I'm quite sure that running 1 task per card will very much stop producing those errors or invalids. I remember reading it was almost impossible to get AMD RX 4xx/5xx to run succesfully 2x at some point in recent history. That current situation with those errors might not mean that your system specifically had any problems.
I strongly believe that another thing could be the OS. I remember from the past that I've had a hardware with AMD GPU running better 2x when the OS was linux and not Windows. Currently I have two hosts equipped with the same AMD GPU models (different manufacturers though). Both are running 2x. Different computers does mean there's already many variables... but the host with linux is producing almost 100% valids whereas host with Windows is producing some invalids again. Temperature is not the problem with that latter host, but something is causing the difference and this is not the first time with comparable setups.
My RX 580 running 2x in Win7 is producing valid work.
At good speed? 2x on my 560 with W10 was never a stability problem; they just ran ~10x slower for some reason.
I'm quite sure that running 1 task per card will very much stop producing those errors or invalids. I remember reading it was almost impossible to get AMD RX 4xx/5xx to run succesfully 2x at some point in recent history. That current situation with those errors might not mean that your system specifically had any problems.
I strongly believe that another thing could be the OS. I remember from the past that I've had a hardware with AMD GPU running better 2x when the OS was linux and not Windows. Currently I have two hosts equipped with the same AMD GPU models (different manufacturers though). Both are running 2x. Different computers does mean there's already many variables... but the host with linux is producing almost 100% valids whereas host with Windows is producing some invalids again. Temperature is not the problem with that latter host, but something is causing the difference and this is not the first time with comparable setups.
My RX 580 running 2x in Win7 is producing valid work.
At good speed? 2x on my 560 with W10 was never a stability problem; they just ran ~10x slower for some reason.
Yes, I am aware of the slowdown you're referencing. I get that with 3x E@H tasks at once but 2x complete faster than 2 single tasks in serial. Some projects I get it with 2x tasks at once. 650k RAC for that GPU but that's elevated from the short tasks awhile back. Typically close to 500k or so from the single RX580.
With all the "computation errors" that my GPU is producing, will it hurt the project, or will these tasks be resent to others? Will it hurt if I keep running Einstein with whatever valid tasks I can get, and the error tasks will get resent?
... will it hurt the project, or will these tasks be resent to others?
Yes, the tasks you return which are notated as computation errors are sent to others. There is a limit (20 here, last time I looked) but that limit is unlikely to be breached in current conditions unless the task itself is faulty.
There is some harm, in that the servers must spend extra computational resources dealing with your task, and, storing a record of the outcome. For the Gamma-ray pulsar detection work in question here, the project resource to send you a specific task is extremely low in terms of transmitted information. We have had other applications here for which files adding up to quite a bit of transmission were distinct for each individual task.
For the more subtle sorts of errors which are only detected when your results are compared to first one from another host--noticed not to compare--and then compared to a yet another, one could suggest that there is some risk if people deliberately tolerate high error rates that two of them might get to compared to one another, match, and lead to Einstein accepting a false result. So long as your situation creates bad results in the form of labeled computation errors, that risk seems very small.
My personal sense of tidiness and order in the universe is much happier with participants whose machines produce very low error rates, but although I can identify "hurt" from participation as you query, I think it is well overbalanced by the benefit of your contribution to the total computational effort here.
Thanks, that's what I thought. I tried underclocking the RX570 from 1280 mhz to 1120 mhz but still got errors, as was suggested earlier. I have no clue why it's not working. I noticed using MSI Afterburner that the GPU core clock jumps around a lot from 1180-1280 mhz during computing Einstein tasks on the RX 570. My RX 470 stays relatively constant between 1203-1205 mhz. However, on SETI, both GPUs ran awesome at 1280mhz with no invalid results.
Newegg.com has new mining RX470's on sale for $100 each. I'm considering buying another to replace this 570.
I'm not sure exactly what you mean by 'new' :-).
Are these brand new (as in never used previously) but just happened to be 'configured' for mining - eg a special mining VBIOS of some sort. Perhaps 'new' means 'newly refurbished' and the cards have previously been used for mining and have perhaps been traded in for whatever reason. If the latter, just be aware that at the height of the mining boom, some people would have thrown caution to the wind in the search for higher performance. I'm not just talking about overclocking or voltage mods. Apparently there were lots of experimental BIOS mods floating around and a traded in card may have been 'butchered' in some way with unsuitable modifications. This might only become apparent under crunching loads.
Of course, this could just end up being scaremongering on my part but hopefully you would make your own inquiries so that you could at least go into the transaction with open eyes.
With regard to continuing on at Einstein, don't be afraid about current computational errors hurting the project in some way. Historically, something like 20% or more of tasks issued never get successfully returned. The project is designed to cope with this and is grateful for what is returned. The main thing is to keep looking for why your errors are happening.
I notice your host currently shows as having a single RX 570, so I guess you did end up removing a card? Are the errors only at 2x concurrency and if you run just a single task, is everything always OK? I have around 8 hosts with single RX 570/580 cards all running 2x for months at a time without producing comp errors. These cards seem to be nice stable crunching platforms. These were all just GPU upgrades to older machines that previously did CPU crunching only. I'm talking about base systems that date back to 2008-2010 era. If such old stuff can support something like a 570 or 580, there has got to be something unusual that is affecting your system. I'm quite hesitant to blame Windows but that's the main difference - I run Linux, you run Windows.
