With tlb trashing the dualminer will mine more decred. =)
When you finally release it... Still waiting... DCR still devaluing.
Topic: CCminer(SP-MOD) Modded NVIDIA Maxwell / Pascal kernels. - page 675. (Read 2347641 times)
When you finally release it... Still waiting... DCR still devaluing.
Only on windows 10.
For GM20x and GK110. GM107 and GK10x already have issues in Win7. Who knows what they did with GP10x..
Like cbuchner rightfully commented in the nvidia forum thread about this issue, we should be allowed to get exclusive GPU access with direct physical addressing.
Hellow world 
I repiar my home PC!
Swap burned (all by my wife...) intel Z68 MB to Z170, cpu 3770K to 6700K, psu ocz 1250 to aerocool 1200.
I AM ALIVEEE!
I repiar my home PC!
Swap burned (all by my wife...) intel Z68 MB to Z170, cpu 3770K to 6700K, psu ocz 1250 to aerocool 1200.
I AM ALIVEEE!
Only on windows 10.
For GM20x and GK110. GM107 and GK10x already have issues in Win7. Who knows what they did with GP10x..
Like cbuchner rightfully commented in the nvidia forum thread about this issue, we should be allowed to get exclusive GPU access with direct physical addressing.
As soon as I put that card into the rig its core clock reads 1519 the other 1492 and it will mine for a few seconds then drivers crash not ccminer. I tried every combo I could think of. Thats on any algo. I'm thinking of putting it on this computer and take out 2 750ti move them to the 970 rig. As soon as I can. See if that works.
There is not a single gtx 970 card that have a default boost clock of 1519 or 1492 that I know of. You must be overclocking.
Different algos can be clocked higher.. Decred can mine for weeks at 1600, but quark cannot. My 970 g1 gaming is clocked at around 1390 for quark on the default clocks. The hashrate is around 18.4 with sp-mod private #5
Or your driver is crasy.
With tlb trashing the dualminer will mine more decred. =)
Only on windows 10.
For GM20x and GK110. GM107 and GK10x already have issues in Win7. Who knows what they did with GP10x..Like cbuchner rightfully commented in the nvidia forum thread about this issue, we should be allowed to get exclusive GPU access with direct physical addressing.
Like this?:
https://github.com/sp-hash/ccminer/commit/6313154da0fb3dd154896779f6c584bb5cf9e626
faster on cuda 6.5. Slower in cuda 7.5
I don't know
. But direct physical addressing would simply eliminate the need for a TLB. It would probably be a pain to get working on different types of hardware, but as somebody who seems old enough to know a bit of 68000 or 6502, you'll get the idea More useful would be support for larger TLB page sizes. WDDM 2.0 even has builtin support for it, but I haven't figured out yet to what extent this is supported by Nvidia drivers: https://msdn.microsoft.com/en-us/library/windows/hardware/dn894180(v=vs.85).aspx
As soon as I put that card into the rig its core clock reads 1519 the other 1492 and it will mine for a few seconds then drivers crash not ccminer. I tried every combo I could think of. Thats on any algo. I'm thinking of putting it on this computer and take out 2 750ti move them to the 970 rig. As soon as I can. See if that works.
There is not a single gtx 970 card that have a default boost clock of 1519 or 1492 that I know of. You must be overclocking.
Different algos can be clocked higher.. Decred can mine for weeks at 1600, but quark cannot. My 970 g1 gaming is clocked at around 1390 for quark on the default clocks. The hashrate is around 18.4 with sp-mod private #5
Or your driver is crasy.
As soon as I put that card into the rig its core clock reads 1519 the other 1492 and it will mine for a few seconds then drivers crash not ccminer. I tried every combo I could think of. Thats on any algo. I'm thinking of putting it on this computer and take out 2 750ti move them to the 970 rig. As soon as I can. See if that works.
There is not a single gtx 970 card that have a default boost clock of 1519 or 1492 that I know of. You must be overclocking.
Different algos can be clocked higher.. Decred can mine for weeks at 1600, but quark cannot. My 970 g1 gaming is clocked at around 1390 for quark on the default clocks. The hashrate is around 18.4 with sp-mod private #5
Only on windows 10.
For GM20x and GK110. GM107 and GK10x already have issues in Win7. Who knows what they did with GP10x..Like cbuchner rightfully commented in the nvidia forum thread about this issue, we should be allowed to get exclusive GPU access with direct physical addressing.
Like this?:
https://github.com/sp-hash/ccminer/commit/6313154da0fb3dd154896779f6c584bb5cf9e626
faster on cuda 6.5. Slower in cuda 7.5
Only on windows 10.
For GM20x and GK110. GM107 and GK10x already have issues in Win7. Who knows what they did with GP10x..
Like cbuchner rightfully commented in the nvidia forum thread about this issue, we should be allowed to get exclusive GPU access with direct physical addressing.
Only on windows 10.
so this mean that they are even worse with tlb trashing, because they should at least match the 970, instead they are mining at 60 speed
I wonder if the new gtx 1070/1080 cards will beat the 980ti in the quark algo.
