I have started AstroPulse v7 but I’m seeing a huge difference from v6.

V6 would take about 1 hour per task, use 99% GPU and CPU/RAM usage was constant.

With v7 the tasks are taking 3 times longer, GPU usage is about 30%, CPU 100% and RAM is up and down like a yo-yo.

Also with v7 the lag is making my PC unusable, I never had this problem with v6.

Is this happening to anyone else?

Thanks

The default tuning parameters have changed considerably for high end cards like your 590's. That seems the most likely thing to be giving the different performance. To test my guess, you could find the empty (0 byte) text file named ap_cmdline_7.01_windows_intelx86__opencl_nvidia_100.txt in your setiweb.ssl.berkeley.edu_beta directory, and put this line in it:

-unroll 2 -ffa_block 1024 -ffa_block_fetch 512

Those were the AP v6 OpenCL defaults. Be sure to work with a plain text editor (Notepad will do if you have nothing better). The file is read whenever an NVIDIA OpenCL task is started, so no other special action is needed to make it effective.

I doubt those settings are ideal for your 590's, but the test will show whether going back to the older defaults gives performance similar to AP v6.

                                                                   Joe

Wow, that was like turning a switch on, back to normal now.

98-99% GPU usage, CPU and RAM are constant and zero lag.

Run time looks a bit longer than v6, about 75 mins.

Also, I have an app_config.xml because of my dual GPU card, what do you recommend for the <cpu_usage> setting?

Thanks for your help

<app_config>
<app>
<name>astropulse_v7</name>
<max_concurrent>2</max_concurrent>
<gpu_versions>
<gpu_usage>.5</gpu_usage>
<cpu_usage>1</cpu_usage>
</gpu_versions>
</app>
</app_config>

I think that <cpu_usage> set to 1 is right for a system where GPU is far more productive than CPU. It's really a matter of having the CPU available to tell the GPU what to do next.

                                                                   Joe

I think that <cpu_usage> set to 1 is right for a system where GPU is far more productive than CPU. It's really a matter of having the CPU available to tell the GPU what to do next.

It is more the OpenCL thing to do. In former days CUDA and CAL+ applications only needed a fraction of the CPU to be told what to do next.

From my observations I have found that the Astropulse Nvidia tasks for versions 6 and 7 run about the same amount of time, next time I run Nvidia astropulse v7 I will take a look at the GPU and memory usage. However, I have observed that when I run the OpenCL Intel GPU version 7 program, my display does get a little laggy (it is noticeable), compared to other OpenCL graphics programs (like Einstein@home and Collatz@home)... (I haven't run astropulse v6 OpenCL for Intel GPU in quite a while, so I don't really remember the lag I get from it, if any).

On my Computers:

http://setiweb.ssl.berkeley.edu/beta/show_host_detail.php?hostid=72196
http://setiweb.ssl.berkeley.edu/beta/show_host_detail.php?hostid=70437

all the tasks of AstroPulse v7 v7.02 (opencl_nvidia_100) ended with errors

On my Computers:

http://setiweb.ssl.berkeley.edu/beta/show_host_detail.php?hostid=72196
http://setiweb.ssl.berkeley.edu/beta/show_host_detail.php?hostid=70437

all the tasks of AstroPulse v7 v7.02 (opencl_nvidia_100) ended with errors

Спасибо.
Этот баг я пофиксил уже, попробуй установить отсюда:
https://www.dropbox.com/s/wkpg7ndei2cifsz/AP7_win_x86_SSE2_OpenCL_NV_r2605.7z
А вот ветка для обсуждения конкретно этого бага: http://setiweb.ssl.berkeley.edu/beta/forum_thread.php?id=2195

---

News about SETI opt app releases: https://twitter.com/Raistmer

А пофиксенная версия не появится для автоматической загрузки?

А пофиксенная версия не появится для автоматической загрузки?

Появится со временем. Что-то Эрик малость подтормаживает с релизом. Если нет желания возиться с анонимной платформой - можно и подождать.

---

News about SETI opt app releases: https://twitter.com/Raistmer

Загрузилась 7.03 - считает без ошибок!

СПС!

Загрузилась 7.03 - считает без ошибок!

СПС!

Ну отлично! Спасибо за подтверждение! А то никак не могу добиться регулярных отчетов :)

---

News about SETI opt app releases: https://twitter.com/Raistmer

I've got Seti Beta running GPU units on 6 computers, but one of them, an Intel, hasn't gotten an AP WU since 7.01 - and it HAS been requesting them!

---

I've got Seti Beta running GPU units on 6 computers, but one of them, an Intel, hasn't gotten an AP WU since 7.01 - and it HAS been requesting them!

For OpenCL work to be sent at Seti and Seti Beta you need OpenCL 1.2 drivers because of precision problem with OpenCL 1.1 drivers,

You're running OpenCL 1.1 drivers on your Intel GPU host, upgrade them to OpenCL 1.2 drivers and you'll get work.

Computer 71336

Claggy