I believe many 1GB cards will currently be unable to run these. Some time ago I tried to run earlier test version at Albert with 1GB GT 610. It wasn't able to run those tasks because there was error in the log indicating that a tiny amount more GPU RAM was required. If it would be possible to relax that RAM requirement just a bit I'm sure there would be a galactic squadron of 1GB cards ready to enter.
Yes, i am also successfully running FGRPopencl work on 1 GB card (Radeon 7850).
I even manage to run TWO tasks in parallel on 1 GB and they work fine (without errors), but in this case a huge speed drops - at least 5 times slower compared to single task, so i revert back to 1 task per GPU immediately after this test and run 2 tasks only on 2 Gb cards.
So probable reasons:
1. May be VRAM usage was lowered in final app after tests @Albert
2. Or Linux app version use slightly more VRAM (i am testing on Windows 7 x64)
3. Or some of other programs (including OS itself) use lot of VRAM(>200 Mb), so for OpenCL app available <800 Mb from 1 Gb total.
Have the requirements changed in the past few years?
I have a computer with a OpenCL 1.2 video card, a Radeon HD 5450, and it is not able to get any GPU work units.
Have the requirements changed in the past few years?
I have a computer with a OpenCL 1.2 video card, a Radeon HD 5450, and it is not able to get any GPU work units.
Your GPU has 512MB of memory. That's not enough these days. 1GB is required.
Thank you all for your input.
)
Thank you all for your input. I love this project!
I believe many 1GB cards will
)
I believe many 1GB cards will currently be unable to run these. Some time ago I tried to run earlier test version at Albert with 1GB GT 610. It wasn't able to run those tasks because there was error in the log indicating that a tiny amount more GPU RAM was required. If it would be possible to relax that RAM requirement just a bit I'm sure there would be a galactic squadron of 1GB cards ready to enter.
https://einsteinathome.org/ta
)
https://einsteinathome.org/task/596251454
The stderr doesn't show it any longer unfortunately, but this is a working GT730 with 1GB DDR3. It's slow, but it works...
Yes, i am also successfully
)
Yes, i am also successfully running FGRPopencl work on 1 GB card (Radeon 7850).
I even manage to run TWO tasks in parallel on 1 GB and they work fine (without errors), but in this case a huge speed drops - at least 5 times slower compared to single task, so i revert back to 1 task per GPU immediately after this test and run 2 tasks only on 2 Gb cards.
So probable reasons:
1. May be VRAM usage was lowered in final app after tests @Albert
2. Or Linux app version use slightly more VRAM (i am testing on Windows 7 x64)
3. Or some of other programs (including OS itself) use lot of VRAM(>200 Mb), so for OpenCL app available <800 Mb from 1 Gb total.
I have a evga GT 730 1GB ddr3
)
I have a evga GT 730 1GB ddr3 card that is crunching here one task at a time
its the 384 cuda core model takes about two hours to crunch a FGRP oCL 1.13
using 89% ram 908MB, 99% GPU its a fanless card running at 53c
and constantly uses a full cpu core Q9550 to do it, Debian Linux
and the only reason i have such a `new` card as this is that the GT9500 started making too many errors.
i will use them till they die .......
i have to heat my house with electric anyway so i may as well run it through a`roomheater` first :)
winter only cruncher
Have the requirements changed
)
Have the requirements changed in the past few years?
I have a computer with a OpenCL 1.2 video card, a Radeon HD 5450, and it is not able to get any GPU work units.
kb9skw wrote:Have the
)
Your GPU has 512MB of memory. That's not enough these days. 1GB is required.
https://einsteinathome.org/host/12664762
Last time contacted server:
...
Richie_9 wrote: Your GPU has
)
Thanks for the quick answer. I guess the Vega 11 in the Ryzen is all that I get.
New GW tasks have relatively
)
New GW tasks have relatively low VRAM consumption (< 150 МБ per task for current version of app).
So probably old GPU will be able to run it.
http://science.n-helix.com/20
)
http://science.n-helix.com/2018/01/integer-floats-with-remainder-theory.html
Possibly double float will be worked out in open drivers ?