It looks like you have dropped a core in the bios. I was suggesting you run 4 cores but use the local option of the Boinc Manager to drop it to 75% of the cpu. This provides that "extra" cpu core whenever it is needed without you having to re-boot the system.

No, I did not drop a core in the bios, I just restricted Boinc to 3 cores. Windows still has the ability to use that spare core. If I want to switch back to 4 cores all I have to do is edit app_config and restart boinc.

The website says you have 3 cores. Since when you drop to 75% of your available cores in BOINC, the Seti website still reports 4 cores, I suspect that something odd is going on.

Tom

Oh, I didn't even notice that. No, literally I just added the ncpus line to app_config. I did not tell Seti to use only 75% of processors in the computing preferences (it is still set to 100%). I don't think this is a problem, it is just a "feature". And considering I'm using the R3 as a dedicated Seti machine, I don't think there is a difference whether I change ncpus or say use xx% of processors. Completely different story if I had several projects running concurrently on that computer, though.

---

Seti@home classic: 1,456 results, 1.613 years CPU time

It looks like you have dropped a core in the bios. I was suggesting you run 4 cores but use the local option of the Boinc Manager to drop it to 75% of the cpu. This provides that "extra" cpu core whenever it is needed without you having to re-boot the system.

No, I did not drop a core in the bios, I just restricted Boinc to 3 cores. Windows still has the ability to use that spare core. If I want to switch back to 4 cores all I have to do is edit app_config and restart boinc.

The website says you have 3 cores. Since when you drop to 75% of your available cores in BOINC, the Seti website still reports 4 cores, I suspect that something odd is going on.

Tom

Oh, I didn't even notice that. No, literally I just added the ncpus line to app_config. I did not tell Seti to use only 75% of processors in the computing preferences (it is still set to 100%). I don't think this is a problem, it is just a "feature". And considering I'm using the R3 as a dedicated Seti machine, I don't think there is a difference whether I change ncpus or say use xx% of processors. Completely different story if I had several projects running concurrently on that computer, though.

Ok, I just hadn't had any experience with BOINC reporting less than the full number of cores except when the core # was restricted in the bios. You might confirm that all your cores are in enabled in the bios on general principles. GP works hand in fist with "Huh?" etc.

Tom

---

A proud member of the OFA (Old Farts Association).

Turning off a core using the BIOS will cut out that core from all use.
Of course if the core was not turned off in the BIOS, but has "died" that could be indicative of a problem with the CPU itself....

---

Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

Turning off a core using the BIOS will cut out that core from all use.
Of course if the core was not turned off in the BIOS, but has "died" that could be indicative of a problem with the CPU itself....

Rumors of a CPU core death are greatly exaggerated. HWinfo showed activity on all four cores.

---

Seti@home classic: 1,456 results, 1.613 years CPU time

I think you actually added ncpus into your CC-config file. That method is a little different than restricting to 75%.

<ncpus>N</ncpus>
Act as if there were N CPUs; e.g. to simulate 2 CPUs on a machine that has only 1. To use the number of available CPUs, set the value to -1 (was 0 which in newer clients really means zero).

---

I think you actually added ncpus into your CC-config file. That method is a little different than restricting to 75%.

<ncpus>N</ncpus>
Act as if there were N CPUs; e.g. to simulate 2 CPUs on a machine that has only 1. To use the number of available CPUs, set the value to -1 (was 0 which in newer clients really means zero).

Yes, I did do that. Look earlier in this thread:

Since yesterday morning, I have set the 2200G to <ncpus>3</ncpus>.

What I didn't realize is that it would report my PC to have only 3 cores, but that makes sense.

---

Seti@home classic: 1,456 results, 1.613 years CPU time

@Bill
--- sorry to take so long posting to your query -- (I couldn't find my "round tuit")

1. 100% CPUs & 100% time
well, 100% of the CPUs I've told boinc that I have. 8 CPUs via cc_config but 10 of the 16 hardware threads are otherwise enabled in the OS.

2. CPU not OC'd (Ryzen 7 1700 with base 3.0 Ghz and boost spec is 3.7 Ghz) I have set the BIOS parameters for 3.3 Ghz and disabled boost; from all that I can understand of the Ryzen boost architecture it can only boost one core to the 3.7 speed and when more cores are actively running the boost speed is less. All the way down to 3.2, I believe, in a fully loaded 16-thread situation. When looking at OS cpu frequency reports I never see anything higher than 3.19. Unsolved puzzle as to why not 3.3 as set in BIOS.

3. OC DRAM? No. Just using the DRAM at mfg. stock speed of 2667 Mhz via the BIOS profile.

4. OC GPU? Just slightly - Graphics clock offset to +50; Memory transfer offset to +200.
Mode set to: Prefer maximum performance in nvidia-settings. I am aware of the Nvidia built-in limits on clock speeds in the P2 state for cards in the compute mode of operation. Willing to fudge a little bit but not brave enough to pursue Petri33's workaround.

