Thanks for taking the trouble to make all those eminently sensible observations and recommendations. I really appreciate the fact that you did that. Whilst I have absolutely no background in electronics in general and PSU design in particular, I do read quite a bit and do understand (in general terms) all of the points you make.
In 2006 when I decided to buy the Celeron 1300 boxes, I did understand that those would have a quite limited life, perhaps 2yrs if I was lucky. One factor in deciding to make the purchase was that the PSUs had ATX12V 4pin connectors zip tied out of the way since the Gigabyte PIII boards had no place to plug them in. The PSUs were already 'P4 ready' and could supply the majority of power at 12V. I saw this as a positive point in their favour.
The machines lasted well and performed very well so I was able to delay upgrading until around 2008-2009. That gave me time to hunt around for PSUs to handle future multicore CPUs. There was no indication of GPU apps at that stage. I came across the 300W (OEM SeaSonic) units - mentioned in the bit you quoted - and was quite excited by the fact they were SFX form factor and could supply their rated output at 12V. I bought a job lot of 65 of them and these are the ones still in use today.
I had initially 'retired' all of the original 175W PSUs but when GPU apps first appeared, I did a few 'cross loading' experiments to see how they would perform at supplying 12V only. They actually work very well that way. So as I started to add GPUs to existing machines powered by the 300W SeaSonics, I powered the GPU using an extra 175W PSUs sitting in a spare corner of the case and using a dual molex to 6pin PCIe adapter. I ran a couple of '2 PSU' test machines for several months with no issues before deciding that this was going to be viable. It's still viable today, although I've been searching for a suitable single PSU alternative (that doesn't cost an arm and a leg) for years now. Australia is not the US and I'm yet to find any suitable "deal sites to lurk on" :-).
In the meantime, I monitor the fleet quite closely and repair as required - essentially cap replacement and fan lubrication. I don't mind too much - I find it enjoyable to solve problems and it keeps me off the streets while I continue looking for the next bulk purchase that seems 'right' :-). I've been having quite a bit of fun playing with RX 460 GPUs recently. They don't need external power so my old CPU only boxes with the 2009 era dual cores and quads can take a RX 460 and still use the existing 300W PSUs only and become very productive.
I was intending to retire some of these some time ago but I worked out how to use the OpenCL libs from the AMDGPU pro driver with the standard non-pro driver that my distro uses so I've been upgrading with RX 460s rather than retiring. The power use is quite modest, around 135-140W at the wall with a E6300 CPU running 1 FGRPB1 task and a RX 460 crunching 2x FGRPB1G tasks. At idle with the desktop running, the consumption is 67W. I aim to use no more than 70% of a PSUs rated output and am quite happy when it's around 50%, as in this case.
My power bill is actually quite a bit lighter these days despite the extra GPUs that have been added, because of my solar power station :-). It continues to be on track to pay for itself completely in around 3 years or so.
... Strangely enough, all Einstein@home GPU tasks fail on the Windows 10 PC after the Creators Update ...
Tullio,
It's not strange at all. It's just that the Creators Update removes your previous driver and doesn't install a new one suitable for crunching. It's probably just missing the OpenCL compute libraries. I don't know the details since I don't run any Windows machines and have no intention of ever trying to run one :-).
I've seen Archae86 post about this several times, giving instructions on how to resolve the issue. Do a search for "Creators Update" and look for any posts of his. Follow the instructions. Fix it instead of continuing to complain about it :-). Go do it. Right now - then enjoy the rest of your day :-).
As I said, I am getting nVidia drivers through Geforce. Latest I installed was 385.41.
Tullio
My RAC has dropped since Creators Update, even if I have always installed the latest nVidia driver. Now I am running Einstein@home only on my old SUN WS with SuSE Leap 42.2 and GTX 750 Ti.
As I said, I am getting nVidia drivers through Geforce. Latest I installed was 385.41.
You also said in that previous message, "all Einstein@home GPU tasks fail on the Windows 10 PC after the Creators Update". I assumed (since you are still mentioning it) that such tasks continue to fail. Perhaps I am wrong :-). You didn't actually mention whether or not you had finally resolved the issue. You just said that you stopped running Einstein on that machine. If the issue is not resolved, please consider taking a few extra steps.
