Hello fellow crunchers...
I've had for a while two MSI 780's running with 3 WU, and when one GPU gave up a year ago, I was left with 1 780 and rolled back to standard configuration (boinc, no lunatics, 1WU for the GPU). And today the GPU is dead, something blew up in it and smoke came out, and some tears came down.

As it is Christmas and I am in a forgiving mood (still blaming MSI), I am planning to get another card (or two, or three) and below are my options:
Asus ROG STRIX-GTX1070-O8G-GAMING
Asus DUAL-GTX1070-O8G
Asus TURBO-GTX1070-8G
Asus DUAL-GTX1060-6G
Asus DUAL-GTX1060-3G

There's also Zotac1050 Ti's:
Zotac GeForce GTX 1060 Mini 6GB GDDR5 Super Compact
Zotac GeForce GTX 1060 Mini 3GB GDDR5 Super Compact
Zotac GeForce GTX 1050 Ti OC 4GB GDDR5
Zotac GeForce GTX 1050 Ti Mini 4GB GDDR5

I am not using this machine for gaming anymore, it is only a media server and crunching SETI. I am assuming 6G won't have any advantage over 3G?

The big question is, which card will be the best for crunching? 1x1070 or 2x1060 or 3-4x1050Ti's?

---

The 1060 3GB version has 1152 CUDA cores enabled whereas the 6GB version has the full 1280 CUDA cores enabled, so I would do research first, also take note of the graphs in this message thread:
GPU FLOPS: Theory vs Reality

Hello fellow crunchers...
I've had for a while two MSI 780's running with 3 WU, and when one GPU gave up a year ago, I was left with 1 780 and rolled back to standard configuration (boinc, no lunatics, 1WU for the GPU). And today the GPU is dead, something blew up in it and smoke came out, and some tears came down.

As it is Christmas and I am in a forgiving mood (still blaming MSI), I am planning to get another card (or two, or three) and below are my options:
Asus ROG STRIX-GTX1070-O8G-GAMING
Asus DUAL-GTX1070-O8G
Asus TURBO-GTX1070-8G
Asus DUAL-GTX1060-6G
Asus DUAL-GTX1060-3G

There's also Zotac1050 Ti's:
Zotac GeForce GTX 1060 Mini 6GB GDDR5 Super Compact
Zotac GeForce GTX 1060 Mini 3GB GDDR5 Super Compact
Zotac GeForce GTX 1050 Ti OC 4GB GDDR5
Zotac GeForce GTX 1050 Ti Mini 4GB GDDR5

I am not using this machine for gaming anymore, it is only a media server and crunching SETI. I am assuming 6G won't have any advantage over 3G?

The big question is, which card will be the best for crunching? 1x1070 or 2x1060 or 3-4x1050Ti's?

. . Having just bought my second MSI card I hope it is not an issue with the brand :(

. . But you are very wrong about there being no advantage in the 6GB version of the 1060 over the 3GB version. The 6GB has 10 CUs over 9 in the 3Gb version, and correspondingly more CUDA cores(shaders). Figures I see on other systems using them are quite impressive. Sadly in my rig with a very humble Pentium D CPU the numbers are not so good. But having just added a 1050ti card (MSI low profile) to my Core2 Duo unit to replace the previous MSI GT730 I can say the performance is excellent. It is achieving better times than the GTX950 in my i5 unit. For my money, bang for your buck and power usage probably would suggest 2 1060-3GB over a single 1070, especially if the rig is a bit older or has a less powerful CPU. IF you are working with a much older unit (mobo/cpu) I think the 1050ti cards would be a good option too.

Stephen

.

