I wonder how their support in Linux world will be with their Polaris cards?
The AMDGPU beta drivers just released definitely support Polaris10 and Polaris11 (just had a peek in the firmware deb - just released) which covers ubuntu 16.04 and steamos.
I looked at AMD's driver page. they have an extensive PDF file regarding the installation of drivers on Ubuntu and RH. It reminds me of the very early days of the floppy disk shuffle on Windows and Windows apps. Microsoft learned early on that the whole process needed to be automated/streamlined if they were to be successful. They automated the install of apps and their OS. I think that this must also be the way for driver installs on Linux. This should be a joint venture between the driver vendor and the distribution vendor. There is too much room to "miss" a step and end up with a failed install. It sounds like I am complaining but I am not. If the Polaris series has excellent performance with E@H I will commit to buying one and follow the PDF install instructions. The first step is Polaris performance with E@H and if satisfactory getting it to work on my distribution of Ubuntu. As for Nvidia's current offerings they seem to have missed our expectations as a viable "crunch" option.
I wonder how their support in Linux world will be with their Polaris cards?
The AMDGPU beta drivers just released definitely support Polaris10 and Polaris11 (just had a peek in the firmware deb - just released) which covers ubuntu 16.04 and steamos.
I looked at AMD's driver page. they have an extensive PDF file regarding the installation of drivers on Ubuntu and RH.
Which PDF / page are you looking at? I didn't see one on the GPUPRO beta page.
Quote:
As for Nvidia's current offerings they seem to have missed our expectations as a viable "crunch" option.
I think i will wait for a few driver releases to shake down before committing, i recall after the previous GPU hardware - driver performance upgrades came along.
As I'm not game oriented the main thing I was looking for was indications of cooler adequacy, artificial clock rate limiting, etc. As I was starting out worried based on many 1080 reviews finding the FE 1080 was easy to heat up to the point of clock rate substantial limitation, and the 1070 cooler being asserted less capable, I mostly thought his reports to be encouraging. Sadly, I suspect he made a "speak-o" in asserting what memory clock rate he overclocked to, forgetting that with the 1070 he was starting at a base of "8000" rather than the "10,000" for the 1080. While the lesser "non-X" memory seemed not to cripple the 1070 relative to 1080 on his gaming tests, only direct measurement will tell what it means on Einstein BRP6 CUDA55 code.
As the non FE 1080s are taking some time to appear, I currently think I'll by one (only) 1070 FE as soon as I can, hoping some direct testing can help clear up some issues, and believing the power efficiency is likely to be good enough to keep me using it.
If the AMD 480 is as much better a deal as early hopes suggest, I have at least two slots ready to accept them. Unless something puts me off, most likely I'll buy one promptly on availability, and start by subbing it directly into the system running the 1070, allowing direct comparisons for Einstein BRP6 work.
While the lesser "non-X" memory seemed not to cripple the 1070 relative to 1080 on his gaming tests, only direct measurement will tell what it means on Einstein BRP6 CUDA55 code.
That's quite expected: the 1070 has almost the same drop in chip throughput as it has in memory bandwidth, so both are equally balanced and neither is crippled. and for Einstein I bet there'll be about the same performance difference, because the app relies so heavily on memory bandwidth. On my GTX970 it's showing ~70% memory controller utilization.
Personally I wouldn't want to "take one for the team" and pay extra for the FE and encourage nVidia to keep this business model.
The Einstein listing for Dave's machine says it is currently running 368.25. That is the one the Nvidia driver page specifies for use with the 1080. I checked the general customer driver download page several times before "first order day" and up to the day before they did not list a driver as suitable for the 1080. I don't know whether 368.25 is the first version listed for the 1080, and I don't know what driver the various reviewers used, though I had the impression they go a specific driver through a back channel.
Nvidia has reported that they have reproduced annoying fan speed variation behavior, found a driver fix, and will be distributing it in a subsequent driver version.
It appears likely Dave has been trying out variations in number of simultaneous tasks, or possibly other settings. Perhaps he will share some more of his findings with us soon.
This video shows 1 6 pin
)
This video shows 1 6 pin connector. Fast forward to 4:30. Confirms $200 price.
AMD's HUGE Announcement - the Best Value Ever??
https://www.youtube.com/watch?v=eqCo0LTUhsE
Some specs:
http://www.anandtech.com/show/10389/amd-teases-radeon-rx-480-launching-june-29th-for-199
I'm seeing posts now that 4GB
)
I'm seeing posts now that 4GB model is $199; 8GB is $239.
RE: I'm seeing posts now
)
+1
RE: robl wrote: I wonder
)
I looked at AMD's driver page. they have an extensive PDF file regarding the installation of drivers on Ubuntu and RH. It reminds me of the very early days of the floppy disk shuffle on Windows and Windows apps. Microsoft learned early on that the whole process needed to be automated/streamlined if they were to be successful. They automated the install of apps and their OS. I think that this must also be the way for driver installs on Linux. This should be a joint venture between the driver vendor and the distribution vendor. There is too much room to "miss" a step and end up with a failed install. It sounds like I am complaining but I am not. If the Polaris series has excellent performance with E@H I will commit to buying one and follow the PDF install instructions. The first step is Polaris performance with E@H and if satisfactory getting it to work on my distribution of Ubuntu. As for Nvidia's current offerings they seem to have missed our expectations as a viable "crunch" option.
RE: RE: robl wrote: I
)
Which PDF / page are you looking at? I didn't see one on the GPUPRO beta page.
I think i will wait for a few driver releases to shake down before committing, i recall after the previous GPU hardware - driver performance upgrades came along.
JayzTwoCents has posted a
)
JayzTwoCents has posted a video 1070 Founders Edition review.
As I'm not game oriented the main thing I was looking for was indications of cooler adequacy, artificial clock rate limiting, etc. As I was starting out worried based on many 1080 reviews finding the FE 1080 was easy to heat up to the point of clock rate substantial limitation, and the 1070 cooler being asserted less capable, I mostly thought his reports to be encouraging. Sadly, I suspect he made a "speak-o" in asserting what memory clock rate he overclocked to, forgetting that with the 1070 he was starting at a base of "8000" rather than the "10,000" for the 1080. While the lesser "non-X" memory seemed not to cripple the 1070 relative to 1080 on his gaming tests, only direct measurement will tell what it means on Einstein BRP6 CUDA55 code.
As the non FE 1080s are taking some time to appear, I currently think I'll by one (only) 1070 FE as soon as I can, hoping some direct testing can help clear up some issues, and believing the power efficiency is likely to be good enough to keep me using it.
If the AMD 480 is as much better a deal as early hopes suggest, I have at least two slots ready to accept them. Unless something puts me off, most likely I'll buy one promptly on availability, and start by subbing it directly into the system running the 1070, allowing direct comparisons for Einstein BRP6 work.
RE: While the lesser
)
That's quite expected: the 1070 has almost the same drop in chip throughput as it has in memory bandwidth, so both are equally balanced and neither is crippled. and for Einstein I bet there'll be about the same performance difference, because the app relies so heavily on memory bandwidth. On my GTX970 it's showing ~70% memory controller utilization.
Personally I wouldn't want to "take one for the team" and pay extra for the FE and encourage nVidia to keep this business model.
MrS
Scanning for our furry friends since Jan 2002
What Driver are people using
)
What Driver are people using for these 1080?
Zalster wrote:What Driver are
)
The Einstein listing for Dave's machine says it is currently running 368.25. That is the one the Nvidia driver page specifies for use with the 1080. I checked the general customer driver download page several times before "first order day" and up to the day before they did not list a driver as suitable for the 1080. I don't know whether 368.25 is the first version listed for the 1080, and I don't know what driver the various reviewers used, though I had the impression they go a specific driver through a back channel.
Nvidia has reported that they have reproduced annoying fan speed variation behavior, found a driver fix, and will be distributing it in a subsequent driver version.
It appears likely Dave has been trying out variations in number of simultaneous tasks, or possibly other settings. Perhaps he will share some more of his findings with us soon.
I'm wondering if there is an
)
I'm wondering if there is an incompatibility of that driver with older Nvidia GPUs.
ie, it won't allow older GPUs to run alongside of the 1080