AMD A-2 thru A-10 and new AMD cpu/gpu integrated chips

Message boards : Number crunching : AMD A-2 thru A-10 and new AMD cpu/gpu integrated chips
Message board moderation

To post messages, you must log in.

1 · 2 · 3 · 4 . . . 7 · Next

AuthorMessage
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 3041
Credit: 131,702,280
RAC: 758,645
United States
Message 1957308 - Posted: 26 Sep 2018, 11:38:43 UTC

This thread is being started up because there is a need to collect in one place discussions about the specific hardware issues when running the A-2 thru A-10 Amd cpu/gpu systems. AMD has and is releasing a new generation of the same concept but using the Zen/Vega architecture. Many of the issues appear to apply to this new generation.
I will find or others will find older threads and post the URL's to them here so that we can re-read my fumbling attempts to solve some of the problems.

I still own an A4 and an A6.

Tom
I will stop procrastinating tomorrow.
\\// Live Long & Prosper (starting tomorrow ;)
ID: 1957308 · Report as offensive     Reply Quote
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 3041
Credit: 131,702,280
RAC: 758,645
United States
Message 1957379 - Posted: 26 Sep 2018, 22:20:38 UTC

Here is 1 old thread: [url]https://setiathome.berkeley.edu/forum_thread.php?id=75264
[/url]

There are 40+ AMD Ryzen 3 2200G with Radeon Vega Graphics [Family 23 Model 17 Stepping 0] crunching on Seti.

There are 80+ AMD Ryzen 5 2400G with Radeon Vega Graphics [Family 23 Model 17 Stepping 0] crunching on Seti.

AMD A10-6800K APU with Radeon(tm) HD Graphics [Family 21 Model 19 Stepping 1] 60+ * 26 distinct listings

AMD A8-6600K APU with Radeon(tm) HD Graphics [Family 21 Model 19 Stepping 1] 60+ * 27 distinct listings

AMD A6-7400K Radeon R5, 6 Compute Cores 2C+4G [Family 21 Model 48 Stepping 1] 20+ * 23 distinct listings

And so on.

So it is clear that AMD-based CPU/gpu systems have a clear presence here.

I am ignoring all the AMD netbooks out there with E-350's etc which are also integrated CPU/GPU's.

I have seen that AMD has announced some new low-end zen/vega integrated chips especially for low-end laptops.

So the presenting issue is "how to get more production out of your integrated cpu/gpu system?
I will stop procrastinating tomorrow.
\\// Live Long & Prosper (starting tomorrow ;)
ID: 1957379 · Report as offensive     Reply Quote
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 3041
Credit: 131,702,280
RAC: 758,645
United States
Message 1957381 - Posted: 26 Sep 2018, 22:31:19 UTC - in response to Message 1957379.  

So the presenting issue is "how to get more production out of your integrated cpu/gpu system?


Examples that may work include.

A better CPU/gpu cooler so the integrated chip won't throttle down due to getting too hot.

Try reducing the number of CPU cores that are crunching at the same time the gpu is, to reduce the load on the internal bus.

The A10, A6 and apparently the 2200G will slow down the gpu processing when the cpus are fully engaged and vice versa. My impression is if the gpu is fully engaged it slows down the cpus.

One thing you can try is idling one of the cpus. This reduces the pressure on the bus and appears to allow 3 out of 4 cpus to run at full speed and the gpu to run at full speed. You can use the local configuration in the Seti Manager to try this out.

Or you can use the project_max_concurrent parameter in the app_config.xml file in the Setiathome data directory that is hidden within the hidden "ProgramData" directory.

In theory, if the software app is adequately written to use gpu parallel processing and the gpu has sufficient parallel processing available, the gpu should always process tasks faster than the cpus do.

