… but you haven’t told them about the power consumption
Sounds like your k34s scale well @ x2 my k33s. I’d love to see logs to understand how it treats the phases Have you done k33s?
What do you perceive gives most advantage … RAID on the ssds or strength of the M1 CPU or … ? … the 980 Pros are pretty awesome little guys! Is Thunderbolt the best connection of the M1 … no nvne slots?
My setup and plot times for chia(k32/33/34) and chives(k25/29/30/31)
CPU/MB: Ryzen 5900X, MSI B550 Tomahawk, under volting with PBO2_neg20 → ~4.4MHz
MadMax: run with Ubuntu 20.04 inside Windows 10
Temp -t: Crucial P5 1TB (PCH pcie lane)
Temp -2: Raid0 (2x Intel DC P4600 1.6TB) (CPU pcie lane)
|plot k size||k25||k29||k30||k31||k32||k33||k34|
|Phase 1 total (sec)||69.1381||108.841||162.107||299.93||632.386||1584.98||3672.97|
|Phase 2 total (sec)||6.52367||46.1537||96.4817||207.041||470.602||1102.19||3694.84|
|Phase 3 total (sec)||87.1587||124.85||163.397||254.573||443.403||1228.47||2615.59|
|Phase 4 total (sec)||5.96225||18.415||34.8181||65.13||141.858||316.992||789.181|
|Total Plot time (sec)||168.858||298.344||456.888||826.768||1688.38||4232.8||10772.8|
|Total Plot time (min)||2.81429||4.9724||7.61481||13.7795||28.1396||70.5466||179.457|
|Plot time ratio||k29/k25||k30/k29||k31/k30||k32/k31||k33/k32||k34/k33|
Just done my first K34, 2 hours 37 minutes (157 minutes)
Phase 1 took 2926.86 sec
Phase 2 took 3625.05 sec
Phase 3 took 2601.7 sec, wrote 87508814944 entries to final plot
Phase 4 took 296.165 sec, final plot size is 461559538237 bytes
Dell T7910 with dual E5-2699v3 CPU’s, 512GB DDR4 ram, and 6 x S7310 200GB drives in Raid 0
CPU’s are not fully utilised, so if I had the disk space I could probably do two in parallel for a quicker per plot time.
PS. I do suffer with the random phase two slow down, so not sure if this is in slow mode or fast mode.
With 72 cores seems something’s not right. Just finished 7x K34s @ ave 127 min (2hr 7 min) w/16C doing MM two in parallel w/P1 stagger. P2 seems off (my ave times in italics). Maybe P2 was in your ‘random slow’ mode?
Phase 1 2926.86 sec > 3611
Phase 2 3625.05 sec > 1622
Phase 3 2601.7 sec > 1969
Phase 4 296.165 sec > 411
What MM parameters are you using? That can made big changes. And two (maybe more for your cores?) in parallel knocks time off as it uses the CPU more evenly and esp. when you stage Phases.
Its 36 cores, 72 threads.
chia_plot_k34 -k 34 -x 8444 -n 1 -r 36 -K 2 -u 256
Can’t do two in parallel as only have1TB of temp space.
Quite a few people have posted to the thread for random phase two slowness, but unfortunately no answers or fixes.
Right 36C. Might try r 18… esp if and when you can do two parallel. Otherwise, that’s same as my setup. Thought you just got a couple 1tb SSds?
No, I got another two Intel S3710 200GB SSD’s, I now have six in raid 0, gives about 1.1 TB of space.
I would use R 18 when running two in parallel, which is what I did when doing K32 in parallel, but would it make any difference with only one plot - I’ll try it anyway.
Not sure. I’ve found 1/2 cores is about like full cores, sometimes better. Times are a so variable it’s hard to tell what is the best setup.
I just tried it with R 18, phase 1 was looking considerably slower so I stopped it.
P1 T1 53s
P1 T2 325s
P1 T3 497s
P1 T1 76s
P1 T2 368s
P1 T3 773s
I try another run with the same settings I used earlier just to see if it was a slow phase 2, unless of course I get two slow phases 2’s in a row.
Do you have any PCI-E slots?
Yes, four x16 slots free, no I don’t want to buy any NVME drives, or single NVME expansion cards (I probably still have some), or even Asus Hyper M2 cards - system probably does support bifurcation.
A year ago I spent a fortune on various NVME drives, some have been sold, some have been repurposed. I can knock out K32’s entirely in ram one every 12 minutes, and K33’s in 57 minutes*, I really have no need to spend substantial amounts of money on plotting hardware - it would be better spent on spinning hard disks for plot storage. Only bought the last two 200GB drives just so I could see how long K34’s take, and I deleted that K34 as I have no where to put it at the moment
*Hopefully the new version of Bladebit will do K33 and K34 plots when it arrives, in a combination of ram and storage space.
I hear you. I have to go over to Bones’s loft and wash his car a couple more times to get some Quid then buy some more hard disks…
Seems P2 was off, just did another and that was 142 minutes (2 hours 22 minutes), so 15 minutes faster than previous one using exactly the same settings.
Phase 1 took 2939.22 sec
Phase 2 took 2734.24 sec
Phase 3 took 2581.65 sec
Phase 4 took 294.904 sec
Still quite a bit off from your times though, some phases faster, others a lot slower.
I am not aware of a 36 core ThreadRipper.
Did you mean 32 cores?
From AMD’s web site:
|AMD Ryzen™ Threadripper™ PRO 5995WX||Discrete Graphics Card Required||64||128||Up to 4.5GHz||2.7GHz||280W|
|AMD Ryzen™ Threadripper™ PRO 5975WX||Discrete Graphics Card Required||32||64||Up to 4.5GHz||3.6GHz||280W|
|AMD Ryzen™ Threadripper™ PRO 5965WX||Discrete Graphics Card Required||24||48||Up to 4.5GHz||3.8GHz||280W|
|AMD Ryzen™ Threadripper™ PRO 5955WX||Discrete Graphics Card Required||16||32||Up to 4.5GHz||4.0GHz||280W|
|AMD Ryzen™ Threadripper™ PRO 5945WX||Discrete Graphics Card Required||12||24||Up to 4.5GHz||4.1GHz||280W|
Is this what your buying me for my birthday???
Neither am I, but it is 36 cores, well actually 2 x 18 cores, think you missed this bit though
Just started plotted k34 on windows 10 using chia built-in cli for mad max with a new zfs qnap nas with 20TB drives in 2 disk raid 0 (no straight jbod on zfs).
On k32 i used stoiks(sp?) And did two plots at once with different tmp nvme and no different tmp2 and averaged around 2 hours for each process. No ramdisk.
Using a single E5-2699v3 @ 2.3ghz.
On my first pass of k34 today i averaged 4.5 hours each with two plotters at once. I used tmp 1 nvme 1 and nvme 2 tmp 2 for plotter 1 and reverse for plotter 2.
Using all 18 cores, K value of 2, 512 buckets. No other tweaks.
Was it switching the tmp1 and tmp2 that helped or the K to 2?
Im surprised my k34 is only 2.25x slower than k32 as ive seen people on here have longer times.
I have a more powerful nas on the way with sas expansion units and a new plotter which will run debian and tmp1 as nvme and tmp2 as ramdisk. Cpu is 2x e5-2690, excited to see what plot times i can get out of that. Worth doing tmp2 as ramdisk or just straight nvme on different disks like this one? Any other tweaks i should look at on either system?