First off, I'm an embedded software engineer. I don't have much in-depth knowlege of GPUs (If you need info on I2C or SPI busses, I'm your man.)

Up until know, I've been using GPUs scavenged from dead machines around the office, so in generally, I've assumed "higher numbers = faster computing".

But now I think it's time o throw out the old hardware and actually spend my own money on a fast GPU for SETI@Home.

My slowest box has a Quadro K2000 and a GTX 750. It's due.

But I don't really know what to look for. I can't afford to just go out and buy the latest and greatest, so I need to compare different cheaper cards. Should I only look at CUDA cores? How important is clock speed? Is memory size or bandwidth more important?

-- RK

---

https://www.boincstats.com/signature/-1/user/23070/sig.png

Should I only look at CUDA cores? How important is clock speed? Is memory size or bandwidth more important?

-- RK

For this project memory and bandwidth are not as important. The developers of the applications used here have gone to great lengths to make sure that
those issues are not restricting people on this project. With other projects (examples Einstein and GPUGrid) Memory and bandwidth do play a significant role.

So we are very fortunate to have them make sure that those aren't factors when crunching the data.

The compute capacity is probably the biggest factor in deciding which card to get. It's safe to say any card after the nvidia 700 series is going to be ok for this project. Best bang for the buck is probably the 1050ti or the 1060 (preferably the 6GB version) If money is no object then the higher end card will produce more per hour but use excessively more power =$$ to your pocketbook.

Ask questions if you are not sure, plenty of people around willing to help you.


Z

---

most power efficient per credit earned/work done (at a reasonable purchase price) would probably be the RTX 2070 or RTX 2060.

i think the TitanV still holds the record for overall most power efficient, but about as productive as a 2080ti but only using like 150w. but its also a 3000 dollar card.

---

Seti@Home classic workunits: 29,492 CPU time: 134,419 hours

For power/price/performance the GTX 1660Ti would probably be the current sweet spot (roughly on par with the GTX 1080 for output, but uses much less power). The RTX 2060 is on par with the GTX 1080Ti when it comes to output, and uses less power.

And Shaggies graphs are always extremely useful when looking at new video cards.

---

Grant
Darwin NT

I can't afford to just go out and buy the latest and greatest, so I need to compare different cheaper cards.

It looks like some have missed this part of your post.

Look into 2nd hand GTX 750/750Ti/1050/1050Ti/1060 cards (from lowest performance to highest) cards as they'll give you your best bang per buck at somewhere around your budget. ;-)

Cheers.

Best buy has a PNY 1050ti brand new for $120 .

https://www.bestbuy.com/site/pny-nvidia-geforce-gtx-1050-ti-xlr8-gaming-overclocked-edition-4gb-gddr5-pci-express-3-0-graphics-card-black-red/5711724.p?skuId=5711724

Newegg has some refurbished 1060 for $120
https://www.newegg.com/Product/Product.aspx?Item=9SIADFR6V56543&Description=1060&ignorebbr=1&cm_re=1060-_-9SIADFR6V56543-_-Product

---

So a 3GB 1060 versus a 4GB 1050TI? Based on the earlier post that said memory and bandwidth wern't as important with SEIT@Home, should I go for the 1060?

---

https://www.boincstats.com/signature/-1/user/23070/sig.png

The 1060 will process more work than the 1050Ti.

Cheers.

So a 3GB 1060 versus a 4GB 1050TI? Based on the earlier post that said memory and bandwidth wern't as important with SEIT@Home, should I go for the 1060?

The 1060 used to hold the top spot in Shaggies graph for credit efficiency per watt/hour/dollar until the GTX 1660 series usurped it. The 1060 3GB would be a good inexpensive choice.

---

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

A proud member of the OFA (Old Farts Association)

I got used GTX 1060 6GB from ebay for 150 USD. I got 6 GB version for 4k gaming, the textures take more than 3 GB at 4k resolutions.
I run one task on this card and most tasks take 8-9 minutes.
So.
If you don't game at 4k, get used 3 GB version for 120 or so.
Keep in mind, you may need new power supply.

Refurbished 750 with 1 year warranty is 90 USD.
Refurbished 1050 with 1 year warranty is 100 USD.
Refurbished 1060 with 1 year warranty is 130-ish USD on sale, often 150 USD, commonly 180 USD; may need new power supply.
Used 1070 is 250-ish on ebay, may need new power supply.

---

I got used GTX 1060 6GB from ebay for 150 USD. I got 6 GB version for 4k gaming, the textures take more than 3 GB at 4k resolutions.
I run one task on this card and most tasks take 8-9 minutes.
So.
If you don't game at 4k, get used 3 GB version for 120 or so.
Keep in mind, you may need new power supply.

Refurbished 750 with 1 year warranty is 90 USD.
Refurbished 1050 with 1 year warranty is 100 USD.
Refurbished 1060 with 1 year warranty is 130-ish USD on sale, often 150 USD, commonly 180 USD; may need new power supply.
Used 1070 is 250-ish on ebay, may need new power supply.