Tom
I will stop procrastinating tomorrow.
\\// Live Long & Prosper (starting tomorrow ;)
ID: 1957381 · Report as offensive     Reply Quote
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 3041
Credit: 131,702,280
RAC: 758,645
United States
Message 1957444 - Posted: 27 Sep 2018, 11:22:56 UTC

@Kissagogo27 mentioned this research which I had lost track of.

http://lunatics.kwsn.info/index.php?topic=1735.0

It talks about loading in the A10 series APU.
I will stop procrastinating tomorrow.
\\// Live Long & Prosper (starting tomorrow ;)
ID: 1957444 · Report as offensive     Reply Quote
monospice

Send message
Joined: 20 Sep 18
Posts: 30
Credit: 1,133,943
RAC: 2,451
United States
Message 1957537 - Posted: 28 Sep 2018, 0:22:02 UTC

Based on my experience and research so far with an A8-7650K, power consumption and heat are the major bottlenecks in these APUs.

AMD seemingly designed these architectures with the expectation that computing tasks will need a greater share of CPU or GPU resources at one time, but not the maximum of both at the same time. This allows AMD to produce chips with smaller power and cooling requirements that fit the constraints of third-party motherboard manufacturing more easily. The motherboard's VRM/PPM that supplies the processor with a variable amount of power needs to support the full range of processor states from idle to 100% utilization, and the addition of a GPU greatly extends that range.

To facilitate this, APU designs cap the minimum and maximum power requirements to a reasonable range for the MB manufacturers. This results in relatively high--and less efficient--minimum clock speeds (a whopping 1.9 GHz on my 7650K), and, on the upper end, the APU reduces CPU frequency when the GPU is under full load to avoid exceeding the maximum power specification. This can support most general computing needs just fine, but presents disappointing performance when trying to achieve maximum production with a full load of CPU+GPU Boinc tasks.

The effect is especially obvious on Tom's A4-5000 mobile APU with a TDP of 15 W that seems to throttle the CPU to 22% to redirect enough power to the fully-loaded GPU. An APU's TDP seems to provide a rough prediction about how much it may throttle the CPU to compensate for the GPU's power consumption. Tom's A10-5700 at 65 W throttles to 60-70%. By contrast the A8-7650K with a 95 W TDP throttles to no less than 75% of the CPU's maximum clock speed. Its newer architecture may also contribute to its ability to better balance the GPU load.

I tried the suggestion to limit the number of concurrent CPU tasks to eliminate the CPU throttling. On a Linux system, even with just one CPU task/core, the CPU still maxed-out at 75% while loading the GPU. This may result from the OS balancing the scheduling of the process across all four cores or an iffy driver. I'll retry the experiment next time I boot into Windows on this box.

One other approach that I'm curios to try: I wonder whether undervolting the CPU saves enough of the power budget to allow the CPU to run at a higher clock speed without compromising the performance of the GPU. From what I can tell, the CPU and GPU take voltage independently, so if the APU triggers CPU throttling based on a power consumption threshold alone, and not a load-based threshold, the reduction of CPU power consumption and heat production may increase the envelope in which it runs at full speed. Unfortunately, my MB doesn't allow me to change CPU parameters, so I'll need to play with AMD Overdrive next time I'm in Windows. I'm interested to know about anyone's experience undervolting an APU.

I remember reading somewhere that APUs control this special throttling by putting the the CPU into a particular P-state, so a program capable of overwriting the frequency in this P-state entry may eliminate the throttling if the motherboard can sustain power and cooling at full load.
ID: 1957537 · Report as offensive     Reply Quote
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 3041
Credit: 131,702,280
RAC: 758,645
United States
Message 1957700 - Posted: 29 Sep 2018, 0:02:42 UTC

I am currently running an A4-5000 http://setiathome.berkeley.edu/show_host_detail.php?hostid=8558045 as part of my attempt to collect some current useful information about AMD's APU line.

It has been slow going because the cpu's seem to be taking around 5-6 hours per task. The good news is the gpu is taking down near 3 1/2 hours. I commend the "tasks" link on the website for this computer if you are interested in the "results" as sparse as they are.

I am currently running the following command for all three of the gpu tasks I appear to be getting.

-spike_fft_thresh 2048 -tune 1 2 1 16 -period_iteration_num 10 -cpu_lock -tt 1500 -high_perf


Since I installed this yesterday I am uncertain how well it is working. I am experiencing intermittent screen lag. That I was not experiencing before.
I am currently running 3 out of 4 cpu cores. And I am using 1/3 of a cpu to drive the gpu.

