CCminer(SP-MOD) Modded GPU kernels.

[sp_]

Try to overvolt.
perfcap reason..  does your card have 6+6 pin power or 6+8pin?

[bensam1231]

Same, I have Elpedia, Hynx, and Samsung and there is no difference between them except for the OS (W10 and W8).

Then I do not understand what affects me at 23. Maybe screenshot hash map will explain.

z.gif

You have somehow limited your power to 60%, hashrate drops 25% and power usage 40%.

He's posting a screenshot of him hashing at 23Mhs for Ethereum with a 970, which doesn't really make sense... more so because he has a tdp cap at 60%

[sp_]

the g1 gaming cards 970 have tdp of 250W because they come with a 8pin + 6pin connector.. When mining the power consumption is sometimes down to 60% of tdp. Depends on algo..

[Grim]

ETH is memory constrained thats why you cant really optimize it. (which was a VERY smart move)

All that goes on is that his card is in P 2 state.

If the timing table of memory is done for all straps up from very high oced memory down to the lowest possible state (eg. 150 Mhz) the timings of the memory actually IMPROVE if you downclock ...

That's why I suggested to eg. HEX Edit the bios timings from a LOW frequency state to the next higher one.

It is just funny because elpida is normally known to have the worst timings and the least oc headroom.

That tells me now that there is probably a BIG headroom for hynix and samsung untapped by us.

With bios timing hacks we could improve nvidia hash for eth by probably a lot.



This model is MSI 970 100 ME right?

[eth100tx]

Newbie here.  Running a 980ti and only seeing 8 MH/s on W10 (it benchmarks at 2x+)
   ethminer-cuda-0.9.41-new2 -U F xxx --cuda-grid-size 8192 --cuda-block-size 128 --cuda-schedule sync

It's unclear if I'm hitting the memory wall or have a problem given I see recent posts with higher numbers in the 22-23 970 range.  I'll dig in, but in guidance is welcome.

[bathrobehero]

I'm still curious of the hashrate difference of people overclocking their memory because everytime I do, the hashrate is the same and all I get is slighty more power consumption. For me, for every algo 100% of the improvements are coming from core OC and memory OC doesn't make a difference at all.

Anyway, Eth is pretty low at the difficulty chart as far as I'm concerned so I'm mainly asking this for other memory hard algos (lyra2, lyra2v2, scrypt-jane).

[Grim]

I'm still curious of the hashrate difference of people overclocking their memory because everytime I do, the hashrate is the same and all I get is slighty more power consumption. For me, for every algo 100% of the improvements are coming from core OC and memory OC doesn't make a difference at all.

Anyway, Eth is pretty low at the difficulty chart as far as I'm concerned so I'm mainly asking this for other memory hard algos (lyra2, lyra2v2, scrypt-jane).

GTX 970 at lyra2v2

same clocks core and memory

samsung is faster by ~200 khash compared to hynix memory

[Vita1ico]

Try to overvolt.
perfcap reason..  does your card have 6+6 pin power or 6+8pin?

6+8pin
Increase the voltage to 1.212 below the hash.(17-20)

, which doesn't really make sense... more so because he has a tdp cap at 60%

You're wrong, importance, hash really shows the pool, you can see the pool early screen 970 + 960 approximately 32Mhs

This model is MSI 970 100 ME right?

She modified the BIOS . Our people have helped to edit the BIOS on low consumption and high performance .1430 watch \ 1.137v
ps Now I understand why others can't 22-23Mhs . Elpida is not the reason ..

[Vita1ico]

Trust me the sole reason is the memory configuration!

I know how to access the BIOS on any tab to see the memory configuration ?
I'm just curious , change the memory to which side increased the hash

[Grim]

The bios tools available are lucky to tell you the brand and let you adjust the frequency only.
You can not adjust vram voltage neither can you adjust vram timings. (sadly)

see post #9155
https://bitcointalk.org/index.php?topic=826901.msg13591064#msg13591064

[Vita1ico]

see post #9155
https://bitcointalk.org/index.php?topic=826901.msg13591064#msg13591064

However any changes of numbers BIOS to do increase +5 Mhs. Because the factory BIOS is not more than 17 .

[Grim]

run this and compare to your other cards

oclmembench.exe
https://drive.google.com/file/d/0B9OoHSmkeSeNNEE3ZkpsSWlnZ1k/view

[Grim]

Also this link provides info on how to hex edit a amd bios for tighter vram timings.

http://www.overclock.net/t/1554360/tahiti-memory-timings-patch-for-hynix-vram

[sp_]

ETH is memory constrained thats why you cant really optimize it. (which was a VERY smart move)

Yes you can. The current miner is not hashing as fast as the copyspeed limit.
Creating the growing dagger file was a stupid move. Soon it will grow to big for the GPU, and then Etherum will go POS or be food for the botnets.

[Grim]

ETH is memory constrained thats why you cant really optimize it. (which was a VERY smart move)

Yes you can. The current miner is not hashing as fast as the copyspeed limit.
Creating the growing dagger file was a stupid move. Soon it will grow to big for the GPU, and then Etherum will go POS or be food for the botnets.

While the TLB trashing sucks I still think it is one of the most brilliant algos around.
More modern gpu designs seem to push the TLB trashing limit upwards.

ETH goes POS anyway... (now that is a mistake, imo)
dunno about Factom tho...

[Vita1ico]

the last mhash (you said 23-24)

22-23 , not 23-24.

@ Vita1ico do you have any other 970 to test?

Unfortunately, others do not. Only 970 + 960 (two cards)

[Vita1ico]

what gpu-z version is this?
It seems your vram actually IS overclocked!! at 1900 mhz (stock is 1750)

GPU-Z 0.8.6

you have another 970 to test

one card 970..

[djm34]

run this and

DSC03604.JPG

time to clean the room  

[djm34]

Really quick test, all cards p0 and memory +500 = 4010 MHz

asus strix samsung ~199 GB
asus strix elpida ~199 GB
palit reference hynix ~190 GB

Samsung and elpida are hashing 22MH but hynix only 19MH. Hynix needs +750 (!!!) on memory before it crosses 20MH line.

thanks for the stats

My cards are hashing in P 2 state but at stock memory clock (bios mod).
BUT in P 2 state the oc for the memory doesn't stick, it only works in P 0 state to oc successfully.

Thats why even at +500 my bandwidth didn't change  

explanation on how to switch to p0 state were given in the pledge thread, it should apply to your card as well

[pallas]

Isn't there a better thread to talk about eth mining on nvidia?
I don't mine eth and I had to skip a couple pages at least.