There is plenty of evidence in previous forum posts that it may be necessary to do a bit of cleanup before reinstalling. I remember comments about 'clean install', 'use DDU', etc. In other words, you may need to do extra work to make sure there are no old remnants left behind that are interfering.
As it turns out, Archae86 has posted again today about this. Also, I just did a quick search for "DDU" (you would probably get the same by searching for "Creators Update") and found this thread which talks about using DDU. And here's another post, that also suggests using DDU and gives a link for where to get it.
I don't follow Windows stuff at all but I believe there are more posts by Archae86 giving even more detail about what to do. They shouldn't be too hard to find if you are still needing to resolve this issue.
Since I am a Windows novice, I use only tools offered by Geforce. SETI@home GPU tasks run perfectly after all nVidia driver updates. After the WOW competition of SETI Germany which ends tomorrow, I shall try running again Eintein@home GPU tasks. Before the Creators update my Einstein RAC was more than 100k. Not it is 26k or less.
Dunno what's faster but I'm seeing 1070's, 1080's and even R280's achieving much better than my Vega 56.
I might be able to squeeze more if I ran more than 1 WU at a time? Not sure how to do that though. I tried changing the project settings for FGRP apps to 0.5 but it hasn't changed anything :?
Since I am a Windows novice, I use only tools offered by Geforce. SETI@home GPU tasks run perfectly after all nVidia driver updates. After the WOW competition of SETI Germany which ends tomorrow, I shall try running again Eintein@home GPU tasks. Before the Creators update my Einstein RAC was more than 100k. Not it is 26k or less.
Tullio
Turn off the Gamma Ray workunits, they are erroring out for you!
...settings for FGRP apps to 0.5 but it hasn't changed anything :?
This guy SMASHES it somehow with an R9 series??...
You need to wait for the settings to propagate, it takes a while. Or try the "update" button.
The guy has two R9 cards, that probably explains why - each card does maybe 150 FGRPB1G jobs a day? The open cl code might not be optimised for the latest AMD cards either.
Hi Dan, Thanks for taking
)
Hi Dan,
Thanks for taking the trouble to make all those eminently sensible observations and recommendations. I really appreciate the fact that you did that. Whilst I have absolutely no background in electronics in general and PSU design in particular, I do read quite a bit and do understand (in general terms) all of the points you make.
In 2006 when I decided to buy the Celeron 1300 boxes, I did understand that those would have a quite limited life, perhaps 2yrs if I was lucky. One factor in deciding to make the purchase was that the PSUs had ATX12V 4pin connectors zip tied out of the way since the Gigabyte PIII boards had no place to plug them in. The PSUs were already 'P4 ready' and could supply the majority of power at 12V. I saw this as a positive point in their favour.
The machines lasted well and performed very well so I was able to delay upgrading until around 2008-2009. That gave me time to hunt around for PSUs to handle future multicore CPUs. There was no indication of GPU apps at that stage. I came across the 300W (OEM SeaSonic) units - mentioned in the bit you quoted - and was quite excited by the fact they were SFX form factor and could supply their rated output at 12V. I bought a job lot of 65 of them and these are the ones still in use today.
I had initially 'retired' all of the original 175W PSUs but when GPU apps first appeared, I did a few 'cross loading' experiments to see how they would perform at supplying 12V only. They actually work very well that way. So as I started to add GPUs to existing machines powered by the 300W SeaSonics, I powered the GPU using an extra 175W PSUs sitting in a spare corner of the case and using a dual molex to 6pin PCIe adapter. I ran a couple of '2 PSU' test machines for several months with no issues before deciding that this was going to be viable. It's still viable today, although I've been searching for a suitable single PSU alternative (that doesn't cost an arm and a leg) for years now. Australia is not the US and I'm yet to find any suitable "deal sites to lurk on" :-).