This leaves me with 2 2/3's cpu's devoted to crunching Seti tasks. BoincManager shows three cpu tasks and one gpu task. GPU-Z is firmly convinced my gpu is pegged at 100% mostly.

Tom
I will stop procrastinating tomorrow.
\\// Live Long & Prosper (starting tomorrow ;)
ID: 1957700 · Report as offensive     Reply Quote
monospice

Send message
Joined: 20 Sep 18
Posts: 30
Credit: 1,133,943
RAC: 2,451
United States
Message 1957726 - Posted: 29 Sep 2018, 3:28:21 UTC
Last modified: 29 Sep 2018, 3:34:37 UTC

I found some time to test the theories in my previous post. It appears that APUs do throttle the CPU by switching it into a higher P-state when the GPU is fully-loaded, and that overclocking these P-states allows the CPU to run at its maximum clock speed in conjunction with the GPU.

On my A8-7560K, the CPU moves up two P-states with the GPU at full load which causes it to throttle to 75%. I adjusted the two P-states above the processor's standard max (non-turbo) state to bump the frequency up and slightly decrease the voltage (to reduce heat). The CPU and GPU now both run at 100% capacity. Here's a table of the current configuration for the 7650K:

Pstate Status CpuFid CpuDid CpuVid CpuMult CpuFreq CpuVolt IddVal IddDiv CpuCurr CpuPower
     0      1     22      0     16  19.00x 3800MHz  1450mV    167     10  16.70A   24.22W
     1      1     22      0     16  19.00x 3800MHz  1450mV    167     10  16.70A   24.22W
     2      1     22      0     16  19.00x 3800MHz  1450mV    167     10  16.70A   24.22W
     3      1     17      0     30  16.50x 3300MHz  1362mV    175     10  17.50A   23.84W
     4      1     17      0     44  16.50x 3300MHz  1275mV    145     10  14.50A   18.49W
     5      1     17      0     44  16.50x 3300MHz  1275mV    125     10  12.50A   15.94W


The first three entries in the table are the processor's turbo P-states. The fourth entry (#3) is the processor's standard maximum, and I adjusted the last two as described above so that they match the maximum frequency with reduced voltage. I used this tool (Linux) to perfrom the adjustment which worked without a hitch on this Kaveri (Steamroller/Bulldozer) APU. My reading suggests that AmdMsrTweaker may work for Windows users, but I did not try it. [[ Obligatory warning: be careful using these tools--they can damage hardware if used improperly! ]]

On my particular system, I also needed to adjust the process priority (nice) of the GPU tasks to 4 (from 10) before the OS delivered enough CPU time to drive the GPU at 100%. As expected, heat output dramatically increases with this configuration despite the reduced voltage.

The last column in the table provides an interesting perspective into power consumption/heat production--it shows the per-core wattage. Although the last two rows show incorrect values (the tool doesn't seem to update these values after changing frequency and voltage parameters), the CpuPower value for the standard max frequency on this chip (3.3 GHz, P-state #3) suggests why APUs throttle the CPU while fully-loading the GPU: 23.84 watts × 4 cores = 95.36 watts, the TDP specified for this chip. The integrated GPU adds power and heat dissipation requirements that exceed this threshold, so APUs throttle the CPU to stay within this limit.

My system seems to handle the extra load just fine so far, but lower-TDP/mobile systems may become unstable after changing P-state values. On these systems, it may still be possible to partially boost P-state performance to some greater percentage of the maximum speed.
ID: 1957726 · Report as offensive     Reply Quote
monospice

Send message
Joined: 20 Sep 18
Posts: 30
Credit: 1,133,943
RAC: 2,451
United States
Message 1957729 - Posted: 29 Sep 2018, 3:46:12 UTC - in response to Message 1957700.  

I am currently running the following command for all three of the gpu tasks I appear to be getting.