It only has one DVI-D connector as there barely was any need for monitors. Also fans on these might be different from the regular models... and other minor changes in specs.
It only has one DVI-D connector as there barely was any need for monitors. Also fans on these might be different from the regular models... and other minor changes in specs.
mmonnin wrote:Richie
)
At good speed? 2x on my 560 with W10 was never a stability problem; they just ran ~10x slower for some reason.
DanNeely wrote:mmonnin
)
Yes, I am aware of the slowdown you're referencing. I get that with 3x E@H tasks at once but 2x complete faster than 2 single tasks in serial. Some projects I get it with 2x tasks at once. 650k RAC for that GPU but that's elevated from the short tasks awhile back. Typically close to 500k or so from the single RX580.
DanNeely wrote: At good
)
Did you leave a cpu core free just for the gpu to use? If not try it and see if the slowdown goes away.
I'd like to keep supporting
)
With all the "computation errors" that my GPU is producing, will it hurt the project, or will these tasks be resent to others? Will it hurt if I keep running Einstein with whatever valid tasks I can get, and the error tasks will get resent?
Joshua wrote:... will it hurt
)
Yes, the tasks you return which are notated as computation errors are sent to others. There is a limit (20 here, last time I looked) but that limit is unlikely to be breached in current conditions unless the task itself is faulty.
There is some harm, in that the servers must spend extra computational resources dealing with your task, and, storing a record of the outcome. For the Gamma-ray pulsar detection work in question here, the project resource to send you a specific task is extremely low in terms of transmitted information. We have had other applications here for which files adding up to quite a bit of transmission were distinct for each individual task.
For the more subtle sorts of errors which are only detected when your results are compared to first one from another host--noticed not to compare--and then compared to a yet another, one could suggest that there is some risk if people deliberately tolerate high error rates that two of them might get to compared to one another, match, and lead to Einstein accepting a false result. So long as your situation creates bad results in the form of labeled computation errors, that risk seems very small.
My personal sense of tidiness and order in the universe is much happier with participants whose machines produce very low error rates, but although I can identify "hurt" from participation as you query, I think it is well overbalanced by the benefit of your contribution to the total computational effort here.
Thanks, that's what I
)
Thanks, that's what I thought. I tried underclocking the RX570 from 1280 mhz to 1120 mhz but still got errors, as was suggested earlier. I have no clue why it's not working. I noticed using MSI Afterburner that the GPU core clock jumps around a lot from 1180-1280 mhz during computing Einstein tasks on the RX 570. My RX 470 stays relatively constant between 1203-1205 mhz. However, on SETI, both GPUs ran awesome at 1280mhz with no invalid results.
Newegg.com has new mining
)
Newegg.com has new mining RX470's on sale for $100 each. I'm considering buying another to replace this 570.
Joshua wrote:Newegg.com has
)
I'm not sure exactly what you mean by 'new' :-).
Are these brand new (as in never used previously) but just happened to be 'configured' for mining - eg a special mining VBIOS of some sort. Perhaps 'new' means 'newly refurbished' and the cards have previously been used for mining and have perhaps been traded in for whatever reason. If the latter, just be aware that at the height of the mining boom, some people would have thrown caution to the wind in the search for higher performance. I'm not just talking about overclocking or voltage mods. Apparently there were lots of experimental BIOS mods floating around and a traded in card may have been 'butchered' in some way with unsuitable modifications. This might only become apparent under crunching loads.
Of course, this could just end up being scaremongering on my part but hopefully you would make your own inquiries so that you could at least go into the transaction with open eyes.
With regard to continuing on at Einstein, don't be afraid about current computational errors hurting the project in some way. Historically, something like 20% or more of tasks issued never get successfully returned. The project is designed to cope with this and is grateful for what is returned. The main thing is to keep looking for why your errors are happening.
I notice your host currently shows as having a single RX 570, so I guess you did end up removing a card? Are the errors only at 2x concurrency and if you run just a single task, is everything always OK? I have around 8 hosts with single RX 570/580 cards all running 2x for months at a time without producing comp errors. These cards seem to be nice stable crunching platforms. These were all just GPU upgrades to older machines that previously did CPU crunching only. I'm talking about base systems that date back to 2008-2010 era. If such old stuff can support something like a 570 or 580, there has got to be something unusual that is affecting your system. I'm quite hesitant to blame Windows but that's the main difference - I run Linux, you run Windows.
Have you thought about this earlier post from Dan Neely?
Cheers,
Gary.
I think they are brand new.
)
I think they are brand new. GPU manufacturers made available a little bit simplified products compared to regular RX480's. An example from Sapphire:
https://www.newegg.com/Product/Product.aspx?Item=1FT-000B-000E2
It only has one DVI-D connector as there barely was any need for monitors. Also fans on these might be different from the regular models... and other minor changes in specs.
Richie wrote:I think they are
)
At 2048 stream processors that's pretty good!