In the meantime, I monitor the fleet quite closely and repair as required - essentially cap replacement and fan lubrication. I don't mind too much - I find it enjoyable to solve problems and it keeps me off the streets while I continue looking for the next bulk purchase that seems 'right' :-). I've been having quite a bit of fun playing with RX 460 GPUs recently. They don't need external power so my old CPU only boxes with the 2009 era dual cores and quads can take a RX 460 and still use the existing 300W PSUs only and become very productive.
I was intending to retire some of these some time ago but I worked out how to use the OpenCL libs from the AMDGPU pro driver with the standard non-pro driver that my distro uses so I've been upgrading with RX 460s rather than retiring. The power use is quite modest, around 135-140W at the wall with a E6300 CPU running 1 FGRPB1 task and a RX 460 crunching 2x FGRPB1G tasks. At idle with the desktop running, the consumption is 67W. I aim to use no more than 70% of a PSUs rated output and am quite happy when it's around 50%, as in this case.
My power bill is actually quite a bit lighter these days despite the extra GPUs that have been added, because of my solar power station :-). It continues to be on track to pay for itself completely in around 3 years or so.
Cheers,
Gary.
Gary Roberts wrote:tullio
)
As I said, I am getting nVidia drivers through Geforce. Latest I installed was 385.41.
Tullio
My RAC has dropped since Creators Update, even if I have always installed the latest nVidia driver. Now I am running Einstein@home only on my old SUN WS with SuSE Leap 42.2 and GTX 750 Ti.
tullio wrote:As I said, I am
)
You also said in that previous message, "all Einstein@home GPU tasks fail on the Windows 10 PC after the Creators Update". I assumed (since you are still mentioning it) that such tasks continue to fail. Perhaps I am wrong :-). You didn't actually mention whether or not you had finally resolved the issue. You just said that you stopped running Einstein on that machine. If the issue is not resolved, please consider taking a few extra steps.
There is plenty of evidence in previous forum posts that it may be necessary to do a bit of cleanup before reinstalling. I remember comments about 'clean install', 'use DDU', etc. In other words, you may need to do extra work to make sure there are no old remnants left behind that are interfering.
As it turns out, Archae86 has posted again today about this. Also, I just did a quick search for "DDU" (you would probably get the same by searching for "Creators Update") and found this thread which talks about using DDU. And here's another post, that also suggests using DDU and gives a link for where to get it.
I don't follow Windows stuff at all but I believe there are more posts by Archae86 giving even more detail about what to do. They shouldn't be too hard to find if you are still needing to resolve this issue.
Cheers,
Gary.
So, back on topic, what would
)
So, back on topic, what would theoretically be the fastest GPU for this project?
Since I am a Windows novice,
)
Since I am a Windows novice, I use only tools offered by Geforce. SETI@home GPU tasks run perfectly after all nVidia driver updates. After the WOW competition of SETI Germany which ends tomorrow, I shall try running again Eintein@home GPU tasks. Before the Creators update my Einstein RAC was more than 100k. Not it is 26k or less.
Tullio
Someone benched his AMD Vega
)
Someone benched his AMD Vega Frontier Edition. Results are between 370 and 373 seconds per WU. Is there anything faster?
source (german): https://www.planet3dnow.de/vbulletin/threads/428323-Usertest-AMD-Radeon-Vega-Frontier-Edition-Bilder-Benchmarks-Energiemessungen
Dunno what's faster but I'm
)
Dunno what's faster but I'm seeing 1070's, 1080's and even R280's achieving much better than my Vega 56.
I might be able to squeeze more if I ran more than 1 WU at a time? Not sure how to do that though. I tried changing the project settings for FGRP apps to 0.5 but it hasn't changed anything :?
This guy SMASHES it somehow with an R9 series??
https://einsteinathome.org/account/74042/computers
tullio wrote:Since I am a
)
Turn off the Gamma Ray workunits, they are erroring out for you!
Why? They run perfectly on
)
Why? They run perfectly on the old SUN computer with SuSE Linux 42.2 and nVidia driver 384.69.
Tullio
Chooka wrote:...settings
)
You need to wait for the settings to propagate, it takes a while. Or try the "update" button.
The guy has two R9 cards, that probably explains why - each card does maybe 150 FGRPB1G jobs a day? The open cl code might not be optimised for the latest AMD cards either.