How many cores? Why not SMT, etc.? The main Seti production is from the GPU, of course, so I tried a few configurations of "N" CPU work units and 1 core "reserved" for GPU support. I tried 6+1, 7+1, 8+1, and even 9+1. The result was that 7+1 gave the best overall production. In the Ryzen, two threads in one core have to contend for resources so once all 8 cores are in use adding more will begin to affect the responsiveness of the core that is supporting the GPU and adversely affect its production. It is true that more threads increases the CPU production but it doesn't make up for the loss in the GPU.
The Linux OS execution scheduler is a bit fuzzy to me. It appears that all the enabled threads (using that term to refer to cores and SMT threads) are treated identically - i.e. no distinction is made between two threads executing in one core (while another core is idle vs. two threads executing in two cores (with their respective SMT resources unused). The second case being preferred.
Why 10 threads instead of just 8? I have a second instance of boinc-manager running and from 1 a.m. to 6 a.m. it is configured to use one thread for two other boinc projects (CPU apps only).

For reference: GPU app is x41p_V0.97b2 and CPU app is MBv8_8.22r3711_sse41 if/when you look through my tasks.

So, that's my story and I'm sticking to it!

---

The Linux OS execution scheduler is a bit fuzzy to me. It appears that all the enabled threads (using that term to refer to cores and SMT threads) are treated identically - i.e. no distinction is made between two threads executing in one core (while another core is idle vs. two threads executing in two cores (with their respective SMT resources unuse

I too have noticed the fuzzy behavior of the Linux execution scheduler, at least how it pertains to Ryzen. I don't see the same behavior with my Intel i7. But the Ryzens and the Threadripper behave strangely when you try to set priority, nice levels and affinities.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

Thank you to all for providing input!

I have been running an experiment for about 10 days now to see how much I am able to crunch with all four cores and the GPU (that is 3 cores running CPU tasks and (1 CPU + 1 GPU) running a GPU task) versus three cores and the GPU (same as above, but 2 CPU tasks instead of 3). This is for a Ryzen 3 2200G, with no discrete graphics card. I am also running stock applications and Windows 10.

At arbitrary intervals during the day, I have collected information on tasks that were completed and validated via the website. I have filtered out any duplicate tasks, and I have excluded certain tasks (see “inconsistencies” below). I wanted to compare how efficient I was in completing tasks for the credit awarded. Since completion time and credit awarded varied for each task, I valued this as how long it takes for a task to run for 1 unit of credit (seconds/credit). The lower the value, the faster the task is completed, which means more tasks can be crunched in a given period of time.
Averaging for the tasks, my results were as follows:
For 3 cores used: 49.95 seconds/credit for CPU tasks, 30.78 seconds/credit for GPU tasks.
For 4 cores used: 49.65 seconds/credit for CPU tasks, 55.03 seconds/credit for GPU tasks.

As can be seen in the data above, there is a 44% improvement in GPU tasks simply by restricting the APU to 3 cores. However, this does not mean that running one less core provides an overall improvement in RAC. If we attempt to figure out the amount of credit per day, the equation would be:

Average credit per day (total for computer) = [(QTY of CPUs running tasks) x (86,400 seconds per day) / (average CPU seconds/credit)] + [(1 GPU) x (86,400 seconds per day) / (average GPU seconds/credit)]

With the equation and data above, a 2 CPU + (1 CPU + 1 GPU) setup provides about 6,266 credits per day, but a 3 CPU + (1 CPU + 1 GPU) provides 6,791 credits per day.

Conclusion:
As Tom has mentioned before, the GPU for the Ryzen APUs are able to work better when it does not have to share lanes/resources with a CPU core (I’m paraphrasing). This is proven with the lower seconds/credit when either 75% of the CPUs are allowed to crunch tasks or if NCPUS = 3 in the cc_config file. However, the overall output of the Ryzen APU is less due to one less core crunching. Even though the GPU is choked down, more credit is crunched in a day by 8% with all CPU cores crunching.

Inconsistencies:
-I did not clearly delineate between disabling one core in cc_config vs setting computing preferences to 75%. However, based on cursory observations, there was not a significant change between the two.
-I probably should not have updated the Adrenalin driver midway through, but I did. GPU tasks at one point were taking 2+ hours to complete when they had taken 75 minutes in the worst case. I reinstalled the driver and it fixed the problem. I have removed those long tasks from the averaging.
-Short bursts were removed from the data pool as they could provide an advantage to the CPU task average.
-I only had one Astropulse task in the entire pool that was completed. I eliminated it from the pool to keep it all similar.
-GPU tasks consisted of both v8.22 opencl_ati_nocal and v8.22 opencl_ati5_SoG_nocal. I did not make a distinction between the two in my averaging.

