrenaline version (one with a Vega gpu, the other with a 570) and confirmed working. Check OP for link. Please post results here. That's all i'm asking  Cheers guys! @Eliovp, once again thanks for your efforts here getting it onto windows. Much easier for traditional Vega Users with PPTs and Overclocking tools to really test the limits of your tool! For my part, I'm getting pretty decent results on ETH; Reference Powercolor Vega 64 on latest Adrenalin 19.3.3 drivers. Thanks to Windward for getting me started on values for parameters. https://i.imgur.com/6bedvOO.jpgGiven the generosity of Eliovp in even making this tool available, I'm not gonna be the guy that takes and doesn't give back. For those curious, here's what I put in a batch script after I honed in on parameters. (PATH to folder where MemTweak is stored)\WinAMDTweak.exe --CL 16 --RAS 30 --RCDRD 14 --RCDWR 14 --RC 44 --RP 14s --RRDS 3 --RRDL 5 --RTP 5 --FAW 20 --CWL 7 --WTRS 4 --WTRL 9 --WR 16 --WRRD 1 --RDWR 19 --REF 17550 --MRD 8 --MOD 15 --PD 8 --CKSRE 10 --CKSRX 10 --RFC 244 Run before starting the miner. For some reason, injecting the timings when mining would crash the miner if FAW>17, but FAW>17 is okay if injected prior. Also, RC<44 led to incorrect shares being found. No guarantees of stability... I was simply looking for accepted shares w/o rejects in short runs. I need to read up as well on compatible RAM timings... lower RC seems to drive HR the most. Happy to answer any Q's; also, memclock was 1107 and was using PhoenixMiner4.2c Thanks @Eliovp for your research and releasing this tool! I'm liking your 12 GPU Vega system -- how does it fare for compute use? I worry not all the slots are x16. Thanks for sharing @N2DCRYPT! 19.4.2 got me stuck at 37 Mh/s and temp was in the 70s and climbing! I'd advise anyone not to upgrade to those. I'm getting 48.5 Mh/s with a Reference Gigabyte Vega 64 on water @ 59 C with 19.3.3 drivers. Was getting around the same @ 54 C with the block chain drivers, I'll probably downgrade back to those.
|
|
|
The pictures  is it possible to upload somewhere else ? +1 all the pics unavailable and they were an awesome reference
|
|
|
Don't spread FUD.
If your'e scrappy and hustle you can get an 8kW array (US or EU panels) with hybrid inverter + mounting for $8000 USD. The battery depending on chemistry will be another $2000 - $6000 - also factoring in sizing, depth of discharge, etc.
Theres no maintenance if you install properly, maybe occasionally dusting off the panels if your climate is like that.
The payback ~2 years at 95%+ utilization if you're paying $0.10/kW from the utility
You have to change the battery often, it will last only few cycles, your 2 years is misleading, more like 10 years if you pay $0.10 per khw and 20 years if $0.20 per khw. Panels last only 10 years or so, some trolls say 20 years which is also misleading cause okay 20 years and 30% of what used to offer at beginning, panels degrade with time, you have to change them. There is no end to solar power maintenance. Well if you're not careful you'll still blow something up but if you do you're research, power from the sun aint bad. Use Lithium-ion or newer chemistries, you'll get quite a few years and 80%+ DoD, no battery changing Panels made in 2013+ degrade 10% after 6+ years and are rated to produce at least 80% after 20+ years or replaced free under warranty. This is why new solutions like GRID-TIED solar arrays are more profitable and easier to maintain. Just freakin plug the inverter into your normal circuit and you are good to go. I feel that solar setups using batteries are not profitable and are for those doomsday preppers who expect the whole world to explode.
Solar can be profitable, if you know what you are doing.
