Dear friends, does anyone know what the current state of the art is for: the most power efficient SETI@Home crunching hardware platform? Id like to build a scaled rig at home (as a hobby). Keen if anyone has done empirical research for maximum compute productivity-per-watt

Thank you,
Mark

---

Have a look at this thread:
https://setiathome.berkeley.edu/forum_thread.php?id=81962#1891052
It may answer your question

---

Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

This question comes up all the time regularly. We always refer everyone to Shaggie76's often updated graphs.
GPU FLOPS: Theory vs Reality

The GTX 750Ti was always the king until the GTX 1050Ti usurped it.

---

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

A proud member of the OFA (Old Farts Association)

Dear friends, does anyone know what the current state of the art is for: the most power efficient SETI@Home crunching hardware platform? Id like to build a scaled rig at home (as a hobby). Keen if anyone has done empirical research for maximum compute productivity-per-watt

Thank you,
Mark

. . I have heard the term scaling used in many different contexts, may I ask just what it means in this context?

. . PS I would think something like a Ryzen 5 - 2600 with a GTX 1070 running Linux and Special Sauce (Zi3V Cuda90) would give pretty close to the highest productivity with lowest energy demand. But it is a bit of a juggling act trying to pick the winning combo.

Stephen

Stephen

? ?

This question comes up all the time regularly. We always refer everyone to Shaggie76's often updated graphs.
GPU FLOPS: Theory vs Reality

The GTX 750Ti was always the king until the GTX 1050Ti usurped it.

I have both a GTX 750 Ti on a Linux box and a GTX 1050 on a Windows 10 PC. The 750 makes no errors in any app, while the 1050 makes errors in Einstein@home GPU tasks and GPUGRID GPU tasks. But they may depend on Windows 10. I download the nVidia drivers using Geforce.
Tullio

---

Mark,

For Seti at Home, at least, the most processing for amount of power (and bang for the buck) is currently (and likely to stay this way) GPU-based processing not CPU-based processing. The following discussion is based on Windows. Linux has some massive speed improvements if you use the experimental cpu? applications.

For example, a GTX 750 TI may run 3-4 tasks / hour. CPU cores run 1 task in maybe an hour and a half (at best) and also get slower as the data and application change (I remember some of my cpu's taking 7 hours, a couple of years ago with different data sets). So the faster the GPU (say a GTX 1060) the more tasks it manages. I have a 1060 that often processes at 7 odd minutes per task. So that is around 8 tasks an hour.

So the best power efficient Seti processor is going to be a MB with a lot of GPU slots and the most effecient power supply available. You then setup as many of the "best" GPU's based on Shaggies list as you can on that motherboard. You may want to devote up to 1 cpu / gpu but I get lots of processing with 0.5 cpu / gpu.

Note: This will not produce the MOST tasks / hour. It will produce the MOST tasks / kilowatt.

HTH,
Tom

---

A proud member of the OFA (Old Farts Association).

Modern cpus (Intel or AMD) crunch the BLC cpu tasks in about 30 minutes with the latest Linux cpu apps. Arecibo cpu tasks right around an hour. The fastest on Intel uses AVX or AVX2 SIMD. The fastest on AMD is either SSE41 or AVX SIMD.

---

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

A proud member of the OFA (Old Farts Association)

My "nothing special" Intel(R) Core(TM) i5-7400 does them in 45 minutes on Windows 10.

A single TITAN V does 105 tasks/hour. (blc guppi vlar is 34 seconds). It uses typically 134W and on occasion 160W.

---

To overcome Heisenbergs:
"You can't always get what you want / but if you try sometimes you just might find / you get what you need." -- Rolling Stones

I see I should have noted the times I was quoting for my cpu are based on "stock" SETI apps. The Lunatix beta installer with the AVX toggled on runs at least 30% faster on my hardware. (But loses all the pretty graphics :( ).

Tom

---

A proud member of the OFA (Old Farts Association).