What’s the total writes per C33 plot on a temp drive with 128GB RAM?
EDIT: Ah, just read it, it’s about 52GB as mentioned on their website.
nossd 3.1 GPU plotting under linux is very fast. 1 minute 40 second a plot C33.
Dell R620, 768GB, 2070 super. Target is 10K mechanical RAID 0. I could do dual CPU plot in parallel also. Haven’t tried it.
Is the new version improve performance existing plots? Based on documentation, with C30 you get more compression and less GPU usage for 4090 over c14. Is this correct? But for 1080TI, there is no such improvement. This is confusing. What is the optimal set up here?
im just going to repeat back to you what you just said
Why does my gpu released almost a decade ago not get linear performance gains. DOnt you developers care about REAL GAMERS on REAL GPUS? There are tens, if not DOZENS of us still using bronze age gear and we demand that the developers expend resources to optimize performance for a gpu 2, nay, THREE generations old. No matter that the devs cant actually go out and BUY one of these gpus because they have not been sold in stores for more than five years and there would be no way for them to test these things… but I DEMAND that the devs fix my unique hardware combo for a $70 gpu I found in the bin. GAMERS RISE UP
I think you need to ask that from NVidia, cuda support is rough for older generations.
Time to invest on new version of GPU, if you plan to continue CHIA. Old ones are just too much power consumption and weak on calculation.
1 Like
You could have been more thoughtful in your reply. If your goal was to shut down the person you replied to, I think it worked.
My favorite gpu as far as bang for the buck is Rtx 3060m. It is a mobile gpu set to a
desktop board. It only uses 80w max and is very close to the desktop version in performance. The one caveat is you need to use third party modified Nvidia drivers. You can google Nvidia Frankendriver. To find the downloads. You can get an Nvidia 3060m on eBay at anywhere from 150 and up last time I checked. Or Aliexpress for a little over 200 last I looked. I have 4 of them. Once you get the drivers figured out they work well.
1 Like
2x Nvidia RTX A2000 12GB
Starting benchmark
Results are shown for plot filter 256 and for all mining GPUs combined
This PC is capable to mine on:
C1: 2134653 plots (1ms/quality)
C2: 1212919 plots (2ms/quality)
C3: 608237 plots (3ms/quality)
C4: 346941 plots (6ms/quality)
C5: 186136 plots (11ms/quality)
C10: 212635 plots (10ms/quality)
C11: 80736 plots (25ms/quality)
C12: 46764 plots (46ms/quality)
C13: 29093 plots (72ms/quality)
C14: 15799 plots (129ms/quality)
C15: 6879 plots (253ms/quality)
C30: 20774 plots (101ms/quality)
C31: 16649 plots (124ms/quality)
C32: 12479 plots (163ms/quality)
C33: 8874 plots (224ms/quality)
C34: 5976 plots (319ms/quality)
C35: 3927 plots (455ms/quality)
C36: 1942 plots (770ms/quality)
C37: 0 plots (unsupported with current GPU configuration)
C38: 0 plots (unsupported with current GPU configuration)
Benchmark ended
Nvidia 3060Ti 8GB
Starting benchmark
Results are shown for plot filter 256 and for all mining GPUs combined
This PC is capable to mine on:
C1: 3111277 plots (1ms/quality)
C2: 1573614 plots (1ms/quality)
C3: 702963 plots (3ms/quality)
C4: 405413 plots (5ms/quality)
C5: 207997 plots (10ms/quality)
C10: 189528 plots (11ms/quality)
C11: 73416 plots (28ms/quality)
C12: 41869 plots (51ms/quality)
C13: 25852 plots (81ms/quality)
C14: 14116 plots (143ms/quality)
C15: 6041 plots (284ms/quality)
C30: 19201 plots (109ms/quality)
C31: 15050 plots (137ms/quality)
C32: 11183 plots (181ms/quality)
C33: 7981 plots (246ms/quality)
C34: 5246 plots (357ms/quality)
C35: 0 plots (unsupported with current GPU configuration)
C36: 0 plots (unsupported with current GPU configuration)
C37: 0 plots (unsupported with current GPU configuration)
C38: 0 plots (unsupported with current GPU configuration)
Benchmark ended
NoSSD 3.1
Older hardware is EASILY available from places like Amazon and EBay.
And after the Etherium move to Proof of Stake, there is a TON of it available.
Per the May 2024 Steam GPU survey, the GTX 1060 is still one of the 10 most widely used GAMING cards (a little over 3%, the outright leader only has about 5), with the GTX1050Ti at 11’th at ALSO a little over 3%.
This older hardware is NOT difficult to acquire NOT anywhere close to unique, even for GAMERS that usually insist on fairly high performance.
I don’t even want to think about how many cryptocoin miners still have these level of cards floating around from their GPU mining days.
I would not recommend buying NEW GPU hardware for CHIA at this point, as it’s not likely to repay the cost before the new plot format shows up.
If you already HAVE the higher-end hardware, or have a different reason to buy it, different story.
1 Like
Chia 3.0 still works well to use CPUs to plot GPU-only farmable plot levels.
Seeing 7mins 20 seconds per plot on C33 plotting with 3.1, pair of E5-2680 v2 CPUS to a 3 drive 10K RPM raid0, 256GB RAM (8 x 32GB modules, though it’s only wanting 162 or some such - but NUMA prefers equal RAM on each CPU).
As I recall, C14 and C15 were almost identical plotting times (perhaps a hair slower, I think I’m remembering 7 min 30 secs).
NO gpu in the machine (other than whatever low level GPU chip is mounted on the motherboard).
I use a small modification to the PLOW script to move the plots from the RAID temp-drive array to the final plotter machine drives.
I did have to add the “-no-gpu-plotting” flag that 2.x didn’t need to make it work, no other changes (other than the -c level) to my Nossd command line.
No, there’s no way to put a GPU that’s worth plotting with in that older HP 1u server - it doesn’t have space for even an A2000.
How can you mine with different plot compressions? I now have c15 and c33 plots on my farm, do I need to open multiple instances of the miner, one for each plot compression level?
No, that is a not the idea. Unless you want to use a weaker gpu for the less compressed plots…no
Just re-read your question. It just works, no special setup is needed for multiple different compression levels
The only time you have to configure the software for C level is when you are plotting.
Farming just handles the C levels internally, makes a note of which plots have which compression (I presume via a flag in each plot) when the farming starts up, and does not care if the compression levels are mixed.
Hm, ok then, I thought the new version doesn’t support mining different compression levels, because I got errors when I tried it.
But it seems my system just can’t handle plotting and mining at the same time with the new version, at least when mining on remote plots, I’m getting errors when reading remote plots now if plotting at the same time.
(I have Win10, 3080 Ti and 128GB RAM, and I also set VRAM limit when mining, so maybe there’s a bottleneck somewhere? Or maybe a Windows 10 thing?)
I suspect it’s more that the new version REPORTS when plots aren’t getting read fast enough, not that it’s a new issue.
Plotting remotely DOES put a lot of load on the LAN, which ties it up under heavy load while plotting - that is probably your bottleneck.
I see the same thing on the machine I have to NFS mount to remotely farm the plots, when I’m plotting to it - it’s NOT a Windows specific issue.
Yeah I agree, although I’m not plotting remotely, I’m only plotting locally. Seems the 3080 Ti, at least my 3080 Ti is not able to handle plotting and mining at the same time. The load isn’t near the limit too, maybe half of the max load (based on the benchmark).
But now I’m done replotting, I see no more warnings or errors.