What you are not understanding is that in every system we have, the issue you see does not happens... which means there is something not working as intended in your system...(but it doesnt imply that something is wrong exclusively in your system, just means that you found a certain behaviour that is not widely spread or commonly reported)
And the data Ive asked to you is about how much time it took to complete 1 task doing only that and how much time it takes to complete 2, 3 and 4 tasks using respectively 2, 3 and 4 cores, allways with the "Use at most xx% of CPU time" set at 100% and doing the whole task, not part under one condition and other part under another condition...
I cant see those times reported in any post here, all your comparisons include taks that were crunched partially under a condition and partially under other condition and the variables changed were the CPU time usage and number of cores, so all the results are hard to compare...
If you did it, no matter if its in this thread or not, and those times still reflect the issue you are seeing, then the next step is to see if there is some service or other app running in your system that can be stopped to see if it is affecting the crunching times... or else it could be that you have the BIOS set to be "green" so it could be downclocking the CPU when it is under heavy load to reduce power consumption... (Ive returned an Asrock mobo due to something like that, it was downclocking the CPU at 1/3 of the nominal clock when the CPU was above 25% of overall load and there was no way to avoid that behaviour)
First of all I want to thank every single person who has spent time reading, analysing and replying my few posts.
I have read all your replies.
Clearly I have a problem concurrently running tasks from this project.
Clearly no-one can describe this problem.
On my PC this project does not work at all efficiently on my PC, as I have explained in excrucial detal.
I respectfully withdraw from this project and have unsubscribed from this thread.
Please don't PM me as your message won't be read.
I have a nearly identical problem. As I was reading OP's posts I kept thinking, "Yes, yes, this is happening on my computer in almost exactly the same way as well!"
I haven't tackled the issue yet, but it has been bothering me a bit. I paid a lot of money to optimize my computer so that I can give as much processing power to BOINC as possible. It's a bit disheartening to see a good chunk of that processing power (that I paid for), essentially leak.
Anyway, I'm just posting to say that I understand OP's frustration here.
My system is an Intel i7-990X with 24GB RAM and 3 ATI Radeon 5770 GPUs.
I have a nearly identical problem. As I was reading OP's posts I kept thinking, "Yes, yes, this is happening on my computer in almost exactly the same way as well!"
I've just had a look through your tasks list and yes, just as with the OP's system, there is something not quite right with the performance of your system. There is no magic bullet here. Nobody is clairvoyant and nobody will be able to close their eyes and imagine the perfect solution. You will get a solution if you are prepared to change and test in an iterative cycle, changing one variable at a time and looking for what things give progressive improvement. The OP didn't think anyone was paying attention to him. The truth is that the OP wasn't prepared to do the hard yards of being very systematic in his testing to identify the real problem. People were indeed trying to help and I'm sure they will try to help you as well.
Quote:
I haven't tackled the issue yet, but it has been bothering me a bit. I paid a lot of money to optimize my computer so that I can give as much processing power to BOINC as possible. It's a bit disheartening to see a good chunk of that processing power (that I paid for), essentially leak.
If you want assistance, you really need to be thorough and complete with all the details of your system and what you have already tried (if anything) to improve things. So start with things like mobo specs - can you run the three PCIe slots as x16, x16, x8? If any slots are less than x8 you will suffer a severe performance penalty.
GPU tasks need quite a bit of CPU support and you are likely to improve performance by deliberately freeing up CPU cores. Are you running GPU tasks singly or are you using the GPU utilisation pref setting to run multiple tasks per GPU? Details, details, details, please. What are you running on your CPU cores and how many are reserved to feed the GPUs?
Another point is that 5700 series GPUs are not exactly high in performance these days. You don't say if you are attempting to run multiple tasks per GPU but I assume not. Most of your tasks are taking around 4-5 ksecs, a few somewhat longer and a few a lot longer - up to 20 ksecs. Also quite a few are ending as 'Validate errors' which to me seems to be an indication of perhaps one GPU being faulty in some way. Alternatively it could just be an indication that you are having cooling issues with three tightly packed GPUs. Do you track GPU temps? Do you have enhanced cooling? Details, details, please.
