Well you know that RAM speed is directly coupled to the Infinity Fabric speed and the overall speed of the processor, right? Faster RAM = Faster Computer.
I saw links to Amazon and Newegg for B-die 3200CL14 RAM for $125 in the past couple of weeks. Fantastic prices for the best overclocking RAM there is for Ryzen.
The wisdom file should work on your 3900X as I generated that file on my 3950X. Same processor basically. But Windows might interpret the cpu capabilities slightly different than what Linux does. So might have slightly different numbers I guess. Try it and see, either it improves the times, makes them worse or does nothing.
Re-installed the 2700 and after some tinkering, I am running about 80% CPU utilization with PBO auto, core boost auto, and Cool'N Quiet off. It is running as high as 3.97 Mhz.
Cpu volt set to 1.25v
9 gpu tasks 0.5 cpu / 0.333 gpu.
9 CPU tasks.
I will re-run the Wisdom utility. I am feeling a bit paranoid about your wisdom file because that was the last thing I changed and ran before the 3900x CPU quit. Will get an RMA started.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
Running a wisdom file can't make a processor crap out. It does no processing of its own. It is simply a list of optimized FFTW routines. It can't make a processor do something it is not allowed to do. It will either speed up, slow down or do nothing for the computation.
I have a box which is only running Gamma Ray #5 CPU tasks. I have also noticed that some tasks finish in 3-4 hours, while others take 11-12 hours to complete. I'm not playing around with custom configs, or wisdom files. This box is a HP DL380 Gen8 with a single 8 core CPU. Someone on the thread mentioned the data sample of the tasks being different which would make sense to me. Anyway, whatever the cause, I'm seeing it too.
Running a wisdom file can't make a processor crap out. It does no processing of its own. It is simply a list of optimized FFTW routines. It can't make a processor do something it is not allowed to do. It will either speed up, slow down or do nothing for the computation.
After I got the 3900x RMA sent off I suddenly had the same problem with the 2700.
I guess the MB was cranky about the slot/expansion card I was using because when I swapped gpus the problem gpu didn't follow. I was able to boot with 2 gpus but not 3.
So I resorted to the 8 port card in the 2nd long slot and suddenly I can boot 3 gpus. Again.
In the process I reverted to 1 cpu thread per gpu thread.
So it looks like I sent back a working 3900x cpu.
I will try out the 0.5 cpu per gpu again tomorrow.
Re-ran my Wisdom file.
Have to see how the 2700 performs.
I couldn't find the B-Die based on what you posted but I have found the same ram I have in the 3950x box for a moderate price. I will probably buy a pair of 8GB sticks after I confirm the 2700 baseline again. I expect Newegg will bounce me out a replacement or return the 3900x cpu in the near future.
Tom M
==edit== Ram ordered.
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
I have a box which is only running Gamma Ray #5 CPU tasks. I have also noticed that some tasks finish in 3-4 hours, while others take 11-12 hours to complete. I'm not playing around with custom configs, or wisdom files. This box is a HP DL380 Gen8 with a single 8 core CPU. Someone on the thread mentioned the data sample of the tasks being different which would make sense to me. Anyway, whatever the cause, I'm seeing it too.
You might get up to a 10% or more speed up if you run the Wisdom utility.
Just fire up a command line and cd into the /programdata/boinc/project/E@H folder and run this: fftwf-wisdom_FGRP5_1.08_windows_intelx86.exe -o FGRPB1wisdom.dat rib67108864
Wait till it comes back to the command prompt.
Any GR #5 tasks that start after that should run (in general) a little faster.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
A couple of days ago, I was getting nothing but the 1045 series of tasks. Those were crunching in 4600-5000 seconds. But those ran out and now I am crunching the 1044 series and those are crunching in 14000-14200 seconds. Big difference just in the different task type and the parameter sets used for each.
But the wisdom file does help in shaving 10% off the non-use of a wisdom file as Tom mentions.
I had some Gamma Ray #5 that were running near 10 hours but it went down to more like 4 odd hours.
What you are seeing is absolutely nothing to do with wisdom files. Changes of this magnitude are also not being caused by a change in CPU. It's just primarily from changes in the data file being analysed.