-spike_fft_thresh 2048 -tune 1 2 1 16 -period_iteration_num 10 -cpu_lock -tt 1500 -high_perf

Tom, please forgive my ignorance...is there a reference for these command-line options? I can't seem to find on that describes what they do.

Since I installed this yesterday I am uncertain how well it is working. I am experiencing intermittent screen lag. That I was not experiencing before.
I am currently running 3 out of 4 cpu cores. And I am using 1/3 of a cpu to drive the gpu.

The GPU handles compositing for Windows' desktop environment and drives the attached display. I imagine that the screen lag results from the Boinc tasks hogging the GPU too long for these to perform well...which sounds like a good thing in this case--it seems to indicate that the tasks are processing as quickly as possible!
ID: 1957729 · Report as offensive     Reply Quote
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 3041
Credit: 131,702,280
RAC: 758,645
United States
Message 1957735 - Posted: 29 Sep 2018, 4:58:17 UTC - in response to Message 1957729.  

I am currently running the following command for all three of the gpu tasks I appear to be getting.

-spike_fft_thresh 2048 -tune 1 2 1 16 -period_iteration_num 10 -cpu_lock -tt 1500 -high_perf

Tom, please forgive my ignorance...is there a reference for these command-line options? I can't seem to find on that describes what they do.

Since I installed this yesterday I am uncertain how well it is working. I am experiencing intermittent screen lag. That I was not experiencing before.
I am currently running 3 out of 4 cpu cores. And I am using 1/3 of a cpu to drive the gpu.

The GPU handles compositing for Windows' desktop environment and drives the attached display. I imagine that the screen lag results from the Boinc tasks hogging the GPU too long for these to perform well...which sounds like a good thing in this case--it seems to indicate that the tasks are processing as quickly as possible!


In Windows, there are text files in the hidden ProgramData\BOINC\projects\Setiathome folder. You are looking for any text file that says "ati ReadMe".

Tom
I will stop procrastinating tomorrow.
\\// Live Long & Prosper (starting tomorrow ;)
ID: 1957735 · Report as offensive     Reply Quote
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 3041
Credit: 131,702,280
RAC: 758,645
United States
Message 1958105 - Posted: 2 Oct 2018, 0:34:18 UTC

Someone in another thread referred me to "AnandTech.com" as having really robust, well documented articles and reviews.

Here is the 3rd review on the APU's that caused me to start this thread:
https://www.anandtech.com/show/13041/amd-ryzen-5-2400g-and-ryzen-3-2200g-integrated-graphics-frequency-scaling

The previous article I read was a systematic testing of ram speed on both APU's. While I haven't read the first article it was referenced as testing the cpu.

One of the conclusions is you can overclock the cpu or the gpu but you can't overclock the cpu and the gpu..

Tom
I will stop procrastinating tomorrow.
\\// Live Long & Prosper (starting tomorrow ;)
ID: 1958105 · Report as offensive     Reply Quote
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 3041
Credit: 131,702,280
RAC: 758,645
United States
Message 1958107 - Posted: 2 Oct 2018, 0:49:17 UTC

This is the other reason besides a PM about years old postings about my A-10 5700.

https://setiathome.berkeley.edu/forum_thread.php?id=83371

While the crashing that started the thread apparently self corrected. It looked like the CPU's were out producing the gpu.

And as of the moment the highest Gflops# is coming off the Seti 8.08(alt) [35+ Gflops] and the gpu is running about 27 Gflops.

The system is here: https://setiathome.berkeley.edu/show_host_detail.php?hostid=8535523

There are at least 30 other CPU/gpu's of this model on the CPU list. So several someone's are happy crunching along.

I sure wish I could come up with a good idea that James could try that would allow the gpu to run at 100% and most of the CPU's run at 100%. I am not sure if he has tried the "run 75%" of your CPU's trick yet. He has reported that is what it is currently doing, but I don't know if that is how it settled out or how he set the parameters.