I'll just point out here that gt40 runs Linux Mint on his rigs so I think that he'd rather like to run the special sauce and not worry about games at 4k, but then again maybe I'm wrong. ;-)

Cheers.

I'll just point out here that gt40 runs Linux Mint on his rigs so I think that he'd rather like to run the special sauce and not worry about games at 4k, but then again maybe I'm wrong. ;-)

Cheers.

The recommendation here on the forum was to run one or two tasks on 1060, so 1 GB would have been enough. Too bad they don't make 1 GB cards anymore.

---

I'll just point out here that gt40 runs Linux Mint on his rigs so I think that he'd rather like to run the special sauce and not worry about games at 4k, but then again maybe I'm wrong. ;-)

You're not wrong. I grew out of video games LONG ago. (I'm almost always up for a good game of
Go, however.)

So in the end, I begged, pleaded, and cooked my wife dinner (the sacrifices we make,) and got a GTX 1660ti with 6GB of memory. We'll see how my numbers change...

(Beware, if you simply pop out the old 660ti and pop in the new 1660ti, it REALLY pisses off the nvidia kernel module and you'll end up with a display stuck at 1024x768, a desktop manager that constantly crashes. It havs something to do with security keys and Secure Boot. The solution was to ssh into the machine, klill the X-server and install the latest nvidia driver by hand.) It might be a better idea to revert to the nouveau driver forst, replace the hardware, then reinstall the nvidia driver.

---

https://www.boincstats.com/signature/-1/user/23070/sig.png

The solution was to ssh into the machine, klill the X-server and install the latest nvidia driver by hand.) It might be a better idea to revert to the nouveau driver forst, replace the hardware, then reinstall the nvidia driver.

That is how I do it in Windows.

---

I can't afford to just go out and buy the latest and greatest, so I need to compare different cheaper cards.

It looks like some have missed this part of your post.

Look into 2nd hand GTX 750/750Ti/1050/1050Ti/1060 cards (from lowest performance to highest) cards as they'll give you your best bang per buck at somewhere around your budget. ;-)

Cheers.

And I really like those used gtx 1060 3GB cards. They bring a good bang for the buck.

Tom

---

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

Hi gt40,

Since you are running on Linux you might be interested in TBar's OneForAll or smthg solution. You could easily threefold your daily output when compared to 'old' CUDA or SoG for NVIDIA on Linux.

The link to the software is ...

--
Petri

---

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

Hi gt40,

Since you are running on Linux you might be interested in TBar's OneForAll or smthg solution. You could easily threefold your daily output when compared to 'old' CUDA or SoG for NVIDIA on Linux.

The link to the software is ... http://www.arkayn.us/lunatics/BOINC.7z

--
Petri

---

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

A proud member of the OFA (Old Farts Association)

Hi gt40,

Since you are running on Linux you might be interested in TBar's OneForAll or smthg solution. You could easily threefold your daily output when compared to 'old' CUDA or SoG for NVIDIA on Linux.

The link to the software is ... http://www.arkayn.us/lunatics/BOINC.7z

--
Petri

Thanks Keith! I kind of knew who was the one to fill my sentence out.

--
Petri

---

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

First off, I'm an embedded software engineer. I don't have much in-depth knowlege of GPUs (If you need info on I2C or SPI busses, I'm your man.)

If you are going to run linux on https://setiathome.berkeley.edu/show_host_detail.php?hostid=8623189 you probably should be running Tbars All-in-one. It should allow that gtx 750 to run one task in 3-5 minutes.

If you run the All-in-one on https://setiathome.berkeley.edu/show_host_detail.php?hostid=8664585
You should get gpu tasks under 3 minutes. And once you tinker with your cpu settings you may be able to get your cpu tasks running near an hour or under.

I don't think you can use the "All in One" on this gpu: https://setiathome.berkeley.edu/show_host_detail.php?hostid=8695718
But I am fairly confident you can get the cpu tasks down near an hour with a little bios tinkering.

This one looks like you could use the "All in One" on it https://setiathome.berkeley.edu/show_host_detail.php?hostid=8623189 and if you want, you could also run the APU too. You would have to manually add the apu to the "all in one" info_config.xml file though. As well as download the AMD app(s).

To recap.
All of your AMD 8c/16t cpus should probably be able to approach a 1 hour or faster cpu task processing speed.
Any system with a gtx 750 or higher gpu should be able to run Tbars All-in-One.
You can run the APU along side a discrete Nvidia gpu if you want. (I have been running a 2400G iGPU along with a gtx 750Ti under Windows 10).

Tom

---

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

Remember as well the difference between 3GB/6GB on the 1060s is 9/10 CU as well.

So with the 3GB models you get 9 CU, 6GB gives you 10.
OR 10% loss over the 10CU.

This may not hold true on every model/brand of the 1060, but it's something to watch for.

---