Next Steps:
One could conclude that running the Vega 8 GPU is not worth it for the Ryzen 3 2200G. The seconds/credit average for the GPU is worse than the CPU, so why not run just CPU tasks and no GPUs? Although that may be true, there may be ways to improve GPU performance without removing one CPU core from productivity. Command lines could help, as well as overclocking RAM frequencies. I will be experimenting with both in the future to see where any performance can be improved. Anonymous applications will likely help as well, but for the moment I am keeping it simple and excluding those.

---

Seti@home classic: 1,456 results, 1.613 years CPU time

Next Steps:
One could conclude that running the Vega 8 GPU is not worth it for the Ryzen 3 2200G. The seconds/credit average for the GPU is worse than the CPU, so why not run just CPU tasks and no GPUs? Although that may be true, there may be ways to improve GPU performance without removing one CPU core from productivity. Command lines could help, as well as overclocking RAM frequencies. I will be experimenting with both in the future to see where any performance can be improved. Anonymous applications will likely help as well, but for the moment I am keeping it simple and excluding those.

Based on the reviews I have read, speeding up memory should have a positive effect on the iGPU. Maybe even the cpu cores. Besides tinkering with the command lines for the iGPU you can also try "turning up" the iGPU frequency.

I have not tried any of these with all cores running and 1 core dedicated to the gpu.

I suspect that 3 out of 4 cores running with 1 dedicated to the iGPU is likely to be a bit more productive even with all the different tweaks than 4 cores running with the iGPU sharing a core. Effectively you should be able to get at least the equivalent of 5 cpu cores out of the system. (4 cpu at X time, 1 gpu at at least X time).

Tom

---

A proud member of the OFA (Old Farts Association).

Something else that explains why the CPU processing doesn't even come close to the CPU processing of my 2700.

Apparently, the AMD (3)2200G/(5)2400G are running Zen 1, not Zen 1+. And if I am understanding it right, the upgrade will be to Zen 1+ for these not Zen 2. Which is a real shame but these are the "odd man out" cpus.

Tom

---

A proud member of the OFA (Old Farts Association).

Haven't done any exact testing, but I've found that setting % of CPUs to "Reserve" one to feed the GPU really helps with the GPU processing. Now how much it helps likely depends on how good the GPU is. If you turn off one core of an i5 just to feed the inbuilt HD 4600, you are probably losing performance. It might be better to just not use the GPU at all if it takes longer than a CPU core to do the work.

My "new" machine https://setiathome.berkeley.edu/show_host_detail.php?hostid=8717979 is an i5 with HD graphics AND a basic Nvidia 1030 GPU. That seems to work best with 75% of cores in preferences, so one core left over to feed the 2 lower power GPUs properly. The combined improvement for GPUs seems to more than make up for the idle CPU core.

More powerful the GPU becomes, the more critical I think this would become.

Oh, and I generally run my stuff at stock speeds, but 100% CPU time and Cores, unless it's a regular use machine. Then I might cut it back to leave a couple of cores / threads free for other things. More of an issue with a lower powered machine, my old Xeon X3330 starts to stutter a bit on videos if all 4 cores are in use. Likewise the wifes old i3 only runs 50% of cores, don't want to interrupt those Facebook games :D

I propose that since this thread was started with the Amd 2200G in mind it is a good place to continue our public conversation.

Yeah, we have encountered before that your 2400G seems to perform slower than my 2200G, and I don't think we ever got to the bottom of it. Yes, you have SMT available, whereas it isn't available for my processor. Just curious, how fast are you running your ram? Number of DIMMs? And I'm assuming they're in the optimal slots? I don't want to hijack this thread, so perhaps we continue this portion of the conversation somewhere else.

I have recently discovered that no matter what I set my bios at, the Windows task manager doesn't reflect the different speeds.
I am running my iGPU at 1600 Ghz with about 1.2volts.
I have been experimenting with reducing the total # of cores I have running. I had it down to 1 for the gpu and 2 others. Bumped it back up to 4 and I think the iGPU either slowed down or a different task made a difference.

Let me take some notes on where my bios settings are right now. This is an ASROCK B450 Pro+ right now. I am considering moving it to a 350 chip set MB so I can play cpu hop-scotch when the Ryzen 3000's come out.