. . Having just bought my second MSI card I hope it is not an issue with the brand :(

. . But you are very wrong about there being no advantage in the 6GB version of the 1060 over the 3GB version. The 6GB has 10 CUs over 9 in the 3Gb version, and correspondingly more CUDA cores(shaders). Figures I see on other systems using them are quite impressive. Sadly in my rig with a very humble Pentium D CPU the numbers are not so good. But having just added a 1050ti card (MSI low profile) to my Core2 Duo unit to replace the previous MSI GT730 I can say the performance is excellent. It is achieving better times than the GTX950 in my i5 unit. For my money, bang for your buck and power usage probably would suggest 2 1060-3GB over a single 1070, especially if the rig is a bit older or has a less powerful CPU. IF you are working with a much older unit (mobo/cpu) I think the 1050ti cards would be a good option too.

Stephen

.

My main concern now is durability, MSI might be good and my experience might be an isolated one but still, I won't go for them again. It would be useful to have the input of others about brands and durability.

I've got a z87 motherboard and a 4770K CPU and it was doing well with 2x780's.

2 1060's might end up with the same price as 1 1080... 3GB, 6GB, 1060 1070 1080... decisions decisions :(

---

I just got 2 ASUS DUAL-GTX1060-O3G's 3 weeks ago and you can see the results so far here after running 2 GTX 660's at a 20K RAC (basically 2 of those 3GB 1060's is the same price here as a GTX 1070).

Cheers.

I'm a happy owner of a MSI 1070 Armor.
GTX 1070 have the best ratio price/performance.
https://setiathome.berkeley.edu/forum_user_posts.php?userid=9158331

It's quite expensive ... so I will buy the second one later.
Don't buy high OC model, they are more expensive for a little extra speed.

---

the 2x1060's seems to have a lower RAC than 1 1070 comparing both your machines. Are you running anything custom?

I am currently in favor of ASUS Turbo GeForce® GTX 1070 as it is cheaper than 2x1060 and the cheapest 1070 I can find here.

---

the 2x1060's seems to have a lower RAC than 1 1070 comparing both your machines. Are you running anything custom?

I am currently in favor of ASUS Turbo GeForce® GTX 1070 as it is cheaper than 2x1060 and the cheapest 1070 I can find here.

Wiggo runs the openCL app on Windows while Laurent runs a (significantly faster) Cuda 8 testing app on Linux.
Also Laurents machine has a 1050ti besides the 1070.

---

the 2x1060's seems to have a lower RAC than 1 1070 comparing both your machines. Are you running anything custom?

I am currently in favor of ASUS Turbo GeForce® GTX 1070 as it is cheaper than 2x1060 and the cheapest 1070 I can find here.

Wiggo runs the openCL app on Windows while Laurent runs a (significantly faster) Cuda 8 testing app on Linux.
Also Laurents machine has a 1050ti besides the 1070.

Yeah you can't compare Laurent's numbers to mine, but you can better compare mine to these 2 Win machines though, i7-6950X w/ 1070 and i7-4770K w/ 1080. ;-)

Cheers.

Thanks guys for your input, I ended up ordering 2x1070's like and self conscious addict would... We'll see how it goes. Will this motherboard x8/x8 thing affect seti? what if I add a 3rd card and it becomes x8/x4/x4?

And considering linux... or a dedicated cruncher...

---

Thanks guys for your input, I ended up ordering 2x1070's like and self conscious addict would... We'll see how it goes. Will this motherboard x8/x8 thing affect seti? what if I add a 3rd card and it becomes x8/x4/x4?

And considering linux... or a dedicated cruncher...

No.

Cheers.

alright I've got a couple of 1070's installed and running for few days now and the usual i7 4770k at 62.5%. I've finished the cuda42's and intel gpu WU and disabled iGPU WU for seti (it is still active serving the monitor but that's it). and each 1070 is running 1 WU.

what caught my eye was the cpu usage and gpu usage. even with 62.5, which leaves me with 3 cores free, the cpu usage is 98-100%. there are 5 cpu WU, 0.407 (by default) per GPU WU, which should leave me with 2 free cores... but still I am seeing 100%... am i missing something here?

GPU load is also around 100%... i remember running 3 WU on my 780's before i could get them close to 100% load.