In this message earlier in this very thread, I showed the big effect on crunch time that a change in the data file can make. I showed actual times for tasks using different data files that were crunched on a single machine with no hardware changes.
Once again, I've had a look at that same machine and can see tasks for different current data files having a similar ratio of 3 to 1 in the crunch time. Tasks that have LATeah1043F or LATeah1044F in their name crunch in around 5.5 hours. Tasks with 1042F or 1045F or 1038F take around 16.5 hours.
I tried to point you towards the real cause for the big changes in an even earlier message, which you apparently either didn't see or chose to ignore. I thought a table of actual results might get the point across, if you hadn't seen the first message.
This will be my last attempt to get you to understand that the single most important variable affecting crunch time is the actual data file being analysed. If you really want to know the effect of the wisdom file or of a different CPU, make sure you test tasks from the same data file with the frequency term being as close as possible. The table given in the first link demonstrates the small change that different frequencies can make.
If you adjust your cache size to get a batch of new tasks all at once, chances are they will most likely be for a single data file and have a very similar frequency component in their name. Then you can do valid comparison tests on those tasks, knowing that whatever you change (CPU or wisdom file) will be solely responsible for any differences you see.
Once again, I've had a look at that same machine and can see tasks for different current data files haIn this message earlier in this very thread, I showed the big effect on crunch time that a change in the data file can make. I showed actual times for tasks using different data files that were crunched on a single machine with no hardware changes.ving a similar ratio of 3 to 1 in the crunch time. Tasks that have LATeah1043F or LATeah1044F in their name crunch in around 5.5 hours. Tasks with 1042F or 1045F or 1038F take around 16.5 hours.
In the Gamma Ray #5 tasks I have currently running, the 1045F tasks are taking 11-12 hours to finish. I have another task 1039F that ran in about 3.5 hours. I suspect the difference between Gary's numbers and mine on the 1045F is caused by the processors used.
At any rate, I believe Gary is correct, in order to compare crunch differences, one must use files from the same family, in other words, compare the results of a 1045F without wisdom, to a 1045F using wisdom.
Well you know that RAM speed
)
Well you know that RAM speed is directly coupled to the Infinity Fabric speed and the overall speed of the processor, right? Faster RAM = Faster Computer.
I saw links to Amazon and Newegg for B-die 3200CL14 RAM for $125 in the past couple of weeks. Fantastic prices for the best overclocking RAM there is for Ryzen.
The wisdom file should work on your 3900X as I generated that file on my 3950X. Same processor basically. But Windows might interpret the cpu capabilities slightly different than what Linux does. So might have slightly different numbers I guess. Try it and see, either it improves the times, makes them worse or does nothing.
The 3900x just crapped out.
)
The 3900x just crapped out. Won't boot.
Re-installed the 2700 and after some tinkering, I am running about 80% CPU utilization with PBO auto, core boost auto, and Cool'N Quiet off. It is running as high as 3.97 Mhz.
Cpu volt set to 1.25v
9 gpu tasks 0.5 cpu / 0.333 gpu.
9 CPU tasks.
I will re-run the Wisdom utility. I am feeling a bit paranoid about your wisdom file because that was the last thing I changed and ran before the 3900x CPU quit. Will get an RMA started.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
Running a wisdom file can't
)
Running a wisdom file can't make a processor crap out. It does no processing of its own. It is simply a list of optimized FFTW routines. It can't make a processor do something it is not allowed to do. It will either speed up, slow down or do nothing for the computation.
http://www.fftw.org/fftw-wisdom.1.html#:~:text=Description,Fourier)%20transforms%20of%20various%20sizes.
I have a box which is only
)
I have a box which is only running Gamma Ray #5 CPU tasks. I have also noticed that some tasks finish in 3-4 hours, while others take 11-12 hours to complete. I'm not playing around with custom configs, or wisdom files. This box is a HP DL380 Gen8 with a single 8 core CPU. Someone on the thread mentioned the data sample of the tasks being different which would make sense to me. Anyway, whatever the cause, I'm seeing it too.
