Topic: Claymore's CryptoNote AMD GPU Miner v11.3 - page 279. (Read 2145076 times)

After some research I can confirm that the reason of bad performance is Elpida memory type on GPU, not only 280x is affected but all cards.
I compared several 290x cards with Elpida and Hynix memory:
Stock 290x Elpida - 650 h/s.
Stock 290x Hynix - 740 h/s!
The bottleneck of CryptoNote PoW implementation on GPU is memory latency, Hynix memory is really much faster than Elpida. Latest GPU-Z shows memory type so you can check it easily.

Thanks for the suggestion but that utter crap Elpida memory Gigabyte always uses will never be able to do 1700Mhz  At least not on my cards.

@trip96
now I see, I had old version of GPU-Z, so I've seen just GDDR5 in memory type,
so I can tell you now, that most of cards are with Hynix memory, just two Sapphires are with Elpida memory.
All ASUSes 280x, Sapphires 7970 have Hynix, but Sapphires 280x, comes with both types of memory.

@kqpahv
I don't know your clocks, but when you set memclock to 1700 it shoud be better,
let me know then, because I have one GB 7970 and with stock (1375) or even 1500 Mhz it was slow,
but after that increase, I start to get normal hashes.

I had the exact same problems, but found them gone once I started mining without overclock. I've 290 cards though.

Please send me complete log so I can check it in details.

My GB 280x's are also with Stilt's bioses and I am still getting 360h with them...

I should add, I believe from what other miners have told me that if any backup pools are configured then it will start submitting shares again after starting a new pool connection, so its not as bad as it looks here.

Hello, sorry to say but v3.4 still seems to have the Monero pool disconnect issue, been testing it against my pool (cryptonotepool.org.uk) so I see from both ends

Running cpuminer-multi and GPUminer 3.4 on same Win8x64 PC, with 270X

The stratum stops responding for about 10 secs whilst it waits for the daemon to backup its blockchain. Claymore miner doesnt report disconnect at this point, but stops hashing shares, 30 minutes later it says "no new job received for 20min", despite having received jobs for last 30min and not doing anything with them I then restarted the pool which caused a proper stratum disconnect, which Claymore registered, reconnected and then waited 30min before reconnecting and receiving new jobs, but not hashing any shares. Throughout that period cpuminer-multi only complained when the pool was restarted and it had to reconnect.

Looks to me like the problem is caused by it being unable to submit a share, but I'm not sure, since there is a logfile here are relevant excerpts, please let me know if you need more details.

Thanks!


Hashing away happily for over 8hrs until this last round, which is when daemon freezes but the pool stratum is still running
02:14:46:425   b10   Time gpu (0): 2802113
02:14:46:435   b10   Round
02:14:46:542   b10   t1: 95598
02:15:25:771   1958   got 253 bytes
02:15:25:774   1958   buf: {"jsonrpc":"2.0","method":"job","params":{"blob":"010094c5a89d0521cb17e4e0ba112c6642c1f473f12d3341c37cc8233e728fbc1ccd051e4cd9e50 0000000d60bc56f561459813cf1eabb4bf8752ca4a4eec13d74549573a67c93440a761e01","job_id":"985668597230687","target":"26310800"}}

02:15:25:777   1958   parse packet: 253
02:15:25:780   1958   new buf size: 0
02:15:25:784   1958   DevFee: 06/25/14-02:15:25 - New job received from cryptonotepool.org.uk:7777
02:15:26:126   18e4   got 253 bytes
02:15:26:129   18e4   buf: {"jsonrpc":"2.0","method":"job","params":{"blob":"010094c5a89d0521cb17e4e0ba112c6642c1f473f12d3341c37cc8233e728fbc1ccd051e4cd9e50 0000000c7f8c9c5c834dc86a5fe0071fc6811fe919b6f03ebb2392fb995f877017b027b01","job_id":"966067351540550","target":"33ec0a00"}}

02:15:26:133   18e4   parse packet: 253
02:15:26:136   18e4   new buf size: 0
02:15:26:139   18e4   df has same pool, skip
02:15:26:143   18e4   06/25/14-02:15:26 - New job received from cryptonotepool.org.uk:7777
02:15:26:146   18e4   Speed: 226 h/s, TotalHashes: 2691K, DevHashes: 133K Mining time:
02:15:26:150   18e4   03:18
02:15:30:849   1958   got 253 bytes
02:15:30:852   1958   buf: {"jsonrpc":"2.0","method":"job","params":{"blob":"010099c5a89d05296c5ab1430803bfab458b509b76e68f6cb7492857a997a0942b1a5466bd26d50 000000011416ef1a2688b77172d313e433bdc520dbb726156e274114b3596948197613b01","job_id":"972674444760195","target":"26310800"}}

