[ANN] ccminer 2.3 - opensource - GPL (tpruvot)

[myagui]

From my first tests with cuda 8 rc, sm61 doesnt improve anything... over 5.2, maybe this rc is not yet optimized for the GTX, but only for the Tesla 6.0

To resume, the 10xx are more like overclocked maxwells with a lower power usage :p Remains the 1080 special memory which seems a crap for now on mem algos

They are tweaked maxwells with lower power usage, according to some in-depth analysis from ixbt
It seems Nvidia focused on game/VR improvements
1 thing that should be faster - context switching/ switching between compute/graphics or different

Well that's exactly what we needed: maxwell with lower power usage.

Yes and No Pallas. It would have been great if we had faster & low latency memory access (there's no significant improvement there over maxwell, if at all). Still, for compute-heavy stuff, great improvement.

[1715690509]

1715690509

[1715690509]

1715690509

[bensam1231]

From my first tests with cuda 8 rc, sm61 doesnt improve anything... over 5.2, maybe this rc is not yet optimized for the GTX, but only for the Tesla 6.0

To resume, the 10xx are more like overclocked maxwells with a lower power usage :p Remains the 1080 special memory which seems a crap for now on mem algos

Yup, but seemingly NeoScrypt is performing like shit on the 1070, regardless of it having faster GDDR5 then the 9xx series. Ethminer wont even run on the 1070 in w8.1

Both the 970 and the 1070 have a 256bit bus.

neoscrypt is performing very well on a 1070, around 11150 on LINUX, with the correct oc, this is 70% more than a 970

So only in Nix? Then it's still broken.


Yeah, the 256bit bus on the current generation of GPUs is rather disappointing. That seems to be a largely contributing factor to current poor performance in memory hard algos.

[joblo]

From my first tests with cuda 8 rc, sm61 doesnt improve anything... over 5.2, maybe this rc is not yet optimized for the GTX, but only for the Tesla 6.0

To resume, the 10xx are more like overclocked maxwells with a lower power usage :p Remains the 1080 special memory which seems a crap for now on mem algos

They are tweaked maxwells with lower power usage, according to some in-depth analysis from ixbt
It seems Nvidia focused on game/VR improvements
1 thing that should be faster - context switching/ switching between compute/graphics or different

Well that's exactly what we needed: maxwell with lower power usage.

Yes and No Pallas. It would have been great if we had faster & low latency memory access (there's no significant improvement there over maxwell, if at all). Still, for compute-heavy stuff, great improvement.

A lower price to performance ratio would also have been nice. The lower power usage is the only advantage over Maxwell ATM.

[Amph]

From my first tests with cuda 8 rc, sm61 doesnt improve anything... over 5.2, maybe this rc is not yet optimized for the GTX, but only for the Tesla 6.0

To resume, the 10xx are more like overclocked maxwells with a lower power usage :p Remains the 1080 special memory which seems a crap for now on mem algos

Yup, but seemingly NeoScrypt is performing like shit on the 1070, regardless of it having faster GDDR5 then the 9xx series. Ethminer wont even run on the 1070 in w8.1

Both the 970 and the 1070 have a 256bit bus.

neoscrypt is performing very well on a 1070, around 11150 on LINUX, with the correct oc, this is 70% more than a 970

1150 ? not more
my 970 gives 710kh/s peaked with Neoscrypt

you think a 70% increase is low? are you joking, also it appear that it can be overcloked even more, this was done with only +200 on the core(1750), these card can reach crazy value when overcloked, like 1800-1900

to me it's more than enough, in the end it's the 2x perf/watt that everyone was expecting since the beginning....

A lower price to performance ratio would also have been nice. The lower power usage is the only advantage over Maxwell ATM.

uh what? it's freacking 70% faster on both etheruem and neoscrypt, how this is only "power advantage"

[joblo]

A lower price to performance ratio would also have been nice. The lower power usage is the only advantage over Maxwell ATM.

uh what? it's freacking 70% faster on both etheruem and neoscrypt, how this is only "power advantage"

Most algos are around 50% higher than their Maxwell predecessor and so is the price. When Maxwell came out it
had an immediate price advantage in addition to the power saving over Kepler.

[Amph]

there is also a power advantage besides the 70% more hash, because a 970 at 700khs mining neoscrypt, will consume like 170w against 145w of a 1070 with 1150khs

[Epsylon3]

what i meant is about the compiled cuda arch, is sm61 was not faster than sm52 on these cards... at least for now. So "older" ccminer versions should be enough for them.

The main hardware improvements made for Tesla are not present in the GTX (or very limited/emulated) and are not related to mining (FP16x2 instructions)

[antonio8]

I replaced just my 970 with a 1080 and the temp in my room has been night and day difference.

The only thing I hate is the price but oh well.

As Epsylon3 said above maybe down the road the cards can be taking advantage of for mining.

[joblo]

I replaced just my 970 with a 1080 and the temp in my room has been night and day difference.

The only thing I hate is the price but oh well.

As Epsylon3 said above maybe down the road the cards can be taking advantage of for mining.

The price/performance may not be ideal yet but there should be much more potential for SW optimizations.
And the power savings alone is worth it unless Maxwell prices drop significantly.

I also wonder about the possible 32 GB barrier. I'm not aware of anyone getting more than 5 980ti's working.
With the 8 GB on the 1080 & 1070 this could be limited to 4. Hopefully this is just a simple driver fix and not something
more fundamental.