To get an idea of what you should expect, performance wise, I found host 5363079 in the top hosts list. This host has dual 5700 series GPUs and is taking a pretty regular time per task of around 5.4 ksecs with a CPU time component of close to 900 secs. I don't know if this host is running GPU tasks singly or not. As a guide, I worked out the theoretical single task GPU only RAC for that host and got a figure around 16K. It has 8 CPU cores at least some of which are doing CPU tasks. So the current RAC of 19K for that machine is quite possible for x1 GPU use. Or it could be doing tasks x2 and not have reached a stable RAC yet.
Further inspection reveals host 5883532 which also has dual 5700 series GPUs but a somewhat weaker CPU. GPU tasks are more than 1 ksecs slower so this host might be doing tasks x2. And I've just now found host 5683942 which has a GTX560 as well as a 5700 series GPU. The tasks are definitely being done singly (x1) and for the 5700 series GPU the run time is 3K secs and the CPU component is only 300 secs. So the earlier examples may well have been x2 and possibly x3 (or slow x2) respectively.
These examples show you what you are aiming for. I'm guessing that you are suffering from lack of bandwidth in trying to support 3 GPUs. As a test, why don't you remove one GPU and make sure the two remaining are running single tasks @x16 and see what happens? You should be able to get crunch times around 3 ksecs. You should experiment using preferences to free up CPU cores in order to find optimal running conditions.
Quote:
Anyway, I'm just posting to say that I understand OP's frustration here.
Everybody could understand his frustration. What he didn't understand is that it is extremely hard to solve his problem 'from a distance' if he's not willing to accept that there just might be something wrong at his end that only he can properly diagnose with rigorous and painfully time consuming testing under properly controlled conditions. He was expecting the 'magic bullet' that nobody actually has :-).
Quote:
My system is an Intel i7-990X with 24GB RAM and 3 ATI Radeon 5770 GPUs.
Which we can already see from your host details page. It's all the other little details that are much more important :).
RE: RE: may be there is
)
What you are not understanding is that in every system we have, the issue you see does not happens... which means there is something not working as intended in your system...(but it doesnt imply that something is wrong exclusively in your system, just means that you found a certain behaviour that is not widely spread or commonly reported)
And the data Ive asked to you is about how much time it took to complete 1 task doing only that and how much time it takes to complete 2, 3 and 4 tasks using respectively 2, 3 and 4 cores, allways with the "Use at most xx% of CPU time" set at 100% and doing the whole task, not part under one condition and other part under another condition...
I cant see those times reported in any post here, all your comparisons include taks that were crunched partially under a condition and partially under other condition and the variables changed were the CPU time usage and number of cores, so all the results are hard to compare...
If you did it, no matter if its in this thread or not, and those times still reflect the issue you are seeing, then the next step is to see if there is some service or other app running in your system that can be stopped to see if it is affecting the crunching times... or else it could be that you have the BIOS set to be "green" so it could be downclocking the CPU when it is under heavy load to reduce power consumption... (Ive returned an Asrock mobo due to something like that, it was downclocking the CPU at 1/3 of the nominal clock when the CPU was above 25% of overall load and there was no way to avoid that behaviour)
First of all I want to thank
)
First of all I want to thank every single person who has spent time reading, analysing and replying my few posts.
I have read all your replies.
Clearly I have a problem concurrently running tasks from this project.
Clearly no-one can describe this problem.
On my PC this project does not work at all efficiently on my PC, as I have explained in excrucial detal.
I respectfully withdraw from this project and have unsubscribed from this thread.
Please don't PM me as your message won't be read.
I have a nearly identical
)
I have a nearly identical problem. As I was reading OP's posts I kept thinking, "Yes, yes, this is happening on my computer in almost exactly the same way as well!"