---

You're running the OpenCL SoG tasks which by default will always use 1 full CPU core to support 1 GPU task. It's just the nature of the beast. Nothing you can do about it. Your 1070's have enough horsepower and memory to support running 2 GPU tasks concurrently. You would then have to let the CPU support both cards with 4 free CPU cores. I believe 2 GPU tasks per card are the sweet spot for the 1070. I would reduce the number of concurrent CPU tasks because the GPU's will do more work on their own in comparison. The OpenCL tasks use a lot more of the GPU than the CUDA tasks beforehand. Unfortunately, they also use more CPU. On my 8 core CPU, I have the max_project_ concurrent setting at 7 which allows 3 CPU tasks along with 4 GPU tasks on my 1070's and that leaves one free CPU core to handle the desktop. I also set affinity to real cores for the CPU tasks with ProcessLasso.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

alright I've got a couple of 1070's installed and running for few days now and the usual i7 4770k at 62.5%. I've finished the cuda42's and intel gpu WU and disabled iGPU WU for seti (it is still active serving the monitor but that's it). and each 1070 is running 1 WU.

what caught my eye was the cpu usage and gpu usage. even with 62.5, which leaves me with 3 cores free, the cpu usage is 98-100%. there are 5 cpu WU, 0.407 (by default) per GPU WU, which should leave me with 2 free cores... but still I am seeing 100%... am i missing something here?

GPU load is also around 100%... i remember running 3 WU on my 780's before i could get them close to 100% load.

You can use the -use_sleep option to (dramatically) reduce CPU usage.

There should be a file named mb_cmdline-8.22_windows_intel__opencl_nvidia_SoG.txt in the ($BOINC/projects/setiathome.berkeley.edu).

Add the following two lines:

-hp
-use_sleep

Now CPU usage will drop significantly. Unfortunately it will also reduce GPU load (and efficiency), but this can be easily solved running multiple GPU tasks at the same time. To do this you will need to create a file named app_config.xml in the same directory.
For me (running two tasks on a GTX-750) it contains the following lines (running at most two MB tasks or one MB task and one Astropulse):

<app_config>
<app>
<name>astropulse_v7</name>
<gpu_versions>
<gpu_usage>0.51</gpu_usage>
<cpu_usage>.1</cpu_usage>
</gpu_versions>
</app>
<app>
<name>setiathome_v8</name>
<gpu_versions>
<gpu_usage>0.49</gpu_usage>
<cpu_usage>.05</cpu_usage>
</gpu_versions>
</app>
</app_config>

If you want to run 3 MB tasks you should change the 0.49 to 0.33.
Changing the command line options does not require a restart (the next task that is started will be using the new options), but after changing the app_config.xml you will need to restart BOINC for changes to take effect.

Tom

You're running the OpenCL SoG tasks which by default will always use 1 full CPU core to support 1 GPU task. It's just the nature of the beast. Nothing you can do about it. Your 1070's have enough horsepower and memory to support running 2 GPU tasks concurrently. You would then have to let the CPU support both cards with 4 free CPU cores. I believe 2 GPU tasks per card are the sweet spot for the 1070. I would reduce the number of concurrent CPU tasks because the GPU's will do more work on their own in comparison. The OpenCL tasks use a lot more of the GPU than the CUDA tasks beforehand. Unfortunately, they also use more CPU. On my 8 core CPU, I have the max_project_ concurrent setting at 7 which allows 3 CPU tasks along with 4 GPU tasks on my 1070's and that leaves one free CPU core to handle the desktop. I also set affinity to real cores for the CPU tasks with ProcessLasso.

I didn't know that, the WU says 0.407 CPUs so I assumed 1 core is more than enough for both 1070's... Anyway, I have the settings on 62.5% so I have 3 cores free. Still, all cores are at 100%, which shouldn't be the case?

---

thanks for the info, I was planning to tweak my GPUs once my RAC reaches its peak. I will do some reading meanwhile on what you've provided.

---