Well my reply was for off-grid, grid tied totally eliminates the battery if done correctly, many power providers offer the service however most want to buy your electricity for much less than what they sell to you, they are such trolls hehe If you grid tie with solar you can usually negotiate time of day rates and negotiate <0.06/kW at night depending on the area. Yea non-1:1 net metering is a huge rip off. gotta email your local politician to vote against that shit.
|
|
|
Wish you all the best. Hope it works out.
|
|
|
These are just cheap Chinese panels. You do realize that to get stable electricity working you need a shitload of other expensive stuff? Batteries and all the controlling electronics etc. Solar power is not cheap, it's one of the most expensive ways to get electricity at the moment actually. People that go for it are mostly interested in being "off the grid", doing a fun tech project, and trying to be eco-friendly. No one in their right mind switch to solar power to save on electricity bills, cause solar is way more expensive than other sources in most parts of the world.
So true and like you said, mainly off-grid cause is stupid going solar if you pay less than $0.20 per khw but i guess some people will never understand what you said because for them once they break even they get free power which is a lie cause there a constant maintenance for that electricity and in the end is tends to be a lot more expensive than most power providers offer and a lot more work, mining is the same thing, too much work for peanuts right now, why mine while you can buy coins cheaper than you would pay to produce, these prices will not last long cause the people selling right now are losing everytime they sell/cash out for usd or other currency. Don't spread FUD. If your'e scrappy and hustle you can get an 8kW array (US or EU panels) with hybrid inverter + mounting for $8000 USD. The battery depending on chemistry will be another $2000 - $6000 - also factoring in sizing, depth of discharge, etc. Theres no maintenance if you install properly, maybe occasionally dusting off the panels if your climate is like that. The payback ~2 years at 95%+ utilization if you're paying $0.10/kW from the utility
|
|
|
if nvidia have kept quiet last year and found a way to not report the oversold gpus then last year and this year would be positive, now people saw that last year was a record win and might see this year a record loss hehe,
As a public company, I think thats illegal per SEC rules on public disclosure. They have to report P&L and inventory so sure they can play number acrobatics but they'll have a huge 'loss' somewhere. I mean they ONLY made 3.21+ BILLION in revenue and 1+B in profit lol edit: also there are other industries that have a larger demand for GPUs vs. mining - nvidia will be fine. If anything, they should improve how fast they can ramp up and down production lines, rather than try to hold stock and play the market.
|
|
|
Nvidia really is just spinning this. Deny that any sales will go to crypto so that wall street won't tie the ups and downs of bitcoin to stock price. Then release a new GPU that is a hashing monster causing miners to buy them up. Enjoy record profits, while still trying to convince wallstreet that sales were from gamers living in their mom's basement, and not miners who buy 8 GPUs to fill a Rig. uh what? https://www.tradingview.com/chart/?symbol=NASDAQ:NVDA + BTC-USD overlay => no real correlation Remember just a couple months ago the CEO of Nvidia was trying to tell everyone that new GPUs were a very long ways off. Now we have the announcement in a little over a week.
He was/is trying to prevent a fire sale on existing stock (10XX series GPUs) and having channel partners eat losses and/or return stock and have nvidia eat losses.
|
|
|
nvidia just disowned crypto going forward in their earnings call happening now lol
|
|
|
110 V outlet --> kilowatt -> surge protector -> PSU
|
|
|
Did some more testing and had a crash with the pure ATX power supply. Also swapped out the C13 power cable. The issue is definitely the motherboard/cards -- when the system crashes, the cards draw way too much power if not completely powered down. Current setup now: 110 V -> DPS-1200 -> Power GPUs and 12v EPS |---> Pico PSU ---> 24 pin ATX + SATA 100 Mh/s @ 427 W 
|
|
|
dps-1200 + atx = 1st boot 435 W -- crashed on 2nd remote login 2nd boot 539 W
Only First boot 427 W - reboot Second boot 431 W
HWiNFO64: CPU: 23 W VEga 100W 1080 Ti 160W
I'd say its most likely power supply line quality and this chinese mobo not mixng very well.
|
|
|
|