Tom
I will stop procrastinating tomorrow.
\\// Live Long & Prosper (starting tomorrow ;)
ID: 1958107 · Report as offensive     Reply Quote
monospice

Send message
Joined: 20 Sep 18
Posts: 30
Credit: 1,133,943
RAC: 2,451
United States
Message 1958122 - Posted: 2 Oct 2018, 4:32:38 UTC - in response to Message 1958107.  

...besides a PM about years old postings about my A-10 5700.
I sure am glad that you posted it all those years ago...


I sure wish I could come up with a good idea that James could try that would allow the gpu to run at 100% and most of the CPU's run at 100%. I am not sure if he has tried the "run 75%" of your CPU's trick yet. He has reported that is what it is currently doing, but I don't know if that is how it settled out or how he set the parameters.

Tom, I'm using the process_priority_special directive in my cc_config.xml to keep the GPU maxed-out:

<cc_config>
  ...
  <options>
    <process_priority_special>3</process_priority_special>
  </options>
</cc_config>

A value of 3 sets the process priority for GPU tasks to "above normal, " which seems to give them as much CPU time as they need and keeps the other CPU tasks from competing with it. No issues or GUI lag to report yet.

Aside from this, overwriting the processor's P-states (with strong cooling) as I discovered above appears to be the only way to eliminate all throttling because of how APU designs intentionally throttle the CPU to squeeze as much bang as possible into the heat allowance for the chip.
ID: 1958122 · Report as offensive     Reply Quote
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 3041
Credit: 131,702,280
RAC: 758,645
United States
Message 1958141 - Posted: 2 Oct 2018, 11:27:41 UTC - in response to Message 1958122.  
Last modified: 2 Oct 2018, 11:29:33 UTC

...besides a PM about years old postings about my A-10 5700.
I sure am glad that you posted it all those years ago...


I sure wish I could come up with a good idea that James could try that would allow the gpu to run at 100% and most of the CPU's run at 100%. I am not sure if he has tried the "run 75%" of your CPU's trick yet. He has reported that is what it is currently doing, but I don't know if that is how it settled out or how he set the parameters.

Tom, I'm using the process_priority_special directive in my cc_config.xml to keep the GPU maxed-out:

<cc_config>
  ...
  <options>
    <process_priority_special>3</process_priority_special>
  </options>
</cc_config>

A value of 3 sets the process priority for GPU tasks to "above normal, " which seems to give them as much CPU time as they need and keeps the other CPU tasks from competing with it. No issues or GUI lag to report yet.

Aside from this, overwriting the processor's P-states (with strong cooling) as I discovered above appears to be the only way to eliminate all throttling because of how APU designs intentionally throttle the CPU to squeeze as much bang as possible into the heat allowance for the chip.


That is VERY interesting. I may hold off on suspending my A4-5000's Seti participation to see how that works "naked" and then with 3 cores. Hmmmmm…..

Tom
I will stop procrastinating tomorrow.
\\// Live Long & Prosper (starting tomorrow ;)
ID: 1958141 · Report as offensive     Reply Quote
monospice

Send message
Joined: 20 Sep 18
Posts: 30
Credit: 1,133,943
RAC: 2,451
United States
Message 1958205 - Posted: 3 Oct 2018, 1:22:05 UTC

I experimented a bit with my Intel-based integrated-GPU laptops. The same behavior that we see with AMD's APUs seem to occur for any device with integrated graphics. Both platforms stuff more compute power into the CPU+GPU package than the chip is designed to run at once and defer to the system to distribute the limited power budget to one component or the other as needed in the moment. As with APUs, the Intel devices cannot run a full load on the CPU and GPU at once. The system throttles the frequency of the CPU and iGPU when the combined load approaches the chip's TDP.

The effect is much more obvious on a laptop with a lower TDP. Just as with Tom's mobile A4-5000 APU, the Intel chips heavily scaled back the performance of the CPU while any major task executed on the GPU. Where AMD's APUs seem to favor the GPU, the Intel chips appear to prefer balance in the distribution of power between the CPU and the GPU. Despite everything I tried, I couldn't get the GPUs to run full speed continuously unless the CPU load was small enough for the power consumption of the whole system to fall below the chip's TDP. In fact, on my Surface tablet at 8 W TDP, just one CPU process interferes with the GPU performance. On the other laptop, a Dell at 15 W, I managed to bring the clock speed of both components back up to around 95% by under-volting the CPU enough to deliver sufficient power to CPU and GPU at once.