Here is sp-mod #6 private compute 5.2 unreleased:
Here is sp-mod #6 private compute 5.2 unreleased:
Got my hands on 1080
with ccMiner 1.5.80-git SP-MOD fork on Quark at default clokcs at ~170W it does 32000
unfortunetly it needs optimization for neoscrypt as it did only 0.450 with default settings
at ethereum it crashes with Genoil build - says doesnt recognizes the device
but at original Ocl version it gives... 12.5mhs with low tdp (need to double check but it is around 30W!) - needs an optimization as well
Let's assume for a moment that you actually have access to the device, which is not a completely ridiculous claim
The crash kind of makes sense, it may require a CUDA 8 (Compute 6.0) build. Don't have access to it yet.
The 12.5MH/s in OCL smells like TLB trashing. Not as bad as Maxwell, but the problem is it can't be optimized. Well unless Nvidia allows us to play with TLB page size in CUDA 8.
Try installing win7, 8 or Linux and run again.
the consumption is telling me that there is actually room for optimization, it can't be only 30w, pascal is not so efficient compared to maxwell
Low power consumption is in line with my TLB trashign theory. I made this a while ago while digging into the GTX750 TLB issues with ETH mining:
On the left of the blue line you see TLB trashing in action, on the right no trashing (left = DAG > 1GB, right DAG== 1GB). check power consumption (and bus interface load).
I wonder if the new gtx 1070/1080 cards will beat the 980ti in the quark algo.
Here is sp-mod #6 private compute 5.2 unreleased:
Here is sp-mod #6 private compute 5.2 unreleased:
Got my hands on 1080
with ccMiner 1.5.80-git SP-MOD fork on Quark at default clokcs at ~170W it does 32000
unfortunetly it needs optimization for neoscrypt as it did only 0.450 with default settings
at ethereum it crashes with Genoil build - says doesnt recognizes the device
but at original Ocl version it gives... 12.5mhs with low tdp (need to double check but it is around 30W!) - needs an optimization as well
Let's assume for a moment that you actually have access to the device, which is not a completely ridiculous claim
The crash kind of makes sense, it may require a CUDA 8 (Compute 6.0) build. Don't have access to it yet.
The 12.5MH/s in OCL smells like TLB trashing. Not as bad as Maxwell, but the problem is it can't be optimized. Well unless Nvidia allows us to play with TLB page size in CUDA 8.
Try installing win7, 8 or Linux and run again.
the consumption is telling me that there is actually room for optimization, it can't be only 30w, pascal is not so efficient compared to maxwell
I wonder if the new gtx 1070/1080 cards will beat the 980ti in the quark algo.
Here is sp-mod #6 private compute 5.2 unreleased:
Here is sp-mod #6 private compute 5.2 unreleased:
Got my hands on 1080
with ccMiner 1.5.80-git SP-MOD fork on Quark at default clokcs at ~170W it does 32000
unfortunetly it needs optimization for neoscrypt as it did only 0.450 with default settings
at ethereum it crashes with Genoil build - says doesnt recognizes the device
but at original Ocl version it gives... 12.5mhs with low tdp (need to double check but it is around 30W!) - needs an optimization as well
Let's assume for a moment that you actually have access to the device, which is not a completely ridiculous claim
The crash kind of makes sense, it may require a CUDA 8 (Compute 6.0) build. Don't have access to it yet.
The 12.5MH/s in OCL smells like TLB trashing. Not as bad as Maxwell, but the problem is it can't be optimized. Well unless Nvidia allows us to play with TLB page size in CUDA 8.
Try installing win7, 8 or Linux and run again.
why are you leaving sp? you can still receive reward if you release another fork for the new card, maybe do the same thing as djm34 with his neoscrypt fork
collect few bitcoin from many user and then release the miner to the public
collect few bitcoin from many user and then release the miner to the public
but it need optimization you can't compare an optimized card with soemthign not optimized, it was the same when 970 were luanched i think
But who will do the job?
Genesis mining seems to be using 380x cards to mine etherum. Is it the r9 Nitro?
Got my hands on 1080
with ccMiner 1.5.80-git SP-MOD fork on Quark at default clokcs at ~170W it does 32000
unfortunetly it needs optimization for neoscrypt as it did only 0.450 with default settings
at ethereum it crashes with Genoil build - says doesnt recognizes the device
but at original Ocl version it gives... 12.5mhs with low tdp (need to double check but it is around 30W!) - needs an optimization as well
with ccMiner 1.5.80-git SP-MOD fork on Quark at default clokcs at ~170W it does 32000
unfortunetly it needs optimization for neoscrypt as it did only 0.450 with default settings
at ethereum it crashes with Genoil build - says doesnt recognizes the device
but at original Ocl version it gives... 12.5mhs with low tdp (need to double check but it is around 30W!) - needs an optimization as well
Neoscrypt is doing 1MHASH @ 1290 core on my 980ti.
Build etherum with compute 5.0/sm 5.0 in the projectfile/makefile. Then it will work.
but it need optimization you can't compare an optimized card with something not optimized, it was the same when 970 were luanched i think
Jump to: