If it works well at 2X, you might find it interesting to check 3X. I suspect it won't run out of memory.
I agree. It appears the 3GB balloon for Nvidia Gravity Wave gpu may not be present in the Linux versions.
There is a GW ati-beta version (under Linux) available. I need to step through testing it after I get some more 3 task baseline created for GW.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
I agree. It appears the 3GB balloon for Nvidia Gravity Wave gpu may not be present in the Linux versions.
There is a GW ati-beta version (under Linux) available. I need to step through testing it after I get some more 3 task baseline created for GW.
Tom M
you keep saying "Linux versions", implying that you are comparing Nvidia to Linux? seems like these are two totally non related things in terms of GPU use.
do you mean to say that the AMD GW GPU app doesn't every get sent 3+GB tasks like the nvidia cards do? or the "linux version" of the AMD app doesn't where the AMD "Windows version" does?? because the nvidia "linux version" of the app certainly does.
is this "raw" individual task time? or scaled average "per WU" time?
for example, if your "raw" time was 33mins, and you were running 3 concurrently, that's a scaled average of 11mins per WU.
just want to know if you are factorinI am posting the actual time each task took with each # of tasks active on the gpu. The reason I am so happy with them is 1 task raw time X 3 =~ 56 minutes yet when I run the 3 in "parallel" they total to only g in the 2x/3x behavior into your times like this or not.
if your 3 task 18.9 min value listed is raw, that's a scaled average of ~6:20 per task. you mean this right?
Raw times.
I am assuming (and maybe I am confused here) that if running three tasks takes only an additional 2.7 more minutes per task then I am gaining more production by running all 3 tasks at once.
On my Nvidia cards I was going from say 12 minutes per task to 24 minutes per task when I tried running 2 tasks. If I am not confused, on the AMD 5700 cards, by going from 16 odd minutes per task to 17 (or 19) minutes per task means running two (or three) is not a linear scale up like it was with the NVIDIA cards.
Respectfully,
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
I agree. It appears the 3GB balloon for Nvidia Gravity Wave gpu may not be present in the Linux versions.
There is a GW ati-beta version (under Linux) available. I need to step through testing it after I get some more 3 task baseline created for GW.
Tom M
you keep saying "Linux versions", implying that you are comparing Nvidia to Linux? seems like these are two totally non related things in terms of GPU use.
do you mean to say that the AMD GW GPU app doesn't ever get sent 3+GB tasks like the nvidia cards do? or the "linux version" of the AMD app doesn't where the AMD "Windows version" does?? because the nvidia "linux version" of the app certainly does.
I have not yet caught a "computation error" that was caused, apparently, by a lack of memory under Linux. But I have not ever run a 3GB card on E@H (that I am aware of).
Yet.
I have "computation errors" for user aborts or server aborts on ati-beta tasks. But haven't (yet) caught any memory failures.
Respectfully,
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
Eventually (maybe today) I will probably switch back to Gamma Ray only on the R5700 box because that will maximize my RAC for that system over the long run.
My plan is to run mixed or pure Gravity Wave on the RX 570's box. I have 3 RX 560's on order from over seas that got tied up in the international shipping channels. And seem to have stopped to rest at Chicago International Trade Zone.
After my success with the 570's I tried to cancel that international order but it was too late. When they get here I will have to decide if I want to use them or sell them and get more Rx 570's or switch all the way to up to 5700's (money should not get in the way of honest hobbies. Right?;)
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
Eventually (maybe today) I will probably switch back to Gamma Ray only on the R5700 box because that will maximize my RAC for that system over the long run.
My plan is to run mixed or pure Gravity Wave on the RX 570's box. I have 3 RX 560's on order from over seas that got tied up in the international shipping channels. And seem to have stopped to rest at Chicago International Trade Zone.
After my success with the 570's I tried to cancel that international order but it was too late. When they get here I will have to decide if I want to use them or sell them and get more Rx 570's or switch all the way to up to 5700's (money should not get in the way of honest hobbies. Right?;)
Tom M
Just caught the GW tasks ballooning to a hour plus. Apparently the "memory load" parameters on the Linux AmdGpu show the load was 17%. This was very low compared to earlier. So I dropped the GW to two tasks. The memory load resumed higher/variable activity and it looks like the tasks are processing faster again.
The Gamma Ray is still set to 3 tasks.
I have switched to Gamma Ray only but it will take a while to finish the cache of GW off.
--edit--
Confirmed GW gpu tasks@2 per gpu have resumed "normal" processing speeds as previously reported. So have Gamma Ray@3 tasks.
----edit----
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
1. I don’t know where you get the idea that the memory problem is only affecting Linux. It affects the Nvidia cards on Windows too if they have 3GB of memory or less. The problem is how much memory is required vs how much the card physically has. Not what app is being used or what OS it’s on. The only difference seems to be the handling of what happens when you run out. Seems like the nvidia cards crash the task, and AMD cards overflow into system memory or hard disk.
2. as to your slow run time. Did you happen to look at how much memory was being used? I think the “memory” metric you’re looking at might be the memory controller load, not how much space is being used. I have a feeling you filled the memory with 3x 3GB tasks like I warned you about and it overflowed into system memory, slowing it all down.
2. as to your slow run time. Did you happen to look at how much memory was being used? I think the “memory” metric you’re looking at might be the memory controller load, not how much space is being used. I have a feeling you filled the memory with 3x 3GB tasks like I warned you about and it overflowed into system memory, slowing it all down.
I haven't managed to figure out where the memory used metric is. I will study the docs some more. In any case it looks like you are right. GW gpu uses more memory than Gamma Ray. Period
Still really liking the great bang for buck/production!
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
2. as to your slow run time. Did you happen to look at how much memory was being used? I think the “memory” metric you’re looking at might be the memory controller load, not how much space is being used. I have a feeling you filled the memory with 3x 3GB tasks like I warned you about and it overflowed into system memory, slowing it all down.
I haven't managed to figure out where the memory used metric is. I will study the docs some more. In any case it looks like you are right. GW gpu uses more memory than Gamma Ray. Period
Still really liking the great bang for buck/production!
Tom M
I have not been able to locate a amd gpu memory utility. The docs on the amd gpu utilities don't clearly point to how to get it to display memory usage :(
And "radeontop" doesn't seem to see Navi 10 or Vega 10 cards at all.
On the positive side the R5700 continues to munch 3 Gamma Ray/Pulsar Search#1 tasks without stopping, getting too hot etc.
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
archae86 wrote:If it works
)
I agree. It appears the 3GB balloon for Nvidia Gravity Wave gpu may not be present in the Linux versions.
There is a GW ati-beta version (under Linux) available. I need to step through testing it after I get some more 3 task baseline created for GW.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
Tom M wrote:I agree. It
)
you keep saying "Linux versions", implying that you are comparing Nvidia to Linux? seems like these are two totally non related things in terms of GPU use.
do you mean to say that the AMD GW GPU app doesn't every get sent 3+GB tasks like the nvidia cards do? or the "linux version" of the AMD app doesn't where the AMD "Windows version" does?? because the nvidia "linux version" of the app certainly does.
_________________________________________________________________________
Tom M wrote: 1 task ~ 16.2
)
is this "raw" individual task time? or scaled average "per WU" time?
for example, if your "raw" time was 33mins, and you were running 3 concurrently, that's a scaled average of 11mins per WU.
just want to know if you are factoring in the 2x/3x behavior into your times like this or not.
if your 3 task 18.9 min value listed is raw, that's a scaled average of ~6:20 per task. you mean this right?
_________________________________________________________________________
Ian&Steve C. wrote: Tom M
)
Raw times.
I am assuming (and maybe I am confused here) that if running three tasks takes only an additional 2.7 more minutes per task then I am gaining more production by running all 3 tasks at once.
On my Nvidia cards I was going from say 12 minutes per task to 24 minutes per task when I tried running 2 tasks. If I am not confused, on the AMD 5700 cards, by going from 16 odd minutes per task to 17 (or 19) minutes per task means running two (or three) is not a linear scale up like it was with the NVIDIA cards.
Respectfully,
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
Ian&Steve C. wrote: Tom M
)
I have not yet caught a "computation error" that was caused, apparently, by a lack of memory under Linux. But I have not ever run a 3GB card on E@H (that I am aware of).
Yet.
I have "computation errors" for user aborts or server aborts on ati-beta tasks. But haven't (yet) caught any memory failures.
Respectfully,
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
Eventually (maybe today) I
)
Eventually (maybe today) I will probably switch back to Gamma Ray only on the R5700 box because that will maximize my RAC for that system over the long run.
My plan is to run mixed or pure Gravity Wave on the RX 570's box. I have 3 RX 560's on order from over seas that got tied up in the international shipping channels. And seem to have stopped to rest at Chicago International Trade Zone.
After my success with the 570's I tried to cancel that international order but it was too late. When they get here I will have to decide if I want to use them or sell them and get more Rx 570's or switch all the way to up to 5700's (money should not get in the way of honest hobbies. Right?;)
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
Tom M wrote:Eventually
)
Just caught the GW tasks ballooning to a hour plus. Apparently the "memory load" parameters on the Linux AmdGpu show the load was 17%. This was very low compared to earlier. So I dropped the GW to two tasks. The memory load resumed higher/variable activity and it looks like the tasks are processing faster again.
The Gamma Ray is still set to 3 tasks.
I have switched to Gamma Ray only but it will take a while to finish the cache of GW off.
--edit--
Confirmed GW gpu tasks@2 per gpu have resumed "normal" processing speeds as previously reported. So have Gamma Ray@3 tasks.
----edit----
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
1. I don’t know where you get
)
1. I don’t know where you get the idea that the memory problem is only affecting Linux. It affects the Nvidia cards on Windows too if they have 3GB of memory or less. The problem is how much memory is required vs how much the card physically has. Not what app is being used or what OS it’s on. The only difference seems to be the handling of what happens when you run out. Seems like the nvidia cards crash the task, and AMD cards overflow into system memory or hard disk.
2. as to your slow run time. Did you happen to look at how much memory was being used? I think the “memory” metric you’re looking at might be the memory controller load, not how much space is being used. I have a feeling you filled the memory with 3x 3GB tasks like I warned you about and it overflowed into system memory, slowing it all down.
_________________________________________________________________________
Ian&Steve C. wrote: 2. as to
)
I haven't managed to figure out where the memory used metric is. I will study the docs some more. In any case it looks like you are right. GW gpu uses more memory than Gamma Ray. Period
Still really liking the great bang for buck/production!
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
Tom M wrote: Ian&Steve C.
)
I have not been able to locate a amd gpu memory utility. The docs on the amd gpu utilities don't clearly point to how to get it to display memory usage :(
And "radeontop" doesn't seem to see Navi 10 or Vega 10 cards at all.
On the positive side the R5700 continues to munch 3 Gamma Ray/Pulsar Search#1 tasks without stopping, getting too hot etc.
The Monolith (its in a black Cube case)
Tom Miller
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!