I have done the benchmark with the ethminer PTX branch but for CUDA the miner it's only detecting one graphic card even using --cuda-devices 0 1 2 Is there any known issue about this?
Similar to many, I am having low hash rate on Windows 10 with my 750 Ti's. My question is..
Can I install win 7 on a virtualbox and expect a higher hashrate?
I am not Linux savvy... but would an ubuntu install on a virtualbox give me the hash rates I am looking for? I really don't want to run a dual boot set up . Thanks.
Finally on a dual GPU rig. Fixed multi-gpu mining on two branches. Devices flag now actually works...at least with two GPUs
cudaminer branch: --gpu-devices 0 1 3 should not mine on 2 and 4 cudaminer-frontier: --cuda-devices 0 1 3 should not mine on 2 and 4
Few notes: on frontier branch, -t flag is broken. Just use --cuda-devices instead. cudaminer branch will always say "mining on platform...1st GPU, blah blah. It isn't relevant. The "Using device" lines make more sense.
Got a very weird 750Ti bug again that is on both branches. It benchmarks at full speed (~8.4MH/s), but when actual mining, the speed halves. On the GTX780 that is sitting next to it, got no problems. Perhaps it is the combination of two different GPUs, perhaps Windows, perhaps a bug in the measurement. From what I've seen on earlier reports, Linux users don't have issues with this...
-edit- posted wrong binaries. if you downloaded binaries ethminer-cuda-02092015.zip, download again.
It's been a while since I posted some actual speedups of cudaminer. Found this nice little gem in CUDA7.5, giving a 2% speedup on 750Ti. Other Maxwell cards should benefit as well.
Now in cudaminer-frontier branch. Make sure to compile with CUDA7.5. Attached binary is Compute 5.0 only. 5.2 will follow later.
--edit-- some more lop3 optimization in the lop3 branch. up to 8,8Mh/s now on a stock GTX750Ti reference design.
It's been a while since I posted some actual speedups of cudaminer. Found this nice little gem in CUDA7.5, giving a 2% speedup on 750Ti. Other Maxwell cards should benefit as well.
Now in cudaminer-frontier branch. Make sure to compile with CUDA7.5. Attached binary is Compute 5.0 only. 5.2 will follow later.
--edit-- some more lop3 optimization in the lop3 branch. up to 8,8Mh/s now on a stock GTX750Ti reference design.
Pretty consistent mining speed. 1x 750Ti compiled with CUDA 7.5 on Ubuntu.
@Genoil Oh yes that's for 1x 750Ti. So weird the miner didn't launch and use both cards. Now have to manually specify with --cuda-devices. Isn't it supposed to use all cards by default?
@cryptlet: by default just 1 card will run. @Genoil: I will try on my rig later when I get home. Now, my gtx 750 ti after OC is almost the same hashrate with my gtx 960. I got about the same hashrate on both Windows 7 and Ubuntu 14 with gtx 960.
@Phantom The previous few builds launch both my cards by default. That's why I don't use any additional parameters and got the above screenshot without thinking much.
@cflush: I've never tried 750ti on Windows 10, but I did with Win 7 and 8.1. When I ran just one card, it ran very smooth. Then I added full rig with 6 cards, never go above 28 Mh/s. It took me about 2 days before I installed Ubuntu. In my experience:
GTX 750 ti: runs smooth and can overclock on Ubuntu (Hashrate is higher than in Windows on Ethereum and Quark) GTX 960: runs smooth on Ubuntu and Windows, but can overclock only on Windows (seem no different btw stock and overclock in Ethminer) GTX 970: I will test it tonight, but I guess it will be the same gtx 960.
@Genoil A little more testing and puzzling on 750ti's. All machines using your latest Frontier-lop3 and exact same batch file to run but the number of devices change per rig for the number of cards.
Rig 1 has 6 750ti's Windows 7 64 bit Nvidia driver 353.38 AMD FX-6350
Cards 0,1 and 2 Gpu Loads 100% but Power Loads 30% and lower. Cards 3,4, and 5 Gpu Loads 80-100% but Power Loads 80-92%
The hash rate is about 28 Mh/s so about 4.5 per card which is still low. This machine will solo mine with no problem but as soon as I run it on a pool the cpu cores peg to 100% and locks up rig and I have to manual reboot. Solo mining the cpu cores stays well under 100%.