Based on these observations, it seems clear that we can't make the same performance recommendations to mobile APU users that we can for those on a desktop, or even for mobile users with different APU models. My desktop APU happily runs full-speed with the P-state tweak described above, but laptops probably don't have power and cooling systems capable of supporting the increased load. For these computers, limiting the number of CPU tasks as others suggested may be the strongest advice. Power budget is probably the biggest limiting factor for all integrated-GPU packages discussed here.
ID: 1958205 · Report as offensive     Reply Quote
monospice

Send message
Joined: 20 Sep 18
Posts: 30
Credit: 1,133,943
RAC: 2,451
United States
Message 1958206 - Posted: 3 Oct 2018, 1:27:04 UTC - in response to Message 1958141.  

Tom, I'm using the process_priority_special directive in my cc_config.xml to keep the GPU maxed-out...
That is VERY interesting. I may hold off on suspending my A4-5000's Seti participation to see how that works "naked" and then with 3 cores. Hmmmmm…..

Tom


By the way, process_priority_special doesn't seem to work for Seti GPU tasks on my Windows machines as it does in Linux. The Seti tasks always start with a "below normal" priority. The directive does work for Einstein@Home GPU tasks, though. Maybe it's a bug...
ID: 1958206 · Report as offensive     Reply Quote
Profile tullio
Volunteer tester

Send message
Joined: 9 Apr 04
Posts: 7510
Credit: 2,495,680
RAC: 4,412
Italy
Message 1958226 - Posted: 3 Oct 2018, 7:33:56 UTC
Last modified: 3 Oct 2018, 7:35:57 UTC

I have a Windows 10 HP desktop with an A10-6700 running up to 4.18 GHz out of its 3.70 maximum. It has also a GTX 1050 Ti GPU board running SETI@home and Einstein@home GPU tasks. GPUGRID GPU tasks seem to overheat it (not overclocked) and they crash. The SETI@home screensaver, on CPU tasks, with the BREAKTHROUGH LISTEN stamp seems to cause reboots but no crashes of the tasks, which complete and validate.
There is a funny thing about the A10-6700: although sold as a 4 cores CPU, the Windows 10 Task manager says it has two cores and 4 logical processors, so only two BOINC tasks can run simultaneously.
Tullio
ID: 1958226 · Report as offensive     Reply Quote
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 3041
Credit: 131,702,280
RAC: 758,645
United States
Message 1958242 - Posted: 3 Oct 2018, 11:01:44 UTC - in response to Message 1958206.  

Tom, I'm using the process_priority_special directive in my cc_config.xml to keep the GPU maxed-out...
That is VERY interesting. I may hold off on suspending my A4-5000's Seti participation to see how that works "naked" and then with 3 cores. Hmmmmm…..

Tom


By the way, process_priority_special doesn't seem to work for Seti GPU tasks on my Windows machines as it does in Linux. The Seti tasks always start with a "below normal" priority. The directive does work for Einstein@Home GPU tasks, though. Maybe it's a bug...



Rats. The A4-5000 is a windows machine and I have no plans to run Linux on it.

I do have another APU out on loan. Its a little bit higher AMD. Need to get it started.

Tom
I will stop procrastinating tomorrow.
\\// Live Long & Prosper (starting tomorrow ;)
ID: 1958242 · Report as offensive     Reply Quote
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 3041
Credit: 131,702,280
RAC: 758,645
United States
Message 1958293 - Posted: 3 Oct 2018, 22:00:35 UTC - in response to Message 1958226.  