BRB.
==edit==
Two DIMMs in 4 slot, 2 channel MB. Both are in the A slot for each channel.
--Had to turn my "-period_iterations_num" from 1 to 10 because for the first time, the keyboard/screen was "laggy".---

Ram XMP profile: DDR4-3000 15-17-17-35 1.35volts.
Turned it up earlier today to: DDR4-3133 15-15-15-36 1.41v
---
iGPU 1600Ghz, 1.21volts
---
Cpu volts 1.376
---
Cool'n Quiet - disabled.
Global C-state - disabled.
--
The "SMT" bios toggle doesn't seem to be working. But I just flashed the bios for Ryzen 3000 support so it could be a "bug". :)
===edit===
Tom

---

A proud member of the OFA (Old Farts Association).

Two DIMMs in 4 slot, 2 channel MB. Both are in the A slot for each channel.

!!!!!! :-} That is incorrect for any motherboard. If you have two sticks of RAM to put into a 2 channel motherboard, the preferred slots are A2 / B2.
No wonder that board is underperforming. READ the manual

Right now you have unequal DIMM slot trace lengths to each stick leading to timing errors and low memory clocks.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

Two DIMMs in 4 slot, 2 channel MB. Both are in the A slot for each channel.

!!!!!! :-} That is incorrect for any motherboard. If you have two sticks of RAM to put into a 2 channel motherboard, the preferred slots are A2 / B2.
No wonder that board is underperforming. READ the manual

Right now you have unequal DIMM slot trace lengths to each stick leading to timing errors and low memory clocks.

It gets worse. I have downloaded the manual and the slots are not labeled on the motherboard layout. Nor at first glance were they labeled on the motherboard. I am going to go a get a better flashlight.
BRB.
And with a better flashlight, I can report that I found the slot labels. I had to take the ram out to see the labeling.
And my memory is just bad enough I am not sure if they were in the 1st and 3rd slots or the 2nd and 4th slots.
They are now in the #2/#4 slots.
A2B2
0X0X

Thank you for catching that Keith.

Tom

---

A proud member of the OFA (Old Farts Association).

Common rookie mistake. You fell into that one innocently.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

And disabling the SMT in the bios still doesn't cause the Windows task manager to think I have 4 cores/4 logical cores. It is still showing 4c/8t.
---Edit---
I reset the Bios to its "defaults" on general principles.
--Edit--

Tom

---

A proud member of the OFA (Old Farts Association).

Common rookie mistake. You fell into that one innocently.

Hi Keith,

This is a problem with mobo design engineers. They do not think the way that the rest of the world thinks. To them, placing 2 DIMMs, 1 in A2 and 1 in B2 is logical to them. If it weren't for the manual that we get with a new mobo, everyone would think to put the DIMMs in A1 and A2. That is logical to the rest of us. Well, at least to me anyway. ;) And why do they make it A2 and B2 for 2 DIMMs, why not A1 and B1 instead. Even that would be more logical. ;) I'm sure that there is a good reason for why the engineers do it the way they do. Heck, they could just silkscreen the board to name the slots more logically if there's a logical reason for the 2 slots used. Oh well... :)

Have a great day! :)

Siran

---

CAPT Siran d'Vel'nahr - L L & P _\\//
Winders 11 OS? "What a piece of junk!" - L. Skywalker
"Logic is the cement of our civilization with which we ascend from chaos using reason as our guide." - T'Plana-hath

Has to do with the physical location of the slots with respect to the socket for the motherboard trace lengths for the signal timing. And whether the motherboard uses T-topology or daisy-chain for the memory design.

I guess they could silk screen the slot labels out of order for their physical location. But its been done this way forever, so industry standard and doubt that it will ever change.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

Has to do with the physical location of the slots with respect to the socket for the motherboard trace lengths for the signal timing. And whether the motherboard uses T-topology or daisy-chain for the memory design.

I guess they could silk screen the slot labels out of order for their physical location. But its been done this way forever, so industry standard and doubt that it will ever change.

Hi Keith,

The actual design may be industry standard, but there is at least one mobo manufacturer that does it right...

I used to watch this guy build computers on YouTube. I don't anymore basically because his videos are mostly 2 or more hours long and he has taught me stuff. He always complains about Asus and their engineers when a customer requests that manufacturers board. Most of the time his builds are done with Gigabyte boards. I just looked at a Gigabyte board on Newegg. The RAM slots are silk screened in a more logical manner. The slots are numbered, from the CPU socket out: 4, 2, 3, 1. That's easy to follow. Put 2 sticks in slots 1 and 2. Does that not make more sense than the way Asus does it? But that will not stop me from using Asus boards. I have always used them and always will. :) Asus should follow suit and change their silk screen to be more logical.

Have a great day! :)

Siran

---

CAPT Siran d'Vel'nahr - L L & P _\\//
Winders 11 OS? "What a piece of junk!" - L. Skywalker
"Logic is the cement of our civilization with which we ascend from chaos using reason as our guide." - T'Plana-hath