Rig 2 has 5 750ti's Windows 10 64 bit Nvidia driver 355.82 AMD Phenom II X6
All cards Gpu Loads 95-100% but Power Loads 28% and lower.
I updated to latest driver and the cards went from 1.2 to 2 Mh/s each. i can solo or pool mine with now issue or system lock up and the cpu load never pegs out like Rig 1. Again same miner and bat.
Rig 3 has 3 960's and 1 750ti Windows 7 64 bit Nvidia driver 353.30 Intel i7 4790k
The 750ti stays Gpu Load 98% or higher and Power Loads up to 100%
No issues at all on this rig and even using 2 cores of cpu to solo mine a different coin.
@antonio8: you could easy hit 52 Mh/s (stock clock) or 58 Mh/s (overclock) with 6 x Gtx 750 ti on Ubuntu. Also, you can mine another coins (quark, Lyrav2...) with ccminer. ATM, Ethereum diff is increasing every block. If you are still testing with Windows, I think you will miss Ethereum soon.
@antonio8: you could easy hit 52 Mh/s (stock clock) or 58 Mh/s (overclock) with 6 x Gtx 750 ti on Ubuntu. Also, you can mine another coins (quark, Lyrav2...) with ccminer. ATM, Ethereum diff is increasing every block. If you are still testing with Windows, I think you will miss Ethereum soon.
I know Ethereum is just about over for us small miners but I am just determined to find out what is causing these issues on Windows.
I can't code to save my life but I like to try and help however I can.
@antonio8: I understand what you are doing now. I also test a lot with gtx 750 ti. However, I just use only one rig whenever I like to test. That rig is running dual boot Win/Ubuntu.
I think if I split the 960's and 750ti's like rig 3 maybe it might work correct. I am thinking maybe it is a compute_50 and compute_52 thing, but like I said I am no dev.
@antonio8 thanks a lot. I hope to find some time today to dig into the Windows issues. I'm beginning to think that it has to do with multi threading after all. At least for Win7.
One last thing. I tried to run a second instance of the miner to get both gpu load and power load over 80% but the second instance of the miner crashes immediately.
@Phantom thx for feedback. I am terrible with Linux but I want to give Ubuntu a go. Is there a precompiled cuda ethminer for Ubuntu? Link? I am pool mining on ethpool. Thx for your assistance.. -cflush
I am new to mining and had a quick question. From what I've read some have 6 750 Ti's per machine. Does that mean that in order for my one 750 ti machine to generate any eth I will need to join a pool?
Here is my hashrate it was nice that I got it to work on the first go, which means many props to those that figured all this out so I could get it done on the first go.
@TheFiendishOne you should join a pool. There is a bug with ethminer on windows, 1x 750Ti will get almost 9MHs on linux. See my screenshot several posts above for the 750Ti mining speed. And yes the, dev Genoil is aware of this windows miner issue.
The difficulty is skyhigh right now. Very difficult to solo-mine.
Comments
Is there any known issue about this?
Benchmarks for one single GPU:
Using CUDA -cuda-device 0:
Using device: GeForce GTX 750 Ti(5.0)
Trial 1... 8738133
Trial 2... 8738133
Trial 3... 8738133
Trial 4... 9087658
Trial 5... 8738133
min/mean/max: 8738133/8808038/9087658 H/s
inner mean: 3029219 H/s
Using OpenCL and -t 1:
Trial 1... 8388608
Trial 2... 8388608
Trial 3... 7689557
Trial 4... 8388608
Trial 5... 7689557
min/mean/max: 7689557/8108987/8388608 H/s
inner mean: 2796202 H/s
I dont know why inner mean shows lower values.
Similar to many, I am having low hash rate on Windows 10 with my 750 Ti's. My question is..
Can I install win 7 on a virtualbox and expect a higher hashrate?
I am not Linux savvy... but would an ubuntu install on a virtualbox give me the hash rates I am looking for? I really don't want to run a dual boot set up . Thanks.