This repeats until pool is restarted, then everything stops for 20min before getting going again

02:48:55:758   18e4   df has same pool, skip
03:07:35:690   1958   No new jobs were received for 20 mins - something wrong with this pool, disconnect.
03:07:36:698   1958   Stratum - connecting to 'cryptonotepool.org.uk' <80.71.13.55> port 7777
03:07:36:715   1958   Stratum - Connected

But now it isnt doing any work
03:20:35:784   18e4   parse packet: 253
03:20:35:786   18e4   new buf size: 0
03:20:35:788   18e4   df has same pool, skip
03:20:35:791   18e4   06/25/14-03:20:35 - New job received from cryptonotepool.org.uk:7777
03:20:35:794   18e4   Speed: 170 h/s, TotalHashes: 2691K, DevHashes: 133K Mining time:
03:20:35:797   18e4   04:23
03:20:36:391   1958   got 253 bytes
03:20:36:394   1958   buf: {"jsonrpc":"2.0","method":"job","params":{"blob":"0100dae3a89d0501cf7e5bb637c34471ea05373fbbd1f35cf4d6399a1a7020bd5b537526436af80 0000000ad58fff8b7e82e5667b80a2532066383e9e147fa2756cad58c41128bbd05ebc402","job_id":"318763095210306","target":"26310800"}}

03:20:36:397   1958   parse packet: 253

anyone try this out yet?

@mirny GPU-Z - Graphics card tab (the default tab that opens) on the left hand side 10 or so rows down is the field "Memory Type"
Mine says GDDR5 (Elpida). I would assume yours will say GDDR5 (Hynix).


I have an update for people as well with the 280x issue. I have found a repository of STILT's BIOS for the 280 cards here:

https://mega.co.nz/#!45pk0ZiB!UXqe0mwS4HepxmIKAOpNtWNNRyoci_jD-bo49DTRb6o


Also, I confirmed that the flash was successful using GPU-Z and the card performed slightly worse and REALLY bogged my system up so that it was definitely unusable while mining. SO STILT'S BIOS is not working for me. Or perhaps the BIOS was not actually STILT's but the flash was successful according to the difference in BIOS enumeration and the fact that the BIOS loaded. I tried another BIOS I found online and it wouldn't even flash.

Anyways, that's all for my update. Anyone else having issues with the 280x feel free to try the link I provided and see if the BIOS in there works for you. I used the ATI flash utility to flash it. That I found here:

http://www.techpowerup.com/downloads/2306/atiflash-4-17/

709 H/s per 290x here, problem is after 16 hours of mining at minergate at 3.098 H/s i was able to make only 1,64 XMR (Monero Coin)

I switched to stratum+tcp://monero.graymines.net:3333 for the last 17 hours, 100% uptime and very fast, but still no block found on that pool cause im alone mining there, so if you people seeking for an stable monero pool, this one is a good one: stratum+tcp://monero.graymines.net:3333...

but still want to know how many XMR are you people making at minergate per day and your hash rate please

My rig: 4 x R9 290x 709 H/s per card at stock clock (1000 Mhz - 1250 Mhz)
CPU: i7 4770k
RAM: 8 GB

@trip96
where exactly in gpuz?
voltage help with throttling

@BitRock
750-760 khs

@spec84
good one

@d-unknown
yes, old cards are working, but what depends is memory size of gpu in this case

Claymore do u have any 280x with elpida vram chips to test our problem?
In other algos we use to lower the difference trought different setings for thread concurency and intensity, and gpu-vram clock.
Here we cant do nothing about that and it is allmost 25% difference.

There was a guy called "THE STILT", he did miracles with 7950/7970/280x bioses and vga memory timings, correcting bios (modding) he pulled out better hashrates from cards with elpida memory.

Don't know if he is still in business, but hoping he could help.

Voltages did not help with hashrate. I am now thinking it has to do with 280x specifically elpida memory.

Any more ideas?

http://benchcryptopools.myvnc.com/monero/#
0% fee pool
24/7 online

What is the core clock and memory clock you use?

What's up with this?

Using minergate for XDN (DuckNote)

because this miner is clearly not designed for old cards, be happy that it works, and dev has not intention to provide support for those gpu's, it's sad, because, they perform as well as new gen cards on other algo's

imho claymore miner use fixed intensity, 5xxx & 6xxx cards don't appreciate it, it's like using -i 21 with a 5xxx card on scrypt mining, hashrate will drop

I have 7950x4 rig, undervolted it today from 1.031 to 1.009, all work nice. Hashrate 470h/s from card.