I haven't tackled the issue yet, but it has been bothering me a bit. I paid a lot of money to optimize my computer so that I can give as much processing power to BOINC as possible. It's a bit disheartening to see a good chunk of that processing power (that I paid for), essentially leak.
Anyway, I'm just posting to say that I understand OP's frustration here.
My system is an Intel i7-990X with 24GB RAM and 3 ATI Radeon 5770 GPUs.
RE: I have a nearly
)
I've just had a look through your tasks list and yes, just as with the OP's system, there is something not quite right with the performance of your system. There is no magic bullet here. Nobody is clairvoyant and nobody will be able to close their eyes and imagine the perfect solution. You will get a solution if you are prepared to change and test in an iterative cycle, changing one variable at a time and looking for what things give progressive improvement. The OP didn't think anyone was paying attention to him. The truth is that the OP wasn't prepared to do the hard yards of being very systematic in his testing to identify the real problem. People were indeed trying to help and I'm sure they will try to help you as well.
If you want assistance, you really need to be thorough and complete with all the details of your system and what you have already tried (if anything) to improve things. So start with things like mobo specs - can you run the three PCIe slots as x16, x16, x8? If any slots are less than x8 you will suffer a severe performance penalty.
GPU tasks need quite a bit of CPU support and you are likely to improve performance by deliberately freeing up CPU cores. Are you running GPU tasks singly or are you using the GPU utilisation pref setting to run multiple tasks per GPU? Details, details, details, please. What are you running on your CPU cores and how many are reserved to feed the GPUs?
Another point is that 5700 series GPUs are not exactly high in performance these days. You don't say if you are attempting to run multiple tasks per GPU but I assume not. Most of your tasks are taking around 4-5 ksecs, a few somewhat longer and a few a lot longer - up to 20 ksecs. Also quite a few are ending as 'Validate errors' which to me seems to be an indication of perhaps one GPU being faulty in some way. Alternatively it could just be an indication that you are having cooling issues with three tightly packed GPUs. Do you track GPU temps? Do you have enhanced cooling? Details, details, please.
To get an idea of what you should expect, performance wise, I found host 5363079 in the top hosts list. This host has dual 5700 series GPUs and is taking a pretty regular time per task of around 5.4 ksecs with a CPU time component of close to 900 secs. I don't know if this host is running GPU tasks singly or not. As a guide, I worked out the theoretical single task GPU only RAC for that host and got a figure around 16K. It has 8 CPU cores at least some of which are doing CPU tasks. So the current RAC of 19K for that machine is quite possible for x1 GPU use. Or it could be doing tasks x2 and not have reached a stable RAC yet.
Further inspection reveals host 5883532 which also has dual 5700 series GPUs but a somewhat weaker CPU. GPU tasks are more than 1 ksecs slower so this host might be doing tasks x2. And I've just now found host 5683942 which has a GTX560 as well as a 5700 series GPU. The tasks are definitely being done singly (x1) and for the 5700 series GPU the run time is 3K secs and the CPU component is only 300 secs. So the earlier examples may well have been x2 and possibly x3 (or slow x2) respectively.
These examples show you what you are aiming for. I'm guessing that you are suffering from lack of bandwidth in trying to support 3 GPUs. As a test, why don't you remove one GPU and make sure the two remaining are running single tasks @x16 and see what happens? You should be able to get crunch times around 3 ksecs. You should experiment using preferences to free up CPU cores in order to find optimal running conditions.
Everybody could understand his frustration. What he didn't understand is that it is extremely hard to solve his problem 'from a distance' if he's not willing to accept that there just might be something wrong at his end that only he can properly diagnose with rigorous and painfully time consuming testing under properly controlled conditions. He was expecting the 'magic bullet' that nobody actually has :-).
Which we can already see from your host details page. It's all the other little details that are much more important :).
Cheers,
Gary.