-Cflush
cudaminer branch: --gpu-devices 0 1 3 should not mine on 2 and 4
cudaminer-frontier: --cuda-devices 0 1 3 should not mine on 2 and 4
Few notes:
on frontier branch, -t flag is broken. Just use --cuda-devices instead.
cudaminer branch will always say "mining on platform...1st GPU, blah blah. It isn't relevant. The "Using device" lines make more sense.
Got a very weird 750Ti bug again that is on both branches. It benchmarks at full speed (~8.4MH/s), but when actual mining, the speed halves. On the GTX780 that is sitting next to it, got no problems. Perhaps it is the combination of two different GPUs, perhaps Windows, perhaps a bug in the measurement. From what I've seen on earlier reports, Linux users don't have issues with this...
-edit- posted wrong binaries. if you downloaded binaries
ethminer-cuda-02092015.zip, download again.
Now in cudaminer-frontier branch. Make sure to compile with CUDA7.5. Attached binary is Compute 5.0 only. 5.2 will follow later.
--edit-- some more lop3 optimization in the lop3 branch. up to 8,8Mh/s now on a stock GTX750Ti reference design.
@Genoil: I will try on my rig later when I get home. Now, my gtx 750 ti after OC is almost the same hashrate with my gtx 960. I got about the same hashrate on both Windows 7 and Ubuntu 14 with gtx 960.
http://www.etherwall.com/
(Source: https://bitcointalk.org/index.php?topic=1155284.640)
How is the performance of the 750ti on Windows 7 versus Windows 10 using cuda miner?
@Phantom : Been using etherwall for a couple days. Seems to work fine. Very basic with no frills.
Thanks again Genoil for all your hard work.
-Cflush
GTX 750 ti: runs smooth and can overclock on Ubuntu (Hashrate is higher than in Windows on Ethereum and Quark)
GTX 960: runs smooth on Ubuntu and Windows, but can overclock only on Windows (seem no different btw stock and overclock in Ethminer)
GTX 970: I will test it tonight, but I guess it will be the same gtx 960.
A little more testing and puzzling on 750ti's. All machines using your latest Frontier-lop3 and exact same batch file to run but the number of devices change per rig for the number of cards.
Rig 1 has 6 750ti's Windows 7 64 bit Nvidia driver 353.38 AMD FX-6350
Cards 0,1 and 2 Gpu Loads 100% but Power Loads 30% and lower.
Cards 3,4, and 5 Gpu Loads 80-100% but Power Loads 80-92%
The hash rate is about 28 Mh/s so about 4.5 per card which is still low. This machine will solo mine with no problem but as soon as I run it on a pool the cpu cores peg to 100% and locks up rig and I have to manual reboot. Solo mining the cpu cores stays well under 100%.
Rig 2 has 5 750ti's Windows 10 64 bit Nvidia driver 355.82 AMD Phenom II X6
All cards Gpu Loads 95-100% but Power Loads 28% and lower.
I updated to latest driver and the cards went from 1.2 to 2 Mh/s each. i can solo or pool mine with now issue or system lock up and the cpu load never pegs out like Rig 1. Again same miner and bat.
Rig 3 has 3 960's and 1 750ti Windows 7 64 bit Nvidia driver 353.30 Intel i7 4790k
The 750ti stays Gpu Load 98% or higher and Power Loads up to 100%
No issues at all on this rig and even using 2 cores of cpu to solo mine a different coin.
I can't code to save my life but I like to try and help however I can.
I think if I split the 960's and 750ti's like rig 3 maybe it might work correct. I am thinking maybe it is a compute_50 and compute_52 thing, but like I said I am no dev.
Going to try it in the morning.
No problem.
One last thing. I tried to run a second instance of the miner to get both gpu load and power load over 80% but the second instance of the miner crashes immediately.
In Lubuntu 15.04 is aroun 28 Mh/s.
Any advice or link about how to OC my 750Ti's in ubuntu??
-cflush
Here is my hashrate it was nice that I got it to work on the first go, which means many props to those that figured all this out so I could get it done on the first go.
Windows 7 x64 GTX 750 Ti
The difficulty is skyhigh right now. Very difficult to solo-mine.