[Amph]

I replaced just my 970 with a 1080 and the temp in my room has been night and day difference.

The only thing I hate is the price but oh well.

As Epsylon3 said above maybe down the road the cards can be taking advantage of for mining.

The price/performance may not be ideal yet but there should be much more potential for SW optimizations.
And the power savings alone is worth it unless Maxwell prices drop significantly.

I also wonder about the possible 32 GB barrier. I'm not aware of anyone getting more than 5 980ti's working.
With the 8 GB on the 1080 & 1070 this could be limited to 4. Hopefully this is just a simple driver fix and not something
more fundamental.

there is already one guy on the ethereum forum running 3 x rig with six 1070 each, without problem, so apparently this is not a real issue, or maybe only with windows? he is using linux

[joblo]

there is already one guy on the ethereum forum running 3 x rig with six 1070 each, without problem, so apparently this is not a real issue, or maybe only with windows? he is using linux

Thanks, that's good to know. Either it's fixed or the 980ti problem was a different issue.

[bensam1231]

BTW this is what it looks like mining NeoS on a 1070 in W8.1 x64. Newest Nvidia drivers.



The MCU isn't being maxed out, which is usually what happens if something is memory bound (MCU usage is actually less vs 970). There is some weird sort of tuning issue with NeoS that needs to be ironed out, pretty certain once it is performance is up to par.

[Epsylon3]

could you return in sp thread for noise about eth and neoscrypt, tx

[bensam1231]

You also have a integrated NeoS miner, thought it was worth mentioning where discussions are currently happening.

[Epsylon3]

i still dont have imported the last djm version... but its on my todo list for the 1.8

[antho281]

I replaced just my 970 with a 1080 and the temp in my room has been night and day difference.

The only thing I hate is the price but oh well.

As Epsylon3 said above maybe down the road the cards can be taking advantage of for mining.

The price/performance may not be ideal yet but there should be much more potential for SW optimizations.
And the power savings alone is worth it unless Maxwell prices drop significantly.

I also wonder about the possible 32 GB barrier. I'm not aware of anyone getting more than 5 980ti's working.
With the 8 GB on the 1080 & 1070 this could be limited to 4. Hopefully this is just a simple driver fix and not something
more fundamental.

there is already one guy on the ethereum forum running 3 x rig with six 1070 each, without problem, so apparently this is not a real issue, or maybe only with windows? he is using linux

On Linux :


So 48 GB seems to be good here too

[Amph]

i still dont have imported the last djm version... but its on my todo list for the 1.8

are you planning to optimize it for pascal?

I replaced just my 970 with a 1080 and the temp in my room has been night and day difference.

The only thing I hate is the price but oh well.

As Epsylon3 said above maybe down the road the cards can be taking advantage of for mining.

The price/performance may not be ideal yet but there should be much more potential for SW optimizations.
And the power savings alone is worth it unless Maxwell prices drop significantly.

I also wonder about the possible 32 GB barrier. I'm not aware of anyone getting more than 5 980ti's working.
With the 8 GB on the 1080 & 1070 this could be limited to 4. Hopefully this is just a simple driver fix and not something
more fundamental.

there is already one guy on the ethereum forum running 3 x rig with six 1070 each, without problem, so apparently this is not a real issue, or maybe only with windows? he is using linux

On Linux :


So 48 GB seems to be good here too

perhaps is exactly windows the problem...again, what algo is that at 40MH for each 1080? lyra2v2?

[antho281]

I replaced just my 970 with a 1080 and the temp in my room has been night and day difference.

The only thing I hate is the price but oh well.

As Epsylon3 said above maybe down the road the cards can be taking advantage of for mining.

The price/performance may not be ideal yet but there should be much more potential for SW optimizations.
And the power savings alone is worth it unless Maxwell prices drop significantly.

I also wonder about the possible 32 GB barrier. I'm not aware of anyone getting more than 5 980ti's working.
With the 8 GB on the 1080 & 1070 this could be limited to 4. Hopefully this is just a simple driver fix and not something
more fundamental.

there is already one guy on the ethereum forum running 3 x rig with six 1070 each, without problem, so apparently this is not a real issue, or maybe only with windows? he is using linux

On Linux :


So 48 GB seems to be good here too

perhaps is exactly windows the problem...again, what algo is that at 40MH for each 1080? lyra2v2?

yeah, this is Lyra2v2 with old version and unoptimized setup.
With a current screen, I'm more about 45 MH / 1080

[crysx]

hi all ...

everything coming along smoothly ...

i will be away for a week or so - and pop in when i can - starting monday ... ill be in the process of setting thefarm - and organizing for it to be connected within the following two weeks ...

this means i will have no internet connectivity for that period of time - and will be doing nothing else but working on thefarm ...

so if there is any contact that is required - please pm or skype me ...

the development is underway for the projects we have - and will continue to do so ...

those who know - understand what is happening with thefarm - and its rebuild ... those who dont - 'watch this space' ...

redevelopment of the miner will be based on ccminer-tpruvot ... which is why im posting here as well ...

a toast to progress ...

#crysx

[tbearhere]

I replaced just my 970 with a 1080 and the temp in my room has been night and day difference.

The only thing I hate is the price but oh well.

As Epsylon3 said above maybe down the road the cards can be taking advantage of for mining.

antonio What wattage does it draw?  Thx