Keith Myers wrote:Running a
)
After I got the 3900x RMA sent off I suddenly had the same problem with the 2700.
I guess the MB was cranky about the slot/expansion card I was using because when I swapped gpus the problem gpu didn't follow. I was able to boot with 2 gpus but not 3.
So I resorted to the 8 port card in the 2nd long slot and suddenly I can boot 3 gpus. Again.
In the process I reverted to 1 cpu thread per gpu thread.
So it looks like I sent back a working 3900x cpu.
I will try out the 0.5 cpu per gpu again tomorrow.
Re-ran my Wisdom file.
Have to see how the 2700 performs.
I couldn't find the B-Die based on what you posted but I have found the same ram I have in the 3950x box for a moderate price. I will probably buy a pair of 8GB sticks after I confirm the 2700 baseline again. I expect Newegg will bounce me out a replacement or return the 3900x cpu in the near future.
Tom M
==edit== Ram ordered.
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
Matt White wrote:I have a
)
You might get up to a 10% or more speed up if you run the Wisdom utility.
Just fire up a command line and cd into the /programdata/boinc/project/E@H folder and run this: fftwf-wisdom_FGRP5_1.08_windows_intelx86.exe -o FGRPB1wisdom.dat rib67108864
Wait till it comes back to the command prompt.
Any GR #5 tasks that start after that should run (in general) a little faster.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor)
A couple of days ago, I was
)
A couple of days ago, I was getting nothing but the 1045 series of tasks. Those were crunching in 4600-5000 seconds. But those ran out and now I am crunching the 1044 series and those are crunching in 14000-14200 seconds. Big difference just in the different task type and the parameter sets used for each.
But the wisdom file does help in shaving 10% off the non-use of a wisdom file as Tom mentions.
That was a stupendous price
)
That was a stupendous price for a weekend at both Amazon and Newegg. Looks like the prices have bumped up about $10. Now $135.
G.SKILL Trident Z Neo (For AMD Ryzen) Series 16GB (2 x 8GB) 288-Pin RGB DDR4 SDRAM DDR4 3600 (PC4 28800) Desktop Memory Model F4-3600C16D-16GTZN
Definitely B-die. These are very popular on the RAM overclocking forums at OCN. Most everybody is able to get them to 3800Mhz MCLK and 1900Mhz FCLK.
Tom M wrote:The Wisdom file
)
What you are seeing is absolutely nothing to do with wisdom files. Changes of this magnitude are also not being caused by a change in CPU. It's just primarily from changes in the data file being analysed.
In this message earlier in this very thread, I showed the big effect on crunch time that a change in the data file can make. I showed actual times for tasks using different data files that were crunched on a single machine with no hardware changes.
Once again, I've had a look at that same machine and can see tasks for different current data files having a similar ratio of 3 to 1 in the crunch time. Tasks that have LATeah1043F or LATeah1044F in their name crunch in around 5.5 hours. Tasks with 1042F or 1045F or 1038F take around 16.5 hours.
I tried to point you towards the real cause for the big changes in an even earlier message, which you apparently either didn't see or chose to ignore. I thought a table of actual results might get the point across, if you hadn't seen the first message.
This will be my last attempt to get you to understand that the single most important variable affecting crunch time is the actual data file being analysed. If you really want to know the effect of the wisdom file or of a different CPU, make sure you test tasks from the same data file with the frequency term being as close as possible. The table given in the first link demonstrates the small change that different frequencies can make.
If you adjust your cache size to get a batch of new tasks all at once, chances are they will most likely be for a single data file and have a very similar frequency component in their name. Then you can do valid comparison tests on those tasks, knowing that whatever you change (CPU or wisdom file) will be solely responsible for any differences you see.
Cheers,
Gary.
Gary Roberts wrote: Once
)
In the Gamma Ray #5 tasks I have currently running, the 1045F tasks are taking 11-12 hours to finish. I have another task 1039F that ran in about 3.5 hours. I suspect the difference between Gary's numbers and mine on the 1045F is caused by the processors used.
At any rate, I believe Gary is correct, in order to compare crunch differences, one must use files from the same family, in other words, compare the results of a 1045F without wisdom, to a 1045F using wisdom.