I have a Windows 10 HP desktop with an A10-6700 running up to 4.18 GHz out of its 3.70 maximum. It has also a GTX 1050 Ti GPU board running SETI@home and Einstein@home GPU tasks. GPUGRID GPU tasks seem to overheat it (not overclocked) and they crash. The SETI@home screensaver, on CPU tasks, with the BREAKTHROUGH LISTEN stamp seems to cause reboots but no crashes of the tasks, which complete and validate.
There is a funny thing about the A10-6700: although sold as a 4 cores CPU, the Windows 10 Task manager says it has two cores and 4 logical processors, so only two BOINC tasks can run simultaneously.
Tullio


http://www.cpu-world.com/CPUs/Bulldozer/AMD-A10-Series%20A10-6700%20-%20AD6700OKA44HL.html

I went and looked again. As far as I can tell the A10 series has four(4) distinct cpu cores without hyperthreading on the chip. It also only has two (2) floating point processors. This means each cpu core processes Seti cpu tasks at roughly half the speed it would if there was 1 floating point processor per cpu core. For non-intense number crunching this allowed AMD to lower the cost of the chip.

I am wondering if you have tried running the Radeon APU gpu while also running the Gtx 1050? Some MB's will let you others will not.

It sounds like your Gtx 1050 is over heating. Sometimes you can download an appropriate gpu overclocking utility and use it to increase the speed of the video card fan. It might help keep it from overheating.


HTH,
Tom
I will stop procrastinating tomorrow.
\\// Live Long & Prosper (starting tomorrow ;)
ID: 1958293 · Report as offensive     Reply Quote
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 3041
Credit: 131,702,280
RAC: 758,645
United States
Message 1958294 - Posted: 3 Oct 2018, 22:13:32 UTC

I got this article in my feed on my cellphone today.

https://www.tomshardware.com/reviews/amd-vs-intel-under-500-pc-build,5825.html

One of the things it did was match a small intel cpu with a Gtx 1050 video card against the faster model of the Amd APU's I have been reading about and talking about here.

The bottom line was the APU (Ryzen 5 2400G) even if you overclocked the gpu, was not the equal in gaming performance to the Gtx 1050. Don't get me wrong, the 2400G produced playable performance on a lot of the games that were used in the benchmark. Just not as fast a frame per second as the Gtx 1050.

If you can map that to Seti processing, it appears that we can't even come close to discrete video card performance. It would probably still require someone who has either a 2200G or 2300G that have been running at optimum processing speed to see what kind of RAC's each could post.

Tom
I will stop procrastinating tomorrow.
\\// Live Long & Prosper (starting tomorrow ;)
ID: 1958294 · Report as offensive     Reply Quote
Profile Tom M
Volunteer tester

Send message
Joined: 28 Nov 02
Posts: 3041
Credit: 131,702,280
RAC: 758,645
United States
Message 1958295 - Posted: 3 Oct 2018, 22:23:10 UTC - in response to Message 1958242.  

Tom, I'm using the process_priority_special directive in my cc_config.xml to keep the GPU maxed-out...
That is VERY interesting. I may hold off on suspending my A4-5000's Seti participation to see how that works "naked" and then with 3 cores. Hmmmmm…..

Tom


By the way, process_priority_special doesn't seem to work for Seti GPU tasks on my Windows machines as it does in Linux. The Seti tasks always start with a "below normal" priority. The directive does work for Einstein@Home GPU tasks, though. Maybe it's a bug...



Rats. The A4-5000 is a windows machine and I have no plans to run Linux on it.

I do have another APU out on loan. Its a little bit higher AMD. Need to get it started.

Tom


I now have disabled the app_config.xml which allows the gpu to use whatever the default is for the amount of cpu power needed to run the gpu task (it is tiny). I still have 1 core/thread idled on a 4 core/thread cpu. I still have the cc_config.xml paremeter in there.
We will see what happens. I think the last time I ran this A4-5000 with the default cpu driving the gpu number the speed of the gpu processing was near 8 hours. So lets see :)

Tom
I will stop procrastinating tomorrow.
\\// Live Long & Prosper (starting tomorrow ;)
ID: 1958295 · Report as offensive     Reply Quote
1 · 2 · 3 · 4 . . . 7 · Next

Message boards : Number crunching : AMD A-2 thru A-10 and new AMD cpu/gpu integrated chips


 
©2019 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.