Topic: [ANN] TeamRedMiner v0.10.10 - Ironfish/Kaspa/ZIL/Kawpow/Etchash and More - page 94. (Read 211877 times)

Hey guys!

Well, unoptimized code isn't the issue, it's rather that it's optimized so much that it's pounding the gpu more than anything we've produced before, especially the memory subsystem.

I'm not chasing efficiency to the same degree that you professional tuners are since my rig(s) are a combination of test and more serious mining. However, on my 8 x Vega 56 ref cards flashed to V64, win10, 18.6.1, clocks at 1408@900, 1100@900, I have zero stability issues, it's been mining for 14h straight now in the current run, and much longer before that as well. I've just stopped it to reconfigure things a few times.

However, on my Vega 64 Liquid Cooling in my dev workstation, CN-trtl is the first algo I've ever seen that kills that specific card but not my blower Vegas. It dies after 1-2h, also running at 1408@900, 1100@900. Effective clocks+voltages in hwinfo64 look very similar to the V56s.

Right now, I'm doing tests with the single-threaded config support that we also added in 0.4.3. This means I'm running --cn_config=L56+0 instead of --cn_config=L28+28. On my V64 LC, hashrate drops from 19.6 kh/s to 18.8 kh/s, same efficiency. The hashrate is expected to drop a little, but this should be more lean on the gpu, it won't be going full throttle on all parts of the hardware at the same time. Theory vs practice is always a bitch though, I'll report back in a few hours on the results.

Meanwhile, you're of course free to try the same trick: switch your problematic Vegas to L56+0 and see if it helps, I'd love to get some more data here.

i have same situation.unstable only on trtl.try different drivers nothing change.

I can confirm the same behaviour on Vega 64 with samsung memory at L28+28, tried droping to L26+26 and still same results.
Normal : 19.56, and after some random time 19.06

OK. I have been trying over several days to replicate the results from pbfarmer on cnv8_trtl and I can say that I am unable to get the efficiency reported by him. Any attempt to reach close to his reported voltage levels of 805mv to 840mV resulted in random DEAD GPUs(stuck in enqueue) after 20-30 minutes hashing or significant hashrate drop per GPU


Even at 870mV, I am still unable to get a stable system running. Dead GPUs and hashrate drop to 15kH/s still occurs after several tens of minutes of mining. And the weird thing is hashrate drops happens to Hynix mem GPUs only.   Somehow CN-TRTL algo is more taxing that CN_R? I am able to run CN_R without failures for a week and with lower voltage settings (850mV - 870mV). I can't seem to do this for CN_TRTL

My settings are as below. But the hashrates aren't sustainable

GPUs : Ref Vega 64 and Vega 56 reference bios
V64 cclk/memclk 1220/1100 @ 870mV L28+28 (Samsung mem) - 19.5kH/s
V56 cclk/memclk 1220/940 @ 870mV L24+24 (Samsung mem) - 19.3kH/s  (Hynix mem) - 18.7 kH/s
ATW power draw 190W per GPU
Adrenalin Driver 18.6.1


Kerney/Todd,

Any ideas what could possibly be wrong? Unoptimized CN_TRTL code?

no powerplay tables at all. Not needed with latest drivers

i have identical problem.

vega 64LE dead after 1 min (2sec)

i must give 880mv for gpu and mem or gpu work only 1 min or 1 sec

pls any put PPT table this for vega64_LE and vega_56

thx

What is the soft power table used together with the settings shown on this 2.8.0 beta 1 overdriventool screenshot?  Possible to share the file?

Those settings together with a brnsted powertable , amd 19.3.3 driver result in dead MSI Vega64LC (samsung memory)  gpus less than a minute mining in TR. not even changing p7 to 875 can be used to save them from failing.

What you set in OverdriveNTool or in Wattman is not what you will get.
You'll get lower clock and voltage. At least according to gpu-z and similar apps.

Looks like there's a turtle/loki pair on hero.

Usually most miner threads put it on the first announcement, and update it as they modify code

This is what I see for official release v0.4.3, algo cnv8_trtl listed last for the -a options. Did we miss it somewhere else?

trying do this merged mining thing here but whats the flag for CN turtle ? this does not work


the -a flag for turtle is not on your help page

Dear dev,
I really look forward to the function of the failover pool in pools.txt file!

ok thx

you use 1548GPU or 1280gpu?

first you say " run my vegas 64LC at ~1280/1100@800mv with 19.6khs." but in table i see 1548gpu

Yeah, reason for pushing trtl first was rather the Loki fork. We will add the 4MB variants that still are active, I promise. It will probably be 1+ week from now though.

Only memory states disabled. You may need to reset/reboot system. Don't use Afterburner. It seems to conflict with this low settings for vega64lc (strange)

0.4.3 CNR algo GPU initialization(very low hash or no start) problem on half my rigs.0.4.2 no problem so far.

Yes please focus on it, all community is asking it. I'll be happy to return to teamred (my favorite miner) when heavy algo's are here. Turtlecoin is cool, but it's a shitty coin nothing compared to bittube or haven.

how disable? disabe what?
pls copy your overdrivetool profile... and ppt table...

i use 800mv for all p1-5 state but vega64LE vork only with p7 870mv+

Actually it's going to have to be a lot later as I'm currently experimenting more with merge-mining. Funky stuff that. Throws me back to 2014... When we were merge-mining crap like Fantomcoin and Moneta Verde... or Quazarcoin... or Digitalnote, lmao!

Todxx and Kerney, we need some CN Heavy/Heaven love for TRM!